Contributors: Hamish Steptoe (Met Office), Alan Whitelaw (CGI)
Issued by: Hamish Steptoe (Met Office)
Issued Date: 29/03/2018
Ref: C3S_435_Lot3 C3S Windstorm Synthetic Event Set_v1.1
Official refence number service contract: 2017/C3S_WISC_Lot3_CGI/SC2
Acronyms
1. Introduction
1.1. Executive Summary
The C3S_WISC_Synthetic_EventSet is a synthetic event set 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 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 final version of the synthetic event Set took the original event set and added two additional sets as described below. These three sets can be used together to provide an overall event set of 22,980 synthetic storms:
- Original synthetic event set: this contained 7,660 events and was recalibrated against the historic event set using four historical windstorm events; Xynthia, Kyrill, Daria and 87J.
- Synthetic set 2: Rather than downscaling against the cumulative distribution function (CDF) of named events Xynthia, Kyrill, Daria and 87J, a storm severity index combining maximum wind gust speed and land area above a threshold of 25 m/s was used to select the strongest six events from the Met Office historical events database. Events selected were for 19/03/1995, 26/02/1989, 25/03/1986, 24/11/1984, 22/03/1983 and 01/02/1983.
- Synthetic set 3: Downscaling for this set was based on station observations from the four named storms used for set 1.
The historical event set mentioned above refers to the historical windstorm footprints produced during the original C3S_441_Lot3 (WISC) project. These were dynamically downscaled by the UK Met Office from ERA-Interim data from 1979 onwards and ERA-20C data prior to 1979. These footprints, and those of the synthetic event set, were all provided in NetCDF format.
1.2. Scope of Documentation
This document describes the C3S WISC Synthetic Event Set using the standard C3S format for product descriptions, ie in terms of product target requirements, product overview, input data and method. It is based on the earlier WISC project documents, particularly C3S_441_Lot3_WISC_SC2-D3.2.1-CGI- RP-17-0080 (C3S WISC Event Set) produced by Hamish Steptoe on 28/07/2017 and updated to v1.1 on 18/12/2017 to describe the enlarged event set.
1.3. Version History
As noted in the executive summary above, the 'original' event set consisted of only the 7,660 events recalibrated against the historic event set using four historical windstorm events; Xynthia, Kyrill, Daria and 87J. In early use, it was considered that the set was too small. Accordingly two additional datasets were added with calibration against the six strongest storms in the historical set and also against station observations for the four storms used to calibrate set 1. These three sets were designed to be used together to create an overall event set with 22,980 storms overall (ie 3 x 7,660).
Note that some minor updates had been applied to the original dataset before the new elements were added. This is the reason that Set 1 files were originally referred to as v1.2 while the files in sets 2 and 3 were denoted as v2.0 and 3.0 respectively.
2. Product Description
2.1. Product Target Requirements
The WISC Synthetic Event Set was developed as a comparator for the stochastic event sets generally used for windstorm risk analysis in the insurance industry. The aim was to derive a comparable set whose method of derivation was openly available and whose characteristics were physically plausible based on this derivation. The aim was therefore to complement rather than replace existing sources of information by providing perspective and checks.
The rationale for the development of the set was based on a series of discussions with insurance companies during the early user requirements stage of the WISC project and some case studies for the use of the event set were provided as deliverable documents of the project. The event set was also used as the basis for the WISC European windstorm risk assessments at NUTS3 level developed by VU Amsterdam based on open source information.
2.2. Product Overview
2.2.1. Data Description
The C3S WISC synthetic event set consists of 22,980 windstorm footprints in NetCDF format, an example of which is provided in Figure 1. The data are a physically realistic set of plausible events, representative of the period from 1985 to 2011 but not designed to replicate this historical period directly. The event set does not represent variants of real historical storms.
