Contributors: CMCC 

Issued by: CMCC

Issued Date: October 2024

Ref: C3S3_413 – Enhanced Operational Windstorm Service

Official reference number service contract: 2023/C3S2_413_CMCC

Introduction

Executive summary 

European winter windstorms, or extra-tropical cyclones (ETC), are a major cause of losses to the insurance sector. To help the sector better understand this risk, the Enhanced Windstorm Service (EWS) for the insurance sector has been developed. The development of the service follows user feedback from previous phases (operational and proof of concept activities), implemented as part of the Copernicus Climate Change Service (C3S). 

The EWS core consists of detecting and tracking Pan-European potentially harmful windstorms, associated with ETCs. This represents a continuation and enhancement of the original C3S Windstorm Service. The main EWS novelties consist of a temporal and seasonal extension of the dataset characterising events over the whole ERA5 reanalysis period (1940-present) and expanding storm tracking from winter months only to the whole year with monthly updates in an operational fashion. EWS also introduces a second tracking algorithm, TempestExtremes (TE, Ullrich et al., 2021), complementing the one already used in the original service (TRACK, Hodges 1995, 1999, Hoskins and Hodges 2002).    

The Enhanced Windstorm Service is constituted of the following products:

(i) Windstorm tracks. It is the path followed by a storm and consists of a list of points characterised by at least a timestamp, a latitude and a longitude. Several parameters characterising the storm can be reported along this path and are detailed in Table 2.

(ii) Windstorm footprint. It provides a spatial snapshot of a storm allowing one to characterise its spatial extent and intensity represented by the maximum 10m wind gust at each grid point in the domain over a 72-hour time window centred on the time step in which the tracking algorithm identifies the maximum 925 hPa wind speed over land, within a 3-degree radius of the windstorm track point (Roberts et al., 2014). A statistical downscaling, based on a multiple linear regression-based model (van den Brink, 2020), is applied to include orography-driven and wind shear effects on footprint wind gusts.

(iii) Storm summary indicators representing annual-based statistics derived on decontaminated footprints considering both the tracking algorithms. Four wind gust thresholds (0, 15.6, 20, and 25  m s^-1) are considered. 

Datasets will be updated through an automatised procedure running codes underlying data production over the periodic ERA5 dataset new releases.  

Scope of the documentation

Windstorm data documentation describes the production, underlying methodological phases, and publication of the “enhanced operational windstorm service” (EWS) developed in the framework of the ECMWF-implemented C3S2_413 contract.

Dataset Description

Dataset Target Requirements

The present dataset aims to promote a knowledge-based assessment of the nature and temporal evolution of windstorms associated with ETCs. This contract presents a continuation, a temporal extension, and an enhancement of the original C3S Windstorm Service. The original service data content will be extended to the detection and tracking of Pan-European potentially harmful windstorms, associated with extratropical cyclones, for the whole available period provided by ECMWF's ERA5 reanalysis dataset (1940 - present) and to a second tracking algorithm, TempestExtremes (TE, Ullrich et al., 2021), in addition to the previously used TRACK / Hodges algorithm (Hodges 1995, 1999, Hoskins and Hodges 2022) hereafter called Hodges. This documentation traces all the steps made in the production of two main types of windstorm datasets: (i) Extratropical cyclones (ETC) tracks based on the use of two detecting/tracking algorithms and (ii) the associated footprints mapping the maximum value of 10m-height wind gust during a specific event.

Dataset Overview

Data description

Windstorm track. This product contains the Pan-European potentially harmful windstorm tracks, associated with extratropical cyclones, for the whole available period provided by ECMWF's ERA5 reanalysis dataset (Hersbach et al., 2020) spanning from 1940 to present. A storm track is defined as a sequence of longitude–latitude points which track an extra-tropical windstorm over time and is defined by the tracking algorithm. Two tracking algorithms are used: (i) Hodges algorithm (Hodges 1995, 1999, Hoskins and Hodges 2002), already in use in the previous Windstorm service and (ii) TempestExtremes (TE, Ullrich et al., 2021). The two tracking algorithms rely on an automated object identification leveraging two different variables, i.e., relative vorticity and mean sea level pressure, respectively (see "Methods" in section 3 for more details). Spatial and temporal filtering is applied to these variables to isolate key characteristics of low-pressure systems, identifying potential candidates suitable for track definition.

