Contributors: Toon Haer (VU), Janet Wijngaard (KNMI), Alan Whitelaw (CGI)
Issued by: Toon Haer (VU)
Issued Date: 21/06/2021
Ref: C3S_D435_LOT3_KNMI_2020 - Tier-3 (risk) indicators description
Official refence number service contract: 2020/C3S_435_Lot3_KNMI
Acronyms
1. Introduction
1.1. Executive Summary
The Tier 3 indicators, described as such because they describe socio-economic impacts of the storms, are risk and loss indicators. These indicators have been built up using open-source information, with OpenStreetMap building level data as a basis. Each building is associated with a land cover type, a building type and related to these, a reconstruction type. Fragility curves, also from the public domain, are then applied to link the building type to the impact of the storm severity. This information is mapped to the NUTS3 statistical regions associated with the buildings and presented in these NUTS3 categories.
This document describes the Risk indicators; loss indicators are described in an equivalent Product User Guide. The main difference with the loss indicators, is the Risk indicators use a synthetic event set for storm footprints from the C3S Synthetic windstorm events for Europe from 1986 – 2011 from climate projections dataset, while the loss indicators use footprints for historical storm tracks derived from ERA5. Risk, expressed as Expected Annual Damage (EAD), is calculated by using a subset of 1600 synthetic storm tracks from the synthetic event set. As a Tier 3 indicator, the Risk indicator complement the other windstorm products as it describes expected yearly socio-economic impacts of windstorms within Europe. These risk indicators combine original and modelled climate data with additional geospatial and socio-economic information in a way that allows their use by the insurance sector. The risk indicator is complementary to the other C3S Windstorm data products. Particularly the risk data which provides an estimation of the potential losses that would occur within a particular location within a typical year. To aid comparison both the loss and risk data is presented in the same format.
1.2. Scope of Documentation
This document describes the C3S Tier 3 Risk indicators using the standard C3S format for product descriptions, i.e., in terms of product target requirements, product overview, input data and method. It is based on the earlier proof of concept WISC contract documents, particularly 'WISC Risk and Loss Indicator Descriptions' produced by Elco Koks on 21-07-2017. The approach is further described in peer-reviewed publication (Koks & Haer, 2020).
1.3. Version History
Preceding the operational stage of the Windstorm Service for the Insurance sector, the pre- operational stage WISC1 successfully demonstrated the estimation of economic losses for winter storm events over Europe, based on state-of-the-art numerical weather prediction models and economic loss models. The service applies a chain of models, from models to generate Tier 1 windstorm footprints to an economic model for estimating Tier 3 economic losses, where the latter uses the windstorm footprints as input. While for loss indicators (damage per storm) the storm footprints where updated, the risk indicators (average damage per year) are based on the synthetic event set that is unchanged since the pre-operational stage.
2. Product Description
2.1. Product Target Requirements
Extreme wind events are among the costliest natural disasters in Europe, causing severe damages every year. For damage estimates, the community mostly relies on post-disaster data, which is often not publicly available. A few approaches offer more generic tools, but again these are often based on non-disclosed data. To offer a generic, high-resolution, reproducible, and publicly accessible tool, this dataset presents estimates from a wind damage model that is built around publicly available hazard, exposure, and vulnerability data. The model is used to provide the current dataset that assesses building damages related to extratropical storms in Europe, but the methodology is applicable globally, given data availability, and to other hazards for which similar risk frameworks can be applied. The model is distributed as an open-source model to offer a transparent and useable windstorm damage model to a broad audience, and the dataset is provided through the Climate Data Store (CDS).
2.2. Product Overview
2.2.1. Data Description
Table 1: Overview of key characteristics of the Tier 3 windstorm risk indicators.
