Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture

Transcription

1 Agriculture and Rural Development Discussion Paper 46 The World Bank Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Alexander Lotsch William Dick Ornsaran Pomme Manuamorn

2 Agriculture and Rural Development Discussion Paper 46 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Alexander Lotsch, William Dick, Ornsaran Pomme Manuamorn Agricultural Risk Management Team Agriculture and Rural Development Department The World Bank Group January 2010

3 2010 The International Bank for Reconstruction and Development/The World Bank 1818 H Street, NW Washington, DC Telephone Internet ard@worldbank.org All rights reserved. This volume is a product of the staff of the International Bank for Reconstruction and Development/The World Bank. The findings, interpretations, and conclusions expressed in this paper do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development/The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA, telephone , fax , All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA, fax , pubrights@worldbank.org. Cover photo: Edwin Huffman Rice fields, Philippines

4 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Acknowledgements This report was produced with input from background papers produced by Robert Brakenridge (Dartmouth Flood Observatory) and Francesco Holecz (Sarmap), and pilot project feasibility reports by ASDECON (Thailand) and Pasco (Japan). The work was funded by the Swiss State Secretariat of Economic Affairs, the Netherlands Ministry of Foreign Affairs, the Bank- Netherlands Partnership Program, and the Japanese Consultant Trust Fund. The section on the Vietnam feasibility study is based on work performed by GlobalAgRisk, Inc. (under contract with World Perspectives, Inc.) for the Asian Development Bank under the Development of Agricultural Insurance for Vietnam Project. Comments provided by Robert Muir-Wood (Risk Management Solutions), Olivier Mahul (GCMNB), Winston Yu (SASDA), and Steven Jaffee (ARD) helped improve this report and are gratefully acknowledged. iii

5

6 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Acronyms...vii Executive Summary...ix 1. Floods as a Source of Risk in Agriculture Scope and Objectives Sources of Flood Risk River flooding and inundation Flash floods Storm surge and coastal flooding Impact of Floods in Agriculture Current Extent of Flood Insurance Property flood insurance Agricultural flood insurance Flood risk and development Challenges for Flood Risk Insurance in Agriculture Definitional Challenges Direct losses Indirect losses Technical Challenges Modelling flood risk Flood zoning Product design and pricing Operational Challenges Underwriting Adverse selection Loss adjustment Financial Challenges Valuation for insurance purposes Covariate risk, catastrophe exposure, and reinsurance Flood Risk Assessment and Mapping Flood Risk Modelling Wide-area Flood Risk Mapping Risk Assessment and the Assumption of Stationarity Remote Sensing-based Flood Risk Mapping Contemporary satellite remote sensing systems New strategies for flood risk mapping and flood index development Potential Applications of Index Insurance to Agricultural Flood Risks Principles of Parametric Insurance Advantages and disadvantages Elements of Parametric Flood Insurance Design Defining flood-induced crop loss Modelling flood hazard...47 v

7 Agriculture and Rural Development Designing the flood index Insurance operational system Scale Options of Flood Index Insurance Institutional Considerations Organizational arrangements Local institutional capacity Underwriting Considerations Flood index insurance product Zoning and client enrollment Loss assessment Financial Considerations Loss modelling requirements Developing premium pricing Risk transfer: Insurance and reinsurance Financial viability of flood index insurance Feasibility Studies Vietnam: Risk Identification and Conceptualization of a Flood Insurance Product Background Findings Index solutions Lessons learned Thailand: Challenges in Designing Flood Index Insurance at the Micro Level Background Findings Lessons learned Bangladesh: Assessing the Feasibility of Flood Insurance in a Complex Environment Background Findings Lessons learned Summary of Feasibility of Flood Index Insurance Schemes at the Micro and Macro Levels Conclusions and the Way Forward...77 Annex: Technical Background...81 A.1. Floods: A Global Perspective...81 A.2. Flood Modelling Approaches...81 A.3. Space-Borne Satellite Sensors Used for Flood Mapping...84 A.4. Global Flood Monitoring: Dartmouth River Watch...86 A.5. Three Examples of Remote-Sensing-Based Flood Mapping...93 A.5.1. Vietnam...94 A.5.2. Thailand...97 A.5.3. Bangladesh References Endnotes vi

8 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Acronyms ADPC ADB ASTER AVHRR BAAC BARC BWDB DEM ERTS ESA FEMA FIRM GIS GDP GPS INPE JAXA LIDAR MFI MODIS MPCI NAIC NAIS NFF NASA NFIP NOAA OECD PML PMF PMP SPOT SAR SIWRP USGS VAR VBARD Asian Disaster Preparedness Center Asian Development Bank Advanced Spaceborne Thermal Emission and Reflection Radiometer Advanced Very High Resolution Radiometer Bank for Agriculture and Agricultural Cooperatives Bangladesh Agricultural Research Council Bangladesh Water Development Board Digital Elevation Model Earth Resources Technology Satellite European Space Agency Federal Emergency Management Agency Flood Insurance Rate Maps Geographic Information System Gross Domestic Product Geographic Positioning Systems Instituto Nacional de Pesquisas Espaciais Japan Aerospace Exploration Agency Laser Light Detection and Ranging Microfinance Institution Moderate Imaging Spectroradiometer Multi-Peril Crop Insurance National Agricultural Insurance Company National Agricultural Insurance Scheme National Flood Frequency National Aeronautics and Space Administration National Flood Insurance Program, US National Oceanic and Atmospheric Administration Organisation for Economic Co-operation and Development Probable Maximum Loss Probable Maximum Flood Probable Maximum Precipitation Satellite Pour l Observation de la Terre Synthetic Aperture Radar Southern Institute for Water Resources Planning United States Geological Survey Value at Risk Vietnam Bank for Agriculture and Rural Development vii

9

10 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Executive Summary Floods are a major source of risk for the agricultural sector. Flood risk in the agricultural sector primarily arises from river flooding, flash floods, and coastal flooding. The impacts of floods can result in sizable agricultural damages at the local level. Floods in agricultural zones expose agricultural producers, agricultural supply chains, rural financial institutions (such as agricultural banks), and governments to financial risks due to the loss of crops, delinquency on seasonal production loans, damage to infrastructure and loss of public revenues. The costs associated with these damages are often absorbed by households directly or governments that provide compensation to agricultural producers in the aftermath of catastrophic flood events. Rural financial institutions also absorb the cost of floods through loan rescheduling or, in catastrophic cases, loan cancellation. In many developing countries, floods are dealt with in a reactive, rather than proactive, manner and little is done to be financially prepared for a catastrophic outcome of floods. The penetration of agricultural insurance in developing countries is relatively low, and flood risk is generally not insured. Recent initiatives by the World Bank have promoted financial strategies for agricultural weather risk management including protection against severe droughts (e.g., India, Malawi) and livestock mortality risk (e.g., Mongolia). Many of these projects have used simple index-based insurance products that indemnify farmers based on loss proxies (such as observation of rainfall or livestock mortality rates). However, little advance has been made to expand agricultural insurance to flood risk and the development of flood insurance products that are technically, operationally, and financially viable. While in some countries, flood insurance is included in a broader crop insurance program that includes coverage for tropical cyclones and the floods associated with them (e.g., in the Philippines), there is generally little availability of flood insurance in the agricultural sector. Providing agricultural flood insurance is inherently challenging for a number of reasons. First, delineating flood risk is difficult as floods cause agricultural damages both directly (e.g., crop and livestock losses) and indirectly (e.g., interruption of business due to damaged infrastructure). Both the characteristics of floods, such as level and duration of inundation, as well as their associated impacts need to be well delineated and quantifiable to make flood insurance possible. Second, the quantification of flood risk requires data and models that produce estimates of the likelihood and severity of flooding in agricultural production zones. Such modelling requires relatively detailed information about terrain and hydrological characteristics in the region of interest to group agricultural farmers into zones of similar risk. Flood risk mapping is further complicated ix

11 Agriculture and Rural Development as potential agricultural damages vary considerably according to the timing of floods throughout the crop production cycle. The design of flood insurance program depends on the accuracy and precision with which flood risk can be quantified. Third, flood insurance is difficult to operate. Flood damages are often localized and can be mitigated to some extent through structural intervention by agricultural producers. Farmers in flood-prone areas are often aware of the flood potential on their land, whereas farmers outside perceived flood zones may not be aware of the potential for catastrophic flooding. As a result, a voluntary insurance scheme may only attract farmers in high-risk areas and there may be little demand in low-risk zones (known as adverse selection). Also, to underwrite flood risk, insurers need to be able to evaluate the risk and exposure of potential clients. To do this, flood risk zones need to be identified to group potential clients according to their likelihood and severity of experiencing floods. Lastly, loss assessment (the verification of a loss) and loss adjustment (the financial payment made) are the cornerstones of an agricultural insurance program but can be difficult to achieve in a systematic and cost-effective manner with on-the-ground verification. Fourth, agricultural flood insurance is difficult to manage financially. Unlike in property insurance, in which indemnities are based on the estimated repair or replacement costs, the valuation of crop loss needs to be determined either on the basis of the input costs at the date of the loss or the loss of expected revenue. Thus, the key challenge in modelling and compensating for flood arises from the timing of flood and the valuation of its impacts at the date of occurrence. A further financial challenge arises from catastrophic flooding that affects many clients simultaneously, resulting in claims that outweigh insurance premiums and capital reserves of insurance. This requires the support of reinsurers to protect insurance companies against potentially large financial exposure. Many of these challenges can be overcome by harnessing flood modelling, flood remote sensing, and an index-based insurance approach, thereby creating a potential to increase the feasibility and availability of flood insurance in the agricultural sector. Provided sufficient data is available, flood modelling can generate information on the location, likelihood, and severity of flood risk and thereby identify flood risk zones that are critical for insurance underwriting, pricing, and loss adjustment. Direct observations of flooded areas derived from satellite remote sensing have become readily available and support quantification of flood risk as well as the verification of flood losses following an event. Lastly, index-based insurance can reduce transaction and operating costs of a flood insurance program and reduce the financial challenges associated with loss measurement and valuation and reinsurance. The technical complexity and the novelty of the approaches described in this report require the support and technical assistance from donors to promote and pilot agricultural insurance in developing countries. The following considerations are to guide any future development of agricultural flood insurance. x

12 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Institutionally, several stakeholders need to be mobilized, and a pilot arrangement typically requires a phased approach to organizational capacity building. Organizational structures can be guided by the experience in developing index-insurance programs for drought, though the complicated nature of flood risk requires more involvement of technical partners and supporting services. Aside from a technical support unit formed by an insurer (or group of insurers), key technical inputs are required from national flood management agencies, remote sensing centers (national or international), and technical experts (e.g., from academic institutions). Challenges in developing a flood insurance program imply that involvement from both the public and private sector is needed. Insurance is best operated in the private sector, where sound insurance principles and needed technical capacity exists. At the same time, major institutional support, data, and expertise exist in government organizations. Operationally, insurers and governments need to develop a clear strategy to manage adverse selection. The main considerations for the design of an agricultural flood insurance program include the feasibility of creating homogenous risk zones, whether the insurance is operated on a voluntary or compulsory basis (the latter would reduce adverse selection, but is politically more difficult to implement), and develop clear underwriting rules that define the types of risk and zones eligible for insurance and the periods and limits covered by the program. Equally important is the development of a clear strategy for loss assessment and loss adjustment that is practical, transparent, and fair. If the insurance policyholder is a risk aggregator (such as a bank or a government entity), a clear strategy is needed for the utilization of insurance payments. Financially, the key considerations relate to loss modelling requirements, premium pricing, and reinsurance. The financial concerns start with an understanding of the expected average losses, the volatility of losses and confidence in loss estimates. Because there is generally little or no experience in flood insurance in developing countries, loss modelling cannot use historical market data and has to rely on primary data (river discharge or rainfall). Loss modelling at the micro (or farm) level is often complex and constrained by detailed data. Similarly, the pricing of insurance is simplified at more aggregate levels and can rely on aggregate indicators such as river level measurements rather than detailed loss modelling at the micro level. Lastly, depending on the expected loss frequency and severity, reinsurance is likely to be required. The research presented here suggests that reinsurance may be more accessible if catastrophic flood risk is aggregated and indexed at a macro level. Translating the theory of agricultural flood insurance into practice in developing countries is difficult, as evidenced by the initial work documented in this report. The demand for technical assistance on flood insurance that has motivated this work led to feasibility studies in Vietnam, Thailand, and Bangladesh that were carried out during the period. The experience from these studies show that the development of agricultural flood insurance at the farm level remains challenging largely due to the lack of data to support xi

13 Agriculture and Rural Development the modelling of flood risk and assessment of potential losses. Many of the technical difficulties that characterize the micro level can be reduced by developing insurance products at the meso level or macro level for rural financial institutions or government entities. While still technically complex, the risk modelling performed at that level can be simplified and performed faster. However, additional procedures would have to be put in place to target insurance compensation paid through these risk aggregating entities to affected agricultural producers after a destructive flood. Going forward, government and donors can play an important role to facilitate the development of risk spreading mechanisms in general and agricultural flood insurance in particular. First, this includes investment in the generation of public goods to support disaster risk reduction and recovery, risk management, and ultimately insurance applications. Second, awareness building and risk education are essential for better risk management and insurance. In that vein, identifying and assessing flood risk are critical first steps. Third, many of the technologies described here have applications beyond insurance, including for better planning, risk reduction, early warning, and disaster response. Insurance can complement such activities, but is only viable if carried out jointly as part of a broader risk management framework. Fourth, more research and technical assistance is needed to develop simple and financially viable products for flood risk transfer at aggregate levels; there is increasing demand expressed for such products from flood-prone countries. Finally, donors and government can support international and regional centers involved in flood modelling and facilitate a platform that convenes the technical expertise required for flood risk insurance development. Several of such centers and core expertises were identified through this work. xii

14 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture 1. Floods as a Source of Risk in Agriculture Flood plains have traditionally supported high population densities due to the advantages for agricultural practices, such as deep fertile alluvial soils and water availability. For instance, the Netherlands and Bangladesh are among the most densely populated countries, and the Netherlands has the highest gross domestic product (GDP) per square kilometer in Europe. At the same time, floods affect more people than any other weather-related peril worldwide (Table 1). Six out of the 10 worst natural disasters in 2007 were floods. 1 While economic damages and loss of life are pronounced in urban and coastal areas due to the concentration of infrastructure and people, floods in rural areas are both closely linked to agricultural production and a major underlying source of systemic risk for farmers. Table 1 Disasters Number of events and economic losses. Floods are a large share of total losses due to natural hazards. Note that a portion of windstorm loss is due to flooding. ( Other combines earthquakes, volcanoes, landslides, and mudslides.) DISASTER PERIOD LOSSES DEATHS AFFECTED ( ) EVENTS (billion $) (1000) (millions) Drought Flood Windstorm Other Total Source: CRED (2009). Flood risk in agricultural areas emanates from a variety of natural and manmade causes, including upstream dams, artificial levees, status of soil drainage improvements, conveyance status of the local channels, trends in contributing-watershed land use that affect runoff, the topography of the watershed, and regional trends in climate that alter runoff frequencies and magnitudes. Climatic trends may affect flood frequency and severity, whereas drainage improvements to an agricultural field may reduce flood duration without any changes to flood peak discharges. Also, agriculture makes preferential use of level to gently sloping lands. Figure 1 and Figure 2 illustrate the distribution of flood-prone land and agricultural land use, and global flood mortality and economic loss risk, respectively. For rural populations, apart from damage to primary assets of crops and livestock, direct flood damage can affect immovable assets (for example, dwellings, storage and processing buildings, equipment, vehicles, land, 1

15 Assessment.qxd:Assessment 3/4/10 7:31 AM Page 2 Agriculture and Rural Development Figure Global distribution of cropland areas (top) and historical flood events since 1985 (bottom) No data 1996 No data No accurate data No data No accurate data 1989 No data Source: DFO (2009). irrigation facilities, and bridges) and stock (for example, fertilizers, crops in processing or in storage, and many other assets). Further, flood impacts can be direct or indirect. Flood can affect primary producers, or may affect supply chain service or credit providers, and ultimately impact government finances. While there are ways to manage seasonal flooding through water and crop cycle management, as well as to mitigate (reduce) risk through planning and engineering, the remaining risk, such as the financial losses incurred by farmers due to extreme flooding, is frequently dealt with in a responsive fashion through ex post interventions. These may include compensating farmers for their losses through government programs or rehabilitating infrastructure after a flood. However, agricultural producers, rural financial institutions, and governments in developing countries have little protection against the financial risks arising from extreme flood. 2

16 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 2 Global Distribution of Relative Flood Risk in Terms of Mortality and Economic Loss. (red high risk, yellow medium risk, blue low risk). Source: World Bank (2005a) Scope and Objectives This paper is concerned with flood and its direct impact on growing crops and livestock assets. It explores how traditional flood risk mitigation and management could be complemented through innovative ways to provide financial protection against extreme event using index-based insurance. Such agricultural insurance products have been tested as a mechanism to provide effective assistance to developing nations in response to losses arising from droughts (e.g., India, Malawi, and Ethiopia). With the work presented here, the World Bank s Commodity Risk Management Group has assessed practical and efficient methods to conceptualize and potentially implement such indexbased insurance for agricultural flood losses. 3

17 Agriculture and Rural Development To do this, this paper explores technological, institutional, and market-related aspects of how flood insurance products can be developed for the agricultural sector in developing countries by harnessing modern technology such as satellite remote sensing, flood modelling, and computer-based risk modelling. First, the paper gives a typology of flood events relevant for agriculture. While short-lived and localized floods can represent an important source of risk, the technologies assessed here are most suitable for widespread river inundation that affects agriculture in flood plain areas in low-lying deltaic regions. Also, the paper focuses on flood risk at the local scale, i.e., the seasonal production risks experienced by farmers and to the financial institutions lending to them. The management of a portfolio of flood risks at the national or regional level is equally important from the perspective of risk reduction and disaster response and provides the framework for effective agricultural flood insurance. Geographically, the focus is on regions in South and Southeast Asia where a relatively large proportion of agriculture is in flood-prone areas. The findings presented here draw on studies performed in Thailand (Pasak River, Petchaboon Valley), Vietnam (Mekong Delta), and Bangladesh, together with local research centers, agricultural banks, and insurance companies, as well as international experts and centers of excellence in remote sensing and flood modelling. These case studies provide insights into the challenges and opportunities to complement flood risk mitigation strategies with flood insurance. The remainder of this section describes the sources of flood risk in the agricultural domain and the current extent of flood insurance, which is currently largely limited to developed countries and property damages. At the same time, floods cause annual losses that are of concern for agricultural producers, financial institutions, and governments in the developing world. The increasing need for better flood risk management that has been expressed by a wide range of institutions in flood prone countries to the World Bank s Commodity Risk Management Group has motivated this work. However, as the paper shows, there are many challenges to use insurance instruments effectively for flood risk, some of which are general to agricultural insurance and others are specific to the risks associated with flooding. These challenges are described in section 2. Technologies that can help overcome these challenges are presented in section 3, and innovative ways to insure flood risk are discussed in section 4. Lastly, section 5 documents some recent case studies where these approaches have been assessed. Technical background is provided in the annex Sources of Flood Risk Flooding occurs in different geographic settings and can be caused by a variety of mechanisms. Floods also occur at various magnitudes (Box 1). The primary cause for floods at the global level stems from heavy rain, followed by brief torrential rain, tropical cyclones, and monsoonal rains (Figure 3). A more detailed global picture is presented in section A.1, including the geographic distribution by flood-causing processes and their seasonal patterning. The most important types of flood risk that affect agriculture include river flood, flash flood, and coastal flood, which are described here. 2 4

18 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Box 1 Levels of magnitude of flood Normal flood (e.g., 1 year flood): Regular inundation of low-lying farmland is common in many tropical countries, for example in Southeast Asia. They occur almost every year and farming practices, especially rice cultivation, are welladapted. Forecasts can be issued to give advice regarding cropping and sowing times to minimize losses. Medium flood (e.g., 1 in 5 year flood): This less frequent flood causes some economic loss, but is not extensive or serious. It affects farmers and people living in low-lying areas and by rivers. Loss of life is unlikely as people are usually prepared for these regular events. Severe flood (e.g., 1 in 20 year flood): River levels continue to rise and affect large geographic areas and people less familiar with this scale of flooding including those living in urban areas. Damages and losses to the physical environment and economic sector are generally significant. Catastrophic flood (e.g., 1 in 100 year flood): Extreme flooding inundates extensive areas. It is extremely devastating with multi-fold impacts to life and property and the economy. Source: ADPC (2005). Figure 3 Global number of floods of various causations, Heavy rain Brief torrential rain Tropical cyclone Monsoonal rain Snowmelt Dam/levy, break/release Ice jam/break-up Extra-tropical cyclone Tidal surge Avalanche related Source: DFO (2009) River flooding and inundation River flooding occurs when the capacity of a river system is insufficient to contain the flow of water in the river, resulting in escape of water from the normal perimeter and submergence of surrounding low-lying land. Prolonged rainfall results in soil saturation, and may occur at times of increased inflow from tributaries. Characteristics of the river flooding are determined by the capacity of river channel(s), slopes, soil permeability, land cover, land use, and control of water flows by any man-made engineering structures (training walls, dams, drainage, etc.). River floods are often slow moving. Increased tributary 5

19 Agriculture and Rural Development flow can affect flood plains, as a result of land degradation in the catchment areas. Flood plains are formed from deposits made by earlier floods. In terms of flood mitigation, river systems range from heavily managed to unmanaged. In practice, although it is possible to take measures to manage floods arising from rivers, it is not possible to control such floods completely. Most river systems have been engineered, for purposes of urban flood protection, agricultural protection, and irrigation management. Flood detention areas may be designated, which generally allow controlled flooding of agricultural areas in order to protect urban regions. Further, in many river systems intentional flooding allows the recharging of soils with nutrient load and moisture. Such irrigation and intentional flooding has evolved around the natural seasonality of the river, and adjustment of the cropping calendar so as to allow windows of cropping and flooding. The deep fertile alluvial flood plains of rivers are favored areas for farming. Many alluvial soils may have been deposited over a long period and no longer are exposed to flood, but by definition, the majority remain potentially exposed. The most favorable and populated areas of a country may therefore lie in flood plains. Further, very substantial areas of river flood plain can exist in a single country, such as in Bangladesh, or the Mekong Delta of Vietnam; typically, many countries have flood plains exposed to river inundation, but often only on more restricted areas, due to more marked topography such as hills, valleys, or slopes (e.g., Mexico, Jamaica, Thailand, Romania, Turkey). These flood plains may be the most intensive agricultural production areas. Although essential crops may also occur along floodplain lands within narrow upland valleys, the aggregate of such land, although locally important, is small compared to the extensive, humid- to semi-arid plains where most global agriculture is practiced. River flooding at any location may be caused by rainfall, or snowmelt, at long distances from the affected location. Distant rainfall from snowmelt or monsoons may be the main drivers of flood on major river systems, rather than localized rainfall. The actual extent of flood will be a combination of all contributing water, whether distant or localized, and is strongly affected by prior waterlogging of soils. By nature of the shallow slopes of a natural floodplain, river inundation flood duration can last for days, or weeks. Recession of the floodwaters is a function of floodplain drainage (natural or artificial), slope, permeability, constrictions to water flow, and so on Flash floods Flash floods arise from intense, localized rainfall, and can happen practically anywhere. Intense rainfall can be measured over any specific period, typically between one hour and a maximum of six hours in the case of flash flooding. Duration of rainfall is longest in slow-moving or stationary storms. A characteristic of flash floods is that flood water rises suddenly, may be fast flowing, may collect in lower lying areas, and normally runs off and ponds rapidly. Residual ponded areas of water (sometimes larger lakes) may be trapped, remaining for long after the flash flood event. The flood impact of 6

20 intensive rainfall is more severe when the ground is already saturated, where soils are impermeable or unstable, and in heavily sloped areas. Sequential intense rainfall events can therefore have a cumulative impact. Where ground is sloping, water is channelled to gullies and temporary watercourses, leading to erosion or landslides, and washing out bridges, culverts, or roads. Flash floods may also impact areas downstream of an intense rainfall event. Within a valley, flash floods can affect foothills, and rivers flood the valley bottoms (e.g., Pasak River, Thailand, section A.5.2). Within a country, regions may be affected by flash flooding (e.g., the northeast region of Bangladesh) whereas river flooding is the main national flood exposure. 3 There is a growing body of evidence that climate change is leading to increased intensity of rainstorms, and therefore to flash flooding (Box 2) Storm surge and coastal flooding Coastal zones are subject to flooding as a result of storm surge increased sea levels driven by tropical storm systems (cyclones) or by strong windstorms arising from intense offshore low-pressure systems. Coastal areas most at risk are low lying, either river deltas or coastal plains. The extent of flooding caused by a coastal sea surge will depend on several factors, especially the topography of the low-lying inland areas, tidal conditions, wind and wave action, extent of inland river flow at the time of coastal surge, and occurrence of localized rainfall associated with the storm event. Torrential rains associated with monsoons and tropical cyclones are also important factors adding to the impact of storm surges. In Asia, severe floods recur during the monsoon and rainy seasons, often with disastrous consequences. Crop losses are generally severe in affected areas but the overall impact at the national level varies among countries. The major cause of the most destructive phenomena is a storm surge a rapid rise of sea level resulting from strong winds driving the water ashore and causing flooding in low-lying coastal areas. Storm surges often accompany intense typhoon systems. In India, storm surges account for more than 90 percent of loss of life and property. Low-lying coastal areas elsewhere, as in Central America, Venezuela, Mozambique, and Madagascar, have also been devastated by storm and flood-related disasters in recent years Impact of Floods in Agriculture Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture There is no global assessment of agricultural flood losses. The numbers presented in Table 1 and Figure 2 reflect only the overall losses due to floods and the frequency of floods relative to other disaster types, but do not provide a separation of losses by sector. However, statistical data at the country level suggests that agricultural flood losses can be substantial locally and regionally: In the Philippines, palay (rice) losses (Figure 5) totaled 3 million tonnes annually between 1994 and 2005, equivalent to 2.1 percent of actual production as a consequence of typhoons and floods (for reference, droughts resulted in further losses, totaling 1 million tonnes or 0.7 percent of actual production, over the same period). At the sub-national level, however, relative 7