Windstorm footprint. Defined as the maximum 10m wind gust over a 72-hour time window centred on the time step in which the tracking algorithm identifies the maximum 925 hPa wind speed (Roberts et al., 2014). Only land grid points are considered. Two spatial configurations of the footprints are provided. (i) “Full domain” and (ii) Storm footprint area referred to as “decontaminated footprint”. The first consists of the event wind gust maxima over the whole domain, whereas in the decontaminated configuration, only the wind gust maxima within a radius of 1000 km from the storm track points are kept. This process aims to avoid the possibility that storm footprints can cluster/overlap in time (maximum gust footprints deriving from two or more events), giving rise to a “contamination” which can obscure features belonging to an individual event. To include orography-driven and wind shear dynamics potentially modulating footprint wind gusts, a “downscaled” version of the same set of footprints is derived through a statistically downscaled ERA5 variables using a multiple-linear-regression-based model (van den Brink, 2020). Moreover, both original resolution and downscaled footprints are subjected to a “decontamination” process to void the possibility that storm footprints can cluster/overlap in time (maximum gust footprints deriving from two or more events), giving rise to a “contamination” which can obscure features belonging to an individual event.

Storm summary indicators. Computed on a yearly basis for each NUTS (Nomenclature of Territorial Units for Statistics) 0, 1 and 2 regions aim at summarise the storm activity of a given region considering four different wind gust thresholds (0, 15.6, 20 and 25 m s^-1). The indicators considered, and defined in the following, are: Storm count, Mean wind gust, Storm Severity Index (SSI) and Normalised Storm Severity Index (NSSI).

Table 1: Overview of the dataset.

Product description 

Windstorm track

The windstorm tracks contain the 6-hourly evolution of a storm event; they are provided in text format (.csv) and include the fields listed in Table 2.

Table 2: Overview and definition of the fields included in the windstorm tracks product.

Windstorm footprint

The windstorm footprints, which are provided in NetCDF format (.nc), include for each storm event the maximum 3-second 10-m wind gust (m s-1) over the 72 hours centred around the footprint central time. The NetCDF file of a footprint contains a single variable which depends on three principal coordinates (latitude, longitude and track_id). The track_id coordinate is included to easily merge several footprints during an analysis. On top of the principal coordinates three secondary coordinates (track_start_time, footprint_central_time and track_end_time) characterise respectively the start, central and end time of the event corresponding to the footprint. All the coordinates of a footprint file are listed in Table 3. 

Table 3: Overview and definition of the coordinates of the windstorm footprints product.

Each footprint NeCDF contains a set of attributes characterising the footprint and summarised in Table 4.

Table 4: Overview and definition of the attributes characterising the windstorm footprints product.

Windstorm summary indicators

The windstorm summary indicators product is provided in text format (.csv) and include the fields listed in Table 5.

Table 5: Overview and description of the fields included in the storm summary indicators product.

Input data

Table 5: Overview of data for input to the Enhanced Windstorm Service. 

Methods 

Tracking algorithms

Hodges and TempestExtremes tracking algorithms use different atmospheric variables for detecting and tracking ETC tracks, 850 hPa relative vorticity and mean sea level pressure, respectively. This determines differences in synoptic-scale dynamics influencing tracks’ detection. In general terms, relative vorticity is better at capturing smaller spatial-scale processes compared to mean sea level pressure based tracking (Hodges et al., 2003). In the context of the windstorm dataset within EWS, this translates into a substantially larger number of events produced by Hodges compared to TempestExtremes. However, not all the events tracked by TempestExtremes are included within the Hodges dataset. This indicates that considering a different tracking variable determines not only a different number of detected events but also events having different dynamical features. In terms of historical events reproducibility, a comparison with a series of 50 windstorms reported in the Extreme Windstorm catalogue (XWS, Roberts et al., 2014) shows larger matching events from Hodges. Hodges reproduces 39 out of 50 XWS events against 17 reproduced by TempestExtremes. However, the comparison between the two algorithms and the reference XWS is suboptimal since this latter leverages the same Hodges algorithm (though a previous release), to detect and track historical events. In this regard, spatial and temporal parameters defining shared events (i.e., at least half of event points with a temporal and spatial discrepancy lower than one day and two degrees respectively), can modulate the number of matching events found in the two tracking algorithms and reference XWS.