Data Description | |
Dataset title | Tier 3 windstorm risk indicators |
Data type | Risk indicators |
Topic category | Natural risk zones |
Sector | Insurance |
Keyword | Windstorm losses |
Dataset language | English |
Domain | Europe, for 21 countries (Austria, Belgium, Czech Republic, Denmark, Estonia, Spain, Finland, France, Great Britain, Germany, Ireland, Italy, Lithuania, Luxembourg, Latvia, Netherlands, Norway, Poland, Portugal, Sweden, Switzerland)
|
Horizontal resolution | Risk values are given for each identified building footprint |
Temporal coverage | Synthetic tracks |
Temporal resolution | Risk values represent the Average Annual Damage |
Update frequency | None (static dataset) |
Version | n/a |
Model | high-resolution wind damage model for Europe |
Experiment | n/a |
Provider | Vrije Universiteit Amsterdam |
Terms of Use | OpenStreetMaps Data made available through the Open Data Commons Open Database License (ODbL) was used in the development of the Tier 3 Loss and Risk indicators. Therefore, works produced from OpenStreetMap, need to attribute the Open Database License (ODbL) https://cds.climate.copernicus.eu/api/v2/terms/static/licence-to-use-copernicus-products.pdf |
2.2.2. Variable Description
Table 2: Overview and description of variables.
Variables | |||
Long Name | Short Name | Unit | Description |
Building | Building | String | Type of building according to OSM. Varies from only "yes" it is a building, to the actual building type. |
Identifier | RISK_ID | Integer | Unique ID of the building. It is a combination of the NUTS ID and the index. |
COUNTRY | COUNTRY | string | The country in which the building is located. |
LATITUDE | LAT | Float16 | Latitude of the centroid of the building. |
LONITUDE | LON | Float16 | Longitude of the centroid of the building. |
LANDUSE CLASS | CLC_2012 | Integer | Land-use code corresponding to the Corine Land Cover classification. |
AREA | AREA_m2 | Float16 | Area of the building footprint. |
Risk estimates | EAD | integer | Risk value for the building for one year |
2.3. Input Data
Table 3: Overview of climate model data for input to Risk indicators, summarizing the model properties and available scenario simulations.
Input Data | ||||
Model name | Model center | Scenario | Period | Resolution |
OpenStreetMap | © OpenStreetMap-auteurs | n/a | n/a | Building level |
CORINE | Copernicus Land Monitoring Service | n/a | CLC 2012 | 30" |
PAGER construction type building stock | U.S. geological service | n/a | n/a | n/a |
Synthetic storm tracks | Climate Data Store | n/a | n/a | 25km x 25 km |
2.3.1. OpenStreetMap (OSM)
All building footprint data are extracted from OSM, which has proven to be the most extensive dataset of publicly available building footprints for Europe. OpenStreetMap is a free, editable map of the whole world that is being built by volunteers largely from scratch and released with an open- content license. The OpenStreetMap License allows free (or almost free) access to our map images and all of our underlying map data. As OSM is user driven, it continuously evolves and improves, improving the building footprint coverage across Europe.
2.3.2. CORINE
CORINE was developed by the European Environmental Agency and distinguishes between 45 different land-use classes, with a known percentage of residential, commercial, industrial, and other land-use classes. This CORINE dataset is used to classify OSM buildings into different sectoral land- use classes; agriculture, residential, commercial/industrial, and transport.
2.3.3. PAGER
PAGER, a US Geological Survey project, stands for the Prompt Assessment of Global Earthquakes for Response. Overall, it is an automated system that takes in seismic data from remote sensors to estimate earthquake shaking and the scope and impact of earthquakes around the world. As part of this, PAGER links with the World Housing Encyclopedia to produce a global database of building stocks. While designed to assess vulnerability to earthquake shaking, the construction type information also provides characteristics that can be used directly to assess windstorm vulnerability.
2.3.4. Synthetic storm tracks
The C3S Synthetic windstorm events for Europe from 1986 – 2011 from climate projections is a synthetic event dataset for windstorms, based on data from the UPSCALE project ("UK on PRACE: weather-resolving Simulations of Climate for globAL Environmental risk"). UPSCALE data cover the period 1985 to 2011 and were produced using the HadGEM3 GA3 and GL3 configurations of the MetUM (Met Office Unified Model) operating at 25km resolution. The project developed five ensembles, some of which are based on RCP 8.5 scenarios, though only standard ensembles were used. The unit of wind intensity, as for the WISC historical footprints, was the maximum 3s gust speed at 10m within a 72-hour period.
The synthetic event set is therefore a physically realistic set of plausible events based on the climatic conditions of the period from 1985 to 2011. It is not designed to reproduce actual historical events of this period, as there is no data assimilation process used to align the model to historical observations, but the simulation ensembles used historical forcings such as sea surface temperatures.