21 Agriculture and Rural Development Box 2 Historical trends in streamflow and climate change Streamflow is the temporally lagged and aggregate effect of precipitation over a river catchment. Streamflow and precipitation patterns have changed in many river basins of the world during the twentieth century, which is largely attributed to humaninduced climate change. Precipitation is generally projected to increase in high latitudes and some tropical areas (e.g., Southeast Asia), and to decrease in some midlatitude regions (the Mediterranean). These changes, together with a general intensification of rainfall events, are likely to increase the potential for flooding and the frequency of flash floods and large-area floods in many regions. 5 This will be exacerbated or at least seasonally modified in some locations by earlier melting of snowpacks and melting of glaciers (e.g., Andes, Himalayas). Regions of constant or reduced precipitation are very likely to experience more frequent and intense droughts, notably in Mediterranean-type climates and in mid-latitude continental interiors. Source: Milly, Weatherald, Dunne, and Delworth, 2002; Milly et al., Figure 4 Relative change in runoff (in percent) during the twentieth (a) and twenty-first (b) centuries. Twentieth-century changes are for the period, and twenty-first century changes are for the period. Changes are relative to the average. Blue areas indicate relative increase in runoff and flood risk. (a) (b) losses have been much higher in some regions of the country than others. Most notably, two of the three major producing regions Regions II and III were also in the top three regions ranked according to losses as a proportion of actual production (approximately 5 percent and 3 percent, respectively). In the main rice-producing area (Cagayan Province, Region II), losses due to 8

22 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture 30 Figure 5 Palay losses due to typhoons, floods and droughts in the Philippines ( ). The top panel shows losses in the main rice-producing area (Cagayan Valley, Region II). The bottom panel shows the relative production losses in all Philippines regions and country-wide. Annual drought, flood and typhoon palay losses in Cagayan Province/Philippines as a percentage of actual production % of actual production Average annual drought, flood and typhoon palay losses by region as a percentage of actual production ( ) in the Philippines 7 % of actual production CAB region I region II region III region IV-A region IV-B region V region VI region VII region VIII region IX region X region XI region XII Congo ABRM Philippines Typhoon/flood Drought floods have exceeded more than 15 percent of annual production in some years (e.g., 1998). Flood is included within the multiple-peril crop insurance coverage provided by the Philippines Crop Insurance. Over the period from 1981 to 2006, 55 percent of all claims for palay were paid for typhoon and flood, and 13 percent for drought. Typhoon damages typically include the combined effects of winds, heavy rains, and flood. In China, between 1982 and 2002, 28 percent of agricultural losses nationally were attributed to flood, compared to 52 percent by drought and 4 percent by 9

23 Agriculture and Rural Development Table 2 Comparison of losses resulting from the 1988, 1998, and 2004 floods in Bangladesh. Loss No. livestock killed 172,000 26,564 8,318 Crops damaged Deaths 2,300 1, Rice production losses (million metric tonnes) No. of people affected 45 million 31 million 36 million Roads damaged (km) 13,000 15,927 27,970 Percent of land inundated No. of homes damaged/ 7.2 million 980,000 4 million destroyed Total losses Tk 82.6 billion Tk 118 billion Tk 134 billion (US$1.4 billion) (US$2 billion) (US$2.3 billion) Source: 2004 Floods in Bangladesh: Damage and Needs Assessment and Proposed Recovery Program. World Bank/ADB joint report, typhoon. Flood and waterlogging in China occurs mainly as a result of both typhoon-induced and seasonal rainfall within the low-lying river basins of the eastern and central provinces. 6 In Bangladesh, 30 percent of the country experiences annual flooding, and extreme floods can extend to 60 percent of the national territory. Major floods occurred in 1988, 1998, 2004, and 2007, and the main impact is on agriculture, on which the majority of the population are dependent. Examples of the damage from three floods are shown in Table 2. In Thailand, the most frequent hazard that affects the country is flood. Compared to other disasters, flood is also associated with the most severe impact both in terms of economic loss and mortality. Floods mainly occur in the monsoon seasons between June and September. Many river basins, such as the Chao Phraya basin in Central Thailand, are sites of intensive agricultural activities while being very prone to flooding due to river swelling and overflow during the rainy seasons. Data from show that flood is a major risk for agriculture in Thailand, annually damaging not only large acreage of crop land but also livestock, poultry, and fishery sectors (Table 3). In Vietnam, all regions of the country are highly exposed to flooding caused by river inundation and storm surge; the exception is in the Northeast and Northwest parts where flash floods are more common. The Mekong River Delta is mostly severely affected by both inundation flood and storm surge, followed by the Red River Delta, the North-Central Cost Zone, and the Coastal Economic Zone. Data from recent flood events demonstrate the magnitude of impact on the agricultural sector in Vietnam. In 2003, highly concentrated torrential rains caused severe inundation in the Northern Delta in the 10

24 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Table 3 Thailand s Agricultural Flood Losses from 2002 to Data from Department of Disaster Prevention and Mitigation, Ministry of Interior, Thailand Agricultural land 1,669, , , , ,889 (hectares) Livestock (number 2,955, ,343 71, , ,211 of animals) Poultry (number n/a n/a n/a 473, ,850 of animals) Fish and shrimp 103,533 22,339 12,884 13, ,260 (number of ponds) provinces of Thai Binh, Ninh Binh, and Nam Dinh. Over 120,000 hectares of rice cultivation of the three provinces were affected, of which 60,000 ha was in Thai Binh alone; half of this area was totally destroyed. The damage caused to aquatic production was also large. In the Mekong River Delta, consecutive severe flooding occurred in 2000, 2001, and 2002, resulting in 1,044 people killed (one-tenth of the total number of deaths in 15 years nationwide); 1.6 million houses submerged; and nearly 500,000 hectares of rice inundated Current Extent of Flood Insurance Globally, the insurance industry is most developed for urban and industrial risks within developed countries. Increasing frequency and financial impact of major flood events have strongly focused the global insurance market on improving the understanding of flood risk for traditional lines of property business. Solutions for rural flood insurance can benefit from these insurance experiences and research efforts. Agricultural flood differs in the assets at risk, which are normally subject to less flood protection than in urban areas and are seasonal in exposure Property flood insurance Flood insurance availability for property risks affecting industrial or household property and contents is highly diverse in different countries. Within Europe, this diversity is most marked and ranges from countries where flood insurance is not available (the Netherlands), to those where it is generally included in all fire insurance policies (U.K.). 8 In Spain, premium for flood is automatically added to private sector fire insurance policies, but the flood insurance is managed by a government insurance fund, Consorcio Nacional de Seguros. Governments may also be involved in providing insurance or disaster relief. In France, flood risk as well as other catastrophic risks are covered under a compulsory catastrophe guarantee provided by the government (triggered by a declared national catastrophe event) and linked to voluntary property insurance. An example of this diversity between countries in Europe is shown in insurance coverage of the 2002 Central European floods 11

25 Agriculture and Rural Development (affecting the Elbe, Danube, and their tributaries). In Germany, 20 percent of economic losses of Euro 15 billion were insured; 20 percent of households in the Czech Republic had flood cover; and in Austria cover was limited to Euro 5,000 to 10,000 per household. 9 The diversity of flood insurance solutions reflects the private sector s concerns over difficulties of insuring flood, the rising incidence of natural disasters including flood, as well as state intervention in some countries. It should be noted that flood is also an important risk in motor insurance and personal insurance branches. The case of the United Kingdom is of interest in that increased incidence of major floods (such as in 2000 and 2007) have led the insurance sector to increasingly question the sustainability of providing flood insurance to all property owners. The insurance sector has proposed stricter conditions, including that government expenditure for flood defences is increased, that building is restricted on flood-prone land, and that other measures are implemented to cope with a significant increase in major flood events. 10,11 In the U.S., property insurance for flood (National Flood Insurance Program) is made available through the Federal Emergency Management Agencies (FEMA). The risk is not carried by the private insurance sector, although is administered by approved insurers. In developing countries, property flood insurance is more limited in penetration, which reflects lower overall insurance penetration, as well as availability of flood insurance within each market. In 2007, Organisation for Economic Co-operation and Development (OECD) countries accounted for 89 percent of global non-life insurance premiums; per capita non-life premiums were US$1,435 in industrialized countries, and US$34 in emerging markets. 12 Within emerging markets, premiums per capita in Asia are 3.4 times higher than in Africa. Flood insurance in developing countries is primarily taken out by industrial and commercial entities in urban areas, with generally more limited insurance by householders, and the insurance sector may not be active in rural areas. In the Philippines, flood insurance is available for business assets, but policies are often only taken out if it is a requirement of a loan. Countries such as Indonesia, Thailand, and China are typical, with flood insurance included for industrial risks (especially those with foreign ownership), but there is low penetration for domestic insurance. A study by OECD of emerging countries showed that, by 2070, Asian cities would dominate global flood vulnerability in terms of assets exposed to flood, particularly to coastal flood. 13 Flood insurance for property risks often offers more restricted compensation than for other fire and allied perils in the policy. Deductibles (the first loss born by the insured) may be higher, and there may be financial limits on the maximum claim payable Agricultural flood insurance There are different types of crop and livestock insurance schemes (Box 3). Flood is normally included as an insured peril in Multiple Peril Crop Insurance (MPCI) together with all other perils, which may cause yield 12

26 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Box 3 Different types of crop and livestock insurance schemes Traditional Crop Insurance Damage-based Indemnity Insurance (Named Peril Crop Insurance). Damagebased indemnity insurance is crop insurance in which the insurance claim is calculated by measuring the percentage damage in the field, soon after the damage occurs. The percentage damage measured in the field, less a deductible expressed as a percentage, is applied to the pre-agreed sum insured. The sum insured may be based on production costs or on the expected revenue. Where damage cannot be measured accurately immediately after the loss, the assessment may be deferred until later in the crop season. Damage based indemnity insurance is best known for hail, but is also used for other named peril insurance products (e.g., frost and excessive rainfall). Yield-based Crop Insurance (Multiple Peril Crop Insurance, MPCI). Yield-based crop insurance is insurance where an insured yield (e.g., tonnes/hectare) is established, as a percentage of the historical average yield of the insured farmer. The insured yield is typically between 50 percent and 70 percent of the average yield on the farm. If the realized yield is less than the insured yield, an indemnity is paid equal to the difference between the actual yield and the insured yield, multiplied by a pre-agreed value of sum insured per unit of yield. Yield-based crop insurance typically protects against multiple perils meaning that it covers many different causes of yield loss. This is because it is generally difficult to determine the exact cause of loss. Index-based Crop Insurance Area Yield Index Insurance. Area yield index insurance is when the indemnity is based on the realized average yield of an area such as a county or district. The insured yield is established as a percentage of the average yield for the area. An indemnity is paid if the realized yield for the area is less than the insured yield regardless of the actual yield on a policyholder s farm. This type of index insurance requires historical area yield data. Weather Index Insurance. Weather index insurance is when the indemnity is based on realizations of a specific weather parameter measured over a pre-specified period of time at a particular weather station. The insurance can be structured to protect against index realizations that are either so high or so low that they are expected to cause crop losses. For example, the insurance can be structured to protect against either too much rainfall or too little. An indemnity is paid whenever the realized value of the index exceeds a pre-specified threshold (e.g., when protecting against too much rainfall) or when the index is less than the threshold (e.g., when protecting against too little rainfall). The indemnity is calculated based on a pre-agreed sum insured per unit of the index (e.g., US$/millimeter of rainfall). Traditional Livestock Insurance Mortality insurance for individual animals is the basic traditional product for insuring livestock. It is very costly for an insurer to provide insurance for individual animals, especially where herd size is small. Premiums are set based on normal mortality rates within the permitted age range, plus risk and administrative margins, and are generally quite expensive. Further, as mortality is, to a considerable extent, influenced by management, the product suffers from adverse selection by the highest risk farmers. (continued) 13

27 Agriculture and Rural Development Herd insurance is a variation on individual animal mortality cover for larger herds. A deductible is introduced, where a certain number of animals, or a percentage of the animals, must be lost before an indemnity is paid. Epidemic disease insurance is offered in only a few countries, most notably Germany. Insurance of government ordered slaughter or quarantine is normally excluded. Epidemic disease insurance carries major and infrequent catastrophic claim exposures necessitating a high reliance on reinsurance for risk transfer. Due to the difficulties of modelling epidemic disease spread and financial exposures, it is difficult to develop this type of insurance and to obtain support from international reinsurers. Index Livestock Insurance Index insurance for livestock has been applied for mortality risk in Mongolia. An aggregate livestock mortality index at the district level is used to compensate individual herders. Weather-based and satellite-based index pasture and rangeland insurance products exist in Canada and the U.S. reduction. Flood insurance is not offered as a stand-alone crop or livestock insurance product. In developed markets, it may sometimes be added as an insured peril to a named peril policy (e.g., to a hail and fire policy). In the U.S., the largest MPCI program globally, flood is included universally as a peril within the Federal Crop Insurance Program, operated by the Risk Management Agency (RMA). There is only limited differentiation between terms and conditions for clients with land that is flood prone ( High Risk Land ), or not exposed to flood, although a higher premium is payable. Farmers can opt not to insure High Risk Land, in which case they cannot insure any of land with that crop against flood in a county. Risks are established by FEMA and the Natural Resources Conservation Service and the information is used by RMA for rating. Excessive moisture, rain, or flood accounted for 30 percent of crop indemnities between 1989 and Excess rain can give rise to several types of loss, for example flood, inability to harvest, crop quality reduction, or disease outbreak. In Spain, while the insurance program was started in 1980, flood was only included as an insured peril in 1999, following extensive research (Box 4). The agricultural insurance system is operated by Agroseguro. In order to minimize adverse selection (i.e., only farmers at high-risk choose to purchase insurance), flood is included as an obligatory peril on all crop insurance product lines ( torrential rain and persistent rain were only added as perils in 2002). 15 In developing countries, the penetration of agricultural insurance is generally very low (Figure 6). However, flood insurance is often included in countries where crop insurance is offered and where typhoon, and flood associated with typhoon, is a predominant risk. In the Philippines, the Philippines Crop Insurance Corporation offers MPCI, including flood. In China, flood is a predominant risk in much of south and central parts of the country as a result of high frequency of tropical depressions, tropical storms, and typhoons. 16 Flood risk is included as a peril in the fast-expanding, 14

28 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Box 4 Agricultural flood insurance in Spain Spain has one of the most developed agricultural insurance systems in Europe. It is managed by a specialist agency, Agroseguro, on behalf of private insurers and the government. Risks are reinsured in the international reinsurance market and, at a catastrophic level, by the government. Features of flood insurance by Agroseguro are: Even though flood is not considered a major risk in Spain, flood insurance is compulsory as an addition to any crop or livestock insurance policy. All clients are eligible for cover. Policy terms and sums insured are as for other perils covered, although higher deductibles may apply for flood. A uniform premium rate is applied per crop type, with no geographical differentiation. Flood was introduced in 1999, well after crop and livestock insurance was well established (Agroseguro was formed in 1980). Flood is one of three catastrophe risks (along with strong winds and persistent rains) handled in a separate catastrophe fund. Inadequate drainage (evidenced, for example, by highly localized flood damage) is not covered. A key point for Agroseguro is that the major penetration of agricultural insurance throughout the country, along with compulsory flood insurance, avoids the problem of adverse selection. Figure 6 Geographical distribution of insurance premium of agricultural insurance programs. Estimated global premium in 2008 is Euro 16.5 billion. 17 China Canada USA Spain Italia France Other India Argentina Mexico Germany Brazil Austria Turkey South Africa South Korea 15

29 Agriculture and Rural Development subsidized, MPCI-based crop insurance market in China. In India, flood is an insured peril under the National Agricultural Insurance Scheme (NAIS), which is an all-peril, area-based crop insurance. In South Korea, typhoon, and flood associated with typhoon, is covered by the National Agricultural Insurance Company (NAIC) Flood risk and development Flood insurance is not viable in isolation and requires an institutional framework for flood risk management. Such a framework establishes the structure and relationships of governmental and non-governmental organizations including government agencies and departments, individuals, and the private sector in assessing and managing flood risk. Key elements of a flood risk management framework include: (1) integration of participations and stakeholders from various sectors; (2) risk-based planning that incorporates risk mitigation, preparedness, response, recovery, and monitoring; (3) integrated watershed management that considers transboundary relationships of water use and flood risk; and (4) a political environment conducive for effective policies and legislation. Flood risk management needs to be considered within the overall development framework. In many parts of the developing world, poverty is a significant contributor to people s vulnerability to flooding; frequent flooding can perpetuate poverty. Despite economic opportunities in urban areas, the livelihoods of many people remain agrarian-based. They live in settlements in low-lying areas, relying on agriculture and depending on the land for their food security. Particularly in Asia, recent decades have also been characterized by a massive migration of people to urban centers in search of employment and better access to services, which has resulted in increased urban flood risk as people settle in high-density urban slums on city fringes or next to flood protection embankments and along riverbanks. Poverty affects people s capacity to cope with floods. Factors contributing to poverty and vulnerability to flood disasters are: low income, weak infrastructure, poor shelter, lack of access to public services, unstable political conditions, conflicts and weak economies, and lack of savings and insurance. During medium and severe floods, poor people are particularly vulnerable and incur disproportionate losses and damages. 16

30 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture 2. Challenges for Flood Risk Insurance in Agriculture Providing agricultural flood insurance has a number of challenges. These include: Delineation of losses caused by flooding; Modelling and quantification of the risk and the impact of floods; Operating flood insurance, including loss adjustment and underwriting; and Managing financial challenges related to risk transfer and reinsurance. The extent of these technical and economic challenges in developing flood insurance is high, illustrating why government is often involved in flood disaster management and compensation (especially in developing countries) compared to insurance provided by the private sector. These supply-side challenges are discussed in more detail in the current section, while the next sections explore ways of addressing them using technology and innovative insurance design. Demand-side challenges for flood insurance are not yet tested in developing countries in particular, the willingness and ability of stakeholders to pay the necessary premiums. Demand-side issues are beyond the scope of this paper Definitional Challenges Irrespective of the underlying causes for floods, the losses associated with flooding can be direct and indirect. At the micro (farmer) level, direct losses relate to the impact on standing crops, livestock, and aquaculture. Indirect economic losses can arise due to the interruption of business until full production can be resumed. Flood damage both at the farm property itself or loss of marketing channels or of transport can indirectly lead to increased costs of working and lower income receipts. At the meso (organizational) level, losses are mostly indirect and affect businesses operating in the rural areas. Examples are the reduction of supply of produce to marketing organizations; reduced demand for services from farmers; and exposure of financial institutions to farmers unable to repay loans or interest on loans after a flood. At the macro (public sector) level, indirect costs may be incurred by governments, in addition to direct reconstruction costs, in order to restore normal government operations and support services to the public. Furthermore, there may be loss of tax revenues by government. For widespread flooding catastrophe, disruption of national or regional public finances can impact future operation or growth of the economy. Shortages of 17

31 Agriculture and Rural Development food may lead to increased prices and cost of living. Public sector assets are rarely insured in the private insurance market, with costs of repair born by central or regional government. In order to mitigate the potential effect of flood, mitigation measures (for example, river draining) are a major budgetary cost both in capital outlay and annual maintenance. Ex ante expenditures on mitigation should reduce ex post costs of reconstruction and rehabilitation, as a result of both structural improvements, and increased efficiencies as a result of disaster preparedness. 18 The primary concern of this work is flood losses experienced directly or indirectly by agricultural producers and the financial and public institutions that provide services to them (e.g., infrastructure and rural lending) Direct losses Direct losses refer to the direct physical loss or damage to assets arising from flood. (i) Crops: Damage to standing crops from flooding depends on several factors: crop type, and its vulnerability to immersion; duration of the flooding; growth stage of the crop and cropping calendar; height of the crop and the depth of water; velocity of water flow (potentially causing soil erosion); sediment deposits that are left on the plant surfaces after the water recedes; sediment deposits burying plants after the water recedes. The impact of flood in crop production is complex due to the number of variables. There may be both a loss of crop volume and/or a loss of crop quality resulting from the flooding. The timing of flood events in relation to crop calendar is therefore particularly important (Figure 7). Crop cycles for most annual crops are typically days from sowing to harvest. The most vulnerable period, depending on crop type, is during flowering and ripening. Furthermore, the crop will only have a value to the farmer if it can be Figure 7 Rice crop cycle 1 Rice crop cycle 2 Rice crop cycle 3 Rice crop cycle 4 Example of rice crop cycles of the Muang Patchaboon district, Petchaboon Province, Thailand. June July Aug Sep Oct Nov Dec Seeding Tillering Booting Flowering Grain Filling Harvest Seeding Tillering Booting Flowering Grain Filling Harvest Seeding Tillering Booting Flowering Grain Filling Harvest Seeding Tillering Booting Flowering Grain Filling Harvest 21 days 5 days days 14 days 14 days 21 days depends on available machines and labors Transplant Avg rice growth stage Avg rice height (cm) Critical water depth (cm) Critical flooding time (days) Seeding >3 Transplant >3 Tillering Growing Booting >4 >4 Flowering >4 Grain filling >4 Harvest >4 Source: ASDECON (2008). 18

32 harvested. Although crops may remain in the field, it is a constructive loss if it cannot be harvested if flood prevents the farmers from reaching the field. Flood duration and depth therefore also affect damage, in addition to flood timing. Turbidity (sediment deposits) also adversely affects post-flooding survival of rice plants. In the case of rice, the dominant crop grown in flood prone areas in Southeast Asia, the flood water depth and submergence tolerance of the crop is the subject of ongoing research. 19 Varieties suited to growing in deep water have been introduced since the 1970s. The new varieties enable rice to better withstand the floods and to exploit production potential in deep-water environments. Also, crop management practices are important to avoid periods of seasonal flooding by adjusting the planting window, the number of cycles grown per year, and by using fast-maturing varieties. All of these factors help reduce direct losses to rice crops from flooding. (ii) Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Livestock: Catastrophic floods can cause direct animal deaths, or indirectly affect the animals through lack of food or instance of disease during the flood duration. There may be an opportunity to mitigate the impact of flood by moving livestock from flood risk zones, if there is sufficient flood warning. This will be easier to achieve for grazing livestock. Intensively housed livestock and poultry cannot easily be moved to alternative locations, and their survival may depend on the housing and associated equipment for feeding, ventilation, and heating or cooling. The more intensive and sophisticated the production system, the less readily can emergency actions be taken. (iii) Aquaculture: Aquaculture production is frequently practiced on floodplains, in dedicated ponds, or in bunded crop production areas, for which controlled water management is required. Flooding of ponds and damage to bunds from flood are important causes of loss in rural areas where inland aquaculture is practiced. Economic loss can be the loss of fish (or other aquaculture) and the damage caused to infrastructure, (e.g., bunds and drainage systems). Costs of reconstruction depend on the type of production system and vary widely Indirect losses (i) Crops: There is a distinction between losses to annual crops and to perennial crops, in relation to the time frame to restore normal production, and therefore the extent of indirect losses. For annual crops, replanting may be possible in the existing crop season if flooding occurs early in the cycle. More normally, flooding results in the loss of a full crop season, and normal planting is only resumed for the next crop cycle, even if there are multiple crops that are grown in a single year. In the case of perennial crops, there is a distinction between loss of the productive asset (for example, the tree) and the annual production of that asset (for example, fruits of the tree). In the event of loss of the tree itself (which is considered as direct loss), there can also be substantial indirect costs, including costs to reestablish the plantation, and loss of income until the asset is productive again, typically 3 to 5 years in the case of fruit trees. 19

33 Agriculture and Rural Development (ii) For a farmer growing annual crops, the direct costs of flood will include the loss of crop, considered previously; plus the repair/replacement costs of buildings, stock, machinery, equipment, as well as dwellings. The indirect costs may include the additional costs of working (e.g., feedstuffs), costs of further disruption (e.g., reduction of future production), loss of markets, or disruption of supplies such as seeds, electricity, or other services. Direct loss of land or topsoil through erosion is a further potential long-term cause of loss. Farmers growing perennial crops may have costs in addition to those cited for annual crops notably, costs to clear and replant trees and the costs of several years of lost income until the trees are back in production. For insurance purposes, annual crops are normally valued either in terms of input costs up to the date of loss, or in terms of expected revenue lost (see section 2.5.1). Given the high risk associated with any form of crop insurance, and the fact that a premium goes up in absolute terms with the sum insured, farmers often find that it is only affordable to insure lost production costs, which is a smaller sum insured amount than one based on lost revenue, which includes an element of consequential loss of profits. For perennial crops, insurance distinguishes between two interests: the loss of the crop (the annual production of the tree) and the loss of the tree. The crop can be valued for insurance in a similar way as for annual crops. If trees are insured, an increasing valuation over time is developed until the tree is back in production; the value reflects either the increasing costs of production incurred over time and/or the value of income lost. Livestock: The indirect impact of flood on pastoral livestock producers is primarily seen in increased costs of feeding or re-housing. For grazing livestock, the flood is likely to lead to loss of grazing for a period; alternative grazing may not be available, depending on the extent of the flood or associated rainfall. For intensive livestock production, flooding is likely to be far more serious, depending on the time period before production can be resumed and the degree of disruption of input or marketing supply chains. Housed livestock, particularly intensive poultry and pig production, are normally insured under more conventional property insurance policies in which such insurance is available. It is rare for flood insurance to be available to intensive livestock producers in developing countries. For a livestock farmer, the direct costs of flood will include the loss of animals and repair/replacement costs of buildings, stock, feeds, machinery, equipment, as well as dwellings. The indirect costs may include the additional costs of working (e.g., feedstuffs) and costs of further disruption (e.g., loss of breeding stock giving rise to delays of follower stock), loss of markets, or disruption of supplies such as electricity. (iii) Aquaculture: The indirect impacts of flood on aquaculture are similar to those of livestock and vary according to production system. Additionally, water pollution or water quality following flood may impact the period before production can be resumed. 20

