Synoptic conditions conducive for compound wind-flood events in Great Britain in present and future climates

Extreme wind is the main driver of loss in North-West Europe, with flooding being the second-highest driver. These hazards are currently modelled independently, and it is unclear what the contribution of their co-occurrence is to loss. They are often associated with extra-tropical cyclones, with studies focusing on co-occurrence of extreme meteorological variables. However, there has not been a systematic assessment of the meteorological drivers of the co-occurring impacts of compound wind-flood events. This study quantifies this using an established storm severity index (SSI) and recently developed flood severity index (FSI), applied to the UKCP18 12 km regional climate simulations, and a Great Britain (GB) focused hydrological model. The meteorological drivers are assessed using 30 weather types, which are designed to capture a broad spectrum of GB weather. Daily extreme compound events (exceeding 99th percentile of both SSI and FSI) are generally associated with cyclonic weather patterns, often from the positive phase of the North Atlantic Oscillation (NAO+) and Northwesterly classifications. Extreme compound events happen in a larger variety of weather patterns in a future climate. The location of extreme precipitation events shifts southward towards regions of increased exposure. The risk of extreme compound events increases almost four-fold in the UKCP18 simulations (from 14 events in the historical period, to 55 events in the future period). It is also more likely for there to be multi-day compound events. At seasonal timescales years tend to be either flood-prone or wind-damage-prone. In a future climate there is a larger proportion of years experiencing extreme seasonal SSI and FSI totals. This could lead to increases in reinsurance losses if not factored into current modelling.


Introduction
Extreme winds and inland flooding are the two largest causes of loss to North-West Europe [1].The individual hazards have been well studied from a loss perspective.However, there is increasing evidence that these hazards can spatially or temporally cooccur over daily [2][3][4] to seasonal [5][6][7][8] timescales.A recent paper by Bloomfield et al [9] confirmed this co-occurrence across all these timescales over the whole of Europe.The relationship between cooccurring meteorological, hydrological and impactdriven wind-flood variables are all different, particularly at short timescales.Daily correlations between GB-aggregate maximum 10 m wind gusts and total precipitation are ≈ 0.7, whereas daily correlations between GB-aggregate maximum 10 m wind gusts and total river flow are ≈ 0.4.The relationships between flood and wind metrics are smaller for more impact focused variables.Understanding these links between hazards, and their possible implications for catastrophe modelling is critical for accurate representations of loss to clients portfolios.Insights can also be gained for infrastructure operators, maintenance crews and government agencies who need to provide warnings on extreme weather events [10].
A key driver of compound wind-flood events at 1-3 day timescales is the presence of extra-tropical cyclones.Previous studies have shown case studies of compound wind-flood events during named storms [11][12][13][14][15][16] or have used cyclone tracking algorithms to show that a storm is present during these compound events [2,4].If considering just the flood hazard, De Luca et al [6] used Great Britain (GB) catchment-level river flow to investigate multi-basin flooding events, finding these are associated with persistent cyclonic and westerly atmospheric circulations (defined using Lamb weather types) and atmospheric rivers.Hillier and Dixon [7] used a large ensemble of climate model data to demonstrate that individual storms impacting north-west GB either tend to be very wet or very windy, not extreme in both.Although co-occurrence is still more likely than pure random chance.At longer-timescales out to two weeks, a multi-storm hazard could be related to clusters of wind storms [5,17].Donat et al [18] show that over central Europe, stormy days are often associated with westerly weather types and positive phases of the North Atlantic Oscillation (NAO+).Zanardo et al [19] also show for Northern Europe the majority of historic winter floods occur during NAO+, and the NAO has a significant impact on flood losses.Although, it is still possible to get extreme storms during other synoptic patterns [18].
Analysis at seasonal timescales showed multi-basin flooding can be correlated with the occurrence of very extreme winds [6], which also extended to potential infrastructure damage from compound wind-flood events [20].The link between seasonal storminess and the NAO over GB, and most of central-Northern Europe was demonstrated in [21].The increased number of storms in NAO+ is explained by the larger area of the North Atlantic with suitable cyclone growth conditions [22].
There is a growing interest among the insurance community in understanding how co-occurring risks may be impacted by climate change, although at present very little research is published on this topic due to a lack of available high resolution future climate projections, and flood simulations.Bloomfield et al [9] highlighted the return period of extreme 1-day wind-flood events may reduce four-fold in a future climate.However, the synoptic conditions associated with these events are still not well understood.Using the same underlying climate models (UK Climate Projections 2018, UKCP18) Pope et al [23] show an increase in cyclonic and westerly wind conditions, suggesting a shift towards warmer, wetter winters.Across the CMIP6 ensemble there is a general trend of wetter winters [24] with an increasing number of extreme cyclones (and stronger associated extreme winds) towards the end of the century [25].The UKCP18 ensemble also sees an increase in extreme cyclones by 2100, particularly in the convection permitting simulations [26].
Pope et al [23] highlight weather patterns likely to be related to winter flooding and see a projected increase in the frequency of these conditions by the end of the century.The winter-mean precipitation during each of these patterns also increases in a future climate [27].Bates et al [28] showed climate change results in significant future UK flood risks when the UKCP18 simulations are used as inputs to a catastrophe model.
Despite the growing body of research on individual hazards, the impact of climate change on the synoptic conditions associated with compound-wind flood events still remains unclear.A key challenge associated with modelling the relationship between wind and flood damage is the need for high resolution meteorological data products [29] and hydrological data that can be matched to meteorological indices [9].The recently develop UKCP18 dataset is therefore an excellent resource to understand the behavior of future wind and flood risks which require high resolution information at as close to asset-level as possible.This study will use the UKCP18 regional climate model simulations to: • Quantify the large-scale drivers of 1-day compound wind-flood extremes over GB in a present climate • Understand potential impacts of climate change on the occurrence and severity of these events.• Understand the seasonal occurrence of compound wind-flood extremes and how this may change in a future climate.