Figure 1: Example synthetic event set footprint showing an event with higher SSI values
Table 1: Overview of key characteristics of the Windstorm Synthetic Event Set
Data Description | |
Dataset title | Windstorm Synthetic Event Set |
Data type | Gridded |
Topic category | Natural risk zones |
Sector | Insurance |
Keyword | Windstorm footprints |
Dataset language | eng |
Domain | Europe defined as follows:
|
Horizontal resolution | 4.4km x 4.4km |
Temporal coverage | The event set is based on data from the following historical period 1985-02-01 to 2011-05-25. Note however that the event set data themselves are not strictly historical representations |
Temporal resolution | Maximum 3s gust for the 72-hour period shown at each grid point based on hourly assessments |
Vertical coverage | Single level |
Update frequency | None (static dataset) |
Dataset version | 1.0 |
Model | Met Office HadGEM3 Global Atmosphere 3 (GA3) |
Experiment | UPSCALE Project |
Provider | UK Met Office |
Terms of Use | Licence to use Copernicus products |
The model outputs 10 m wind speed (m01s03i227) as a 6-hourly variable based on hourly maximum from a spherical Earth, rotated-pole Arakawa C-grid. The global UPSCALE domain is reduced to a limited area domain spanning 25°W to 40.5°E and 34.4°N to 71.5°N (see Fig. 1), unrotated and regridded from 25 km to 4.4 km to match the WISC historical footprint resolution on the WGS84 datum using linear interpolation.
It should be noted that due to issues with the running of the original UPSCALE model there are some periods of missing windstorm data in the resulting event set. These are as follows:
- Oct 1988 - Nov 1998, (Ensemble member 4)
- Dec 1988 - Mar 1990, (Ensemble member 3)
- Oct 2002 – Nov 2002, (Ensemble member 4)
- Dec 2008 – Feb 2010, (Ensemble member 5)
Figure 2: Comparison of domains used for the WISC project: event set domain (pink) and historical footprint domain (blue), with the land-sea mask used for calculating the storm severity index (SSI) (dark red).
Extra tropical cyclone tracking is performed using the XTCTRACK algorithm (Hoskins and Hodges 2002; Hodges 1999; 1995), as in Roberts et al. (2014), and uses 850 hPa relative vorticity. Events are identified based on a 72 hour window centred on the time of maximum vorticity. Only tracks that have their maximum vorticity occurring within the event set domain (as in Fig. 1) are retained. Where multiple events occur within a 72 hour period, only the track with highest relative vorticity is retained to filter out multiple tracks that may belong to the same synoptic system. Figure 2 gives a graphical representation of the event set, showing the temporal distribution of storms for each ensemble member.
Table 2 summarises the filtering process for each ensemble member: the total number of events that meet these filtering criteria is 7,660.
Table 2: Summary of remaining events after each step of the filtering process.
Ensemble Member | ||||||
1 | 2 | 3 | 4 | 5 | Total Events | |
Before filtering | 35,583 | 35,481 | 33,650 | 34,550 | 33,628 | 172,892 |
Domain filtering | 4,039 | 3,934 | 3,780 | 3,820 | 3,723 | 19,296 |
Temporal filtering | 1,594 | 1,571 | 1,484 | 1,517 | 1,494 | 7,660 |
As part of the regridding process, event footprints are recalibrated to match the distribution of historical analysis gust speeds. This step accounts for the conversion needed to derive 3-second gust speeds from the model 10 m wind output. The UPSCALE model did not provide sufficient diagnostic variables to use standard gust parametrisation (e.g. Howard and Clark 2007; Sheridan 2011), so we use a quantile mapping process to account for the energy cascade driving the small scale gust process not captured at the UPSCALE model resolution of 25km. The event set footprints are matched to an amalgamation of Euro4 analysis storms: Xynthia (27/2/2010), Kyrill (18/1/2007) , Daria (25/1/1990) and the Great Storm of 1987 (16/10/1987). Quantile mapping is performed by qmap in R, using a smoothing spline to fit the quantile-quantile plot of observed and modelled data across the whole range of wind speeds. In the context of the WISC event set, the term 'downscaling' is used to refer to the combined process of regridding and quantile mapping. Figure 3 shows an example of this process, with the cumulative distribution of wind/gust speeds before and after downscaling.