34 2.2. Technical Challenges Modelling flood risk Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Flood risk modelling is a well-developed methodology and is a fundamental tool to support flood risk management, flood prevention, and flood warning systems as well as for hydro-electrical energy, integrated water and flood management in catchments, and many other applications. The increased concern of and losses to the global insurance industry has led to major advances in integration of flood hazard and vulnerability modelling for insurance purposes by insurers and reinsurers of property risks. The expertise has been developed by a few major reinsurers, and by specialist companies, notably Risk Management Solutions (RMS), Eqecat, and AIR Worldwide, whose client base is in the property insurance and reinsurance market. The key technical challenges associated with flood risk modelling relate to the estimations and mapping of the physical hazard (the flood) these are further detailed in section 3. Furthermore, remote sensing is a complementary technology to flood modelling, which strongly supports many aspects of flood insurance. In the case of flood modelling for property insurance, vulnerability distributions for buildings (percentage damage, according to depth and duration of flood) have been developed according to building type, business type, contents, and stocks. These vulnerability functions have then been linked to the hazard model in order to determine the financial impact of given flood scenarios. They have been used to allow the development of premium rates and for accumulation control (a process by which insurers calculate the prudent amount of risk that can be accepted by the insurer in any given region). Loss or damage caused to crops by flooding implies either (1) an inability to continue cultivating that crop (i.e., the crop is lost at the time of the flood); or (2) the crop will partially recover if the flood recedes, but there is a loss of yield at harvest time. Furthermore, the flood might give rise to secondary constraints to the final harvest, notably disease as a result of high humidity, loss of quality, contamination by flood residues, or inability to access fields with harvesting machinery. As flood insurance for agriculture is not yet a developed class of business, this sophisticated modelling approach for property insurance, has not been widely applied to estimation of damage to crops or livestock within an insurance scheme. However, there is no reason why the same principles cannot be applied. The development of crop vulnerability functions has specific challenges: Unlike property damage modelling, where fixed assets remain static throughout the year, and generally over the long term, crops are only present for part of the year and are likely to change from year to year according to crop rotation; Damage caused to crops is differentiated according to growth stage. Generally, damage is less at the vegetative stage, but may be high at the reproductive and ripening stages; 21

35 Agriculture and Rural Development The timing of the flood is critical in relation to the crop cycle. Although crop cycles are known, there is variation according to seasonal factors, such as onset of planting rains or temperature, and there is also variation between individual farmers; The duration and depth of flood is very important to the actual damage caused to a crop. Crops adapted to immersion may survive flooding, even as part of their natural growing cycle, (e.g., certain varieties of rice). Other crop types may be very vulnerable and survive only for a short period; Sediment residues on the plant itself, or siltation of the land, may restrict the ability of the crop to recover after recession of the flood; An associated effect of flooding and/or excessive rainfall may be to cause soil waterlog for a period after recession, leading to an inability to access fields with machinery; Minor floods are beneficial to crop production, if they occur at the expected time, bringing nutritious sedimentation and moisture; the impact of flood benefit or flood damage is related to timing and duration of the flood event. Cropping systems and crop varieties have evolved in order to maximize such benefits and to avoid damages caused by flooding. These points reinforce the need, in any insurance system for agricultural flood insurance, to create a GIS-based database, which will allow information on cropping and geographical location of crops to be known. This forms a logical extension of the database of clients that needs to be held by an insurer or by an agricultural bank lending to farmers. River systems are subject to flood management, and the intervention of human activities can be crucial to insurers, especially if these are not planned or known in advance. In particular, agricultural areas are frequently flooded purposefully (flood detention areas) in preference to urban areas being flooded. Such interventions, often based on political decision, are very difficult to model Flood zoning Zoning is an important feature of any flood insurance program and requires a modelling approach that defines the spatial extent of flood relative to agricultural assets. A fundamental advantage of zoning which refers to grouping farmers who will be treated identically for premium and claim calculation is to reduce cost of enrollment and of loss adjustment. A challenge for any insurance agricultural scheme in developing countries is the small economic size of the farmer clients. As a result, high unit administrative costs can be incurred for farmer enrollment and for loss adjustment. Conversely, a disadvantage is that there still is a possibility not all farmers will be equally impacted within the defined zone despite best technical efforts to define the zones. This situation is referred to as basis risk. The resolution (unit area) of zones is a critical issue for flood insurance: to strike a balance between the requirements to assess flood risk in localized areas, yet avoid excessive basis risk. (It is noted that in weather index insurance for drought or excess 22

36 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture rainfall, where measurements made at specific meteorological stations are used to trigger payouts, all farmers within a given distance of the station have the same cover which is a similar principle as zoning for flood insurance.) Initial considerations in the background feasibility case studies for this paper (section 5) were that floodplains would need to be zoned, based on flood modelling outputs, where zonal boundaries would be dictated by elevation or by river training. Such boundaries would be curved. The initial considerations were that unit areas of between 50 and 100 hectares (identified from flood models) could be used as the zoning framework for (1) farmer enrollment; and (2) calculation of payouts, which would be proportional for all farmers within a given zone. As a result of the flood modelling work in the Pasak River Valley in Thailand (see Figure 16 in Section 5.2 in Chapter 5), where possible boundaries from flood modelling were indicated, it became apparent that determination of flood zone boundaries (which did not conform to administrative boundaries) was problematic. Therefore, while bunded floodplain areas may provide obvious boundaries, a simplified approach would seem to be the imposition of a grid system (see section 4). In constructing such flood zone grids, farmers would be enrolled (using GPS measurement from a center point in the farm) into each grid cell. Then, the key issue is the level of resolution of each grid cell that is most appropriate. This gridded flood zoning system has not yet been empirically tested but would likely be determined based on: (1) the resolution of remote sensing imagery to be used (for example, to give a minimum pixel count within each grid cell); (2) the uniformity of the floodplain; and (3) whether a micro level or aggregated interpretation of flood risk was needed. An additional challenge for zoning is the difference between managed floodplains (e.g., those with river training walls) and unmanaged (natural) floodplains, as active water management (irrigation, flood detention areas, etc.) affects flood patterns. Flood risk zone boundaries may, however, be defined more easily if there are training walls, where flooding occurs after overtopping Product design and pricing Agriculture has additional challenges in flood product design, because timing of flood is critical, according to the growth stage and vulnerability of the crop, and whether flood occurs during or outside of the cropping season. Aggregating farmers into zones of homogenous flood risk also implies similarity of the crop calendar for all farmers in that zone. Homogenous payouts require that farmers in a zone are also reasonably synchronized in their cropping calendars. There are technical challenges in establishing premium rates that relate to agricultural flood. Notably, flood modelling according to timing and seasonality of flood (in addition to spatial estimation of annual frequency) is an extra step in flood premium assessment. Premium rating based on risk assessment requires technical information on the hazard (frequency of 23

37 Agriculture and Rural Development occurrence), vulnerability (degree to which assets and activities can be affected by the hazard), and location of crop or other assets. Another issue for agricultural flood is that of duration of flood and of sediment load in flood water. The timing and duration of flood are not normally so critical in building or contents insurance as the financial exposure of these assets to flood do not vary seasonally. Related to challenges of premium assessment is the need to model infrequent but severe losses (normally through flood modelling) in order to determine a probable maximum loss (PML), which may be expressed as a one in 100- or one in 250-year flood loss. The potential for extreme loss is a factor in determining the capital required to operate an insurance scheme and a role of private reinsurance or government backing. Financial planning for flood insurance involves consideration of the volatility of annual result, cost of capital and reinsurance, administrative and loss adjustment costs, subsidies (if any), and research and development costs, all affecting the final premium charged Operational Challenges There are several challenges related to the operation of an agricultural insurance scheme for flood, including: (1) risk-underwriting, which is closely linked to the ability to map flood zones and flood frequency; (2) adverse selection, which potentially arises from an uneven distribution of risk; and (3) loss adjustment, which is typically associated with high transaction costs in many agricultural insurance schemes Underwriting Underwriting refers to the process by which an insurer evaluates the risk and exposures of potential clients in order to determine acceptable risk (eligibility), level of coverage, and premium. The requirement to establish risk zones is critical for underwriting flood risk. Establishing good information of the flood risk is essential for insurers, as characteristics of the expected flood, and its impact, need to be understood. An insurer needs to understand the expected frequency of flooding, expressed as return period in flood modelling terms, for any location. Establishment of flood risk zones within a flood plain based on different return periods is therefore central to the decisions on insurance scheme design. A challenge, therefore, in flood underwriting is to establish flood risk zones, for which purpose flood modelling is the main methodology employed. The applications and limitations of flood modelling and remote sensing (section 3) dictate the confidence in calculation of return periods and the level of confidence which is possible in expected flood frequency and duration at a local level. A main challenge is to be able to identify, for all insured clients, the probability and severity of expected losses and the extent of claims arising from these losses. This is essential in order to derive premium rates required in order for the insurer to be able to meet claims, administrative costs, loss assessment costs, and a margin of profit. 24

38 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture The ability to zone risks, and degree of confidence that results at a local level, are important factors determining the resolution of the insurance that is possible. These considerations dictate whether a scheme can be operated at a micro level of resolution, which targets individual farmers, or whether a more aggregated solution may be needed for example, flood insurance for an agricultural bank that lends to many farmers in a region (see section 4) Adverse selection The key challenge faced by insurers in developing flood insurance is the potential for adverse selection. The risk in most types of flooding is highly localized and may be clearly demarcated. This is different from a highly correlated risk like drought, which is likely to affect all farmers in the same region regardless of the location of their farm within the region. Farmers are aware of the flood potential of their land and of specific fields within their farms. Farmers outside perceived flood zones may not be aware of the potential for extreme flood events (for example, one in 100 years). If they have not experienced major adverse events, farmers may not be willing to pay for flood insurance. Insurers are faced with several challenges in relation to adverse selection, irrespective of whether the focus is on agricultural or property risks. First, if the scheme is voluntary, only farmers in high-risk areas will tend to insure, reducing the available premium pool. Second, it may be difficult for farmers in high-risk regions to afford premiums that are actuarially matched to the risk; in low-risk regions, flood risk may be very low and therefore there is no appetite (demand) to purchase insurance, even at low costs. Minimizing the potential for adverse selection requires that premium rates are related to expected claim costs, or that differential terms, such as higher deductibles, are included. Further, insurance systems with compulsory flood premiums, irrespective of flood risk, can be used to create a risk pool. Compulsory systems normally recognize a degree of cross subsidization between low- and high-risk clients (a solidarity approach). However, they require that there is a large pool of farmers participating. Where no other types of insurance (e.g., property, crop, etc.) exist with which to link flood insurance, establishing a large pool only for flood risk is problematic. Insurers have found ways (in conjunction with government) of creating compulsory enrollment of all policyholders (a mutuality approach ) and adding flood to existing insurance policies. 20 In theory, if it were possible to ensure that premium rates precisely reflect the risk of each buyer, adverse selection can be avoided. Numerous mechanisms are routinely implemented by insurers to try to match premiums to risk in other insurance classes. These measures include no-claim bonuses (an experience rating approach); close analysis of claim statistics to segment the client base; and establishing zonal and individual rating scales. In the case of flood insurance in agriculture, the normal basis of premium rate development is by using flood modelling, backed by any history of flood records (and possibly remote sensing archives), to determine premium 25

39 Agriculture and Rural Development rates per each homogenous risk zone. A major challenge in premium rating is the level of resolution of flood risk zones, which is possible from flood modelling. Flood exposure is complex on a local basis, especially for flood during minor to medium flood events, and the limits of flood modelling are likely to be a constraint on mapping detailed flood risk zones down to farm level. Given that individual farm rating is not likely to be feasible, rates must be set for all farmers within a given zone. If the resolution of the zone is low (i.e., the zone size is large), or if there are other reasons why actual risks cannot be reflected by premium rates charged, there is a high risk of adverse selection. Further, if there is a strong demarcation between areas of either very high risk, which are too frequently flooded to be insured and very low risk, where farmers do not demand insurance, then flood insurance may not be viable. During a feasibility study, it was found that areas of the Pasak River Valley in Thailand had these dichotomous risk characteristics (section 5) Loss adjustment Loss adjustment for flood insurance faces the same types of difficulty typically found in all crop insurance. A key challenge in loss adjustment (whose definition is different from loss assessment, as described in Box 5) for flood is to obtain objective, rapid, widespread, and low-cost information on the extent and duration of flooding. This is very difficult to achieve in a systematic manner using only on-the-ground measurement. Remote sensing offers an opportunity to harness a technology that was not previously available. When linked to ground validation, it provides a very powerful tool for real-time measurement of flood-induced crop losses, which could support an insurance program. Designing a viable loss measurement system for agricultural floods has to be considered in relation to zoning of insured farmers. Individual-farmer flood loss assessment, similar to the process employed for traditional crop insurance, would likely increase costs and also gives rise to increased potential for adverse selection. On the contrary, establishment of area-based loss assessment and zonal payouts for floods would bring many similar advantages that index insurance offers as when the approach is applied to other perils. The application of index approach to flood risk is discussed in detail in section 4. But to apply the index principle to flood loss adjustment, a key is to ensure that areas of reasonably homogenous risk profile can be identified, and based on which, flood risk zones are formed. The resolution of available data becomes critical in this design aspect for flood insurance as in the underwriting aspect discussed previously Financial Challenges The key financial challenges of flood insurance schemes are related to: (1) developing a simple approach to valuate the losses in the field and (2) managing the financial impact of insuring covariate and catastrophic risk effectively. 26

40 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Box 5 Loss adjustment for traditional and index-based crop insurance Loss assessment refers to the measurement of physical loss, while loss adjustment refers to the finalization of the financial payout to be made in an insurance claim. Both are the cornerstone of any crop insurance program. Traditional crop insurance products usually covering named peril(s) or multiple perils are sold to individual farmers. Hence, when there is a loss, measurement of loss or damage for individual farmers has to be made when a claim is reported. Detailed procedures have been established, according to crop type and peril causing the damage, to estimate loss. In the case of damage-based policies (e.g., hail and named peril policies), an estimate of percentage damage is normally made by field inspection soon after the loss event. In the case of yield-based insurance (multiple peril crop insurance-mpci), losses are measured at harvest by comparing final actual yield with a previously-established insured yield. For small farmer communities, the main challenge with conducting loss assessment and loss adjustment for any traditional crop insurance product is the complexity in conducting these exercises at the individual farmer level. Unit areas for individual assessment are too time-consuming, and overall costs of loss assessment operations are high. Skilled manpower is not normally available to undertake such loss assessment in developing countries, particularly for catastrophic perils which require simultaneous assessment of a large number of insured clients. The need to find a simplified and lower cost product led to the interest in index insurance. Under index insurance, no field assessment takes place; instead, payouts are made against a separate proxy measure that correlates with losses the farmers experience. In weather index insurance, the measurement is the amount and/or variation of a meteorological parameter (such as rainfall, temperature, etc.) recorded at a meteorological station. In area index insurance, official average yield assessments made over a specified unit area, such as a defined local district, are used as the basis of identical compensation for all farmers subscribing to the insurance within that unit area. A common feature of all types of index insurance is that all farmers in a given zone are treated identically, thus reducing costs of loss assessment and resulting timely payouts. The same index principle can be applied in flood insurance. If a zone that is expected to be homogenously damaged by a flood could be established, then all farmers in the zone can expect similar damage from a given extent of flooding, thus similar payouts. Flood zoning is a cornerstone of flood insurance product design under this system Valuation for insurance purposes Flood loss assessment in property insurance relies on the principle of indemnity; after a loss, an assessor needs to agree on the damage, the repair or replacement required, and the costs. Hence, modellers of the impact of flood on property needs to make use of vulnerability estimates according to property type and expected repair or replacement cost. For loss assessment in crop insurance, on the other hand, assigning a value to the loss of crop requires a different approach to valuation. Similar principles used in crop valuation for crop insurance internationally can be applied to agricultural flood insurance. This also gives an opportunity to simplify crop 27

41 Agriculture and Rural Development loss assessment for flood. In short, crop valuation that could be applied to flood events is either based on (1) input costs at the date of loss, or (2) loss of expected revenue. Input cost valuation is easiest to establish, as the normal inputs for crop growing, the timing in which these inputs are applied within a crop cycle and the expected cost of such inputs, is not subject to major seasonal variation. Costs, on a unit area basis (such as acre or hectare), can include the variable costs (seeds, fertilizers, herbicide, etc.), fixed costs (allocation of overhead costs per unit area), and some value assigned for both the farmers own labor and hired labor. Incremental values of input costs can be calculated as the season progresses. Total input costs for cultivating one season could form total sum insurance of the insurance contract. Furthermore, in order to simplify insurance administration, an agreed value scale can be established according to growth stage. A flood event that damages the crop towards the later growth stages will result in higher payouts than one that impacts a young crop. This is based on an assumption that farmers would have invested more in crop inputs as the crop has matured. Revenue valuation for crops is more complex and where practiced is also normally done on an agreed value basis. Attempting to determine the yield and price of the crop that would have occurred in a particular season is fraught with uncertainty. An agreed-value approach can establish an expected revenue scale. The scale takes some account for the stage of growth at the time of loss, as there will be a saving of input costs (for example, costs of midseason fertilization or herbicides or harvesting) that will not be incurred for early- or mid-season loss. If a valuation scheme is designed to include some element of loss of profit margin, an approach is to add an agreed amount per unit area to the input cost agreed value scale as indicated previously. Flood insurance for crops can benefit from a simplified agreed-value scale. The key challenge in modelling for flood that affects crops, as opposed to property, therefore arises from the timing of the flood event. The date of occurrence in relation to crop growth becomes critical to the survival of the crop, or loss of yield, and therefore of the valuation of the financial impact Covariate risk, catastrophe exposure, and reinsurance Flood risk insurance is a form of catastrophe insurance, where it is expected that a large number of insured clients will simultaneously be affected by an event. For insurers, this covariate risk contrasts with well-spread risks that are independent of one another. The latter are risks covered in most forms of property and motor insurance, sometimes referred to as idiosyncratic risks. In the case of crop insurance, there is a range of degrees of risk correlation: from completely independent in the case of fire, to partially independent in the case of hail, to highly correlated in the case of drought or non-localized flood. The fact that flood risk is highly correlated does not make it an uninsurable risk; however, it makes the financial management of the risk more complex. As with other catastrophic risks, it is fully expected that in some years, flood 28

42 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture claims will far outweigh premium. For an insurer operating in a country, the opportunity to achieve flood risk spread to reduce these peak exposures is normally limited, although opportunities for risk diversification exist depending on geographical distribution of flood, rainfall patterns, river catchments, and potential clients. Risk diversification (either by a network of mutual insurers or by a single insurance company) depends on the size of country but is more limited for flood than for more independently occurring perils such as hail. In this context, the financial management of peak exposures, through reserve management and through reinsurance, become extremely important. These measures are well established in property insurance and crop insurance and are equally critical for flood risk insurance in agriculture. As a generalization, reinsurance capacity is available in all classes of insurance to support genuine primary insurers who are able to demonstrate viable underwriting and insurance products, and flood insurance is no exception. Provided that flood risks can be quantified, they can be structured for risk transfer in the same way as any other class of business subject to catastrophe exposure. For a market-based flood insurance product, the support of reinsurers is essential to individual insurance companies underwriting flood insurance as it provides a key protection against potentially very large financial exposure. Acceptance of catastrophe or correlated risks is an area of specialization of most reinsurers. They are theoretically better able to build more balanced portfolio of risks, given the ability to source a number of uncorrelated or partially correlated risks globally. This diversification has the important effect of reducing the cost of underwriting these risks. Structuring a reinsurance program with proportional and non-proportional reinsurance is feasible for flood risk. But because flood insurance in agriculture is new and reinsurance for agricultural flood is not yet widespread, risk transfer from domestic insurers to international reinsurers may be easier to achieve using an index basis. This builds on recent experience which has shown that index instruments facilitate the risk transfer as they overcome some of the key underwriting constraints (e.g., adverse selection, moral hazard, and high loss assessment costs, etc.), which had inhibited many reinsurers to participate in conventional crop insurance. If there is also a chance of infrequent but extremely severe flood risk that might exceed the ability of the commercial insurance market to support, then the government can become involved in financing this catastrophic layer of risk. This is sometimes referred to as the market failure layer. 21 In many countries, the government plays the role of insurer of last resort for this risk. 29

43 Agriculture and Rural Development 3. Flood Risk Assessment and Mapping The advances in geo-information technology allow floods to be observed and modeled in a more timely, accurate, and objective manner than in the past. This section describes the key technologies that are being employed to monitor floods and assess flood risk and the manner by which they can be harnessed to address challenges discussed in section Flood Risk Modelling Flood risk assessment is the estimation of the risk of future water discharges and/or levels (stages) that cause damage. It commonly uses data from local and regional river gauging, or hydrometric stations covering the longest possible periods-of-record, and statistical flood frequency analysis, coupled with techniques to estimate flood risk for land parcels of interest. Such work may provide, for example, an estimated 100 year discharge, referring to flow (with a probability of occurrence in any given year of.01 or 1 percent) for a stream or river valley at a particular location. Given this discharge, and an adequate characterization of the geometry, slope, and resistance to flow in the river channel and floodplain, flow hydraulic modelling techniques are used to calculate flood stage. In turn, comparison of stage to local topography provides a map of land parcels under and above this flood level, so inside and outside the 100 year floodplain. The quality of the resulting flood hazard map depends heavily on the necessary input data and can vary widely depending on how well characterized the floodplain topography is. Also, depending on the type of flood, the connectivity of local depressions along a floodplain landscape affects flow routing and actual flooding. For floods occurring through bank overtopping of the local river, parcels below the flood stage may not be inundated unless there is a connection to the water source; for example, levee systems may provide local protection to such low-lying land. However, some floods occur through saturation of floodplain soils; direct precipitation onto waterlogged soils results in the emergence above ground level of the local water table. Such floods occur, for example, repeatedly in many low-lying monsoonal areas of Asia. In these cases, levee systems and hydraulic connectivity become less important, and the land elevation is the critical factor. For either type of flood, the level of detail in the characterization of local topography strongly affects the spatial resolution, accuracy, and precision of the resulting flood hazard map. Hydrologists are currently engaged in improvements in both areas (flood discharge and flood inundation) of flood risk assessment. By using watershed modelling, antecedent soil moisture, and time-series of extreme rainfall, flood events can be modeled or synthesized and their return periods estimated. Box 6 provides an example of flood risk assessment process for flood insurance in the 30

44 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Box 6 Flood risk assessment in the United States Flood risk mapping procedures must be both standardized, if they are to be employed equitably over large areas, and scientifically responsible. Fundamental to modelling floodplain inundation is estimation of the size of very rare flood flows, such as those having, for example, an estimated 0.2 percent chance of being exceeded in any given year (the 500-year flood; commonly used in U.S. flood insurance studies where detailed basis exists for maps of the 1 percent annual excedence, 100-year flood). Alternatively, the probable maximum flood (PMF) is based on assumption of the most severe hydrological and meteorological conditions at a site and uses the probable maximum precipitation (PMP) reports provided by the U.S. National Weather Service. PMP may be used for applications such as dam design. Within the U.S., Congress established the National Flood Insurance Program (NFIP) with the passage of the National Flood Insurance Act of The NFIP was to reduce vulnerability to flood damage and a flood insurance safety net for individuals. Flood insurance was not generally available though the commercial markets and the NFIP enabled property owners to purchase insurance against flood losses. Such insurance shifts risk from federal taxpayers to those whose properties are at risk, and extension of the insurance brings with it certain obligations on the participating communities regarding measures to reduce exposure to flood losses. The 1968 Act authorized the Federal Government to identify and publish information with respect to all floodplain areas, including coastal areas located in the United States that have flood-prone areas, and then to establish or update flood-risk zone data in all such areas and to make estimates with respect to the rates of probable flood-caused loss for the various flood risk zones for each of these areas. More recently, though the Flood Map Modernization Program and its partners, the Federal Emergency Management Agency (FEMA) provides updated flood hazard data and maps for the United States to support NFIP. 22 In regard to estimating flood discharge, the U.S. and many other nations have developed standard techniques or handbooks so that flood size and frequency are estimated in a uniform ( accountable ) manner, and even though standardization alone does not accomplish the desired goals of accuracy and reliability. Within the U.S., the United States Geological Survey (USGS) developed regional regression equations to estimate the flood frequency and magnitude at ungauged locations of a watershed. These equations are based on flood frequency and magnitude from gauged watersheds and on a standard probability distribution (the Log Pearson Type III). Regression equations transfer flood characteristics from gauged sites to ungauged sites through the use of topographic, physical, and climatic characteristics of the ungauged watershed. The USGS also developed a National Flood Frequency (NFF) computer program that estimates the flood frequency and magnitude for ungauged sites by the application of the appropriate regional regression equations. NFF was first released in 1993 and subsequent revisions were made in order to incorporate updated data. Because the majority of stream gauge data used in developing the regression equations were collected from rural watersheds, the applicability of the regression equations is generally restricted to rural watersheds. Urban watersheds exhibit different flow characteristics. Therefore, regression equations for urban watersheds have been developed for some areas, and include parameters such as the percentage of impervious areas and urban development factors. Also, the USGS (continued) 31

45 Agriculture and Rural Development regional regression equations are not appropriate for estimating flood magnitude and frequency in watersheds subjected to flow regulation, or to heavily mined areas or karst areas (limestone terrain with abundant sinkholes, caves, and caverns) where excessive runoff is diverted into or outside the surface water basin. Finally, due to the relatively limited number of years of stream gauge data availability, large estimation errors are, in general, inherent in the peak-flow discharges estimated using USGS regional regression equations. These errors are, in turn, incorporated into subsequent hydraulic modelling and mapping of flood hazard. Hydrological risk zone mapping addresses, among other societal needs, the requirement for flood risk assessment in setting insurance prices. Risk zone mapping produces a set of maps and quantitative, standardized, legally defensible interpretations of flood hazard zone delineations. In developed insurance sectors such as the U.S., these are flood insurance rate maps (FIRMs). The basic data input is detailed topography and map information concerning the local drainage network (streams and rivers). In principle, hydrological risk zone mapping also requires developing numerical models that dynamically describe hydraulic characteristics (i.e., water level, flow, velocity, etc.) for a defined topographical region. These models incorporate historic and current meteorological, hydrological, and hydraulic data, as well as relevant information about natural and man-made physical features including the location and characteristics of flood control works. When the environment in the defined region changes either by natural or human causes, the hydrological risk maps must be revised and updated. Consequently, the underlying numerical models should be capable of incorporating and simulating the effect of changes to a number of variables over defined time periods. United States. Increasingly, new-technology methods of acquiring the needed topographic data are being emphasized because flood risk maps are so dependent on topographic vertical and horizontal accuracy and coverage. Laser light detection and ranging (LIDAR), commonly from airborne platforms, has emerged as an effective method of acquiring the high-verticalprecision topography needed. Finally, for many regions, wide spread availability of digital elevation data and GIS software permit the automation of the time-consuming tasks associated with flood-prone area delineation. The analytical capabilities of the GIS significantly speed the calculation of the flood discharge information and ultimately flood-prone area delineation. Floodplains have long been preferred locations for many forms of agriculture, and some agriculture in fact depends on yearly flood inputs and is adapted to associated damage; this is the case for crops along the Nile River, for example, prior to the Aswan Dam. Insurance schemes, however, must avoid covering frequent losses in favor of providing compensation from unusually damaging events. In developed nations, there has long been a focus in flood hazard evaluation on reducing losses to residential and commercial properties by, for example, delineating the the 100-year floodplain and restricting development therein. However, in developing nations, low-lying lands that are flooded approximately once every 5 years are commonly quite intensively farmed and will continue to be. The frequency of crop losses from such flooding will also continue at a correspondingly high rate. In this regard, focus on a simple 100-year floodplain delineation would not be most appropriate. 32