Data and methods
In the following subsections the meteorological data (section 2.1) proxies for wind and flood risk (section 2.2) and meteorological indices to classify extreme events (section 2.3) are presented.All metrics used in this study are calculated during extended winter (October-March) and nationally aggregated.

Climate data
The UK Climate Projections 2018 (UKCP18) regional model simulations are used in this study.UKCP18 provides hourly 12 km data from 1980-2080 using the representative concentration pathway (RCP) 8.5 climate change scenario, with a 12 member perturbed parameter ensemble [30].Many previous compound wind-flood studies are focused on the historical period (due to lack of appropriately high resolution data in the future period).The UKCP18 simulations allow us to address this issue.Hourly instantaneous wind gusts and daily total precipitation are taken from all ensemble members for two periods, 1981-2000 and 2061-2080.Although having high resolution climate data is important for accurately modeling compound wind flood risk, other factors are also important to address.As discussed in [9], daily total precipitation is not an appropriate metric to capture winter fluvial flood risk, due to the lagged response of large catchments to extreme rainfall.Therefore, daily-mean GB-aggregated river flows from the Grid-to-Grid (G2G) hydrological model [31] driven by UKCP18 data, are used to create the flood severity index.Further details of the UKCPderived river flows are given in [32].
Recent results from [26,27,33] show there are potential issues with the ability of the 12 km UKCP18 simulations to capture extreme precipitation and wind gusts respectively, when compared to the UKCP18 2.2 km convection permitting simulations.However, for this study a key requirement was the availability of corresponding hydrological model simulations created in [31], which were not available for the 2.2 km model.Manning et al [33] show that the 12 km and 2.2 km simulations are largely similar except over orography, which is not the focus of this study (as there are rarely large exposure centres here).However, exploring these results in the 2.2 km convection permitting UKCP18 simulations would be an excellent topic for future work.