The calibration process is design to make the event set equivalent and compatible to the WISC historical footprints. The grid cells of the model, representing an area average gust speed, are not comparable with point-based gust observations, which are strongly influenced by sub-grid scale localised effects, such as topography but also buildings and vegetation.
Table 3: Overview and description of variables.
Variables | |||
Long Name | Short | Unit | Description |
Maximum 3 second wind gust at 10m above the surface | N/A | m/s | Maximum 3s wind gust at 10m over a 72-hour period centred on time of the maximum vorticity at 850 hPa along the storm track. |
2.2.2. Storm Severity Index and Gust Speed Measures
Three measures of storm severity are provided with the footprints:
Max gust ( \( U_{max} \) ) over the land-sea mask (see Figure 1) in meters per second.
Mean gust ( \( \bar{U}_{25} \) ) over the land-sea mask for speeds exceeding a 25 m/s threshold.
- The Storm Severity Index (SSI)
The Storm Severity Index (SSI) is based on the land area exceeding a 25 m/s threshold, scaled by the cube of the mean gust speed. Damaging winds are typically defined as those exceeding a threshold of 25 m/s (e.g. Roberts et al. 2014) with the gust energy varying with the third power of gust speed, a representation of the advection of kinetic energy (e.g. Hennessey 1977; Klawa and Ulbrich 2003). The area of damaging wind over land has also been shown to be a good predictor of windstorm damage (Dawkins et al. 2016). Our SSI definition for windstorm event i is then (after Dawkins et al. 2016):
\[ \text{SSI}_{i} = A(U_{i,25})\bar{U}_{i,25}^3 \]where \( A(U_{i,25}) \) is the grid box area in square kilometers of the LSM (Land Surface Mask) that exceeds 25 m/s.
Table 4 summarises the top 10 storms for each of the 3 storm metrics. Footprints of the top storms are show in Figures 4-6. We note that ranking by SSI is expected to provide the most robust and representative metric of storm severity. Figures 5 and 6 show that the \( \bar{U}_{25} \) and metrics are susceptible to be skewed by small areas of very high gust speeds. A histogram of event set SSIs is shown in Figure 7.
Table 4: Footprint ID for top 10 ranked storms for each storm metric. A full list of footprints and their associated metrics are provided in the accompanying file eventset_summary.csv
Rank | \( U_{max} \) | \( \bar{U}_{25} \) | SSI |
1 | fp_ga3ups_198809241200_0493_004 | fp_ga3ups_200304270600_1455_002 | fp_ga3ups_199905250600_2144_002 |
2 | fp_ga3ups_200710241200_1130_002 | fp_ga3ups_199712230600_0566_002 | fp_ga3ups_200005291800_2248_003 |
3 | fp_ga3ups_199311201800_1708_002 | fp_ga3ups_200201030000_0633_003 | fp_ga3ups_200701250600_1478_004 |
4 | fp_ga3ups_200801211800_1318_005 | fp_ga3ups_198802141800_2036_002 | fp_ga3ups_199910170000_0793_004 |
5 | fp_ga3ups_200109151200_0231_002 | fp_ga3ups_199102220600_2174_005 | fp_ga3ups_200305270600_2200_003 |
6 | fp_ga3ups_199712100000_0007_004 | fp_ga3ups_199202251200_2458_002 | fp_ga3ups_200209181200_0351_001 |
7 | fp_ga3ups_201002020600_1730_001 | fp_ga3ups_200001141800_1135_005 | fp_ga3ups_199909140600_0247_005 |
8 | fp_ga3ups_199801100600_1024_002 | fp_ga3ups_199412171800_0478_004 | fp_ga3ups_199610051800_0625_002 |
9 | fp_ga3ups_199709131800_0292_005 | fp_ga3ups_199603131800_0358_005 | fp_ga3ups_199905030600_1622_002 |
10 | fp_ga3ups_199309100600_0283_004 | fp_ga3ups_199912160600_0415_004 | fp_ga3ups_199109010600_0022_005 |
Figure 3: Dot plot ('Pollock') representation of event set, showing the temporal distribution of events in time. Each circle represents one event in the event set coloured by ensemble. The circle area is scaled by the SSI. Dots with black borders represent the top 200 storms in the event set.