TempestExtremes Algorithm

TempestExtremes algorithm (TE, Ullrich et al., 2021) is run considering ERA5 reanalysis (Hersbach et al., 2020) 6-hourly mean sea level pressure. In the present configuration, TempestExtremes algorithm requires:

• Storms to persist for at least 60 hours;
• Storms to have a maximum gap of at most 18 hours;
• Latitude range [25 °N - 66 °N] and longitude [-80 °E - 30 °W];
• Finally, TE algorithm imposes ETCs moving at least 12° from the start to the end of the trajectory, to eliminate stationary features and spurious shallow lows generated over regions with complex orography.

The track candidates produced by the TE algorithm are further filtered according to spatial and temporal filtering criteria, filtering out tracks characterised by at least one of the following features:

• Tracks with less than 5 points;
• Tracks with more than 3 points north of 65 °N;
• Tracks starting eastern of 5°W;
• Tracks ending west of 5°W.

Catalogue entry with the list of tracks passing filtering criteria consists of text-like [.csv] files, as in the example shown in Table 1.

Each track is identified by a:

• ID;
• Longitude and latitude;
• Mean sea level pressure at each step;
• Date;
• Wind gust value at 10m height;
• Land-sea-mask reporting the percentage of land area of ETC track grid nodes.

Hodges algorithm

The tracking algorithm which has already been used to produce the original Windstorm Service is the Hodges algorithm (Hodges 1995, 1999, Hoskins and Hodges 2022). The application of this algorithm has been extended to the entire ERA5 reanalysis period, so the two algorithms are operationally available for the entire ERA5 period (1940-present).

The input necessary to run the tracking algorithm is a 3-hourly frequency 850 hPa relative vorticity dataset, also from the ERA5 reanalysis (Hersbach et al., 2020). The input is on the native IFS gaussian grid (N320).

The following steps are performed starting from the input data:

• The 850 hPa relative vorticity fields are spectrally filtered to the T42 resolution (corresponding to about 480 km) to discard features related to small-scale background noise;
• Relative vorticity maxima with a value larger than 1×10−5s−1 are retained;
• The adaptive algorithm described in Hodges 1999 is applied to merge vorticity features into a cyclone track;
• The tracks that last less than 1 day or travel less than 1000 km are discarded;
• The mean sea level field is added by searching for the nearest pressure minimum with a 5° radius of the vorticity centre.

Figure 1: Comparison between the output of the two tracking algorithms produced for the exemplary period 2000-2005. Upper panels: annual windstorm tracks and corresponding mean sea level pressure. Central panels: mean sea level pressure of windstorm tracks crossing the domain divided into cells of 3ºx3º during the same period. Bottom panels: Number of unique windstorm track passages binned into each grid cell of a 3°x3° resolution grid.   

Footprints decontamination

European windstorm can cluster in time. In fact, maximum gust footprints could likely be derived from two or more events. To minimise this “contamination”, instead of taking the maximum gust over the whole domain, only gusts inside a 1000km radius of the track position at that time are assumed to be part of the event.

Figure 2: Original and decontaminated footprints for one representative windstorm

Footprint statistical downscaling

Original footprints are statistically downscaled through a multiple linear regression-based model (van den Brink, H. W. (2020)).