Compound event indices
It is important that as well as appropriate meteorological data, that suitable metrics are used to capture the risks associated with wind and flood risks.If not, the relationships presented may not be accurate [9].The metrics used to most directly estimate impact in this study are the storm severity index (SSI, in the form used in [17,34]) and a new flood severity index (FSI) developed in [9].These metrics provide an indication of times when a meteorological hazard exceeds predefined percentile thresholds (which are chosen based on the potential for losses if they are exceeded).The SSI can be given as: Here v i,j is the daily maximum wind gust at longitude i and latitude j. pop i,j is the 2020 population density in a given location, taken from [35].N i and N j show the total number of longitudes and latitudes calculations are performed over respectively.The SSI has been developed based on historical loss and damage data (see [36]).Throughout the paper the SSI and FSI metrics are calculated at daily resolution (t = 1 day) which is the highest temporal resolution common across all climate model inputs.Therefore, v 98 i,j always represents the gridded daily 98th percentile from October-March.
The FSI is defined as: ) 1 otherwise The form is developed from the SSI with parameters defined in the same way, except here the key hydro-meteorological variable is q, the daily total river flow.The cubic behaviour is removed as this was not required to best represent the relationship between wind and flood losses [9].The form of SSI used here is well established in the literature and relates well to European Losses [17].The development and verification of the FSI metric are described in [9].The metric performs particularly well when verified against the number of UK floods (r = 0.74).The FSI does however pick up many small events where no flooding was historically recorded.There is a much better match between the occurrence of extreme events (see [9] for full verification) so the most extreme FSI events are the main focus of this study.
When calculating SSI and FSI for the future UKCP18 period the historical percentiles are used.UKCP18 has been shown to reproduce historical SSI and FSI occurrence and compounding well when compared to the ERA5 reanalysis [37] (see figure 5 of [9]) and is therefore used in this study with minimal comparison to historical observations.However, equivalent results plots can be found in the supplementary material when relevant.
Three types of compound events are considered in the results.Days in which SSI > 0 and FSI > 0 (which gives an indication of relatively strong winds and river flows somewhere over GB).The other two definitions of compounds illustrate various levels of extreme events: SSI > 95th percentile and FSI > 95th percentile, SSI > 99th percentile and FSI > 99th percentile.These were chosen to be large enough that it is feasible that damage may occur, but not so extreme that we run out of events for rigorous statistics.In this study the focus of the compound events analysis is on daily and seasonal timescales.However, details of periods where multi-day compound events (e.g.consecutive days of SSI and FSI exceeding a relevant threshold) are reported in the supplementary material for general interest, as these may be particularly challenging to the insurance community.
Throughout the study more FSI events are reported than SSI events, due to the higher spatial resolution of the Grid-to-Grid model data than UKCP18 (and therefore more days when one of the grid boxes exceeds a given threshold).Previous work found that when comparing ERA5 (30 km) and UKCP18 (12 km) the correlations between wind and flood metrics were consistent, across all resolutions, so we do not believe this to be problematic for the study results.

Weather patterns
Daily synoptic conditions are defined using the UK Met Office DECIDER tool, using the methods outlined in [38] and the pattern identifications from [23].The method assigns each day into a set of 30 weather patterns, which have their own climatological characteristics.These patterns are commonly used for medium-long range ensemble forecasting for multiple applications [38,39].
The original patterns were created using a nonhierarchical k-means clustering algorithm [40] applied to gridded daily MSLP observations [41] for a North Atlantic-European domain (30 • horizontal resolution for the period 1850-2003.The patterns have since been calculated for the ERA5 reanalysis [42] and UKCP18 global model [23].For each UKCP18 ensemble member, daily MSLP anomalies are calculated (compared to the mean of the corresponding day of the year for the anomaly period).The MSLP anomalies are then regridded to 5 • latitude-longitude resolution over the weather pattern domain.Weather patterns are assigned based on their closest distance (defined as the area-weighted sum of squares difference) at each grid point [38].
There is good agreement between the UKCP18 ensemble representation and of patterns when compared to ERA5 (see figures 2 and 3 of [23]), where any large differences seen are likely due to the differing spatial resolution of the datasets affecting their ability to capture extreme MSLP anomalies.
This may appear a very large number of patterns, however this allows for a comprehensive classification of UK weather throughout the year.Lower-numbered patterns tend to occur more often in summer, and higher-numbered patterns tend to occur more often in winter.The higher-numbered weather patterns have some of the larger mean sea level pressure (MSLP) anomalies (both positive and negative) and can represent some severe weather circulation type (occurring around 1% of the time).A brief description of the 30 weather patterns and examples of their MSLP composites can be found in table 1 of [23].
In [38] a classification of eight patterns is also provided (intended to be more useful for seasonal forecasting) which are an an aggregation of the 30 patterns into more large-scale flow conditions (see figure A1 for MSLP composites of these patterns and see [38] for detailed discussion).For visual clarity in the results sections these eight patterns are used.with specific pattern numbers and descriptions given in the text when appropriate.