Figure 4: Example quantile mapping of model wind footprints (blue line) to gust footprints (green line) to match Euro4 analysis footprints (black line) for a single event
Figure 5: Footprint of top ranked SSI.
Figure 6: Footprint of top ranked 
gust. Note the areas of high speeds around the coast of northern Norway.
Figure 7: Footprint of top ranked Umax gust.
Figure 8: Histogram of event set SSIs. Note the logarithmic frequency scale.
2.3. Input Data
The input data to the eventset is summarised in Table 5 and described below in this section.
Table 5: Overview of climate model data for input to C3S WISC Event set, summarizing model properties and available scenario simulations.
Input Data | ||||
Model name | Model centre | Scenario | Period | Resolution |
UPSCALE / Met Office HadGEM3 Global Atmosphere 3 (GA3) and Global Land configurations | Global Areas for which storms were extracted were bounded by West: 25° / East: 40.5° / South: 34.4° / North: 71.5° | 5 Ensembles | 1985-02-01 to | N512 / 25km |
2.3.1. Input Data 1: UPSCALE model data
The model contributing the event set is taken from the UPSCALE project (Mizielinski et al. 2014). UPSCALE provides an ensemble of weather-resolving simulations from the Met Office HadGEM3 Global Atmosphere 3 (GA3) and Global Land configurations, as documented by Walters et al. (2011).
The simulations cover the period from 1 February 1985 to 25 May 2011 over a N512L70 grid, equivalent to 25 km at 50°N covering the lower 85 km of the Earth's atmosphere. Each ensemble member is initialized from one of five consecutive days, starting in February 1985, following a five year model spin-up. In total, the model provides roughly 130 years of data.
The primary prognostic variables from the GA3 configuration derive from a semi-implicit semi- Lagrangian formulation of the non-hydrostatic fully-compressible deep atmosphere equations of motion, as defined by Davies et al. (2005), and include three-dimensional wind components. The effect of local and mesoscale orographic features (hills through to small mountain ranges) not resolved by the mean orography defined by the 25 km grid spacing, and turbulent motions in the atmosphere, are parametrized.
2.3.2. Input Data 2: C3S_441_Lot3 (WISC) historical windstorm footprints
Calibration of Sets 1 and 2 of the synthetic event set was undertaken using the C3S_441_Lot3 (WISC) historical windstorm footprints. These footprints were dynamically downscaled from ERA Interim data by the UK Met Office as part of the C3S_441_Lot3 (WISC) project.
2.3.3. Input Data 3: Station observations
Calibration of Set 3 of the synthetic event set was undertaken using UK Met Office station data corresponding to the four historical windstorm events used for Set 1, ie Storms Xynthia, Kyrill, Daria and 87J.
2.4. Method
2.4.1. Background
The WISC event set is best described as a physically realistic set of plausible events, representative of the period from 1985 to 2011. It is not designed to replicate this historical period: the event set does not represent variants of real historical storms. Data assimilation is not used to align the model to historical observations, and each of the 5 ensemble members has evolved independently from the others from a unique starting point. However, they are forced by historical forcings, including aerosol, ozone, solar and volcanic variability as defined by the Atmospheric Model Intercomparison Project II standards, with daily sea surface temperatures (SSTs) and sea ice derived from the Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) of Donlon et al. (2012).
2.4.2. Model / Algorithm
The model contributing to the event set is taken from the UPSCALE project (Mizielinski et al. 2014). UPSCALE provides an ensemble of weather-resolving simulations from the Met Office HadGEM3 Global Atmosphere 3 (GA3) and Global Land configurations, as documented by Walters et al. (2011).