The simulation in each of the five ensemble members has evolved independently from the others, from different starting points. The dataset is published in the climate Data Store, please refer to the entry for additional details.
2.4. Method
2.4.1. Background
The building level loss estimates are calculated using a conventional risk modelling framework (Fig. 1), where we define risk as a function of hazard – the probability and strength of an event with potential to cause harm; exposure – the value of assets subject to the hazard; and vulnerability – the susceptibility of the asset to hazards of a given severity. An overview of the modelling approach is show in Figure 1 and further described in 2.4.2. Different from the loss indicators, the risk indicators use the synthetic storm tracks (ref) as input.
Figure 1: Overview of the various steps for the loss estimations (Koks, Tiggeloven, et al., 2017)
2.4.2. Model / Algorithm
Hazard – The hazard input was the C3S_WISC_Synthetic_EventSet as described in Section 2.3.4. The unit of wind intensity, as for the WISC historical footprints, was the maximum 3s gust speed at 10m within a 72-hour period. For the risk calculations, a subset of 1600 synthetic tracks from the event set were used.
Exposure - The exposure maps are generated by combining CORINE and OpenStreetMap (OSM) data. OSM data provides building footprints, which can be extracted to shapefiles. While the OSM data provides good coverage for most countries, there might still be buildings missing for others. Despite this limitation, the OSM database offers the most consistent and extensive building dataset for Europe. Since the OSM database is constantly being updated, the model always extracts the most up- to-date OSM data. The OSM data is combined with CORINE to categorize buildings per sector. CORINE is developed by the European Environmental Agency and distinguishes between 45 different land-use classes, with a known percentage of residential, commercial, industrial, and other land-use classes.
This CORINE dataset is used to classify OSM buildings into different sectoral land-use classes; agriculture, residential, commercial/industrial, and transport. Figure two shows in further detail how the exposure maps are created, and how they are used to assess losses by using vulnerability data.
Vulnerability – The vulnerability is described as the susceptibility of an exposed asset to the hazard. This susceptibility is captured in so-called fragility curves, that show the relation between a certain wind speed, and the percentage damage done to an asset which in this case are buildings. We use the windstorm fragility curves by Feuerstein et al. (2011), which distinguishes between different building construction types: (I) weak outbuildings, (II) outbuilding, (III) strong outbuilding, (IV) weak brick structure, (V) strong brick structure, (VI) concrete building. When the fragility curve is combined with an estimated value of reconstruction costs, it can be translated in a vulnerability curve, which describes the damage done to a building at a certain wind speed. For this model, the reconstruction costs are taken from a study by Huizinga and De Moel (Huizinga et al., 2017). These values are corrected for each country following the differences in GDP, and corrected for regionally by the difference between national GDP and regional GDP. The OSM database used for extracting exposed building footprints does not provide information on different building construction types.
Figure 2: Example of loss calculation. Source: (Koks, Tiggeloven, et al., 2017)
To be able to distinguish between construction types for different countries, we use the PAGER database. The PAGER database defines 106 different building types for each country, which are aggregated to the 6 different building types considering in Feuerstein et al. (2011). Most of the European buildings fall into the latter two categories. Using the PAGER database, we obtain the share of each of the building types within a country (for example, 5% weak outbuildings, 30% strong brick structure, and 65% concrete building). The hazard, exposure, and vulnerability data is overlaid to obtain loss estimates. Loss estimates are made for each building footprints, for each vulnerability curve for different building construction type and multiplied by its relative share within the country. The loss estimates are aggregated for each NUTS3 region, for each country, and for each sector.