46 Instead, the most useful information in developing nation agricultural landscapes may be better knowledge of frequently flooded lands (for example, return period 5 yr), and as compared to lands flooded much less frequently (return periods of 10 yr, 50 yr, or 100 yr) Wide-area Flood Risk Mapping Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture There are two distinct tasks in flood modelling: (1) estimating the size and frequency of the flood flows, and (2) determining the inundation depth and mapping extent to be associated with such flows. For agricultural insurance purposes, there is also a motivation to incorporate flood duration into such probability estimates, so that land areas that sometimes experience long duration of flooding can be differentiated from those where the flood wave passes quickly and drying occurs more rapidly. Real world floods are derived from different causations and exhibit different seasons of occurrence (see section A.1). Ideally, recurrent interval-based flood-risk zoning would exclude flood statistics for noncritical seasons and instead map risks for periods during which crops are most susceptible to damage (e.g., during the flowering or maturity stage). Ideally, because of the uncertainties involved, the results of the risk zone mapping for each study area would also be validated by some independent means, such as by comparison to aerial or satellite imagery of known events. The search for more effective and economical methods of assessing flood hazard in developing nations is motivated by these difficulties in applying the traditional approaches. There is also a pressing and immediate need to identify lands flooded more frequently than every century. Although watershed land use or river channel changes may affect the size of the 100-year event to some degree, their most powerful effect is on more frequent flooding. In this regard, traditional flood risk assessment uses the past as a guide to predicting the future: it assumes stationarity in the frequency distribution of peak flows over time. The importance of this assumption merits further discussion in section 3.3, and its general lack of validity suggests that alternative methods of risk mapping should be considered Risk Assessment and the Assumption of Stationarity Statistical analysis of streamflow time series for the purpose of flood risk assessment has long used the assumption of stationarity, i.e., that the sampled time interval is representative of a population of flood peaks from a homogenous population whose recurrence intervals are not changing systematically over time. However, floods along any given river are, very commonly, composed of events caused by different meteorological circumstances. 23 The assumption of stationarity is a convenience for statistical analyses, but it is widely acknowledged to not be true, in the strictest sense. Opinions vary, however, on how important such departure from stationarity is. For larger rivers, the effects of upstream land-use change also involve international relations. For example, flooding along the Mekong River in Southeast Asia causes extensive crop and other societal damage, raising the question as to what degree such flooding can be attributed to upstream, crossborder changes, such as deforestation in China. In this regard, analyses of 33

47 Agriculture and Rural Development individual river flow series from many regions have often demonstrated that the assumption of stationarity is rarely fulfilled: many streamflow series exhibit long-term trends, and including changes in the frequency and magnitude of floods. 24 Some of these trends may be related to global warming or to other climate changes. 25 Recent analyses using both precipitation data and modelling indicate that global precipitation has in recent decades increased, that precipitation intensities have also increased, and that both total annual amounts and per-storm amounts will continue to increase. 26 This in turn will cause, at least in some regions, an increase in flooding. 27 The overarching implication for flood risk assessment and insurance pricing is that application of standard statistical methodologies for determining flood frequency and magnitude do not provide a complete and accurate assessment of flood risk. The past is not, necessarily, a key to the present and future. Because of the variety of causes for non-stationary and non-homogenous flood flow series (see an example of the Mississippi River discussed in Box 7), there are no standard techniques for including such variability for calculating actual flood risk at many locations and over large regions. At a minimum, however, methodologies for determining stationarity or non-stationarity in flow series should be applied. Risk assessment could also take into account the modeled future trends in climate as well as known trends in contributing watershed land use. The incorporation of historical data can provide a longer time scale in some locations and thus further evaluate long-term trends. In some reforesting watersheds, downstream flood risk is decreasing, whereas, in many watersheds subject to deforestation, rapid urbanization, agricultural development, or channel aggradation, flood risk is increasing. Upstream reservoir construction and other engineering works strongly affect downstream flow hydrology and, in areas bordering high mountains, global warming-related glacier melting are today posing entirely new flood risks as alpine lakes expand. 28 Flood risk determinations based on national-standard methodologies for the statistical manipulation of river discharge data are, thus, only a first step. They could be plainly inappropriate for regions in developing nations where the needed long-term instrumental records and channel geometry surveys are lacking, and where watershed changes in recent years suggest that exposure to flood risk is also changing. Such issues affect most contemporary river systems. They become even more pronounced from the perspective of agricultural flood risk. These findings constitute a strong caution against simplistic assumption about stationarity in developing nations. In many of these nations, contributing watershed changes are known to be dramatic, the frequency and magnitude of relatively frequent flooding are of most interest, and there are likely to be other inter-annual, inter-decadal, and greenhouse warming-related changes and trends. In this context, the utility of the spatially extensive data provided by orbital remote sensing is now examined. It provides a time perspective extending only to recent decades, but remote sensing benefits from unbiased and complete geographic coverage. What is observed and measured by remote sending (inundation extent, rather than discharge) is directly relevant to risk zone mapping. 34

48 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Box 7 Examples: Non-stationarity of streamflow in the Mississippi Even within developed nations with abundant instrumental streamflow records, stationarity issues cause significant uncertainties. An illustrative U.S. example concerns the middle and lower Mississippi River, wherein different flood risk assessments result from the application of flood stage versus flood discharge time series. 29 For the Missouri River, channelization during the past ~100 years has caused channel capacity losses. At measurement stations spanning 1000 km from the lower Missouri to the middle Mississippi rivers, increases of up to 3 4 m (in flood stage, for the same discharge) have been linked to the construction of navigational engineering structures and levees. Different results thus occur if the same statistical techniques are applied to flood stages (which are most directly related to floodplain inundation extent) and the more commonly used flood discharges (this latter method causes an incorrect, reduced, estimation of risk). In this case, even if the flood discharge time series exhibits stationarity, the actual flood risk probability is changing. However, careful analysis of the flood series time series indicates that twentieth-century climate change also invalidates stationarity of the flood series. Figure 8 Flood magnitude, in cubic meters per second Illustration of non-stationarity of flood time series for the Mississippi River, U.S. 2% probability flood 5% probability flood 10% probability flood 99% probability flood Year 3.4. Remote Sensing-based Flood Risk Mapping Gauging station-based flood risk approaches, and the associated statistical and modelling infrastructure, all evolved before the advent of orbital remote sensing and the possibility of routine mapping of flood events. In the absence of the latter, flood modelling is the only feasible approach for many regions but, if remote sensing is available, developing an observational record of inundation may be more accurate and also more cost-effective. Unlike the case for gauging station data, the protection of some land by dams or levees is directly recorded by remote sensing of those lands during regional flooding. Flood peak discharges can be high without actual land inundation and relatively low but, if levees fail, still be accompanied by large-scale flooding. Examples of the latter in developing nations occur very often: in 2008, levee failures during relatively modest flows were important in 35

49 Agriculture and Rural Development agricultural land flooding throughout South Asia, including in Nepal and India. 30 It appears that measuring and monitoring flooding via a downward looking approach, from orbit, has strong potential for objective determination of flood impacts. That is, it may be better to measure and monitor what agricultural lands are actually being flooded, than attempt to determine via flood modelling what land should be flooded based on discharge recorded at a gauging station. Orbital remote sensing technology also provides internationally consistent geographic coverage of flood inundation. Through this approach, flood hazard assessments can, in principle, be addressed without ingesting and manipulating any meteorological or river discharge data. Instead, the direct observable is the history and location of actual flood inundation. Figure 9 shows sample rapid response flood inundation map obtained from remote sensing. When intersected with map data regarding croplands, such Figure 9 NASA MODIS and RADARDAT-based flood inundation mapping for a hurricane event in Central America (Dominican Republic and Haiti). Landsat data prior to the flood event provides a reference surface water data layer. The two MODIS sensors image this land several times each day and provide for multiple steps during flooding. Map is shown at reduced spatial resolution (much more detail is present in the full-size versions). 36

50 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture maps offer the opportunity to immediately quantify area of cropland flooded. An associated scientific challenge is how such map data can be transformed into quantitative and reliable evaluation of flood risk. Although realistic (but computationally intensive) modelling of floodplain inundation and drying can be locally accomplished, 31 such work continues to demonstrate the importance of factors other than discharge in affecting inundation patterns and duration (for example, antecedent conditions such as floodplain soil moisture). Depending on such conditions, the same discharge can cause varying inundation durations. The intensive effort invested in predicting the size of flood discharges thus does not translate directly into accurate inundation prediction. Meanwhile, orbital remote sensing is providing still largely-untapped direct observation of exactly the phenomenon that otherwise must be modeled as a function of many variables. Remote sensing can delineate those land areas where flooding is in fact a contemporary hazard, rather than areas where discharge data, channel bathymetry, channel slope, hydraulic theory, topographic data, and modelling indicate where flooding could occur. An entirely different approach in flood risk assessment is possible, given the new technologies Contemporary satellite remote sensing systems Although the usefulness of remote sensing has been recognized and described by many professionals, the transfer of this capability into risk mapping has been slow to occur. This may be due mainly to sensor data acquisition and distribution policies, which have previously made relevant data difficult or expensive to obtain. Orbital remote sensing was first used to map inland river and coastal flooding commencing in the early 1970s (see A.3 for more details on satellites and space-borne sensors). A number of sensors have been used for flood mapping, including the Advanced Very High Resolution Radiometer (AVHRR), the synthetic aperture radar (SAR) satellites (ERS-1, ERS-2, Radarsat, JERS-1, and Envisat), the French SPOT satellites, and the two MODIS sensors aboard NASA s Terra and Aqua satellites. 32 The AVHRR sensors, operated by the National Oceanic and Atmospheric Administration, which provide ground resolution of only 1.1 km (features on the ground smaller than this cannot be distinguished), were early-on appreciated as flood mapping tools, although mainly large rivers were studied. However, the record provided by these long-flying sensors dates back to the mid-1980s, and data archives are available to the public without charge. They contain valuable information about past large floods. Most other sensors are higher in spatial resolution (instead of 1 km resolution, ~30 m to ~1 m) but lower in temporal sampling (several days to ~15 day revisit intervals). They provide snapshots of floods, very close up, commonly at significant per-scene cost, and not always at the time of peak inundation. The significant exceptions are the two MODIS sensors, beginning 37

51 Agriculture and Rural Development in late 1999, which provide wide area, very frequent (twice daily) coverage and at a much-improved spatial resolution of 250 m. The MODIS sensors also include better geolocation information and calibration procedures, which facilitate accurate water/land discrimination and efficient and accurate flood mapping. In regard to temporal sampling (revisit frequency), it is important to note that river discharge varies dramatically, sometimes through several orders of magnitude, over time periods of only a few days (Figure 10). Thus, an incorrectly timed image will not measure peak inundation. This constraint, present even within the MODIS imaging capabilities due to interference by cloud cover, is becoming less important each year as new satellites are launched, including cloud-penetrating SAR satellites, and general orbital imaging capabilities become more widely available and more economical. Instead of snapshots of the Earth, scattered widely in time, it is now possible to monitor land surface changes over large areas on a frequent basis. Because of the prior constraints, remote sensing has previously been mainly employed in a relatively ad hoc manner, usually as a response to very large flood events that require disaster assistance and persist long enough for the high spatial resolution sensors to be employed. Archives specifically designed to store these flood inundation mapping results have not been created up until recently. Satellite images were mainly only used for news stories, to guide relief efforts, or to demonstrate satellite capabilities, but the data and inundation outlines were not routinely published and otherwise preserved for risk assessment purposes. 33 Figure 10 Gauging station-based flood hydrograph for western U.S. showing variability of river discharge over daily time intervals. Discharge changed by an order of magnitude in less than one day and peak discharge lasted only a few hours. Larger rivers in lower gradient terrains exhibit somewhat slower rises to maximum and subsequent declines. Discharge, cubic feet per second 40,000 30,000 20,000 10, Jan 2002 Jan 2003 Jan 2004 Jan 2005 Jan 2006 Jan 2007 Jan 2008 Jan 2009 Provisional data subject to revision Median daily statistic (74 years) Discharge Source: Jennings, M.E., Thomas, Jr., W.O., Riggs, H.C.,

52 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Imaging of the extent of flooding during one particular event, although clearly valuable for identifying flood-prone land, does not directly provide information about probability of recurrence. In most cases, remote sensingbased flood mapping studies have failed to address this issue. With the advent of the MODIS sensors (in late 1999) and their associated data distribution system, remote sensing capabilities dramatically changed and it has become clear that new events can and should now be mapped within the context of previous events (see for example Figure 11). As noted, MODIS is adequate for mapping and monitoring through time many floods along medium to large rivers. As illustrated previously in Figure 9, floods can now be tracked economically, in map view, and on a near-daily basis and, using GIS and related digital technologies, the record of such flooding can be preserved for incorporation into wide area flood risk assessments. However, retrieving archival image data is itself a complex task. As shown in section A.3, many nations have now flown Earth imaging satellites, and data availability from such sensors ranges from non-existent for users outside of those planned, to available at significant expense per scene through commercial providers, to freely available at no cost to an international public (Box 8). Any such list is outdated soon after being compiled, as new sensors are now being launched on an almost weekly basis, and data distribution policies are very often in flux. These new launches do serve to emphasize that the Earth is currently being surveyed by a constellation of imaging satellites and that flood inundation can be directly observed and routinely mapped. Each sensor technology has advantages and disadvantages for mapping flood inundation. In addition to those factors already described, sensors that rely on visible light reflected from the surface (optical sensors) suffer from data gaps caused by cloud cover and nightfall. Even, as the case with MODIS, when frequent overpasses allow maximum use of temporary clear weather, Figure 11 Extent of flooding during the years derived from MODIS data (left) and cropland extent (right) in Southeast Asia. Previous floods Cultivated areas, irrigated agriculture and cropland Source: Dartmouth Flood Observatory s digital World Atlas of Flooded Lands. DFO

53 Agriculture and Rural Development Box 8 Data availability and cost Today, data availability is not an issue. Depending upon the system, spaceborne data are systematically or acquired upon request by the different space agencies, such as European Space Agency (ESA), National Aeronautics and Space Administration (NASA), Japan Aerospace Exploration Agency (JAXA), Canadian Space Agency (CAS), and Instituto Nacional de Pesquisas Espaciais (INPE) along with private companies and consortia. Images can be easily ordered through the Internet and they can be downloaded by file transfer protocol (FTP) or delivered on DVD by surface mail a few days after the acquisition. Ground receiving stations, can be purchased, hence enabling the near-real time product generation. The cost of such ground receiving stations depends mainly upon the area to be covered, however, for limited areas (less then km) the cost is in the order of Euro 300,000. In general, low and medium resolution data are freely available. Increasing the spatial resolution, the cost per scene increases approximately from 0.03 euros/sqkm (30 m resolution) to 14 euros/sqkm (1 m). In case that data are provided by public space agencies, the data are generally distributed free of charge or at reproduction cost. For commercial use, images must be purchased through a data provider. In this case the data cost is generally in the order of few hundred euros. Assuming that the sensor is owned by a private company/consortium, the data cost is often higher, but the cost for commercial data is generally decreasing as well in particular for archived data. For example, in a country such as Bangladesh (150,000 sqkm), in order to secure adequate monitoring for occurrence of flood events, weekly data are required. Here, the use of Wide Swath data (400 km) at a resolution of approximately 85 meters would be suitable for the provision of river conditions and inundated areas at country level. The data cost would be approximately 13,000 euros/year (250 euros/image for 52 weeks) corresponding to less than 0.1 euros/sqkm, if weekly data are required; 26,000 euros/year if data are provided every 3.5 days. If flood events occur, high resolution data (10 20 meters) should be additionally acquired. One hundred images corresponding to 25,000 euros (250 eurosimage 100 images) would be a reasonable data amount. The total cost, therefore, would be in the range of 38,000 euros corresponding to less than 0.25 euros/sqkm/year. sustained levels of cloud cover may constrain imaging a particular flood event. In contrast, the synthetic aperture radar (SAR) sensors return data both day and night, and through even heavy cloud cover (SAR sensors are in italics in the table A3 of the annex). Their data costs remain significant. An unlike AVHRR and MODIS sensors, wide area SAR sensors, in always on mode, do not yet exist. Whereas optical wide area sensors, such as MODIS (scanning a swath of 1600 km in one overpass) generally provide more frequent revisit frequency than narrow swath sensors such as ASTER (which scan 60 km in one swath), the temporal and spatial coverage advantage comes at the cost of spatial resolution (250 m for MODIS versus 15 m for ASTER). Depending on the terrain, 250 m may be too coarse for useful mapping of some floods, even though, when coupled with other information, such data might be used to establish that insurance-triggering flooding has occurred. 40

54 New strategies for flood risk mapping and flood index development Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture The objective monitoring and mapping of floods in support of insurance has two inter-related needs: (1) objective assignment of land parcels and crops into flood risk zones, and (2) consistent and accountable determination, as floods occur, of local excedence of flood index threshold values such that reliable prediction can be made of crop losses. Significant limits have been identified to the accuracy and precision of traditional, non-remote sensing, modelling-based approaches to the first need, especially when large geographic areas are to be included. The problems carry over as well to any precipitation- or river gauging station-based flood index for triggering insurance payments, there may be sufficient meteorological data to predict within certain limits what the flood discharge at a station will be, or a flood discharge may be known. However, it remains a separate and challenging assignment to predict what lands will thereby be inundated or what the level of actual crop damage will be. In developing countries there are several compounding issues. A developing nation water ministry may not have rapid access to dispersed in-country gauging station measurements. Large floods, in any case, quite commonly damage the measurement stations. Also, even when available internally, such data are often not accessible to non-domestic organizations attempting to assess flood damage. It is worthwhile exploring how much can be accomplished independently with remote sensing data and especially data that can be shared freely among nations and used to produce mutually agreeable and verifiable threshold index values. Developing flood damage indexes for agricultural insurance purposes will likely include methods to objectively estimate flood inundation extent, water depth, and flood duration, so that the flooded or non-flooded status of particular land parcels can be determined, and, ideally, the associated depths and durations. Other factors such as water sediment load and water velocity may also affect crop damage but are not discussed further here. In order to suggest paths towards operational remote sensing-based methodologies, two sets of applications are presented in more detail in the technical annex. The first application demonstrates a prototype global flood measurement and risk assessment system (River Watch) operating in pilot form at the Dartmouth Flood Observatory (see section A.4). A second set of applications are country and basin-specific and serve as examples of customdesigned systems that would be needed to effectively monitor floods in different flood regimes and geographic situations. The specific examples include application in Vietnam, Thailand, and Bangladesh (see section A.5). 41

55 Agriculture and Rural Development 4. Potential Applications of Index Insurance to Agricultural Flood Risks There are major drawbacks that characterize traditional MPCI programs as discussed in section 2. As a result, there are strong motivations for both private and public sectors to develop for agriculture parametric or index-based insurance solutions that demonstrate several advantages over MPCI. This section illustrates the possible application of the index approach specifically to an agricultural flood insurance program. It also highlights how the new remotesensing technologies discussed in section 3 could be harnessed to complement traditional approaches such as flood modelling as well as to support the design of such programs Principles of Parametric Insurance Parametric insurance does not indemnify based on actual loss measurement after a claim. Instead, parametric insurance ex ante agrees to make a payment upon the occurrence of a pre-specified triggering event. The triggering event is often a catastrophic natural event that may precipitate a loss or a series of losses. Parametric insurance is suited for low frequency but high intensity covariate losses. For insurers, parametric schemes reduce transaction costs involved in underwriting and administering insurance policies because there is less need for actual loss assessment for payment of claims and less uncertainty in setting premium rates. This administrative cost savings enable insurers to offer the product at less cost. Also, the insured farmers receive timely payouts and claim disputes are minimized. Weather index insurance is a form of parametric insurance. Index-based weather insurance products are contingent claims contracts for which payouts are determined by an objective weather parameter that is highly correlated with financial loss of the insured. For example, rainfall-indexed insurance is well-suited to agricultural production in regions where there are widespread crop losses due to drought or excess rainfall. Temperature and wind speed are other common weather indexes. For index insurance to provide effective protection to agriculture, a basic requirement is that the chosen weather index needs to be measurable, objective and representative of the predominant risk for the crop insured. Publicly available measures of weather variables generally satisfy these properties. To form a basis for insurance, the parameters used in the measurement of a weather index should reflect the state of one or more weather variables during the contract period, and types of parameter are often shaped by the needs and conventions of market participants. 34 New 42

56 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture innovations in technology, including the availability of low-cost weather monitoring stations and sophisticated satellite imagery as described in section 3, are likely to expand the number of areas and risks for which weather variables can be indexed Advantages and disadvantages Index-based insurance is less susceptible to some of the problems intrinsic in traditional MPCI. Apart from the reduction of administrative costs mentioned in the previous section, the product also offers other advantages. 35 First, information requirements for underwriting index insurance are simpler. Because index insurance payouts are not based on the measurement of actual losses incurred, there is no need to classify potential policyholders according to their risk exposure. This is usually a significant informational constraint in traditional agricultural insurance underwriting. In the case of index insurance, no household-level information is needed. For rainfall insurance, the risk assessment uses primarily historic rainfall data to evaluate the impact and frequency of insufficient rainfall. The simpler information requirement makes index insurance more feasible in many low-income countries with data constraints. Second, the objective and exogenous nature of the weather index reduces traditional problems of crop insurance, such as adverse selection and moral hazards. Adverse selection arises from information asymmetries. Because index insurance is based on information widely available to all parties, the product avoids adverse selection by reducing the opportunity that informational asymmetries can be exploited by the insured who have more knowledge about their own risk than the insurer. Index insurance also reduces moral hazards because the indemnity does not depend on the individual s actual losses but on an objective proxy. Therefore, the policyholder cannot influence the likelihood of receiving a payment. Third, indexed products are also likely to facilitate risk transfer to the international markets. Experience suggests that international reinsurers are likely to reduce the portion of the premium charged for uncertainty ( loading ) when the insurance is based on independently measurable weather events. In recent years, more and more global reinsurers have become involved in supporting the development of weather index insurance in lower income countries. However, weather index insurance also has several limitations. The key limitation of index insurance is basis risk, which can cause the mismatch between index payout and the insured s actual loss. Basis risk can affect both parties there can be either no payment when a significant loss occurs or payment when no loss occurs. Basis risk is likely to be lowest when weather index insurance covers catastrophic risk because this level of risk tends to be highly correlated. Basis risk will be high in areas with microclimates where the weather risk is not correlated, making index insurance inappropriate in this environment. In many cases, however, basis risk can be minimized by the installing of more weather stations. 43

57 Agriculture and Rural Development Second, unlike MPCI, weather index insurance generally only provides protection from one peril represented by the chosen index. If another uninsured risk results in a loss, the insured will not receive a payment. Because of single-peril protection and basis risk, weather insurance has not yet gained strong interest in higher income countries where it competes with government-subsidized MPCI. There will also be challenges in explaining both the single-peril protection and basis risk to clients in lower-income countries. Third, weather index insurance requires adequate weather measurement infrastructure. Even though weather data tend to be more available than yield data, many lower income countries have poor records of historical weather data. Some lower income countries have very few weather stations, and building the needed infrastructure to decrease basis risk can be expensive and take time. Finally, product options for different weather risks are still limited. The majority of weather index insurance products have been designed for rainfall risk; however, deficit or excessive rainfall is not the largest weather risk in many areas. Experience insuring other weather risks with new indexes is needed. In some regions, farm losses often result from a complex interaction of perils for example, increased heat that leads to pest problems that make weather index insurance less suitable. Despite these limitations, weather index insurance is often the best option for insurers wishing to offer a form of weather insurance in lower-income countries because it has the potential to address correlated risk affordably and is operationally less challenging. Box 9 and Table 4 reviews recent weather index insurance initiatives in low-income countries Elements of Parametric Flood Insurance Design Extending the index approach from drought risk to flood risk requires concept adaptation as well as expansion into new technical frontiers. While drought indexes can be constructed with rainfall data using relatively established methods, the complicated nature of flood requires using a combination of data sources and methodologies, including river gauge and rainfall data, flood modelling, agro-meteorological modelling, satellite remote sensing, and other geo-information technology in order to design flood indexes that accurately proxy crop losses. It is important to emphasize that the type of flood strongly impacts the feasibility of flood index insurance. Current experience suggests that the index approach is most applicable to inundation flooding affecting large geographic area. For example, an inundation flood in a relatively flat plain surrounding a river delta is more amenable to indexing than a high-velocity flash flood in mountainous valleys. The following section describes key elements and considerations in applying the index approach to agricultural flood insurance. Specifically, there are four main design components of parametric flood insurance each of which requires the innovative application of concepts and technology. 44

58 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Box 9 Examples of index insurance in developing countries Agricultural index insurance products have been implemented in many developing countries. Index insurance pilots have increased in prevalence since These pilots demonstrate how weather index insurance is flexible and how products can be designed for different users, i.e., households, firms, governments, and international organizations. Most index insurance experience to date is mainly for protection of drought risk for micro-level users such as farmers. There is limited experience for meso- and macrolevel aggregate risk transfer, though interest in this level of weather risk instruments financing has been increasing and more pilots implemented. Table 4 Summary of weather index insurance pilot projects since Program Countries level Description (year started) Micro Weather-indexed insurance India (2003), Nicaragua (2006), for smallholder farmers, Honduras (2006), intermediated through Guatemala (2006), Malawi institutions with rural (2005), Ukraine (2003), outreach Thailand (2006) Meso Weather-indexed portfolio India (2004) hedge for rural financial Vietnam (under development) institutions that lend Peru (under development) to farmers Macro Weather insurance or Mexico (2003) weather-indexed Ethiopia (2006) contingent financing for Malawi (2008) governments or The Caribbean Catastrophe international organizations Risk Insurance Facility (2007) Source: See, for example, Manuamorn (2007); Skees, Hartell, and Murphy et al. (2007); Mahul and Cummins (2009) Defining flood-induced crop loss An index insurance scheme quantifies loss by measuring an objective parameter approximating loss. While a single index such as rainfall can well represent an event like drought, in more complicated cases a composite index of multiple parameters (such as rainfall plus temperature) might be used in an attempt to capture the loss more accurately. Unlike rainfall or temperature risks, floods are more complex and a single parameter is not sufficient to fully describe the event. For the purposes of flood index insurance, it is necessary to define a flood event as a combination of various measureable parameters such as extent, peak flows, duration of discharge, volume of discharge, depth of inundation, etc. While aspects of flood that cause loss are relatively well understood in property insurance, it is very important to recognize that defining flood-induced loss for crops is more complicated. For agriculture, the timing of flood in 45