The simulations cover the period from 1 February 1985 to 25 May 2011 over a N512L70 grid, equivalent to 25 km at 50°N covering the lower 85 km of the Earth's atmosphere. Each ensemble member is initialized from one of five consecutive days, starting in February 1985, following a five year model spin-up. In total, the model provides roughly 130 years of data.
The primary prognostic variables from the GA3 configuration derive from a semi-implicit semi- Lagrangian formulation of the non-hydrostatic fully-compressible deep atmosphere equations of motion, as defined by Davies et al. (2005), and include three-dimensional wind components. The effect of local and mesoscale orographic features (hills through to small mountain ranges) not resolved by the mean orography defined by the 25 km grid spacing, and turbulent motions in the atmosphere, are parametrized.
2.4.3. Calibration
The calibration process was designed to make the event set equivalent and compatible to the WISC historical footprints. The grid cells of the model, representing an area average gust speed, are not comparable with point-based gust observations, which are strongly influenced by sub-grid scale localised effects, such as topography but also buildings and vegetation.
The final version of the synthetic event Set consists of three sets as described below. These three sets are designed to be used together to provide an overall event set of 22,980 synthetic storms:
- Synthetic set 1: this contains 7,660 events recalibrated against the historic dataset using four historical footprints for the Xynthia, Kyrill, Daria and 87J windstorm events.
- Synthetic set 2: Rather than downscaling against the cumulative distribution function (CDF) of named events Xynthia, Kyrill, Daria and 87J, a storm severity index combining maximum wind gust speed and land area above a threshold of 25 m/s was used to select the strongest six events from the Met Office historical events database. Events selected were for 19/03/1995, 26/02/1989, 25/03/1986, 24/11/1984, 22/03/1983 and 01/02/1983.
- Synthetic set 3: Downscaling for this set was based on station observations from the four named storms used for set 1.
The event set has been compared with stochastic and historical data in a range of assessments including case studies carried out during the WISC project itself. Some of the comparisons have been conducted on a commercial basis, but some have been presented during sessions of the of European Windstorm Workshop series, in particular those at Karlsruhe (2018) and Birmingham (2019).
2.4.4. Validation
Validation work of the WISC historical footprints, by KNMI, suggested that only footprint data in areas over 500m altitude were likely to require bias corrections towards station observations. Therefore, the event set footprints may be interpreted at face value, i.e. as an assessment of the maximum gust speed at each 4.4 km cell. Given the uncertainty in extreme wind gust point observations from meteorological station networks, and the scarcity of such observations, no general bias correction of footprint data towards station observations was undertaken.
Clustering of the footprints that make up the event set was not explicitly considered, but the footprint metadata contains sufficient information to reconstruct the timeline of storm occurrence in each ensemble member either using the time variable, or by parsing the footprint median time from the file ID. The temporal filtering means that each footprint can be considered independent of each other such that the user can define their own clustering scenarios.
2.4.5. Filename Description
File names are constructed using the following naming convention:
C3S_EvenSet_fp_ga3ups_<Median Time><Track ID><Ensemble ID>_v1.nc
where median time is the median time of footprint and time of maximum vorticity (formatted %Y%m%d%H); track ID is the XTCTRACK algorithm track ID padded to 4 digits, and ensemble ID is the synthetic set number (i.e. 1.2, 2.0, 3.0) to 2 digits and lastly the ensemble number to 1 digit (i.e.
1-5). The final version number (i.e. v1) refers to the version number of this dataset in the Climate Data Store catalogue.
3. Concluding Remarks
The C3S_WISC_Synthetic_EventSet is a synthetic event set 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 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.
Acknowledgements
The Met Office would like to acknowledge Matthew Mizielinski and Malcolm Roberts for useful discussions regarding the GA3 setup, and making the GA3-UPSCALE data freely available for use in this project.