Historical events
The first step in quantifying the meteorological drivers of compound wind-flood events is understanding when these events predominantly occur.Figure 1 shows the seasonal occurrence of extreme wind, flood and compound events for different levels of extremeness.The top row of figure 1 shows statistics for all times when the SSI and FSI indicators are greater than zero.SSI and FSI events are common throughout the extended-winter period with the occurrence of both peaking in January, therefore resulting in the most compound events being present in this month as well.The colours on the bar chart demonstrate the pattern assignment from [38].Rather than retaining all 30 weather patterns the bar charts show a sub-set of 8 key weather types (see figure A1 for examples of the patterns).Across the whole season the most commonly occurring pattern category is NAO+.A similar result is also seen when the metrics are calculated for historical observations (see figure A2).The most common weather patterns for SSI, FSI and compound events have cyclones located over the UK (e.g.patterns 26, 29 and 30 from the full 30 pattern classification).These were all highlighted in [23] as patterns likely to be associated with UK fluvial flooding.However, There is a surprisingly large presence of NAO-events on days with non-zero SSI and FSI values, supporting [18] which showed around 5% of storm days (depending on the specific storm definition) occur during NAO-conditions.The presence of UK High Pressure for the FSI events is also notable.
A positive value of SSI or FSI is not necessarily enough for there to be loss or damage.Figure 1 therefore also shows events that exceed the 95th and 99th percentiles during the 1981-2000 period.The most extreme compound events mainly happen in January and February, and are predominantly associated with cyclonic weather types (see table A1 for more details of the 14 largest events which form the compound P99 composite).A key point here is we see extreme events in the UK High and subsets of the North-Westerly pattern, both of which were not identified  (1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000).Top: all events, middle: events exceeding the 95th percentile of the given metrics bottom: events exceeding the 99th percentile of the given metrics.Colors indicate the weather pattern on the day of the event, characterized using the met office DECIDER tool (see methods for details).The axis in each subplot are fixed in figures 1 and 2 for visual clarity.
in [23] as likely to be associated with fluvial flooding.This shows that the synoptic situation associated with compound wind-flood events are not necessarily the same as just for flood events.We note that when considering the observed period of 1981-2000 in figure A2 only 31 compound P95, and one compound P99 events are present.This shows the value of using the larger UKCP18 ensemble.The distribution of weather patterns across the different levels of SSI and FSI severity is similar in the observed period, supporting the results of [23] which shows the UKCP18 ensemble represents similar seasonal occurrences of each of the 30 patterns to ERA5.However, we note a difference in the timing of the extreme FSI events in observations (where more occur in early winter).This is likely related to the Grid-to-Grid model setup and an interesting topic to understand in future work.
When examining the most extreme compound events in detail there are two periods where there are multiple consecutive days of extremes (see table A1).Although there are two multi-day events (consecutive 1-day events), there are no times when two extreme-compound events are seen in the same October-March season.There is also a good spread of events across ensemble members (table A1).

Future events
Figure 2 shows the distribution and weather patterns present during the SSI, FSI and compound events in the future (2061-2080) period.We see a 9% reduction in the total number of SSI days, but a 26% increase in FSI days.Overall this leads to a 31% increase in compound events of any magnitude in the future climate, driven by the increase in FSI events.This is consistent with the thermodynamic response to climate change, with the future period of UKCP18 showing a much warmer and wetter climate [43], and therefore increased river flows (see section 3.4 for details).The seasonality of each event type is still similar to the historical period (see figure 1) and NAO+ is still the most commonly occurring pattern category.
In the future period we see an increase in compound events in February.There is an increase from 14 historical to 55 future compound events exceeding the 99th percentile of SSI and FSI.The details of these are given in tables A2 and A3.A much smaller proportion of the most extreme events fall outside the December-February period (similar to results from [32] when creating a future flooding event set).In the future period, 20 out of the 55 days are multi-day events (i.e.consecutive 1-day events), suggesting that compound wind-flood events may increase in duration in a future climate.As the events are all ≈ 3 days, this is likely due to slower moving individual storms [44] rather than storm clustering.The cyclonic patterns from the NAO+ classification are the most common patterns present during extreme compound events (with 24 and 29 being new additions from the historical period, see table A1).Increased occurrence of the Northwesterly patterns are seen for the future compound wind-flood events (see also figure A7 for details of changes to the pattern frequencies for all days).The presence of a NAO-pattern is particularly interesting as these conditions are not commonly associated with stormy weather over the UK.Section 3.4 will unpick possible reasons for this patterns' presence.