59 Agriculture and Rural Development relation to stage of crop growth primarily determines the extent of loss (Figure 12). As a crop grows, critical thresholds for when flood results in damages for that particular crop also change. Different loss thresholds also exist among different crops. For example, flood depth is likely to cause more damage to rice varieties that do not elongate with water that ones that do. For insurance, one or a combination of the previously mentioned flood parameters could be chosen to form a proxy for crop loss. Once the key parameters are identified, a flood index insurance policy would also need to define the level of index which triggers payout (for example, flood depth of above 50 cm and/or flood duration of more than 5 days, etc.) and the method by which the index is measured. The payout can be triggered by river gauge Figure 12 Example of flood loss criteria for rice according to different stages of crop growth. The example crop is the white jasmine rice 501 variety grown in the Pasak River Basin in Petchaboon, Thailand. Growth stage of White Jasmine Rice cm 160 cm 110 cm 70 cm Sowing Trans planting 50 cm 25 cm Growth Stage Rice height (cm) June Seeding 0 25 July Transplant Aug Tillering Sep Booting Oct Flowering Nov Reproductive (Grain Filling) 160 Dec Harvesting day 160 Critical water depth (cm) /20* Critical flooding duration (days) >3 >3 >4 >4 >4 >4 >4 Source: Expert based assessment of critical flood parameters identified in the context of a flood index insurance feasibility study project in Petchaboon, Thailand (*The critical water depth during the first two weeks of September is 70 cm while the last two weeks is 20 cm, respectively). See ASDECON (2008). 46

60 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture and/or remote sensing measurements. By definition, payout from such a flood index insurance policy will be based on an agreed value system as discussed in section 2. The basis for the agreed value payout is the value of crop, either in terms of total inputs invested or potential income, at the time that the defined flood event occurs. While it is theoretically possible to design a flood index that combines many aspects of a flood event, such a composite index would also be more complicated to construct and administer. A balance needs to be struck between the technical complexity of the flood index and its practical application for insurance Modelling flood hazard The objective of hazard modelling is to understand the spatial patterns and occurrence intervals of flooding. Before performing flood hazard modelling for insurance product design, it is therefore important to define the key flood parameters, as discussed in the previous section, to be modelled. In addition, the model requires additional quantitative and qualitative information to be gathered to calibrate the chosen flood model. This includes the boundary of geographical areas to be insured, the time period within the production season which is exposed to floods, and so on. In many cases, farmers are also interviewed about their past flood experience. They are often asked to recollect the most catastrophic flood years in memory. Such information is useful for the purpose of setting some benchmark years in the flood model or to validate the model s output. Different modelling approaches can be employed to estimate flood frequency and extent (A.2). The model output forms a basis for defining homogenous flood risk zones for premium pricing. Operationally, insured farmers can be grouped per flood risk zone. A GIS database can be created that contains information on the number and location of insured farmers within each zone. When a flood event occurs, farmers in the same zone will be treated as having been damaged homogenously. As a result, flood index insurance payout can be made on a zonal basis, as in the case of rainfall index payout per weather station Designing the flood index Once the extent and probability of flooding is determined, a quantitative relationship between the characteristics of the flood and the resulting crop damage has to be established. This can be done either through modelling by employing a crop growth model that captures the impact of inundation on crop development, or empirically by using historical field observations or expert assessment (e.g., from agricultural extension services or agronomic research centers). The same principles of index design developed for agricultural rainfall index insurance can be used for flood index design, although these have not yet been able to be applied in practice as part of this study. A flood index has to define a triggering event, payout thresholds, incremental payout scales, and payout limits. The insured values can be set based on available economic data, risk 47

61 Agriculture and Rural Development Table 5 Example of a flood index insurance structure with total production cost as sum insured. Days of inundation of 60 cm. flood Yield damage Insurance payout 3 days No damage No payout 4 days 20% loss 20% of total production cost 5 days 60% loss 60% of total production cost 6 days 80% loss 80% of total production cost 7 days 100% loss 100% of total production cost management needs of the insured and affordability of the insurance. Further, the index design needs to be based on the level of aggregation of the flood index insurance scheme being considered: micro (farm-level clients); meso (intermediate level policyholders, for example, agricultural banks or processors); or macro (government level). Micro level: The majority of experience of weather-based index insurance is at the micro level. A conclusion of this study is that micro-level flood insurance is challenging, but in future cases that micro level, individual farmer flood index insurance is determined to be feasible, a simplified scale based on duration of flooding is one option (Table 5). More simplified alternatives could be foreseen in which payouts are binary (all or nothing), based on measurement parameters such as length or depth of inundation. Furthermore, the area over which such measurements are to be made must be defined, based on the level of aggregation (section 4.3) determined for client enrolment and loss measurement. Meso and macro levels: An index payout scale based on river gauge measurement, reflecting river discharge and associated expected flooding, can be considered. An example of an index for a payout at the meso level is one that is designed (but not yet implemented) for a rice-producing area of the Mekong Delta in Vietnam (section 5.1). Development of an index based on actual stream gauge levels provides an existing measurement against which to develop the index, without employing additional technology (such as remote sensing or in-field depth measurement markers). A payout scale based on river gauge measurements could also be graduated (as in Table 5) or binary. At an even higher level of aggregation, macro-level payments for a disaster compensation system, operated by government, could be based on measurement of the extent (area) of flooded land, within defined geographical areas, and within defined time period and duration, from remote sensing. A government could consider purchasing an insurance contract based on such flood index to finance the extreme layer of risk of its disaster compensation fund. A precedent exists in Mexico where the federal government insures its budget against national drought risk with macro-level rainfall-indexed insurance. Even in cases where insurance is not involved, aggregate flood 48

62 indexes could also serve as a criterion in paying out flood compensation to the public, thus improving both the speed and objectivity of the system. It should be noted that flood indexes can form a basis of other financial instruments apart from insurance. A few property catastrophe flood bonds that used the flood index approach have been placed, allowing insurers to protect against major flood damages. One example was the Blue Wings program placed by Swiss Re for Allianz, an insurer. The flood element of the bond was to be triggered by excedence of flood gauge levels on 50 specified reference locations on rivers in the U.K., with a payout scale devised based on flood modelling carried out by modelling company Risk Management Solutions (RMS). Payouts required that more than four river systems were at the status of severe flood warning by the U.K. Environment Agency. Flood depth was to be verified against buildings by an independent firm of engineers Insurance operational system To implement a flood insurance scheme, an operational system must be created with appropriate support technology. The operational system can be hosted by the insurer if there is adequate technical capacity, or by an external public or private technical service provider. The system can be designed to serve many critical functions before, during, and after the insured period. For example, based on regularly captured remote sensing imageries, the system can monitor the stage of crop growth and detect an onset of the flood event. The satellite images can be continuously captured from sensors with high visit frequency and analyzed in relation to the GIS database of flood zones to determine whether a flood event of payout-triggering duration has occurred. While the insurance payout rules would have been defined ex ante for an index product, using remote sensing in this manner provides the additional benefit of objective verification of flood occurrence and insurance payout determination. This can be done in a cost-effective manner, provided that an appropriate data acquisition arrangement between a remote sensing agency and the implanting institutions are in place. In addition, an operational system of this nature has broader applications beyond the context of insurance. For example, the system could provide early warning before disasters, support objective public disaster payments outside formal insurance, and provide information for food security monitoring. Indeed, there is a general public-good rationale for governments to invest in such an information system regardless of its linkage to insurance programs Scale Options of Flood Index Insurance Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture In principle, flood index insurance for agriculture can be developed for different levels of risk transfer. In a micro level program, policyholders are individual farmers. In a meso-level program, the policyholder is an organization with direct or indirect financial exposure to an aggregated flood risk. In the commercial sector, the most obvious examples of potential users of 49

63 Agriculture and Rural Development index insurance are agricultural credit banks, microfinance institutions (MFIs), or processors in the supply chain, working with clients in flood-prone areas. At the macro-level, the budget of a local government agency responsible for compensation of farmers, or an international organization responsible for flood relief, could be insured in the event of flood. In practice, the feasibility of implementing the scheme at each level depends on a variety of factors such as the specific characteristics of each flood plain; the availability, quality, and resolution of data; the level at which demand for insurance is expressed or aggregated; and legal and regulatory requirements for insurance in each country. A micro-level product would identify flooded areas at high resolution and, in the case of index insurance, reduce basis risk. Given the general lack of data at fine resolutions in developing countries, there is still a significant technical challenge in designing a flood index insurance product at a farmer level in the short run. If such a micro-level scheme is attempted, there may be significant basis risk between a flood index derived from coarse resolution data (such as flood modelling) and actual local damage. The likely basis risk will make the index unsuitable proxy for farmer s risk. One the other hand, meso- and macro-level flood indexes aim to capture the catastrophic risk at an aggregate level. In these cases, basis risk becomes smaller as the index is likely to capture such a widespread event. However, implementing meso-level schemes will require a risk aggregator, which becomes the insured party. The aggregator also needs to set rules for application and distribution of claim payouts if the benefits are meant be shared. There is perhaps a higher potential for such schemes to be developed within the current data and technological constraints. The pilot project in Vietnam provides an example of a creative design of flood index insurance scheme at the meso-level for an agricultural bank using river gauge data (section 5) Institutional Considerations In developing countries, introduction of a flood insurance product would require the mobilization of stakeholders and some from of support from donors. A pilot arrangement can be considered as it usually allows a phased approach to organizational capacity building. The phases are usually composed of a feasibility study, product design and training, and pilot implementation Organizational arrangements In recent years, organizational structures have been developed for many pilots on drought index insurance. 36 A similar structure can be foreseen for a flood index insurance project. But given the more complicated nature of risk, a more important role of technical partners and supporting services is envisioned (Figure 13). This organizational structure for micro-level flood index insurance anticipates that the policyholders are farmers. A suitable rural service organization could 50

64 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 13 Example of organizational structure for a flood index insurance program (micro level). Organizational structure for micro level flood insurance Stakeholder steering committee National flood agency Remote sensing agency Reinsurers Insurer(s) Technical support unit External technical assistance Distributor e.g. MFI, farmer co-operative Extension and training for farmers Farmers in defined flood risk zone Farmers in defined flood risk zone Farmers in defined flood risk zone serve as the linkage between the insurer and these farmers. With existing outreach, the organization could sell the insurance as agent of the insurer, thus reducing the operational cost of reaching the famers. To develop the project, the insurer would form a technical support unit responsible for the flood insurance program, which would coordinate with all technical and extension agencies as required. Marketing campaigns and sales of insurance to farmers could be conducted by the rural organization partner under the control of the insurer. Informing farmers about flood insurance also offers the opportunity to provide advice to farmers on adjusting their practices to avoid or mitigate flood impact. Different resolution levels of flood insurance scheme imply different contractual obligations for an insurer as well as different organizational arrangements for implementation (Figure 14). For meso- or macro-level insurance, the insurer s contractual obligations end with the payment of a claim to the aggregator, and there are no further operational requirements on the insurer s side. Payouts can be triggered by broader (less high resolution) measurement of flood principally river gauge measurement. For micro-level insurance, the insurer is responsible both for organizing loss measurement and for claims payment down to the farmer level. The insurer needs to have the capability to deliver insurance at the local level. The process involves handling farmer enrollment, premium collection, loss assessment, and index payment settlement. Though many tasks can be handled by the agent organization, this still implies significant capacity needed on the part of the insurer. 51

65 Agriculture and Rural Development Figure 14 Key contractual and operational differences between a micro- and macro-level flood index insurance program. Micro flood index program Meso/macro flood index program Insurer Insurer Policies, premiums, claims Policies, premiums, claims Distributor Policyholder is Aggregator Policies, premiums, claims Aggregator sets the payout rules Policyholder is Farmer Farmers Local institutional capacity Apart from the insurer, the development of a flood insurance program needs support of the following key domestic institutions. Coordinating inputs from these external expert organizations could be a key task of the technical support unit formed by the insurance company. 1. National flood management agencies Typically there are one or more ministries responsible for flood-related activities such as forecasting and warning, flood prevention, flood mapping, flood emergency response, and, more broadly, river basin management. Overlapping responsibilities usually exist in water resource management, such as irrigation for agricultural purposes and water supply for industrial and urban use. These interrelationships between various agencies are well recognized as an integrated approach to flood management is required. 37 For development of flood index insurance, the most important national capacity requirement is flood modelling. Even in developing countries, flood modelling is extensively used by many agencies. National capacity in this area frequently exists. It is the application of flood modelling to support extensive flood mapping of agricultural areas, especially at a high resolution, that is less common. Broad categories of flood-prone areas may be zoned, but not at a level of resolution that would be required to provide sufficient information to support decisions to set premium rates by insurers at a highly localized level in a micro-level scheme. Further constraints may be related to available input data for flood models. Information on historical rainfall or river discharge data or digital elevation models is also generally lacking in developing countries. 52

66 Development of flood insurance needs to build on the existing flood modelling capacity. Investment in improved hydrological data collection needs to be made. Once this is done, the work would require significant analytical work in the development phase, in order to establish zoning and support premium rating. Furthermore, existing national institutions engaged in this process should be supported by or linked to technical capacity in the international private sector. This would enhance their access to better flood modelling software, established methodology in premium calculations, and needed consulting services. 2. Remote sensing agencies Regional or national satellite operating agencies (e.g., ESA, JAXA) provide image capture and distribution services. These agencies have entered into many collaborative arrangements with national agencies responsible for remote sensing applications in developing countries. In addition, there is a proliferation of private sector, international providers in application design, training, and outsourcing. The cost of image purchase is decreasing continuously, as more providers supply the market. Independent advice on access to the international market would allow national agencies to benefit from this trend. Considerable capacity building is required for national remote sensing agencies to design a service, specifically for flood detection, which could be provided to insurers that offer flood insurance. Capable and reliable service from these agencies is a key consideration in determining which flood insurance product is feasible. Service definition and appropriate contractual relationship between an insurer and a remote-sending provider needs to be structured before the start of a flood index insurance program that contains a remote sensing-enabled component. 3. Public-private sector partnerships Challenges in developing a flood insurance program imply that involvement from both the public and private sectors is needed. Insurance is best operated in the private sector, where sound insurance principles and needed technical capacity exists. However, the major institutional support of existing government organizations, for example, in providing data and conducting flood modelling are also instrumental. In developing countries, developing flood insurance for farmers will likely benefit from public-private partnerships with clear organizational agreements between the parties, specifying the roles, responsibilities, and reporting channels. Government or donor support, for example, in funding upfront research and development costs, will be important in the initial phases of program design Underwriting Considerations Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture The following section discusses major underwriting considerations for a micro-level flood index insurance for agriculture. Many considerations also apply to meso- and macro-level programs. 53

67 Agriculture and Rural Development Flood index insurance product The flood insurance product features must be described in the insurance policy or insurance certificate, which is a simplified evidence of insurance given to farmers after purchase of the policy. All terms and conditions of the index would be written in the policy document. Key terms to be explained are similar to what would be found in a drought index insurance policy, which include the following: Policy period refers to the time window within which a flood event is insured. Sum insured refers to the value insured, which is normally on a per unit area basis. The sum insured is an agreed value between the insured and the insurer. The most common benchmark of sum insured for agricultural index insurance is production costs per hectare. (For a multi-phase contract, the sum insured may be incremental per phase, reflecting the higher accumulated cost of production in later stages of crop growth). Index payout criteria refer to the pre-agreed definition of flood event, defined according to onset and duration or other parameters of flooding, which would result in payout. Defining the payout rules also includes defining the trigger at which payments start to be made, incremental payment scale (also known as tick size ), and maximum payment limit. Flood loss assessment refers to the methods that will be used to measure flood onset and duration, for example, using remote sensing. Additional triggers refer to any other trigger events that may be set in the policy that must be met to before policy is activated. This might or might not be included in a micro-level product as these triggers are most likely to apply to a macro product: for example, river gauge excedence or rainfall above a given threshold within a specified time Zoning and client enrollment A highly challenging underwriting feature of any type of flood insurance is potential for adverse selection. An insurer must have considered this challenge and, if feasible, found a clear strategy to overcome it through the insurance program design. The main methods are: (i) Establishing homogenous risk zones: The building block for both farmer enrollment and loss assessment, is the zone into which the farmer is allocated at the time of acceptance for insurance. As discussed in an earlier section on the conceptual underpinning of flood index insurance product, it is considered that grouping of farmers into zones of homogenous flood risk is an essential element of design. Floodplains are relatively predictable in flood pattern and duration, so zoning should also be able to predict flood patterns. For unmanaged flood plains, flood models will produce estimated flood recurrence frequency maps, related to topography. For managed river floodplains with river draining, flooded areas delineated will be through planned sequential flood (into detention areas) or by 54

68 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 15 Illustration of gridded flood zones for rice production in the Muang District of Petchaboon, Thailand. Floodplain zoning Medium risk pricing zone etc High risk pricing zone River Low risk pricing zone Grid for enrolment and flood measurement Key issue: grid resolution? overtopping/breakage of bunds. Each type of flood plain requires its own zoning approach for insurance purposes. One option is to create a gridded basis of zoning, where each grid square is considered a homogenous zone. Options of different grid resolution give good opportunity to match factors such as known homogeneity of local flood risk, farm size, and number of farmers captured in specific grid cells. Further, such grid cells could be the basis for remote sensingbased flood loss measurement (Figure 15). This gridded approach gives an advantage that no subjective decisions are needed in deriving boundaries between one flood risk zone and the next. A disadvantage is that gridded boundaries are not marked by physical features on the ground. Alternatives to the gridded approach to zoning would be to demarcate flood risk areas based on the output of flood modelling, which would indicate areas of expected flooding frequency. Boundaries of flood risk frequency are based on models, and the resolution (scale) of interpretation depends on the resolution of the flood model that is used. Allocation of flood risk zones for insurance participation, based on flood risk zones determined from such models, has not been tested. It should be noted that most current farmer zoning systems (e.g., for an agricultural bank) are based on administrative boundaries. Allocation based on a different set of risk criteria, especially where they are produced from models that imply insurance premium rates per zone, would be problematic. 55

69 Agriculture and Rural Development (ii) All farmers falling within a particular homogenous risk zone would pay the same rate of premium. In micro-level flood insurance, establishment of groups of farmers in homogenous areas goes part way to reducing highly localized adverse selection, where there may be very localized variations in flood risk. After grouping, the extent of potential adverse selection remaining depends on how well premium rates reflect the actual flooding risk. In any case, each floodplain area would have been studied during research into product design and premium rating during the preparation phase. The type and characteristics of flooding to which the floodplain was exposed would have been established and the degree of homogeneity of the flood risk determined. Voluntary or automatic (compulsory) insurance: Adverse selection would be reduced in compulsory insurance schemes compared to voluntary schemes. Automatic (compulsory) insurance requires agreement that there are benefits as a matter of public policy. A measure adopted for property insurance in several countries has been to add flood insurance to existing insurance policies for fire and allied perils. Premium differentiation is still needed, but a large premium pool is created by this method. (iii) Underwriting rules for farmer enrollment into insurance scheme: Underwriting rules are needed for flood insurance, as in any class of insurance. These may include: Establishing the eligible zones for insurance. Defining uninsurable risks. Farmers in risk zones considered as carrying unacceptably high risk may need to be excluded from insurance. A guideline for the threshold, above which a risk is likely to be considered uninsurable, is once every seven years. 38 Maximum limits of sum insured would be established in any one risk zone and in all for the whole program. These would be based on consideration of risk accumulation and probable maximum loss (PML), expressed according to the expected return frequency of the event. Risk acceptance period (policy sales period) needs to be well before farmers had a possibility of predicting if it was likely to be a season with increased probability of flood. If not, farmers will buy insurance only in seasons of high flood risk, and not in seasons of expected low risk. If this is the case, on average, the insurer s long-term premium rate (based on equal sales in all years) will be distorted. Actual enrollment process would be simplified (as with drought index insurance). The key attributes required are location of the farmer (defining in which homogenous risk zone the farmer was located) and other simple criteria, such as area of the crop (e.g., rice) due to be planted and units of insurance purchased (such as coverage for five hectares). GIS location of the farmer would be recorded. Note that for small farmer communities, a point source coordinate should be sufficient for the insurer to locate the farmer, as vector field boundaries would be major additional work in establishing a geo-referenced client database. 56

70 Loss assessment Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture The cornerstone of any crop insurance program, around which the entire program needs to be built, is a practical, objective, and fair basis of loss assessment. Flood measurement under a micro-level program can be achieved by two main methods. Remote sensing: detection of areas flooded and duration. Field measurement: detection of flooded areas by ground-based survey. Remote sensing is foreseen as the most rapid and most objective method of detecting flood and a cornerstone of flood index insurance. Issues that are highly relevant to its application are frequency of image capture related to resolution and types of sensor, as well as costs (section 3). It would be a clear advantage to insurers to measure flood measurement and make payments solely using remote sensing. The reality is, however, that some ground-verification of results shown by remote sensing is required. This issue is most significant where a complex index is designed for example, incremental payouts based on duration of flood; and where minor flood events are included. In a catastrophic flood event, it is obviously easier to identify flood as large areas are flooded simultaneously. In reality, all floods have a margin at the edge of the flooded area, where neighboring farmers will be either flooded or not flooded. For example, on a major flood, 90 percent of all homogenous risk zones might be very clearly flooded and the policy triggered; but 10 percent might contain some part of the zone that was not flooded (or only flooded for a short period). This is a form of basis risk. An option would be to simply declare that all farmers in any zone affected would receive payment according to the worst affected part of that zone. This emphasizes that the determination of the actual area covered by each homogenous zone should be small enough to minimize such difficulties associated with the basis risk. Nevertheless, it is recognized that there are likely to be existing systems for recording flood, which can be employed, in the absence of remote sensing, in the context of flood insurance in some capacity. As has already been noted, maintaining control over the human element in these existing systems of measurement, as experienced in traditional crop insurance programs, can prove highly challenging. The introduction of flood measurement using remote sensing can add a substantial element of objectivity to improve the ground-based processes. However, the novelty of this technology will also require extensive education of farmers for acceptance. Therefore, it will be important to still use ground-truthing to validate the correctness of satellitebased measurements, thereby promoting confidence in the application of remote sensing technology. For a macro-level index insurance, payout to the policyholder (holding aggregate risk) is on the basis of triggering the index specified in the policy. There are two options for the measurement of flood to trigger a macro-level index. The first would be to measure flood using remote sensing, in order to determine the area (and possibly duration) of flooding in pre-defined areas. 57

71 Agriculture and Rural Development The second would be to use river discharge measurements, made at nominated river gauging station(s). Payouts would be based on a pre-agreed scale, where the extent of flood measured (by remote sensing) or level of flood water (measured by river gauges), would be based on an ex ante assessment of the expected loss related to area damaged or flood gauge levels Financial Considerations Loss modelling requirements An insurer s financial concerns start with an understanding of (1) expected average loss of a given portfolio; (2) the volatility of expected losses each year; and (3) the confidence of the expected loss projections. These three factors underlie several important criteria of the premium rate required and the capital to be employed by the insurance company and/or the reinsurance required. Loss modelling is the starting point, therefore, in financial management for the insurer. Insurers will not have the option of market statistics for flood insurance in a new program. They must rely on whatever information is available, of which the most important is flood modelling. In addition, the insurer will wish to access any other possible data on past flood losses that were recorded officially and routinely. In the case of very large flood events, there may have been special studies or other press information on the events that could help the insurer in projecting loss. For insurers and reinsurers, the only available primary historical data available for use as inputs of flood modelling are often river discharge data and rainfall data. The insurer is reliant on specialists to interpret the output of flood modelling to project the expected return period of flooding and other characteristics, such as frequency related to date and duration or depth of flooding. If flood has been manually recorded on a local basis for other reasons, such data may also be available. An observation should be made regarding the use of primary data in macroand micro-level programs, and how it could affect the pricing characteristics of the insurance program. In a macro index scheme, the same primary data (i.e., river level or discharge measured at river gauges) can be used to both set premium rates and to trigger the index payout (through measurement of the river level or discharge during the flood season). Therefore, such a macro-level scheme can be described as a true primary index insurance program due to the dual use of primary data. For the insurer, making both premium rates and payout on the same index is very transparent. It also provides more comfort as the payouts in the current season are made based on the same data series as the historical data used for premium rating. This adds a higher degree of certainty that the premium rates would cover the risk being insured. However, for a micro index scheme, payouts will be based on a different measurement from that used to calculate the premium. This is because while the premium is estimated using flood modelling, the payout for a micro scheme will be triggered by loss measurements during a flood event using 58

72 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture remote sensing. This discrepancy makes a micro level scheme not a true primary index insurance. As a result, there is also more uncertainty in pricing, as the index payout is not based on a continuous series of primary data measurements. This type of uncertainty is routinely handled in the underwriting of most classes of insurance by the commercial market (who use modelling and interpretation of multiple data sources to derive premiums) and would need to be handled by the insurer working on micro-level flood index insurance. Research and development activities have shown that loss modelling is a very complex task, especially when applied to design of micro-level flood insurance schemes. It cannot provide high-resolution data for premium development at local levels, although it can provide broad indications. In particular, added difficulties in relation to modelling agricultural risk arise from the fact that timing and seasonality of crops is critical. Flood modelling that needs to include timing of flood risk (frequency of recurrence in given time windows) introduces a new level of technical complexity Developing premium pricing When establishing a price for an index insurance contract, an insurer will take into consideration their own risk appetite, business imperatives, cost of risk, and operational costs. While there are a variety of methodologies for rate setting, in general the pricing for all contracts will contain an element of expected loss, plus some loading or risk margin, as well as administrative costs. Therefore, in general the premium charge for a contract can be broken down as follows: Premium Expected Loss Risk Margin Administrative Costs Expected loss is the average payout of the contract in any given season. In some years, payouts in excess of the expected loss can occur and the insurer must be compensated through the risk margin for this uncertainty and for the internal provisions that must be made in order to honor these potentially large payouts. For flood index insurance, the values of the expected loss and the risk margin must be established from historical river gauge data or output of flood risk models. Box 10 describes how expected loss could be calculated. The approach for determining the risk loading over the expected loss differs from risk taker to risk taker, and many use a combination of methods to determine the risk margin included. A sensible pricing methodology uses a risk measure such as the Value-at-Risk (VaR) of the contract to determine the risk margin. A VaR calculation is aimed at determining the loss that will not be exceeded at some specified level of confidence, often set at 99 percent. It targets the potential extreme negative deviations in payouts for the risk taker and therefore the associated capital charge for taking on the risk. Finally, administrative costs are essentially the costs necessary to run the business, including charges for data, office costs, taxes and brokerage or intermediary charges if necessary, and loss adjustment costs. 59