References
Davies, T, M J P Cullen, A J Malcolm, M H Mawson, A Staniforth, A White, and N Wood. (2005). "A New Dynamical Core for the Met Office's Global and Regional Modelling of the Atmosphere." Quarterly Journal of the Royal Meteorological Society 131 (608). John Wiley & Sons, Ltd.: 1759– 82. doi:10.1256/qj.04.101.
Dawkins, Laura C., David B. Stephenson, Julia F. Lockwood, and Paul E. Maisey. (2016). "The 21st Century Decline in Damaging European Windstorms." Natural Hazards and Earth System Sciences 16 (8): 1999–2007. doi:10.5194/nhess-16-1999-2016.
Donlon, Craig J, Matthew Martin, John Stark, Jonah Roberts-Jones, Emma Fiedler, and Werenfrid Wimmer. 2012. "The Operational Sea Surface Temperature and Sea Ice Analysis (OSTIA) System." Remote Sensing of Environment 116: 140–58. doi:http://dx.doi.org/10.1016/j.rse.2010.10.017.
Hennessey, Joseph P. (1977). "Some Aspects of Wind Power Statistics." Journal of Applied Meteorology 16 (2): 119–28. doi:10.1175/1520-0450(1977)016<0119:SAOWPS>2.0.CO;2.
Hodges, K I. (1995). "Feature Tracking on the Unit Sphere." Monthly Weather Review 123 (12). American Meteorological Society: 3458–65. doi:10.1175/1520- 0493(1995)123<3458:FTOTUS>2.0.CO;2.
———. 1999. "Adaptive Constraints for Feature Tracking." Monthly Weather Review 127 (6). American Meteorological Society: 1362–73. doi:10.1175/1520- 0493(1999)127<1362:ACFFT>2.0.CO;2.
Hoskins, Brian J, and Kevin I Hodges. (2002). "New Perspectives on the Northern Hemisphere Winter Storm Tracks." Journal of the Atmospheric Sciences 59 (6). American Meteorological Society: 1041–61. doi:10.1175/1520-0469(2002)059<1041:NPOTNH>2.0.CO;2.
Howard, Tom, and Peter Clark. (2007). "Correction and Downscaling of NWP Wind Speed Forecasts." Meteorological Applications 14 (2). John Wiley & Sons, Ltd.: 105–16. doi:10.1002/met.12.
Klawa, M., and U. Ulbrich. (2003). "A Model for the Estimation of Storm Losses and the Identification of Severe Winter Storms in Germany." Natural Hazards and Earth System Sciences 3 (6): 725–32. doi:10.5194/nhess-3-725-2003.
Mizielinski, M. S., M. J. Roberts, P. L. Vidale, R. Schiemann, M. E. Demory, J. Strachan, T. Edwards, et al. (2014). "High-Resolution Global Climate Modelling: The UPSCALE Project, a Large-Simulation Campaign." Geoscientific Model Development 7 (4): 1629–40. doi:10.5194/gmd-7-1629-2014.
Roberts, J. F., A. J. Champion, L. C. Dawkins, K. I. Hodges, L. C. Shaffrey, D. B. Stephenson, M. A. Stringer, H. E. Thornton, and B. D. Youngman. (2014). "The XWS Open Access Catalogue of Extreme European Windstorms from 1979 to 2012." Natural Hazards and Earth System Sciences 14 (9): 2487–2501. doi:10.5194/nhess-14-2487-2014.
Sheridan, P. (2011). "Review of Techniques and Research for Gust Forecasting and Parameterisation." Exeter, UK.
Walters, D. N., M. J. Best, A. C. Bushell, D. Copsey, J. M. Edwards, P. D. Falloon, C. M. Harris, et al. (2011). "The Met Office Unified Model Global Atmosphere 3.0/3.1 and JULES Global Land 3.0/3.1 Configurations." Geoscientific Model Development 4 (4): 919–41. doi:10.5194/gmd-4-919-2011.