Risk - With the use of the event set, an approximation of the risk can be estimated. By using the estimated losses for the initial 1600 available storms of the synthetic event set, the approach sets up a frequency distribution of the losses. Setting up the frequency distribution allows to obtain an estimation of the probability of occurrence of a specific storm within a year. Consequently, the probability-loss distribution of the losses allows us to estimate the risk. A common approach to estimate this risk is to integrate the damage over all probabilities. This is done by calculating the cost for n return periods where both hazards and vulnerabilities have been calculated (Olsen et al. 2015). To do this, the trapezoidal rule is often used, leading to the following equation:
\[ EAD = \frac{1}{2} \sum_{i=1}^n \left( \frac{1}{T_{i}} - \frac{1}{T_{i+1}} \right) + (D_{i}-D_{i+1} \ast D_{i} \]In equation, n is chosen so that all relevant return periods are covered from negligible cost of quite frequent events to very rare events. 𝑇 are the return periods and 𝐷 are the damages per storm for each building. It should be noted that for most storms, the damages will be zero.
2.4.3. Validation
Risk is calculated based on the synthetic storm tracks, and therefore validation of loss estimates on observed damages is not possible. However, the loss estimates for observed storm tracks are validated as described in Koks & Haer (2020), showing the performance of the model. We provide a summary here for the SA of the loss estimates. The SA is performed in a Monte Carlo modelling framework following Crosetto et al. (2000) and Helton (1993). to investigate uncertainty and sensitivity related to the parameters. The following steps are performed: (1) assigning distributions to input parameters, (2) generating samples of different combinations of input parameters, (3) evaluating the model using the generated combinations of input parameters, and (4) analysing the results for uncertainty and sensitivity. Using SAlib, a publicly available Python library (Hermann, 2017), we perform a Delta Moment-Independent Measure (DMIM) analysis, as developed by Borgonovo (2007) and Plischke et al. (2013). This type of sensitivity analysis can be interpreted as a global sensitivity indicator which looks at the influence of input uncertainty on the entire output distribution without reference to a specific moment of the output (moment independence) and which can be defined also in the presence of correlations among the parameters. In total, we set up a set of 5000 different combinations of parameter values, focusing on the fragility curves. The sensitivity analysis shows that for each country/storm combination, the fragility curves are the most important driver of the results.
Validation of the synthetic tracks which are used as input for the risk estimates is described as part of the case studies undertaken during the WISC Proof of Concept project.
3. Concluding Remarks
This dataset is built using a high-resolution damage model to estimate the damages to buildings due to extratropical windstorms in Europe. The approach provides flexibility in the derivation by developing the vulnerability curves from building level upwards. The approach is particularly valuable to support insurers' and academic assessments for post-disaster quick-scans and estimates of potential wind damage towards the future, allowing them to use an open-source and transparent approach. While we demonstrate the methodology on a continental scale, it is not bound by a geographic region, and thus can be applied globally provided that data is available.
References
Borgonovo, E. (2007) A new uncertainty importance measure. Reliab. Eng. Syst. Saf. 92, 771–784.
Crosetto, M., Tarantola, S. & Saltelli, A. (2000) Sensitivity and uncertainty analysis in spatial modelling based on GIS. Agric. Ecosyst. Environ. 81, 71–79.
EEA. CORINE Land Cover 2012 (2014). Copenhagen, Denmark: European Environmental Agency.
Helton, J. C. (1993) Uncertainty and sensitivity analysis techniques for use in performance assessment for radioactive waste disposal. Reliab. Eng. Syst. Saf. 42, 327–367.
Herman, J. & Usher, W. (2017) SALib: an open-source Python library for sensitivity analysis. J. Open Source Softw. 2.
Huizinga, J., de Moel, H., Szewczyk, W. & others (2017). Global flood depth-damage functions: Methodology and the database with guidelines.
Koks, E. E., & Haer, T. (2020). A high-resolution wind damage model for Europe. Scientific Reports. https://doi.org/10.1038/s41598-020-63580-w
Plischke, E., Borgonovo, E. & Smith, C. L. (2013) Global sensitivity measures from given data. Eur. J. Oper. Res. 226, 536–550.
Olsen, A. S., Zhou, Q., Linde, J. J., & Arnbjerg-Nielsen, K. (2015). Comparing methods of calculating expected annual damage in urban pluvial flood risk assessments. Water, 7(1), 255-270.
Roberts, J. F. et al. (2014) The XWS open access catalogue of extreme European windstorms from 1979 to 2012. Nat. Hazards Earth Syst. Sci.14, 2487–2501.
Waisman, F. (2015) European windstorm vendor model comparison. in Slides of a presentation at IUA catastrophe risk management conference, London 30