73 Agriculture and Rural Development Box 10 Estimation of expected loss for macro- and micro-level flood insurance program In the case of flood insurance, expected loss has to be estimated. As noted, this calculation is far more problematic for micro- than macro-level flood insurance. Macro level estimated loss Data source: river gauge historical daily discharge volume data for each station Payout criteria: based on river gauge trigger level and exit, period of insurance cover, and duration (number of days) above trigger level. Incremental payout levels can be considered based on increased heights above the trigger. Estimated loss cost: the average of historical payouts that would have been made over multiple years. An actuary will estimate the recurrence of extreme events that might not have been included in the data set to determine a risk margin. Micro level estimated loss Data source: output of flood risk models, spatially presented in flood risk maps. Flood analysis will need to be specified to model expected frequency of flooding in specific time windows. River gauge historical daily discharge volume data for each station can be used to validate and check recurrence frequency for major loss events. Payout criteria: based on flood recurrence frequency for farmers enrolled into specific homogenous area zones. Payout could be total loss only once a flood is measured, or on a graduated and agreed scale, according to the duration of flood, and the growth stage of the plant (also fixed in the policy by agreed dates for each crop growth phase). Estimated loss cost: estimated as the average of historical payouts that would have been made over multiple years. Flood modelling output, especially at low resolution, will not normally provide the necessary information to allow completely objective assessment of estimated loss costs, as there are multiple variables of flood timing and duration. However, flood mapping is capable of assigning farmers into zones, where recurrence frequency (flood events per annum) can be a starting point Risk transfer: Insurance and reinsurance Insurers starting a new program for flood will be faced by an unbalanced portfolio and catastrophe risk exposure if multiple insureds in floodplains are affected by specific events in early years. Studies of major flood events globally have confirmed the catastrophic nature of floods. Therefore, flood insurance is likely characterized by years of zero claims or years of multiple claims during a major event. An insurer operating nationally, especially initially, will have limited opportunity to develop a spread of risk, for example, between different regions of a country. In contrast, an international reinsurer is well placed to build a book of catastrophe programs from different countries and achieve a balanced portfolio. Hence, reinsurance is a very important element of financial management for a domestic insurer offering flood insurance. Flood insurance programs are reinsured in the same way as other catastrophe risks. A review of all reinsurance structures is not part of this report. However, 60

74 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture comment is provided here on the appetite by reinsurance market for flood risk, as well as some key points differentiating between programs which are or are not reinsurable. Factors in a reinsurer s assessment whether to support specific flood insurance programs are likely to include: product technical details: the original policy, loss assessment methods, and resources; risk assessment of proposed insured farmers and marketing plan to achieve portfolio (and pilot introduction, if applicable); assessment of the underwriting plan and proposed portfolio; basis of premium calculation for the original program, degree of uncertainty, modelling of expected payouts over time, and maximum probable loss; basis of reinsurance terms and conditions; assessment of commercial opportunity (including expansion potential, if pilot scale is uneconomic in initial years); existing commercial relationship in other lines of business and other business potential (strategic opportunities of client relationship. There is little track record to assess reinsurers appetite for flood risk insurance. From discussions held with market actors during the course of this study, it is clear that reinsurers are very interested in expanding their business in developing countries, and that they see index insurance as an important development. However, it is also clear that the complexities of flood insurance, including the difficulties for insurers noted in this report and particularly in developing countries with poor data, are a barrier. The objectivity of using remote sensing for loss adjustment is highly attractive to insurers, including if only in providing a verification of in-field loss assessment. In summary, reinsurers like the fact that flood indexes based on river gauge data avoid field loss assessment, even if river gauge data is subject to many river management factors. They also like the objectivity of remote sensing. Actual reinsurance pricing is affected by the probable maximum loss (PML), as this dictates the amount of capital that should be allocated by the reinsurer. As with other drought index (and traditional crop) insurance programs, spread of risk and portfolio building, thus reducing the ratio between premium volume and PML, has the effect of lowering reinsurance costs. Reinsurers have confirmed an interest in the present study, and in the potential for supporting flood risk reinsurance. 39 Reinsurers have indicated a willingness to support macro-level risk transfer, based on river flow indexes. The integrity of the index of river flow would need to be investigated in each case, in particular to assess the significance of any watershed management changes over time. While river gauge data are a primary index used in macro flood index insurance (and gauge data are used as an input for flood modelling in a micro flood index insurance), the river flow is subject to 61

75 Agriculture and Rural Development numerous variables, not least river draining and dam construction. Rainfall data and gauge flow data at several places in a river basin are normally available, and modelling work is either available for other reasons such as flood warning or is likely to be undertaken as part of the development of a flood insurance system. Hence, reinsurers have indexable data and information with which to interpret trends and changes. Furthermore, because loss assessment in field is not part of the payout structure of such a reinsurance, there is reduced requirement for due diligence by the reinsurer of the operational capability of the insurance company. Reinsurers are also interested in micro-level flood risk insurance products, but also are in agreement with challenges and limitations in assessing local level risk and the underwriting issues described in an earlier section. The level of basis risk between risk zoning outputs from a flood model and the actual flood experience of a micro-level insurance scheme are subject to variables that are not straightforward to quantify. However, these variables are no different to those faced in many other types of insurance, where underwriting information available for premium rating and client allocation according to risk are often uncertain. A conclusion arising from this study is that, in the short run, catastrophe risk transfer could more easily be accepted by reinsurers using river gauge indexation. Macro-style flood policies could be sold to aggregators of risk and directly reinsured internationally. An insurer could still develop micro-level insurance programs and obtain index reinsurance. However, the insurer would likely have to retain all risks related to such micro-programs other than the macro-level risk for catastrophe events, which could transfer to reinsurers Financial viability of flood index insurance Several key questions need to be investigated in relation to the financial viability of flood index insurance. 1. Can the insurance program be financially sustainable for the insurer? In the absence of specific examples of agricultural flood insurance scheme, it is only possible to comment in generalities. Key factors for the insurer are: (a) whether premiums can be set at a sustainable level, including potential to adjust the premium based on experience; (b) reinsurance availability (particularly in the maintenance of relationship in the event of a major loss in early years); and (c) ability of the insurer to avoid adverse selection by farmers. Clearly, operational and technical viability is also a pre-requisite to financial viability. 2. Can farmers afford premiums, and are they willing to pay for flood insurance? A guideline of a maximum flood frequency of once every seven years, on average, is considered as being the highest risk area that could be insured. As a guideline for drought index insurance, maximum acceptable premium rates for farmers are considered as to be in the region of 10 percent of the sum insured (i.e., 10 percent of production costs, not 10 percent of revenue). Farmer-demand studies have been undertaken for flood risks in Bangladesh

76 The studies indicate a willingness to pay insurance premium for flood in a general sense. But in reality, very specific design criteria for flood insurance would be required in order for a farmer to make a binding assessment as to whether a product meets his or her needs. A related issue is whether any cross subsidization is allowed in premium rates, especially within a compulsory flood insurance scheme. In property insurance schemes where all property owners are obliged to insure (such as in the U.K.), this type of cross subsidization occurs. A difference between drought and flood index insurance adds another difficulty. With drought (rainfall deficit) insurance, triggers can be adjusted to an extent, as rainfall is a graduated risk, giving an option for the insurer to offer a range of premium rates according to drought severity coverage. This approach is less easy for flood insurance, as flooding tends to be all or nothing as far as the event occurrence is concerned, although graduated payout based on duration of flood can be considered. However, drought index insurance appears more flexible regarding trigger options than is likely to be the case for flood index. 3. What are developmental costs versus operational costs? Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture It is evident that flood index insurance requires significant technical mobilization, particularly for flood modelling and remote sensing. Developmental costs, even if existing organizations are available nationally, may be high. This stresses the importance of donor and/or government contribution to upfront technical costs in developing countries. Costs of assessing the feasibility and then designing a flood insurance product, particularly at a micro level, are high. Given the limited experience of existing flood insurance for agricultural land and the unknown probability of designing sustainable flood insurance, the costs may be difficult to justify and may have to be assessed on a case-by-case basis. However, given the importance of flood as a risk, finding solutions has a high priority. 63

77 Agriculture and Rural Development 5. Feasibility Studies Over the course of , the co-authors of this paper have assessed the feasibility of expanding the concept of index-based insurance that has been widely piloted in a number of countries for drought to flood risk. Work was carried out in collaboration with other institutions and experts with generous support from donors. The activities focused largely on Vietnam, Thailand, and Bangladesh and are summarized here Vietnam: Risk Identification and Conceptualization of a Flood Insurance Product Background Natural hazard risks dominate the landscape in Vietnam. Major typhoons, flooding, droughts, and other natural hazards inflict significant hardships on many households in the country, but particularly on agricultural households. Among others, flooding is a dominant and pervasive risk that can cause severe agricultural production losses. Flood events in Vietnam can result from river flow, originating from distant catchment basins; from flash flooding caused by heavy rainfall, including from typhoon; and from coastal sea surge in the rice and aquaculture production areas in river deltas. The flood regime is part of the natural cycle for farmers, but exceptional flooding or excess rainfall causes severe disruption and economic and livelihood impact. Between , records show that Vietnam suffered 44 major flood events, affecting more than 27 million people and 27 million hectares of land and resulting in over US$1.3 billion of loss. 41 In 2005, the Asian Development Bank (ADB) in conjunction with Department of Insurance of the Ministry of Finance undertook a technical assistance project entitled Developing Agricultural Insurance in Vietnam. Its primary focus was to conduct a feasibility analysis of using index-based insurance product(s), which are designed to mitigate the problems of MPCI, as a new risk transfer instrument for Vietnam s agricultural producers. A second component of the project was to develop prototype index insurance products that could lead to a pilot test depending on the outcome of the feasibility study. A World Bank team contributed technical inputs to this project Findings After undertaking risk assessment, the ADB team decided to focus on designing an index insurance product for the early flooding that negatively impacts the spring summer rice crop in the Mekong Delta. The insurance contract window was proposed to be around June 20 to July 15, which is when early flooding interferes with the unfinished spring summer rice harvest in the focus area, i.e., the Dong Thap Province in southern Vietnam. Dong Thap Province is in the seasonally flooded alluvial plains on the north bank of the Mekong River, bordering Cambodia. The Tien Giang River (one of 64

78 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture the two branches of Mekong River) runs through the province north from Cambodia to the south of the province for 105 km. Rice is a dominate crop in Dong Thap. In some areas of Dong Thap, farm households are obtaining up to 90 percent of their income from rice production. There are three main rice crops in Dong Thap: 1) winter spring season (November to early March); 2) spring summer rice season. (March/April to July/August); and 3) summer autumn season. (July to October). While regular seasonal flooding is beneficial to farmers, the spring summer rice production is heavily affected if flooding occurs early in July or August. While the Mekong Delta floods every year, Vietnam has made massive investments in infrastructure to manage the normal extent of flooding in its early stages to extend the growing season. These include dikes, channels, and other control structures lying below the Tan Chau hydrological gauging station in Dong Thap. A problem occurs when flows increase faster than on average during the later part of June and early July when farmers still have their spring summer rice crop in the fields. In some years such as 2000, large rice-producing areas were affected by the early heavy flooding that occurs in this season. The covered dike systems protected the rice fields when water levels are normal, but the rice fields can be damaged when the dikes are breached by flooding waters (see Figures in Section A.5.1 for map of Dong Thap flood risk zones). In order to design a flood index insurance product for the early flooding, daily water level data obtained from the Tan Chau gauging station from were used to examine water levels exceeding 250 cm for dates between June 20 and July 15. The 250 cm value is the level at which downstream flooding becomes a serious problem while the June 20 July 15 critical period coincides with the main harvest activity in Dong Thap. When this measure is used, four out of 27 years had excess water levels during this period. This threshold has been found to result in significant levels of unharvested rice, resulting in lost income and difficulty faced by rice farmers in repaying loans. The threshold also represents roughly a 1-in-7-year chance, which should be an acceptable level of frequency for most insurance providers Index solutions The previously mentioned technical analysis formed a basis for designing flood index insurance. While recognizing the impact of the early flooding on individual farmers, the design also recognizes current technical and operational constraints for implementing micro-level flood insurance. Based on the analysis, a flood index insurance product termed Business Interruption Insurance was proposed for pilot consideration. Instead of targeting individual farmers, this meso-level product is designed to protect the loan portfolio of the Vietnamese Bank for Agriculture and Rural Development (VBARD), a dominant agricultural lender in Dong Thap. It is envisioned that the contract would be purchased by VBARD against the early flooding event defined as water levels above 250 cm occurring between June 20 July 15. The index insurance contract would be underwritten against recorded water levels at Tan Chau, a main river gauge station in Dong Thap, with correlations made with data from a back-up station upstream in Cambodia. The proposed insurance is expected to help VBARD protect its 65

79 Agriculture and Rural Development portfolio from business interruption costs (e.g., lost interests, administrative expenses, etc.) incurred from large amount of loan rescheduling, thus transferring a big risk out of the system Lessons learned This meso-level insurance is a practical proposal that recognizes technical and operational challenges in directly insuring individual farmers from flood risk. Instead of seeking a solution for individual farmer insurance, the ADB project team developed a product at an aggregated level of risk at VBARD. In essence, this transfers, from insurers to VBARD, the task determining rules to pass on the benefits of payouts from such insurance from VBARD to farmers. The proposal also seeks to transfer risk from an agricultural bank which, through its lending practices, has become the de facto risk aggregator and agricultural insurer for the individual farmers. However, as it was far too late into the early flood window, the flood insurance transaction did not take place in 2008 but is currently under consideration by relevant parties in Vietnam for the current year. The project has raised interest in index-based risk transfer and the costs and benefits of this approach are being discussed. If implemented, this would be the first flood index insurance transaction for agriculture in developing countries. The purchase of this insurance would represent a significant step towards improving VBARD s portfolio risk management using a market approach. It will also provide a demonstration of the application of the index-based risk transfer concept for agricultural insurance in Vietnam and other developing countries. Another advantage of this type of meso-level product is that it does not preclude later development of individual farmer flood insurance, if technical and operational conditions are found to be feasible in future Thailand: Challenges in Designing Flood Index Insurance at the Micro Level Background Thailand was the first country where the World Bank has started full-scale flood index insurance product development research. The Maung Petchaboon District of the Petchaboon Province was selected as the activity site with rice as a focal crop. International and local consultants were contracted to assess flood risk during the rice production season in the Pasak River basin in Petchaboon using flood modelling techniques. The risk assessment was expected to form the basis for designing a flood index that quantifies the relationship between flood events and rice production losses Findings In the early stage, this technical work in Petchaboon was conducted with a primary objective to design a flood index insurance pilot at the micro level. Later, it was found that modelling the impact of inundation flood on rice at such a high resolution was a challenging task. While this first attempt at flood modelling in Petchaboon resulted in a prototype flood duration index for the flowering and pre-harvest time window, the index could not be 66

80 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture implemented in a real pilot scheme due to several limitations in the outputs from the flood model. 43 The key limitations included: (1) insufficient quality of topographic and hydro-meteorological data to parameterize and validate the chosen flood model; and (2) the chosen modelling approach, which was computationally too expensive to generate sufficiently long-time series of outputs for the purpose of premium rating. Possible solutions were identified to overcome these limitations. In the second phase of research, the team decided to use a simplified modelling approach, i.e., a hydraulic model (MIKE 11) that is widely used by local agencies and researchers. This new approach would allow (1) better and more reliable quantification of the long-term historical patterns of flood in the region (which is required for actuarial evaluation) and (2) better characterization of the spatial flood patterns in the flood plain of the Pasak River basin, in particular with respect to flood depth and duration. That is, the model would have to resolve flood patterns at the daily timescale. In addition, more information would be gathered with regards to farming practices, flood regime and management, and socio-economic impacts of flood on rice farmers to better understand the dynamic of flood losses for rice farmers in the study area. The scope of this second phase of work was also broadened to include the potential non-insurance risk management applications of flood modelling outputs, particularly for Bank for Agriculture and Agricultural Cooperatives (BAAC), which is a major lender to farmers in Thailand, including those within the study site Lessons learned For the insurance application, technical outcomes of this second phase research still demonstrated that flood index insurance would be very challenging to implement in the study area, especially at a micro level. First, the model outputs showed that the Petchaboon River Valley is characterized by a large area prone to relatively frequent localized flooding (for example, once in three to five years), while the area prone to infrequent flooding (for example, only once in 10 years) is small (Figure 16). Such concentration of risk lends itself to the adverse selection problem, while frequent flooding would lead to extremely high premiums, making the insurance scheme unsustainable. Second, timing of flooding is critical in relation to the actual damage to crop production concerned. With local knowledge of the flooding pattern, farmers can to some extent manage their own risk exposure due to the regularity and timing of flood events. An insurance system is only feasible when there is the possibility of infrequent and widespread flood that is outside of the farmers ability to influence by management decisions. Finally, BAAC, despite its good database system, has no detailed database of locations of farmers in the flood prone areas. The information was only collected at the administrative district level; village-level data were not available. This makes geo-locating farmers into different homogenously defined flood risk zones (which are smaller than an administrative district) that is a cornerstone of a micro-level flood insurance scheme a very difficult prospect. It was found that the stakeholders are interested in non-insurance benefits of this technical work in Petchaboon, though they are not of immediate application 67

81 Agriculture and Rural Development Figure 16 Flood risk zones for rice production in the Muang District of Petchaboon, Thailand. Flood Hazard Zone Very high risk High risk Moderate risk Low risk Flood depth > 20 cm Flood Hazard Zone Very high risk High risk Moderate risk Low risk Flood depth > 40 cm Flood Hazard Zone Very high risk High risk Moderate risk Low risk Flood depth > 70 cm Flood Hazard Zone Very high risk High risk Moderate risk Low risk Flood depth > 160 cm Source: ASDECON in the study area. At the moment, BAAC cannot directly use the flood risk maps to inform ex ante planning of lending to rice crops, and/or ex post rescheduling of loans following a flood event. This is because as a state bank BAAC cannot alter lending criteria, policy, or practices in only particular areas due to one particular risk, as it would result in changing advantages and disadvantages in borrowing among customers nationwide. In addition, both non-farm incomes and government compensation seem to have played a key role in good loan recovery rates, despite the frequent flood events in the study area. However, experience from the flood risk mapping exercise has strengthened the already increased risk awareness at BAAC, especially on how agricultural lending could benefit from the broader use of weather, agronomic, and geographical data. It has also highlighted the importance of creating a comprehensive GIS database to support the bank s operation. At the corporate level, BAAC is in the process of introducing a credit scoring system. The system is being developed with outside experts to create a systematic ranking of factors affecting credit worthiness. Factors that have been included so far are socio-economic such as incomes, saving rates, household expenses, repayment history, and so on. Allowing BAAC to separate customers into 10 grades, the system is planned to be adopted by BAAC nationally, starting with new 68

82 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture customers in 2009, in order to form a basis for risk-based loan pricing in the future. BAAC is interested in eventually integrating disaster risks into the system in a holistic manner, which implies a need for national assessment and mapping of key disaster risks such as droughts and floods. As BAAC does not develop its own database on disaster zones, the bank will use data collected by various government departments as input into the credit scoring system. This flood risk assessment and index design exercise is therefore a relevant ex - perience both for BAAC and government agencies to collaborate in flood risk identification, in the quantification of agricultural flood losses and impact on loans, and in the use of coordinated technologies to support the process Bangladesh: Assessing the Feasibility of Flood Insurance in a Complex Environment Background Bangladesh is one of the world s most vulnerable countries to natural hazards, including floods, droughts, cyclones, and earthquakes. Eighty percent of the country consists of floodplains created by more than 300 rivers and channels, including three major rivers (the Ganges, the Brahmaputra, and the Meghna). At the same time, Bangladesh is only a small part of a large regional system, as less than 10 percent of the watershed area of these rivers falls within the national territory. The floodplains are home to a large population most of which are rural and poor whose lives are intricately linked to the flooding regime. Annual regular flooding has traditionally been beneficial, while lowfrequency but high-magnitude floods can have adverse impact on Bangladesh s citizens and economy. Recent major floods occurred in 1988, 1998, 2004, and The 2007 event directly affected more than 14 million people, caused more than 1,000 deaths, affected more than 2 million acres of agricultural land, and damaged and destroyed infrastructure (more than 30,000 km of roads) and social and educational facilities as well as private assets, including housing, crops, livestock, and fisheries. The preliminary damage and loss assessment, for the crops, livestock and fishery sub-sectors, totaled about US$648 million. The country subsequently experienced another natural disaster, Cyclone Sidr, in November 2007 that caused estimated damages and losses of Bangladeshi Taka (BDT) billion (US$1.7 billion) equivalent to 2.8 percent of Bangladesh s GDP. Several types of flood occur in Bangladesh (Figure 17). The main type of flooding arises from river inundation flood, as a result of the major river systems flowing into the country, which originates in the Himalayas. Glacial melt waters and monsoonal rains arise outside the country, but flow through Bangladesh. While high river flows and associated flooding are normal annual events, major floods result in massive extensions beyond routine floodplains. Flash floods are frequent in the northeast of the country, in the Hoar Region. Coastal flooding can occur in the south, associated with high tides and high stream flow. Tidal surges may result from typhoons, which bring a combined effect of heavy rains and tidal surges. Agricultural land has been classified and mapped according to its potential for flood and other natural disasters, as well 69

83 Agriculture and Rural Development Figure 17 Flood-prone areas of Bangladesh Panchagarh Flood Prone Areas Bangladesh 26 Thakurgaon Lalmonirhat Nilphamari Dinajpur Rangpur Kurigram Kilometers Gaibandha 25 Nawabganj Naogaon Joypurhat Bogra Sherpur Jamalpur Netrakona Mymensingh Sunamganj Sylhet 25 Rajshahi Nator Sirajganj Tangail Kishoreganj Habiganj Moulvi Bazar 24 INDIA Pabna Kushtia Manikganj Meherpur Rajbari Chuadanga Jhenaidah Magura Faridpur Gazipur Narsingdi Brahmanbaria Dhaka Narayanganj Munshiganj Comilla INDIA Jessore Narail Chandpur Shariatpur Madaripur Gopalganj Lakshmipur Feni Barisal Noakhali Khagrachhari 23 Satkhira Khulna Pirojpur Bagerhat Jhalakati Patuakhali Bhola Chittagong Rangamati Barguna 22 Bay of Bengal Bandarban 22 Legend Severe river flooding Moderate river flooding District boundaries Urban Cox's Bazar Low river flooding Sundarban and reserved forest Severe flash flooding Rivers and Bay of Bengal MYANMAR Moderate flash flooding Kaptai lake and waterbodies 21 Low flash flooding Severe tidal surge Moderate tidal surge Not flood prone Source: Bangladesh Agricultural Research Council BARC/UNDP/FAO GIS Project: BGD/95/006 July 2000 as appropriate rice cropping patterns and varieties that minimize potential exposure to flooding events and food insecurity (Figure 18). Given the impact of flooding in Bangladesh, flood management and disaster prevention and management is highly developed through institutions such as the Bangladesh Water Development Board (BWDB). The success of these 70

84 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 18 Detailed flood types mapping within the northeast Hoar Region of Bangladesh. Flood Types in Haor Districts Netrokona Sunamganj Sylhet Mymensingh Kishoreganj Habiganj Low river flooding Moderate river flooding Low flash flooding Moderate flash flooding Severe flash flooding Moderate tidal surge Not flood prone River erosion District boundary Source: Bangladesh Water Development Board (BWDB) in WFP (2004). efforts is evident from the significant reduction in lives lost during recent major floods and the organization of relief, recovery, and reconstruction Findings There is interest in the role of flood insurance in Bangladesh as well as a recognition of the major challenges that flood insurance would present. In 2006, the World Bank carried out an investigation about the potential for flood insurance in Bangladesh; flood index insurance was considered as an option. The study found that there were many complexities for designing flood insurance (index or non-index) in the country in general which include: a varying degree of influence of human management of flood risk; influences of farmers decision regarding timing of the cropping calendar; definition of beneficial versus damaging flood events; lack of rural insurance penetration; and flood zoning and premium rating challenges. As for flood index insurance in particular, the approach could be technically feasible in those areas infrequently affected by river inundation flood; however, the current feasibility study concluded that flood index insurance at a micro level would be organizationally challenging and would require further research Lessons learned Given the importance of costs of major flood events to government, a further option is to consider the use of indexed flood measurement at a macro level, as a possible mechanism for rapid release of financial resources for the government after a flood or other natural disaster. This could be done either through insurance payouts or through release of contingent loans from international financial institutions (such as the World Bank) to the government. Notably, the use of remote sensing to capture the extent of widescale flooding can provide objective and rapid assessment of flood extent and could be considered to assist linkage of index risk transfer instruments to 71

85 Agriculture and Rural Development government financing of disasters. Such an approach would strengthen the government s existing disaster relief and recovery mechanisms, which are welldeveloped under the National Disaster Management Council, and would not be part of a micro-level insurance program. Further activities on agricultural flood insurance need to be put in the context of gradual agricultural insurance development in Bangladesh. Agricultural insurance was operated in the country based on an MPCI system and on a limited scale between 1977 and 1995, when it was dropped due to financial unsustainability. A new proposal for agricultural insurance in Bangladesh was put forward in 2008 by SBC, the state insurance company, to the Ministry of Finance, suggesting the formation of a specialized agricultural insurance company similar to NAIC in India and making insurance compulsory for accessing agricultural loans. In parallel, the government requested technical assistance from the World Bank in assessing viable insurance products as well as public-private partnership framework in risk financing to support agricultural insurance market development. Flood insurance needs to be considered in such wider context, taking into account the stage of rural insurance in Bangladesh, which is very poorly developed. Other agricultural index products that may be technically easier at a micro level include drought index insurance, hail insurance, and livestock insurance. Development of these micro-insurance products could be a first step to strengthen rural insurance in Bangladesh, while taking advantage the microfinance (MFI) sector that could serve as a distribution channel. In order to add value to the livelihoods of farmers, any agricultural insurance products for Bangladesh must also be developed taking into account not only the extreme complexity of risks but, more important, the intricate system of informal risk management strategies prevalent in the rural areas. Bangladeshi farmers have learned to adapt their farming practices to floods, while flood control structure is also substantially improved. At the same time, agriculture is increasingly becoming diversified out of rice mono-crop cultivation for subsistence consumption, into new high value-added crops (such as fruits, maize, potato, cassava, and so on), livestock, and fishery enterprises as well as other non-farm income generating activities. In addition, some recent farmer interviews show that many of them have increasing ability to alternately access credits from many MFIs due to the concentration of MFIs working in the same area. All of these factors have allowed farmers in many areas of Bangladesh to earn more income and better cope with disasters. In this context, careful risk analysis and demand assessment are needed in order to determine where insurance fits. For example, as crops might not be the most economically valuable activities in many cases, there might be more demand from farmers in insuring other activities such as livestock, poultry, or fish ponds Summary of Feasibility of Flood Index Insurance Schemes at the Micro and Macro Levels Tables 6a and 6b summarize main findings about the feasibility of flood index insurance at the two distinct levels of program resolutions. 72

