Windstorm Information Service CGI IT UK Ltd SIS Workshop, Warbrook House 17-19 October, 2016 Implemented by
Project Team
What are windstorms, why do they matter to the Insurance Industry? Windstorms = Extra tropical cyclones Infrequent, large storms, capable of causing widespread damage to property and large insurance losses. E.g. Storm Daria $8.2bn (indexed to 2012).
Insurance Risk (and Catastrophe Modelling) Re-insurers aggregate risk from many insurers and have large exposure to windstorm risk. (Re-)Insurance industry needs to understand the risk from windstorms (insuring for 1 in 200 year events). Currently only 35 years of reliable historical data is available, including around 70 major storms, making risk difficult to assess. Insurance industry is supported by specialist cat modelling companies. As historical data is limited Cat Models use synthetic event sets large numbers of storms. Methods for generating event sets are proprietary. aims to provide transparent, authoritative data to support understanding of windstorm risk. 4
Overview of the supply process Met Services Weather observations Previous Model field Analysis Model fields ECMWF ERA-INT / 20C Re-analyses Users Storm tracks Storm footprints Stochastic Event set Cat Modellers Tier 1 Tier 2/3 Storage and presentation Windstorm Hazard Exposure / Vulnerability Adaptation strategies Insurers / re-insurers 5
Existing services Extra-Tropical Cyclones less well covered than Tropical Cyclones Existing hazard services Re-analysis limited production due to complexity / processing Storm analysis Covered by XWS using ERA-INTERIM (1979 2014) Cat modellers develop stochastic event sets Insurers also analyse to an extent dependent on organisation scale Hazard indicators Most services present information at the summary level Existing exposure / vulnerability / loss Insurers and re-insurers undertake own analysis General services also provided some shared eg PERILS 6
Gaps Need for a longer, more comprehensive time series Transparency of methods applied is very important Additional downscaling Recognised that some prioritisation of sub-regions and storms will still be necessary to use the resources as efficiently as possible. Concentrate on highest levels of financial exposure which will be defined by both the countries covered and the availability of windstorm cover in each. User customisation important - eg ability to rank and flag storms according to a range of criteria Standard approaches to storm characterisation, such as those applied in XWS are considered valuable, though alternative methods are also considered useful for comparison. Support for in-house simulation platforms as well as cat modelling companies Improved support to stochastic modelling process so that the output event sets are clearly dynamically realistic. Interest in the possibility of a event set Validation support to Cat modelling companies 7
User Interviews User Sector Method Team Oasis Insurance Face-to-Face CGI XL Catlin Insurance Face-to-Face CGI, UoR, UKMO Munich Re Insurance Face-to-Face UKMO Aspen Re Insurance Webex / Teleconf CGI, UoR, UKMO Atkins Civil Engineering Face-to-Face CGI, UoR, UKMO Willis Insurance Face-to-Face CGI, UoR, UKMO AON Benfield Insurance Webex / Teleconf CGI, UoR, UKMO Chaucer Insurance Webex / Teleconf CGI, UoR, UKMO Applied Impact Research (AIR) Catastrophe Modellers Face-to-Face CGI, UoR, UKMO
User Requirements Authoritative, comprehensive, free and open Historical Records to support Understanding of major storms Understanding of present risk Development of cat models, Validation of cat modelling results Experienced users will typically download all data and use off-line show limited interest in indicators show limited interest in visualisation Insurance Users showed limited interest in future climate change projections (also rather uncertain). 9
data products Historical Storm Tracks Database From ERA-INT and ERA-20C Longer time series 1940 present 1900 1939 if reanalysis data quality is sufficient Historical Storm Footprints From ERA-INT and ERA-20C Coverage of at most major storms Bias corrections to be applied Validated against scatterometer and local observations 10
data products (II) Tier 1 Indicators - Historical indicators will provide a quick comparison of ERA-20C to ERA-Interim for re-insurers: Number of cyclones in a given decade Typical intensities, and intensities of extremes, of cyclones in a given decade For ERA-20C only: Decadal variability Tier 3 Indicators Total Insurance losses due to windstorms 11
Products (III) Event set will be based on ensemble runs of UPSCALE climate model giving 130 model years of current climate conditions will attempt to build a windstorm risk model based on the industry-supported, open source OASIS platform. Challenge is to develop a robust vulnerability model that matches industry historical loss data. 12
13
Additionality Use of latest reanalysis datasets (ERA-20C as well as INT) Longer, more comprehensive time series (1900 or 1940 fwd) Lower thresholds so more storms available for analysis Additional downscaling 4km as well as 25km resolution footprints Some prioritisation of sub-regions and storms still necessary Transparency of methods applied Improved validation and re-calibration of data Reference products to support in-house simulation platforms Event set will allow for cross checks with commercial cat models Support for insurance users with less in-house capabilities
V0 Prototype: Portal (1) Sub-site C3S portal
V0 Prototype: Portal (2) Explore for data display
V0 Prototype: Portal (3) Display track data - under construction
Storm Tracks Adrian Champion University of Reading Implemented by
Historical Catalogue Re-insurers require a dataset of historical windstorms to understand risk of future windstorms: Validate against data from catastrophe models Intensity distributions Numbers of cyclones Cyclone intensity Compare to historical records Investigate specific well known events Improve their understanding of extreme events
Historical Catalogue ERA-Interim (1979 to present) High resolution Frequently used by re-insurers and catastrophe modelling companies As used in the European windstorms catalogue ERA-20C (1900 to 2010) Much longer period -> larger sample size Relatively high resolution New dataset, requires analysis Catalogue of around 150 events in a given year
Historical Catalogue Hodges (1994,1995,1999) tracking algorithm Based on 850hPa relative vorticity at T42 resolution Vorticity centres used to calculate trajectory of individual extratropical cyclones (cyclones that occur north of 30N at any point) Extra fields referenced back to vorticity fields at full resolution at each timestep Minimum MSLP within 6 degrees of vorticity centre Maximum wind within 6 degrees of vorticity centre Maximum land-wind within 3 degrees of vorticity centre As used in the European windstorms catalogue Filters require cyclones to last 2 days and travel 1000km
Historical Catalogue
Historical Indicators ERA-20C relatively new re-analysis product Unknown to re-insurers Still be analysed by scientific community Re-insurers want to know how reliable the data is Part of the work done by UoR will be to assess ERA-20C Historical indicators will provide a quick comparison of ERA-20C to ERA-Interim for re-insurers: Number of cyclones in a given decade Typical intensities, and intensities of extremes, of cyclones in a given decade For ERA-20C only: Decadal variability
Climate Change Projections Re-insurers not interested in climate change projections: Policies renewed yearly/3 years Adjust pricing every year Re-insurers are interested in understanding the general trend of the intensity distributions, to protect long-term business: Publication looking at the current understanding of the impact of climate change on North Atlantic storm track