\text{WindGust} = 10.3 + 0.0112 \cdot \text{ERA5}^2 + 0.0148 \cdot \text{wgSLh}^2 + 0.00355 \cdot \text{ELEV}

\text{wgSLh} = u_{10} + \alpha \cdot \frac{u_{100} - u_{10}}{\log\left( \frac{100}{10} \right)}

The destination resolution is imposed by the orography term (ELEV), at 1 km resolution. Three predictors are used to downscale wind gusts at 10m height (WindGust) is derived using three predictors:

• ERA5 denoting predictor 1 (wind gust at 10m height from ERA5);
• wgSLh denoting predictor 2 (wind gust estimated from wind-shear between two height levels, (van den Brink, 2020);
• ELEV denoting predictor 3 (observed elevation derived from the 1 km resolution elevation file).

In CDS, footprints are provided through different NetCDF files identifying:

• Original resolution full (not decontaminated);
• Original resolution decontaminated;
• Downscaled full (not decontaminated);
• Downscaled decontaminated.

Figure 3: Decontaminated footprints for two representative events (Jeanette and Gero). On the left panels, the original resolution and on the right panels statistically downscaled footprints. Windstorms tracked with the TempestExtremes and Hodges algorithms are displayed in the upper and bottom panels respectively. The same event corresponds to a different ID since the two algorithms identify a different number of tracks during the period considered.  

Table 6: Overview of the windstorm tracks and footprints datasets input variables.

Windstorm summary indicators

The catalogue entry contains summary indicators which are aggregated annually over NUTS 0, 1, and 2 regions. All indicators are contained in a single csv file covering the whole period from 1940 to the present. The available indicators are defined in the sections below and are:

• Storm count;
• Mean wind gust;
• Storm Severity Index (SSI);
• Normalised Storm Severity Index (NSSI).

All four indicators have been computed considering four wind gust thresholds 0, 15.6, 20 and 25 m/s.

Yearly storm count

The yearly storm count looks at the occurrence of storms exceeding a given threshold in a given region. A storm event is considered to affect a region if its decontaminated footprint intersects the region. For each storm event, the decontaminated footprint of wind gust exceeding a threshold is considered to count storm events exceeding a given threshold.

Mean wind gust

The mean wind gust indicator looks at the average wind gust speed for storms exceeding a given threshold in a given region. For a given year all the decontaminated footprints are filtered for each threshold to keep the footprint area exceeding the threshold and averaged over time. The resulting average mean wind gust speed is then aggregated spatially in the predefined NUTS regions. Areas of a region that are not affected by a storm are not considered as “no values” and are ignored when performing the spatial average.

Storm Severity Index (SSI)

The Storm Severity Index (SSI, Dawkins et al., 2016) aims at quantifying the severity of the storm affecting a region. To do so the indicator combines both the total area of a region affected by a storm in a year and the mean wind speed gust speed exceeding a threshold. The SSI can be summarised by the following formula:

SSI(threshold) = \cup A_{footprint\cap region}\left[\overline{mean(windgust>threshold)}\right ]^{3}

Where:

• \cup A_{footprint\cap region}

  is the total area of the region affected by storms in a year. This area considers the union of the areas affected by each single footprint;

• \overline{mean(windgust>threshold)}

  is the temporal mean of the wind gust speed exceeding a threshold spatially averaged over the selected region. The overbar stands for “spatial average”, which is performed after the temporal average.

Only regions with an average of at least one storm a year over the period 1991 - 2020 are displayed on the map to focus on statistically significant regions.

The complete time series of any region can be consulted by clicking on the regions even if it shows no values on the map.

Normalised Storm Severity Index (NSSI)

The Normalised Storm Severity Index (NSSI) aims to quantify the severity of the storm affecting a country/region with a dimensionless number. The NSSI is based on the SSI but uses two extra quantities in order to normalise the SSI in terms of spatial extent and wind gust climatology.

The spatial normalisation consists of dividing the total area affected by the storm by the area of the region so that the size of the portion of the region affected by storms is taken into account.