86 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Table 6a Summary Feasibility of Macro-Level Flood Index Insurance Type of flood River flood/inundation Flash flood Coastal storm surge flood Risk zoning and flood mapping Premium calculation Index design Risk zoning at a level of resolution for a macro scheme may be feasible using flood modelling and/or by archive remote sensing data. Challenges include upstream river management, local flood control infrastructure, and detention areas; modelling is data dependent, and archive data may be at low resolution. Premium rates can be calculated for an index based on river discharge data. Risk transfer may therefore be possible using river discharge data as the basis for an index. The challenges to interpreting changes in river discharge data are noted above. A macro index can be designed in which payout is based solely on river discharge data, but this is only suitable for holders of aggregated flood risk. The output of flood modelling and calibration of flooded areas against river discharge data can be used to design an appropriate payout scale according to river discharge. The degree of confidence in flood modeled output for agricultural flood risk insurance is not yet tested and is highly dependent on resolution needed. Risk zoning is infeasible beyond general identification of areas prone to flash flooding. Premium cannot readily be calculated for specific locations. An index product cannot be foreseen for flash flood risk. Risk zoning is difficult but could be carried out with risk modelling based on coastal surge height predictions. Such modelling is practiced in Europe. Broad flood prone areas may be known from past events. Flood patterns caused by coastal surge are less predictable than during river inundation flood. Development of a rating basis for coastal surge, and associated flooding of agriculture, is extremely challenging; methodology has not been developed. Even if coastal surge data is available, calibration of such surges against expected flooding inland would be needed. Coastal flood is not directly related to river discharge, hence an index product based on river discharge cannot be foreseen. As noted, the strongest feature of an index including coastal flood is loss assessment using remote sensing at a macro scale. The most difficult aspect remains the modelling of surge-induced flood risk and associated premium calculation. (continued) 73

87 Agriculture and Rural Development Table 6a (Continued) Type of flood River flood/inundation Flash flood Coastal storm surge flood Farmer enrollment Loss assessment Summary Farmer client locations can be recorded on a GIS database. Holders of a macro policy could create payout rules for the distribution of payouts generated by a river discharge index. Remote sensing can support objective identification of affected areas down to a micro level, and this could be validated and supported where ground inspections exist. An index product for risk transfer by the aggregator to reinsurers, based on river discharge data, is feasible. However objective flood measurement and loss payment, based on remote sensing, would require that the risk aggregator establishes its own payout rules. This approach could allow the transfer of large scale risk, but the need for the aggregator to establish payout rules at local level may make this product suited to a government compensation scheme, or for the holder of aggregated risk, such as an agricultural bank. Farmer client locations can be recorded on a GIS database. Remote sensing cannot provide objective loss assessment at micro or macro level, because runoff is rapid. An index product is not feasible, and objective measurement is not feasible. However, localized nature of flash flood may make conventional field assessment more appropriate, supporting a disaster payment system, or conventional indemnitybased flood insurance. Farmer client locations can be recorded on a GIS database. Remote sensing could provide objective measurement of flood events from coastal flood surge. Indexation of a product for coastal flood is not considered feasible, but objective flood measurement and disaster payment system could be foreseen using remote sensing. If flood is known to be highly correlated to cyclones, and some form of risk transfer based on cyclone occurrence is potentially foreseeable, although with high basis risk to areas actually flooded. Engineering principles are applied to calculate the construction of coastal dikes at given return periods. Practical extension of this work to calibrate agricultural flood risk, using flood modelling, has not yet been undertaken. Risk transfer at macro level to reinsurers for cyclone-induced coastal flood for agriculture is not yet tested. 74

88 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Table 6b Summary Feasibility of Micro-Level Flood Index Insurance Type of flood River flood/inundation Flash flood Coastal storm surge flood Risk zoning Premium calculation Index design Farmer enrollment Risk zoning is feasible using flood modelling and supported by archive remote sensing data. An objective is to define homogenous flood zones for farmer enrollment. However, the size of such zones (resolution) is critical. Risk zoning in heavily flood-managed areas will be dictated by flood training with dikes. Premium calculation using flood modelling is problematic at high resolution, but broad risk zones can be created. Timing and duration of flood are critical to crop vulnerability and the most problematic aspect in developing premium rates. An indexed payout scale would be based on flood timing, extent, and duration, measured by remote sensing. Premium is based on flood modelling, and payouts based on remote sensing brings an element of uncertainty in pricing. Farmer client locations can be recorded on a GIS database, within defined flood zones. Risk zoning is infeasible beyond general identification of areas prone to flash flooding. Premium cannot readily be calculated for specific locations. A micro-level index product cannot be foreseen for flashflood risk. Farmer client locations can be recorded on a GIS database. Risk zoning at a micro level seems unlikely to be feasible. Macro-level premium estimation would be difficult and micro-level would be even more challenging. It is not possible to foresee a true index product for coastal surge flooding. Farmer client locations can be recorded on a GIS database. (continued) 75

89 Agriculture and Rural Development Table 6b (Continued) Type of flood River flood/inundation Flash flood Coastal storm surge flood Loss assessment Summary Flood date and duration would be detected using remote sensing, within defined homogenous flood risk zones. Ground survey could support these observations. Micro-level flood index insurance is highly challenging. A pure index product for flood cannot be developed at the micro level. Objective and independent loss measurement using remote sensing and an agreed payout scale can bring the most important benefits of index to flood insurance. Reinsurance of major risks could be carried out using a macro index based on river discharge, leaving some potential basis risk with the insurer. Remote sensing cannot provide objective loss assessment at micro or macro level, because runoff is rapid. An index product is not feasible. Objective measurement is not feasible, due to rapid onset and dispersal of flood water. However, the localized nature of flash flood may allow more conventional field assessment to be practical, supporting a disaster payment system, or by conventional indemnity insurance. As with river flooding, similar technology could be applied for coastal surge flood. A pure index product for flood cannot be developed at the micro level. Objective and independent loss measurement using remote sensing and an agreed payout scale can bring the most important benefits of index to flood insurance. However, there is no real basis for developing premium rates. 76

90 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture 6. Conclusions and the Way Forward Floods represent a significant and pervasive risk in many parts of the world, both in built-up urban and rural areas. While people in flood-prone areas, such as agricultural producers in South and Southeast Asia, have developed techniques to mitigate normal seasonal flooding, extreme and widespread flooding represents a major source of risk for agricultural producers and the institutions that provide financial services to them. The work presented in this report arose from requests made by rural financial institutions, World Bank project teams, and other development partners in the Southeast Asia region to explore ways in which catastrophic flood risk could be insured. Traditionally flood risk has been very difficult to insure and has been largely limited to property insurance in developed countries; virtually no agricultural flood insurance is available in low- and middle-income countries. This report explored how technologies such as flood modelling and satellite-based remote sensing of flood, combined with index-based insurance, can be harnessed to support flood risk management and flood insurance programs for agricultural producers. The findings presented in the report are derived from several feasibility studies carried out in Thailand, Vietnam, and Bangladesh. Unlike flood damage in urban property, the timing of flooding is critical in relation to the actual damage to agricultural production. Defining agricultural flood events to be insured requires a good understanding of various aspects of flooding, such as flood extent, duration, and depth, in relation to the stages of crop growth. Throughout the crop growth cycle, critical thresholds at which a flood event results in damages also change. This requires relatively detailed modelling and mapping of floods to identify the times and locations of highest flood risk. In principle such detailed flood risk assessment can be performed by flood modelling. However, data limitations that are common in most of the settings explored in this work can severely constrain the reliability of flood model output. These constraints can be partially overcome by direct observations of floods from remote sensing, which can provide timely estimates of flood extent and duration. While remote sensing uses sophisticated data and models, most of these have become readily available today in a reprocessed form and can easily be applied in a geographic information system by local institutions to support flood risk mapping. With knowledge of the historical flooding pattern, farmers can to some extent manage their own risk exposure due to the regularity and timing of flood events. An insurance system is only feasible where there is the possibility of infrequent and widespread flood that is beyond the farmers ability to 77

91 Agriculture and Rural Development influence by management decisions. In many circumstances, risk reduction and mitigation are the priority measures before insurance. In principle, parametric flood insurance can be developed for micro, meso, and macro levels of risks and policyholders (i.e., from a very disaggregated level of individual farmers to more aggregated levels), whereby an entire subregion (e.g., collectives of producers in a sub-catchment) would be insured. In practice, the technical feasibility at each level depends on a variety of factors such as the specific characteristics of each flood plain; the availability, quality, and resolution of data; and the level at which demand for insurance is expressed or aggregated. Organizational feasibility, especially at the micro and meso levels depends on whether flood insurance is provided as a stand-alone product or linked to other inputs or credit. Drought index insurance, for example, has shown to be adopted when bundled with credit and inputs (e.g., seeds). Finally, operational feasibility of flood index insurance has to be assessed in terms of compliance to legal and regulatory requirements. A micro-level flood insurance product would identify flooded areas at high resolution, with an aim to compensate individual farmers. Given the general lack of data at fine resolutions in developing countries, there is still a significant technical challenge in designing and implementing flood insurance at a farmer level. This was demonstrated in the case of Thailand. Meso- and macro-level flood indexes aim to capture the catastrophic risk at an aggregate level, but the schemes will require the risk aggregator, such as an agricultural bank or a government, to set rules for application and distribution of insurance payouts to micro-level beneficiaries. Within the current data and technological environment, there is a higher potential for meso- and macrolevel schemes to be developed. This was demonstrated in the case of Vietnam, where a meso-level insurance product was proposed. Instead of targeting individual risks, the product would transfer risk from an agricultural bank that, through its lending practices, has become the de facto risk aggregator and agricultural insurer for the individual farmers. The product could be seen as a practical proposal that offers a solution while recognizing technical and operational challenges in insuring individual farmers from flood risk. At present there is little flood insurance in developing countries; however, the demand for effective mechanisms to financially protect households, rural lending institutions, and governments against flood risk is increasing. This includes the financing of the direct losses caused by floods (both in agriculture and urban settings) as well as the indirect effects such as the interruption of government services caused by extreme floods. For instance, the Caribbean Catastrophe Risk Insurance Facility, a mechanism that has provided protection to Caribbean countries following hurricanes and earthquakes since 2007, is now exploring ways to expand its coverage to floods as well. Similar efforts are underway for Pacific Island states. At the same time, established agricultural insurance programs such as the National Agricultural Insurance Scheme in India are also currently exploring ways to provide faster and more effective insurance payments for floods to agricultural producers. 78

92 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Going forward, agricultural insurance options for flood risk are best explored and piloted in the context of ongoing technical assistance activities in the rural domain for instance, related to rural finance, land reform, and agricultural supply chains. (This equally applies to drought risk.) While the activities and findings documented here have been largely carried out in response to specific requests by agricultural banks and other stakeholders, the experience shows that the development of an agricultural insurance project requires a sustained effort by multiple stakeholders (including technical agencies, insurers, banks, extension services, etc.). This is best accomplished in the framework of a broader activity. Delivering benefits from insurance to farmers is best demonstrated when a package of measures, such as seeds, inputs, credit, and insurance, can lead to improved productivity at the farmer level. It is equally important that any pilot-testing of agricultural flood insurance is accompanied by activities to systematically identify and assess flood risk and targeted interventions to reduce flood risk through structural measures, crop management, and planning. Without such parallel measures, any insurance scheme is unlikely to be sustainable, both financially and operationally. In many settings, simple insurance solutions will likely not be the appropriate means to protect agricultural producers from flood risk. This includes settings where agricultural areas are used as retention zones to protect downstream urban areas or where human activities are drastically changing the hydrological regime (e.g., through urbanization, changes in upstream land use, or engineered structures). Future efforts to develop and test agricultural flood insurance in developing countries can be supported by government and donors in several ways. First, flood risk assessment requires high-quality data on geophysical characteristics including hydrology, topography, climate, vegetation, and soils. The collection of such data is generally done for purposes other than insurance; however, to make agricultural flood insurance possible, the systematic collection, updating, and provisioning of such data is essential. Donors and governments alike can play an instrumental role in strengthening the capacity of the respective agencies and institutions. Apart from geophysical data, socioeconomic data is also required to assess the potential risk arising from floods in the agricultural domain. This includes systematic inventories of planted crops area and collection of agricultural loss data. For agricultural banks and governments alike, mapping key assets at risk is critical to assess their exposure to flood risk. Second, risk awareness building and education are essential to promote insurance. Identifying and assessing flood risk is a critical step in order to educate potential clients of an insurance scheme (farmers, banks, governments) as well as to promote flood risk mitigation measures. Knowledge and awareness are necessary to identify risk reduction measures that eliminate avoidable risk. This applies at all levels. Risk awareness building informs government with respect to its own risk exposure as well as the policies that can be put in place to reduce risk (e.g., through planning, regulation, and so on). At the organizational (meso) level, risks are often not explicitly dealt 79

93 Agriculture and Rural Development with once identified and quantified, remedial actions and hedging strategies can be explored. Lastly, to have agricultural producers ultimately benefit, they need to be educated about the risks they are exposed to and the options to cope with the financial outcomes. Third, remote sensing provides a rapid and cost-effective way to map and monitor floods and could be used in a more operational manner to help government entities manage floods more effectively. This may be supported even without the explicit objective to develop agricultural flood insurance and promote better disaster preparedness, response, and recovery. The information that can be readily derived from remote sensing (including flood extent) can be instrumental in supporting government schemes to provide compensation to farmers in flood-affected areas. Such schemes would not only systematically generate flood risk information over time, but also a framework for the incorporating of flood risk insurance at a later stage to complement government expenditures. Fourth, given the technical challenges and level of complexity to develop insurance products at the farmer (micro) level and the increasing need to provide flood protection instruments at a more aggregate level, future research should aim at developing simpler indicators of flood risk that are viable and robust at the regional and perhaps national levels. Such indicators (e.g., using rainfall metrics as a proxy for flood risk) will naturally be much coarser and carry a high level of basis risk. Research could address how basis risk can be reduced by improvements in data coverage and quality and be managed through internal mechanism of the risk-taking entity. Fifth, despite advances in technology and innovation in insurance, providing flood risk insurance remains a challenge, in particular in rural settings, and requires participation of a broad set of stakeholders and technical expertise ranging from agronomics and flood modelling to insurance design. To advance flood risk management and flood risk insurance more effectively, a platform that combines such expertise is required, including from international and regional centers of excellence. 80

94 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Annex: Technical Background A.1. Floods: A Global Perspective The significance of flood risk in agriculture varies both regionally and locally, according to characteristics of rainfall, snowmelt, topography, soils, and many other factors. Mitigation of and adaptation to flood risks has shaped the past and present farming practices in any particular region. Diverse geographic regions require recognizing the varied causes of flooding and an awareness of their geographic and seasonal patterning. Mapping historical floods reveals a global geographic picture wherein some regions are more flood-prone than others (Figures 19, 20, and 21). Specifically, flood-prone regions of the world clearly include: (a) the eastern U.S., (b) Central America and Mexico, (c) northwestern South America, part of Amazonia, and mid-southern South America, (d) central Europe and the U.K., (e) an irregular belt of sub-saharan Africa, the Sudan, and coastal southern Africa, (f) Iran, Afghanistan, Turkey, and eastern India, (g) all of southern and eastern China, Indonesia, Malaysia, and the Philippines, (h) southern Japan, (i) high-latitude northeastern Russia, and (j) northern, eastern, and far western Australia. Other areas are either less flood-prone or generate less attention due to smaller human population density. Thus, floods over much of Canada are not widely reported in media due to low impact on human settlements. Floods in parts of Scandinavia, eastern Europe, central Russia, and northern China are also not widely reported. Northern Africa including the Sahara provide only sparse news-reported flooding, and, finally, there are few reports of flooding from desert areas such as the western U.S., parts of South Africa and Australia, western India, most of the Arabian Peninsula, and Mongolia. Examination of reported floods sorted by causation allows estimation of the dominant flood risk in different parts of the world. Figure 19 provides global maps of four genetic classes of floods affecting tropical and subtropical regions. Similarly, there is a distinct seasonal patterning of floods as shown in Figure 20. A.2. Flood Modelling Approaches Hydraulic models used for flood modelling can be divided into one dimensional, quasi-two dimensional, and combined models. 44 1) One Dimensional Hydrodynamic Models: These types of models are based on partial differential equations of hyperbolic type (de Saint-Venant equations) for conservation of mass and momentum in one dimension. In order to be solved, they need boundary conditions such as hydrographs. The main advantage of hydrodynamic models is that they produce water levels and discharges at discrete points without the need to utilize 81

95 Assessment.qxd:Assessment 3/4/10 7:32 AM Page 82 Agriculture and Rural Development Figure 19 Catalogued flood from undifferentiated heavy rain, tropical storm, monsoonal rains, and brief torrential rain. Floods caused by undifferentiated heavy rain since 1985 Floods caused by tropical storm since 1985 Floods caused by monsoonal rains since 1985 Floods caused by brief torrential rain since 1985 Source: DFO additional rating curves. They also produce realistic attenuation of the flood hydrographs so they can be used to reproduce particular historic events. Most of the inaccuracies related to the results of these models come from severe violation of one or more of the basic assumptions, like onedimensionality of the flow, uniform velocity, and horizontal water level 82

96 Assessment.qxd:Assessment 3/4/10 7:32 AM Page 83 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 20 Catalogued floods in January, May, and September. Floods beginning in January (since 1985) Tropic of Cancer Equator Tropic of Capricorn Floods beginning in May (since 1985) Tropic of Cancer Equator Tropic of Capricorn Floods beginning in September (since 1985) Tropic of Cancer Equator Tropic of Capricorn Source: DFO across the cross-section or small bottom slopes. The computational time is sometimes considered to be the limiting factor in the use for flood forecasting. 2) Quasi Two-Dimensional Hydrodynamic Models: These types of models are often described as compartment models. Flow in the main channel is described by the one-dimensional de Saint-Venant equation, while the land surrounding the main channel is divided into compartments due to natural 83

97 Agriculture and Rural Development or man-made obstacles (roads, for example). Water level at each compartment is solved using the continuity equation, while the flows across the compartment boundaries are calculated based on the water levels in the adjacent compartments and the boundary characteristics. This approach has considerable advantages over the one-dimensional models in modelling flow situations that are clearly multidimensional. However, compared with the one-dimensional hydraulic models, computation is more demanding. 3) Raster-Based Quasi 2 Models: These types of models are relatively recent. They are based on the same idea as the quasi two-dimensional models but the size of compartment is reduced down to raster cells. The water levels are updated using the continuity equation, while the fluxes between cells are estimated using water levels in the adjacent cells and some approximations (uniform flow, gradually varied flow, diffusive wave, or a weir equation). These types of model are closely related to the topography, which simplifies input of data and visualization of results. However, they are not yet in wide use as they are not part of commercially available packages. These models are proving to be very useful especially due to their explicit treatment of topography. Their main disadvantages are: (1) approximation of the flow in the main channel as a kinematic wave that is not suitable for flatter rivers; and (2) computational expense due to the small cell sizes (compared with the compartment models). It is also worth noting that these models do not provide velocity estimates. 4) Two-Dimensional Hydrodynamic Models: The types of models are based on hyperbolic partial differential equations, and they are usually solved by the methods of finite differences, finite elements, and finite volumes. The method of finite differences offers a computationally efficient solution but it is restricted to rectangular or curvilinear grids. On the other hand, the method of finite elements can use irregular meshes but requires much heavier computation. The method of finite volumes is a recent development used mainly in the research community. It provides solution for discontinuous flows but it requires exceptionally small time steps due to the use of explicit computational schemes. 5) Combined Models: Combined one- and two-dimensional hydrodynamic models are recent developments especially aimed at flood modelling. Flow within the main river channel is approximated as a one-dimensional flow, while the out-of-bank flow is treated as a two-dimensional situation. Currently, different researches are reporting successful developments of the combined models using a variety of approaches to flow approximations and different grid arrangements. A.3. Space-Borne Satellite Sensors Used for Flood Mapping This table shows satellites and sensors that are potentially available to measure past and present flood events. Optical satellites are shown in Roman font, while imaging radar satellites are shown in italics; currently operating (2009) satellites are shown in bold. 84

98 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Satellite and/or Sensor Name Launch Date Landsat-1/ERTS-1 (ERTS-A) 23-Jul-72 Landsat-2/ERTS-2 (ERTS-B) 22-Jan-75 Landsat-3 (Landsat-C/ERTS-C) 05-Mar-78 Landsat-4 (Landsat-D) 16-Jul-82 Landsat-5 (Landsat-D1) 01-Mar-84 Landsat-7 15-Apr-99 French Satellite Probatoire de l Observation de la Terre (SPOT) series: SPOT-1 (SPOT-A) 22-Feb-86 SPOT-2 (SPOT-B) 21-Jan-90 SPOT-3 26-Sep-93 SPOT-4 24-Mar-98 SPOT-5 03-May-02 Indian Remote Sensing (IRS) satellites: IRS-1A 17-Mar-88 IRS-1B 29-Aug-91 IRS-P2 15-Oct-94 IRS-1C 28-Dec-95 IRS-P3 21-Mar-96 IRS-1D 29-Sep-97 IRS-P4 / Oceansat-1 26-May-99 IRS-P6 / ResourceSat-1 17-Oct-03 IRS-P5/Cartosat-1 tbd IRS-P7 / Oceansat-2 tbd Canadian/U.S. radar satellite: Radarsat-1 tbd Canadian Radar satellite Radarsat-2 tbd European Remote Sensing (ERS) satellites: ERS-1 17-Jul-91 ERS-2 20-Apr-95 Envisat tbd Japanese Remote Sensing satellites JERS-1 tbd ALOS/Daichi tbd U.S. Earth Observation System (EOS) satellites: EO-1 tbd EOS-AM1/Terra (includes MODIS and ASTER) 18-Dec-99 EOS-PM1/Aqua (includes MODIS) 04-May-02 China (PRC) Brazil Earth Resources Satellite (CBERS)/ZY-1 series: ZY-1A/CBERS-1 14-Oct-99 ZY-1B/CBERS-2 21-Oct-03 85

99 Agriculture and Rural Development ZY-1B2/CBERS-2B China (PRC) earth Resources Satellite (CRS)/ZY-2 series: CRS-1/China Resource-1/ZY-2A CRS-2/China Resource-2/ZY-2B CRS-3/China Resource-3/ZY-2C Republic of China (Taiwan) Satellite (RocSat) series: Rocsat-1 Rocsat-2 Israeli Earth Resources Observation Satellite (EROS) series: EROS-1 EROS-2 EROS-C Argentine Satellite de Aplicaciones Cientifico (SAC) series: SAC-A [Endeavour-launched] SAC-B SAC-C German Space Agency/commercial satellites: TerraSAR-X U.S. commercial satellites: Ikonos-2 QuickBird-2 OrbView-1 OrbView-2 / SeaStar / SeaWiFS OrbView-3 WorldView-1 U.S. NOAA Polar Orbiters (AVHRR sensor): NOAA 12 NOAA 14 NOAA 15 NOAA 16 NOAA Sep Sep Oct Nov Jan May Dec Apr-06 tbd 04-Dec Nov Nov-00 tbd 24-Sep Oct Apr Aug Jun Sep May Sep Jun-02 A.4. Global Flood Monitoring: Dartmouth River Watch The Dartmouth Flood Observatory has used satellite data to map global floods and provide quasi-real-time information on flooding. River Watch has been possible due to the availability of high-quality global satellite data and a high degree of automation calibration and mapping. The three activities of the flood observatory include: (1) satellite-based flood detection and magnitude assessment, including extent and duration; (2) rapid flood mapping; and (3) integration of mapped floods into quantitative flood hazard assessment. 86

100 Flood detection Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture As presently configured, the system utilizes two remote sensing data sources: for flood detection and magnitude assessment, cloud penetrating frequency of the NASA/JAXA AMSR-E satellite aboard AQUA, and for rapid flood mapping, the two NASA MODIS optical sensors aboard TERRA and AQUA. Both data sources are freely available in the public domain and provide global coverage on a twice-a-day or better schedule; all have planned follow-on sensors, so that implementation of future operational systems need not depend on these particular satellites. For flood detection and measurement, using the AMSR-E microwave frequency allows data retrieval independent of cloud cover. These data are at coarse spatial resolution: approximately 8 km. AMSR-E is therefore used to monitor rivers in a manner similar to that of traditional ground-based gauging stations. Instead of mapping floods, the AMSR-E portion of River Watch (Figures 21 and 22) provides the capability to detect floods as they are occurring along predetermined river measurement sites, and to thereby monitor flood location, duration, and severity. With additional calibration provided by even Figure 21 Sample satellite gauging site display for River Watch. The AMSR-E remote sensing data has been transformed into estimated discharge. Flows above the red line, at this location, are identified as floods. Map displays (Figure 22) can show the sites within a region or watershed that have exceeded this threshold each day. Site ID: 40 Old no. 77 Lat River: Name: Latest measurement: Hydrologic status: Latest M/C ratio: Estimated current discharge: Ice-cover determination: Mekong 10 Oct-07 4 Major Flood 1: m 3 /sec. Long. Contributing area:. Seven day runoff: Period ending: km mm 18 Oct-07 Ground station information: GRDC# Nellie ( ) This site not yet calibrated to actual discharge or runoff values Estimated Discharge, June 19, 2002 Present (m 3 /sec.) Cambodia Location Map ,000 13,000 15,000 17,000 19,000 1-Aug Aug Aug Sep Sep Oct Oct Nov Nov Dec Dec Jan Jan Feb Feb Mar Mar Apr Apr May May Jun Jun Jul Jul Aug Aug Sep Sep Oct Oct-07 9-Nov Nov-07 9-Dec Dec-07 2-Jul-02 2-Oct-02 2-Jan-03 2-Apr-03 2-Jul-03 2-Oct-03 2-Jan-04 2-Apr-04 2-Jul-04 2-Oct-04 2-Jan-05 2-Apr-05 2-Jul-05 2-Oct-05 2-Jan-06 2-Apr-06 2-Jul-06 2-Oct-06 2-Jan-07 2-Apr-07 2-Jul-07 2-Oct-07 2-Jan-08 2-Apr-08 2-Jul-08 2-Oct-08 Estimated Discharge and Microwave data for Measurement and Calibration Sites, Aug 1, 2006 Present (The Discharge Estimation is AMSR-E measurement/encumbrance and radiance ratio at 36.5 GHz) AMSR-E radiance Measurement Target Calibration Target 5-year flood 1.33-year flood Estimated discharge (m 3 /sec.) 87

101 Sardinia Valencia Majorca Lesbos Ankara Agrinion Izmir Lisbon SPAIN Palermo ITALY TURKEY PORTUGAL Cordoba Sevilla Adana Tabriz Tunis Sicily Athens Algiers GREECE Peloponnesus Aleppo Mosel Gibraltar MALTA Rhodes TUNISIA Vallelta Nicosia Tehran Crete Rabat CYPRUS Beirut SYRIA Madeira Casablanca Damascus IRAN LEBANON Island Tripoli Tel Aviv IRAQ Baghdad Esfahan Amman Alexandria ISRAEL MOROCCO Basra JORDAN Cairo KUWAIT Shiraz Canary Islands SAUDI Kuwait ALGERIA ARABIA LIBYA Al Manamah BAHRAIN EGYPT QATAR Ad Dawhah Riyadh Abu Dhabi U. A. E. WESTERN SAHARA MALI MAURITANIA NIGER Nouakchott SUDAN ERITREA Sanaa Dakar SENEGAL Khartoum CHAD Asmara YEMEN Banjul Niamey GAMBIA BURKINA Bamako Suqutra Ouagadougou NIGERIA DJIBOUTI Bissau N Djamena Djibouti GUINEA BISSAU GUINEA BENIN Addis Abbaba Conakry SIERRA SOMALIA LEONE IVORY COAST Abuja Freetown TOGO Ibadan CENTRAL ETHIOPIA Lome AFRICAN Monrovia LIBERIA GHANA Porto CAMEROON REPUBLIC Accra Abidjan Novo Bangui Douala Yaounde EQUATORIAL GUINEA UGANDA Bata KENYA Mogadishu Libreville Sao Tome CONGO Kampala Mbandaka GABON DEM. REP. RWANDA Nairobi OF CONGO Kigali BURUNDI Brazzaville Bujumbura Kinshasa TANZANIA Matadi Zanzibar Island Dar es Salaam Luanda Likasi ANGOLA Lubumbashi Comoros Huambo MALAWI MOZAMBIQUE ZAMBIA Lilongwe Lusaka Harare MADAGASCAR Antananarivo ZIMBABWE Bulawayo NAMIBIA Windhoek BOTSWANA Walvis Bay Gaborone Pretoria Maputo Johannesburg Mbabane SWAZILAND LESOTHO Bloemfontein Maseru Durban SOUTH AFRICA Umtata Cape Town Agriculture and Rural Development Figure 22 Sample map display of global-coverage AMSR-E-based river discharge data. Red and purple dots are gauging sites experiencing flooding. Enhancements to such displays could include increasing map scale, adding drainage and cultural feature, and delineating floodplain lands. An international river basin such as the Ganges could, for example, be monitored in this way, with a denser array of measurement sites, and thus floodplain areas sensed as exceeding particular flood discharges and levels would be objectively defined. Data for: 26-Dec-07 Not online yet Normal Flood Low flow or ice-covered Major flood Zambezi River Save River Limpopo River A F R I C A Orange River widely scattered ground-based gauging stations, the remote sensing signal at these sites can be transformed to river discharge or to stage, 45 and the signal can also be compared to inundation imaged and mapped by MODIS at various intervals. The AMSR-E period of record is now long enough (six years) to reliably estimate the five-, 10-, or 15-year recurrence interval floods, using statistical assumptions such as the Log Pearson III distribution. With additional work, it will be possible to extend the records at each site (there are 2,583 sites now being measured) back into the mid-1980s using archived data. If accomplished, this would allow estimation of the 1-in-50-year events using the standard flood distributions. Finally, there is also the opportunity to make any operational system more robust by incorporating other sources of similar 88

102 data now available (e.g., from the NASA/JAXA Tropical Rainfall Monitoring Mission satellite and follow-on sensors). The, methodology does not depend on successful operation of any one sensor for its future utility. In short, the River Watch example indicates that it is possible for remote organizations to measure the onset of flood conditions anywhere in the world. If in-country ministries are also cooperative with data sharing, then the quality of such measurements can be enhanced (by, for example, improved and validated calibration to water discharge units). 46 Rapid flood mapping For the second task, rapid flood mapping, the prototype system uses the 250 m (visible and near infrared) bands of the two MODIS optical sensors (example in Figures 9 and also as a multi-year compilation in Figure 11). The steps needed to produce such maps are relatively straightforward and require only modest computational and software resources; they can, in principle, be accomplished by many organizations. They are as follows: (i) determination of the general region of flooding, either from news reports or from sensor measurements such as River Watch provides, (ii) searching online archives, which provide browse images, to select scenes where cloud cover is not obscuring the flooding, (iii) acquisition of the selected scenes (free via the Internet), (iv) geocoding and rectification of the portion of the scenes that include flooded land, (v) digital classification algorithm processing of these rectified subscenes and binning of pixels into land and water categories, (vi) GIS polygon fitting around the classified water areas, again using standard algorithms available in most remote sensing analysis programs, (vii) Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture validation of the outlined water areas by visual comparison to the unclassified image and removal of cloud shadow areas misclassified as water, (viii) lastly, integration of the final flood polygons into GIS workspaces that include other data layers, such as topography, cultural features, permanent water, and previously mapped floods to produce map displays showing the flooded lands. There are efforts underway by various remote sensing science groups to automate this methodology as much as possible. Due mainly to the vagaries of cloud cover, MODIS coverage of any particular flood event may be temporally dense, but irregular: imaging may be possible daily or twice daily for several days, and then several days may be obscured. However, where flooding occurs along a reach also monitored via AMSR-E, the River Watch system accurately determines the timing, magnitude, and duration of the flood peak and can match this discharge to the moreintermittent MODIS record of inundation. Figure 23 provides an example of 89

103 Agriculture and Rural Development Figure 23 Guide to predicting inundation display from River Watch. The archived GIS data base of MODIS-imaged flooding along a ~50 km river reach is compared with AMSR-E data from the same location (AMSR data are from gray line square shown on the maps). The mapped inundation is draped over a shaded relief map produced from global topographic data, which could also be used to model larger, more rare floods and floodplain lands. The map shown on the right is close to the 1- year or 10 percent annual excedence probability floodplain. Gaging reach: Old number: Current hydrologic status: Guide to predicting inundation 36 Irrawaddy Yenangyaung M/C 12-Dec-06 Normal flow Magwe Burma The current hydrologic status and discharge or C/M ratio can be used to estimate present inundation extent. Compare to the maps below: Note: 1) Where possible, maps show exact inundation extent/amsr-e measurement matches 2) Maps are based on water classification of MODIS images 17-Feb M/C 7-Aug M/C 25-Aug M/C Sinbyugyun Sinbyugyun Sinbyugyun Salin Salin Salin Yenangyaung Yenangyaung Yenangyaung Sagu Sagu Sagu Kilometers Minbu Magwe Kilometers Minbu Magwe Kilometers Minbu Magwe combining in this synergistic manner these independent sources of satellite data. With such information and displays as a guide, an analyst can: (a) predict the map extent of ongoing inundation based on the incoming AMSR-E data, even without any new MODIS information, (b) determine if particular flood thresholds have been reached, and (c) assess the severity of the event in comparison to all other events observed since mid-2002, when the AMSR-E data stream began. If MODIS mapping data are then obtained of the ongoing event, the River Watch system provides independent sensor validation, at 250 m spatial resolution, of the estimated flooding. Also, as the new AMSR-E flood information (the flood hydrograph) is incorporated into the annual flood peak time series, the system automatically can re-estimate flood recurrence intervals using the Log Pearson III distribution (or other distribution of 90

104 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture choice). This allows, in turn, statistical recurrence interval estimates to be matched to mapped flood inundations, thus creating flood risk zones. The issue of (lack of) stationarity of flood series, discussed in regard to gauging station flood series, also applies to River Watch data. However, the remote sensing-derived data record period is short and extends to the present: using such data favors acceptance of recent flood history as being indicative of the immediate future and because it is not mixed into a longer-term record. For analyses of these data, it appears that, particularly for determining excedence probabilities for relatively frequent flooding, the most defensible strategy is to assume stationary means and variance, and incorporate new flooding into the time series without adjustment (the recent past being taken as the best guide to the near future). Each time a new flood of record occurs, this approach will cause an upward revision of flood magnitudes and flood risk along the affected river valleys. Alternatively, one could connect the remote sensing record with the regional, longer-term record provided by gauging stations and/or use modelling-based regionalization approaches in the same manner, as is common with in situ gauging station data. Flood hazard assessment The previous examples illustrate how flood detection, near real-time flood mapping, and compilation of such maps over time can all be accomplished via remote sensing supported by available ground-station data and auxiliary data sets such as topography. The 6-year AMSR-E record allows for consistent estimation of the 10-year, or 10 percent annual excedence probability flood, on an automated basis, and as applicable to at least several tens of km upstream and downstream from the measurement site. As a general rule, the floodplain associated with this discharge will be approximated by the maximum inundation imaged by MODIS during this same period of record (in detail, this depends on the actual distribution of flood peaks so far measured). Each MODIS inundation image can also be matched to its coeval discharge value, and thus to its precise flood recurrence probabilities. As the length of this observational record lengthens, the capability to estimate future flood risk for less frequent events improves. Assuming an additional 4 years of record, to mid-2013, the 20-year flood discharge will be constrained, as well as its inundation extent in map view. Such data can also be combined with channel slope, cross sectional area, floodplain topography, and Manning s n roughness estimates, and standard hydraulic modelling techniques, to model inundation in detail, and verify n values: this provides for more reliable extrapolation to 100-year or other low-probability floodplain boundaries. Remote sensing thus offers a relatively efficient and direct path towards identifying both the flood risk of specific land parcels, and designing flood indices that signal when particular flood levels have been obtained and for how long. Finally, remote sensing data might interface with agricultural lands data for objective estimates of large-scale crop damage. Figure 24 is a recent Rapid Response Inundation Map produced at DFO using the MODIS technology already described. This is for 2008 flood of record (according to news reports) flooding in Thailand and Laos and for which international food 91

105 Agriculture and Rural Development Figure 24 Rapid Response Inundation Map for flooding in Laos and Thailand, August In this version of the GIS workspace output, the new flooding is shown in red, above dark blue colors that are the aggregated limits of previous floods; the red color is below the light blue color, which is permanent surface water. Another version could be made to display different information to the end user: by moving the red layer below the maximum observed inundation limit, only flood of record areas would remain in red. DFO Event # Thailand/Laos Mekong River - Rapid Response Inundation Map Universal Transverse Mercator - UTM Zone 48 North MODIS flood inundation limit Maximum observed inundation SWBD reference water: WGS 84 - Graticule: 2 degrees Copyright 2008 August 15, 2008: Limit : DCW Rivers: Urban Areas: Shaded Relief from SRTM data Dartmouth Flood Observatory Dartmouth College Hanover, NH USA Chris A Farmer, E.K. Anderson, G.R. Brakenridge Ban Ray Pak Lay an Hut Ta Phu Ban Na Le Muang Pa Chiang Khan Ban Houaykep Long Ban Houay Pamon Ban Keo Song Lay Ban Hatkham Ban Nong Khay Borikhan Tourakom Ban Keun Ban Thabok Ban Thana Ban Nong Ban Non Ban Tha Ngon Ban Pak Ton Ban Woen VIENTIANE NONG KHAI Ban Wan Ban Don Ban Kokkhay Ban Naxon Ban Hatkonong Ban Sopchat Ban Nape Ban Na lnh Noi Ban Pakha Ban Phayat Ban Phonkho Ban Thong-Noy Ban Thongkha Nam Theun Loei WANG SAPHUNG LOM KAO LOM SAK Ban Nam Duk Nua Ban Kok sa Miles UDON THANI Kranuan KHONKHAEN KALASIN MAHA SARAKHAM Ban Nakok MUANG KHAMMOUAN Ban Nadon Ban Nahoun Ban Nong Bok Ban Nongkhi Ban Khap Phuang Seno Don SAVANNAKHET Ban Ban Naphan assistance was requested (per communication to DFO from the U.N. World Food Programme). In regard to agriculture, Figure 25 illustrates the potential complexity of the field situation and also the value that could be provided by remote sensing such as from MODIS. It shows, at larger scale than Figure 24, a remote sensing-based cropland layer beneath the 2008 flooding, now provided as a partially transparent red shading. The resulting color symbology from adding the agricultural land layer is: red, flooded non-agricultural land; light grey, non-flooded agricultural land; and brown, flooded agricultural land. The associated GIS system can display the latter alone as a map product and also provide numerical measurements of total flooded agricultural land. If the area was subdivided into watersheds, or governmental regions or agricultural risk zone units, then, similarly, the output could include flooded agricultural land acreages within each such unit. Finally, given the capacity of MODIS to provide frequent coverage, duration of flooding can also be numerically addressed, and by programming the GIS to display: (a) agricultural land areas 92

106 Assessment.qxd:Assessment 3/4/10 7:32 AM Page 93 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 25 Comparison of the AVHRR-derived 1 km resolution agricultural land classification to the flooding in August 2008 in Laos and Thailand. Light blue is permanent surface water; intermediate blue are areas previously mapped as flooded (since 2000) and not flooded during this event; light grey is agricultural land; red is currently flooded, non-agricultural land; brown is currently flooded agricultural land; and dark blue is previously flooded agricultural land not currently flooded Kilometers Tha U Then Ban Nakok Muang Khammouan Ban Nadon Ban Nahoun Ban Nonn Bok Ban Kengmun Ban Nongkhiat-Lua Ban Khap Phuang intersecting, for example, flooding in two successive images acquired five days apart, and (b) agricultural land areas only flooded on the earlier image and thus with a shorter duration of flooding. Such comparisons could form an objective and defensible basis for empirical damage estimates; with data sources such as MODIS as input, it also appears that such comparisons could be performed quickly and economically. A.5. Three Examples of Remote-Sensing Based Flood Mapping This section presents the generation and use of different remote sensing products in support of flood risk management in the context of the three case studies described in Chapter 5 (Vietnam, Thailand, and Bangladesh). Unlike the satellite-based sensors used from wide-area flooding mapping described the previous sections, the data used here are of higher spatial resolution (providing information at the farm level), but their acquisition has to be scheduled a priori due to the large data volume arising from the high spatial resolution. These sensors utilize radar wavelengths (synthetic aperture radar 93

107 Agriculture and Rural Development [SAR]) and can penetrate clouds and multiple images from different dates where used to detect changes in surface water. Given the nature an availability of data the examples presented here focus on the mapping of flood extent and, in the Vietnam case, on the comparison between flood extent derived from remote sensing and flood modelling. Flood duration could not be estimated due to the lack of available data at the time of the case study. Data processing was performed in an automated way. At this stage, all data sets are in a selected cartographic reference system already embedded in a Geographic Information System (GIS). The integration with ancillary data is therefore straightforward. A.5.1. Vietnam Dong Thap, located 160 km southwest of Ho Chi Minh City, is an agricultural province whose economic activity is centered on rice production and aquaculture. Based on the local geography and topography, Dong Thap is classified into two areas of inundation (Figure 26). The shallow inundation area is located to the south of Nguyen Van Tiep canal. Flood control works that are likely to be built in this area include: canals, flood discharge culverts, ring dikes to protect industry, towns, and so on. The deep inundation area (blue color in Figure 27) is located to the north of Nguyen Van Tiep canal. Due to the deep levels of inundation experienced here, flood control measures are developed to protect the spring summer rice crop (until mid-august) and to construct flood control works that protect infrastructure within Dong Thap Figure 26 Classification of inundation areas for flood management in Dong Thap province, Vietnam. The polygons represent the boundaries of primary (thick) and secondary (thin) channels. Deep Inundation Area K. Nauren Ven Tiep Shallow Inundation Area 94

108 Assessment.qxd:Assessment 3/4/10 7:32 AM Page 95 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 27 Top left to bottom left: The first three images show time varying extent of inundated area based on satellite image acquisitions from August 5, August 18, and September 9, The red box indicates an example where significant spatial changes occurred. Bottom right: Inundation map showing flooded areas on all dates black, areas flooded on August 5 in cyan, areas flooded on August 18 and September 9 in red, areas flooded on September 9 in yellow. province without increasing the flooding affect either upstream in Cambodia or downstream in other Mekong Delta provinces. For demonstration purposes, three ENVISAT ASAR Wide Swath data (400 km by 400 km, at a resolution of approximately 85 meters) acquired on August 5, 95

109 Agriculture and Rural Development August 18, and September 9, 2005, were used (no additional acquisitions over this area were available during that time frame). The resulting multi-temporal color composite covering the whole Dong Thap area is shown in Figure 28 (top left). Flooded areas in all three dates are well identifiable as black areas, while gray lines correspond to the primary and secondary channels provided by Southern Institute for Water Resource Planning (SIWRP), 47 which created risk maps for Dong Thap as part of the ADB s project in Vietnam discussed in section 5.1 in Chapter 5. As a last step, the results derived from remote sensing were compared with flood model output for the same region. The Vietnam River Systems and Plains (VRSAP) model maintained by the SIWRP is designed to simulate onedimensional hydrodynamic river flow and quasi two-dimensional flow on Figure 28 Top: ENVISAT ASAR Wide Swath sample images of August 5 (left) 18 (center) and September 9 (right) 2005 with Dong Thap boundaries. Flooded areas are visible in dark. Bottom: ENVISAT ASAR Wide Swath color composite with average scenarii: minimum modelled depth (left) and maximum

110 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture floodplains. The following hydrological data sets are available for analysis and use as inputs into the hydraulic model of the Mekong Delta: Water level stations for the calibration of the VRSAP hydraulic model to historical floods; Tidal water level data for the tidal influence of the East Sea and West Sea, downstream of Dong Thap province; Flows into the upstream boundary condition of the hydraulic model at Kratie, Cambodia; Rainfall data within Dong Thap province and upstream in the existing hydraulic model catchments. Figure 28 shows the outcome for minimum and maximum modelled depth for an average scenario. Comparing these results with the information provided by Remote Sensing (refer also to Figure 27 in the bottom left), it can be observed that the VRSAP model scenario has a good agreement with the flood extent map identified in the multi-temporal SAR data. Note, however, that according to the VRSAP model the polygons are assumed to be uniformly flooded, which is not a constraint in the satellite-based map. A.5.2. Thailand Situated in the heart of the country some 350 km from Bangkok, Petchaboon borders on three regions (the North, the Central, and the Northeast). The central part of the province is on the Pasak River basin with mountain ranges running along both the western and eastern sectors. Because of the fertility of the land, Petchaboon has always been an agriculturally productive area. The Pasak River can be divided into four zones: the Upper Basin, the Petchaboon-section Basin, the Middle Basin, and the Lower Basin. The Upper Basin stretches from the river s source in the Petchaboon province. High and steep mountains are characteristic features. The Petchaboon section extends south from the Upper Basin through a landscape of alternating low hills and plains on both sides of the river used for orchards and rice cultivation. The Middle Basin is characterized by hilly terrain. It provides the location of the Pasak Dam and its reservoir. The Lower Basin features low-lying floodplains with fertile soil for agriculture and potential for irrigation development. Heavy rainfall brought in by the southwest monsoons usually occurs from May through October. Annual rainfall averages 1,250 mm/year, with a maximum of 255 mm in September. During the dry season, November through April, total rainfall amounts to only 150 mm, with only 8 mm in January. Consequently, there is a vast difference between the amount of rainwater runoff during wet and dry periods. In the rainy season, the runoff can reach 2.2 billion m 3, with a maximum of 920 m 3 in October. During the dry season, average runoff amounts to some 200 million m 3 per month with only 17 million m 3 in March. For demonstration purposes, two Radarsat-1 images (150 km by 150 km, at a resolution of 25 meters) acquired on September 16 and October 13, 2002, were used. The September image shows widespread flooding, which had receded by October (Figure 29). As in the Vietnam case, no additional acquisitions over this area were available during the given time frame. In addition to these 97

111 Agriculture and Rural Development Figure 29 Radarsat-1 samples (25 meter resolution) of September 16th (top left) and October 13th (top right) Flooding in the September image is visible as black areas, the October serves as a reference (no flood) image. A Landsat ETM+ image (30 meter resolution) from February 9th 2002 (bottom right) was used to map vegetation and permanently inundated areas as a baseline. The map derived differencing the two Radarsat-1 scenes (bottom left) shows inundated areas as a result of the September flooding. images, a cloud free Landsat ETM image acquired on February 9, 2002, only partially covering the Pasak River, was used to map natural vegetation and permanently inundated areas as a baseline (Figure 30). The Radarsat-1 images were reprocessed to have a constant reference height. By careful intercalibration of the images, flood extent and height could be estimated by differencing (subtracting the flood image from the non-flood image, while 98

112 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 30 Full view of the Pasak River valley (extent shown is approximately 135 km 38 km). Inundated areas as a result of the September 16, 2002 flooding are shown in blue tones (see caption for Figure 29 for explanation). accounting for permanently inundated areas), without using ancillary digital elevation data. Legend Classification Landsat ETM+ Green: vegetation and rangeland; brown and gray: barren land; blue: permanent water. Legend Classification Radarsat-1 Black: water-covered areas on both dates; blue: flooded areas on September 16, 2002 (the different blue tones correspond to the 99

113 Agriculture and Rural Development different flood severity); yellow and green: cultivated areas (the two colors correspond to the different growth); pink: uncultivated land. A.5.3. Bangladesh Bangladesh occupies one of the biggest deltas in the world and has an area of about 150,000 km 2. It enjoys sub-tropical monsoon climate and experiences annual average precipitation of 2,300 mm, varying from as little as 1,200 mm in the west to over 5,000 mm in the east. It has 230 rivers including 57 international rivers. Among them, 54 rivers originate from India including three major rivers (the Ganges, the Brahmaputra, and the Meghna). Three minor rivers originate from Myanmar. The rivers both big and small gradually became incapable of draining the huge quantity of silt-laden run-off passing through them during the monsoon period and cause floods. Inundation to the extent of 20 percent area of the country is beneficial for crops and ecological balance. But the flood of more than 20 percent causes direct and indirect damages and considerable inconvenience to the people. The country is extremely flat with low land relief with only a few hills in the southeast and the northeast part of the country. Generally, ground slopes of the country extend from the north to the south and the elevation ranging from 60 m to 1 m above mean sea level at the coastal areas in the south. River floods are an annual phenomenon ranging from April to October with the most severe flooding occurring during the months of July to August. The floods of 1988, 1998, and 2004 were catastrophic, resulting in large-scale destruction and loss of life. For demonstration purposes, two different examples are shown. The first one refers to the 2004 flood event, where the entire country has been flooded. The second one illustrates, for a selected area located around 70 km northeast of Dacca, recurrent significant flooding during In both cases, Wide Swath data (400 km by 400 km, at a resolution of around 100 m) were used in order to cover the entire extent. It is worth mentioning that in Bangladesh, even for limited flooding, inundation can easily reach an extent of 100 km by 100 km. In these cases the standard image acquisition mode would not be sufficient to map on the same moment the entire area. Figure 31 illustrates the flood extent on July 25, 2004, derived from ENVISAT ASAR Wide Swath data. Due to the availability of pre- and post-event ASAR data, using data differencing, it was possible to discriminate between permanent water (dark blue) and heavy (blue)/light (bright blue) flooded areas. Note that the additional differentiation i.e., heavy/light flooded areas could be performed by merging the flood extent map with the Digital Elevation Model (DEM). Areas where flood depth was estimated to be less than a few meters were classified as light flooded. Remaining areas have been assigned as heavy flooded. For the second example four ENVISAT ASAR Wide Swath images acquired on May 14, 2007, May 30, 2007, July 4, 2007, and November 21, 2007, were used. Here, due to the evident and recurrent large inundated extent (around 100

114 Assessment.qxd:Assessment 3/4/10 7:33 AM Page 101 Assessment of Innovative Approaches for Flood Risk Management and Financing in Agriculture Figure 31 Flood extent map of July 25, 2004, combined with days above danger level for 2004 (bar chart) provided by BWDB. Panchagarh Thakurgaon Nilphamari Lalmonirhat Kurigram Dinajpur Rangpur Gaibandha Sherpur Jaipurhat Sylhet Netrokona Naogaon Sunamganj Jamalpur Bogra Nawabganj Mymensingh Rajshahi Habiganj Sirajganj Nator Tangail Gazipur Pabna Meherpur Narsingdi Manikganj Kushtia Chuadanga Jhinaidah Brahmanbaria Dhaka Rajbari Narayanganj Faridpur Munshiganj Comilla Magura Jessore Shariatpur Madaripur Narail Gopalganj Chandpur Khagrachari Laksmipur Barisal Satkhira Moulvibazar Kishoreganj Feni Noakhali Khulna Chittagong Bagerhat Pirojpur Rangamati Patuakhali Jhalkati Bhola Borguna Bandarban Cox's Bazar 10,000 sq km in July), the three flood maps were generated by simply subtracting the images on the flood dates from the reference date when the rivers show the normal situation (May 14 illustrated in Figure 32 in the bottom right). Comparing this product type with the Vietnamese one (Figures 27 and 28), it is evident that the level of detail here is coarser, and the maps emphasize the overall extent of each flood event rather than the detailed patterns. 101

115 Agriculture and Rural Development Figure 32 Inundation maps derived from ASAR Wide Swath samples (100 m resolution, approximately 120 km by 120 km) for June 30 (top left), July 4 (top right), and November 11 (bottom left), As a reference image (i.e., non-flooded situation), the May 14, 2007 (bottom right), has been selected. The blue color represents the flooded areas during the three events. 102