The climatology-based normalisation consists of dividing the mean wind gust speed by the 98^th percentile of the probability distribution function (PDF) built on the local (grid-point-specific) climatology of the wind gusts (climatological period: 1991-2020). This normalisation aims to modulate the SSI of a single storm with the expected climatological severity of storms in a specific region.

It represents what should be considered a severe wind gust for the specific country/region (according to the NUTS considered) relative to the local historical climatology. The 98th percentile is of course different for different territories, which translates into a different probability of being exposed to (relatively) severe wind gusts in the country/region. This, in principle, should correspond to a structural historical capability to cope with what the local PDF defines as a severe wind.

NSSI(threshold) = \frac{\cup  A_{footprint\cap region}}{A_{region}}\left[\frac{\overline{mean(windgust>threshold)}}{\overline{P_{98}}}\right ]^{3}

Where:

• \cup A_{footprint\cap region}

  is the total area of the region affected by storms in a year. This area considers the union of the areas affected by each single footprint;

• A_{region}

  is the total area of the region (it is constant in time);

• \overline{mean(windgust>threshold)}

  is the temporal mean of the wind gust speed exceeding a threshold spatially averaged over the selected region. The overbar stands for “spatial average”, which is performed after the temporal average;

• \overline{P_{98}}

  is the 98th percentile of the wind gusts PDF considering all the decontaminated footprints over the period 1991 – 2020. The overline stands for the spatial average.

Concluding Remarks

EWS represents an enhancement and temporal extension of the original C3S windstorm service. It is operational meaning that it automatically updates windstorm datasets based on ERA5 period new releases on a monthly basis.

EWS leverages two tracking algorithms and derives windstorm footprints at the original resolution and statistically downscaled. Taking advantage of the whole ERA5 time series EWS datasets presents an unprecedented temporal extension, of paramount importance for:

-a better characterisation of large year-to-year fluctuations in terms of frequency and magnitude thus representing a valuable tool for the insurance and reinsurance sectors.

-Providing the basis for statistical inferences about trends involving crucial ETCs-related dynamics, their multi-decadal climate variability and associated climate risk.

-EWS datasets can serve as reference products to evaluate the capability of global and regional climate models to reproduce windstorm features and temporal evolution.

References

van den Brink, H. W. (2020). An effective parametrisation of gust profiles during severe wind conditions. Environmental Research Communications. Institute of Physics. https://doi.org/10.1088/2515-7620/ab5777

Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A., Muñoz-Sabater, J., et al. (2020). The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society, 146(730), 1999–2049. https://doi.org/10.1002/qj.3803

Hodges, K. I. (1995). Feature tracking on the unit sphere. Monthly Weather Review, 123(12), 3458-3465.

Hodges, K. I. (1999). Adaptive constraints for feature tracking. Monthly Weather Review, 127(6), 1362-1373.

Hoskins, B. J., & Hodges, K. I. (2002). New perspectives on the Northern Hemisphere winter storm tracks. Journal of the Atmospheric Sciences, 59(6), 1041-1061.

Dawkins, L. C., Stephenson, D. B., Lockwood, J. F., & Maisey, P. E. (2016). The 21st century decline in damaging European windstorms. Natural Hazards and Earth System Sciences, 16(8), 1999–2007. https://doi.org/10.5194/nhess-16-1999-2016

Roberts, J. F., Champion, A. J., Dawkins, L. C., Hodges, K. I., Shaffrey, L. C., Stephenson, D. B., et al. (2014). The XWS open access catalogue of extreme European windstorms from 1979 to 2012. Natural Hazards and Earth System Sciences, 14(9), 2487–2501. https://doi.org/10.5194/nhess-14-2487-2014

Ullrich, P. A., Zarzycki, C. M., McClenny, E. E., Pinheiro, M. C., Stansfield, A. M., & Reed, K. A. (2021). TempestExtremes v2.1: A community framework for feature detection, tracking, and analysis in large datasets. Geoscientific Model Development, 14(8), 5023–5048. https://doi.org/10.5194/gmd-14-5023-2021

Related articles

Related articles appear here based on the labels you select. Click to edit the macro and add or change labels.