Synoptic details
Although the weather pattern analysis in sections 3.1 and 3.2 is useful for broad understanding it is also useful to examine composites of the synoptic conditions during the most extreme events, to understand the details of the assignments.Figure 3 shows daily maximum precipitation composites for the top 14 SSI and FSI events from the historical period and the top 55 events from the future period (see tables A1-A3 for details).In both the historical and future periods the most extreme SSI events are associated with strong pressure gradients over southern GB and very high surface wind gusts (see figure A4) in the region of largest exposure (note that meteorological variables are weighted by population when calculating SSI and FSI).Examining individual events shows storms are present over GB in all cases (see figure A5 for location of storm centres).These features of the SSI and FSI composites are also seen in the largest events from the observed period (see figure A3).
For the FSI events, in both a present and future climate the MSLP gradient is weaker over GB compared to the SSI or compound events.This is due to the variety of large-scale conditions that are leading to extreme GB rainfall (not shown) and the increased number of events in the composite.Despite the larger variation in synoptic conditions there is still a tendency for storms to be located to the North-West of the UK (see figure A5).The precipitation composites show that extreme FSI events are associated with large rainfall totals over the cities of Northern England, suggesting that many of the compound events may cause damage to different locations across GB rather than spatially co-occurring flood and wind damage (see figure A4 for potential wind damage).
The most striking differences in figure 3 are the changes in daily maximum precipitation in the future FSI and compound events.The location of most extreme precipitation in the future period is centred over Southern England (as reported in [43]), suggesting extreme rainfall may now coincide with locations of potential wind damage (see figure A4).This is also a region experiencing increased coastal flood risks, associated with sea level rise [45].Storms are generally present around GB for the compound windflood events, although in a future climate it becomes more common for the minimum MSLP to be over the North Atlantic (see figure A5).In a future climate the compound event composites look most similar in structure to the future FSI composites.This suggests that in our (much wetter) future climate, the key driver of a compound event is the presence of strong winds on a generally wet day.

Precursors to compound events
Understanding the weather conditions commonly present during a compound wind-flood event is useful as a first step to develop early warning systems.However, identifying any notable precursor behaviour both seasonally, and just a few days-weeks before and event is also helpful to improve our understanding of the meteorological drivers.
Figure 4 shows the 20-year mean October-March seasonal cycles for key meteorological variables used to generate SSI and FSI.Little change is seen in daily max GB-aggregate wind gusts in the future period, with the timeseries being dominated by climate variability (although we note that extreme wind gusts have been shown to increase in 2.2 km spatial resolution UKCP18 simulations, [33]).For precipitation there are notable increases in rainfall from December-February in the future period, which could explain some of the increases in FSI days.The changes in seasonal river flows are more complex, with a notably drier October-November, near-average conditions in December and then a much wetter January-February.The dryer weather in Autumn could be explained by the increases in drier summer-type weather patterns and decreases in stormy weather patterns [43,46].This helps to explain why so many of the future FSI events in figure 2 occur in February.Future work could investigate soil-moisture conditions in the Grid-to-Grid model to identify how saturated catchments are in the lead-up to an extreme FSI event.
Figure 4 shows the GB-aggregate conditions for the three weeks preceding the largest compound events from table A1 (solid lines), tables A2 and A3 (note only the historical events are shown on figure 4 for clarity).In the historical period GB-aggregate Left: Daily-mean climatology's from October-March for the key variables used to produce FSI and SSI for the historical (solid) and future (dashed) periods.Right: Three weeks of GB-aggregate antecedent weather conditions for the historical extreme compound events (>P99 SSI and FSI, solid lines) and future (dashed lines).Lighter colour lines show the individual events for the historical period.wind gusts are slightly above the climatological average for approximately two weeks before the occurrence of the compound event, with steadily increasing anomalies seen for the 5 days preceding the event.We do however note individual events have quite a range of possible responses (light coloured lines in figure 4).A similar situation is seen for GB-aggregate river flows and precipitation, but interestingly the largest GB-aggregate precipitation totals are often seen the day before the compound event, suggesting a lagged response in the SSI-FSI behaviour, and a potential need for catchments to fully saturate before the most extreme events.This supports analysis from [47] showing that GB-flooding is driven by a combination of extreme precipitation and excess soil moisture (which could be related to high river flows).

Seasonal occurrences of events
Although the majority of this work is focused on individual days, we also examine the seasonal (October-March) joint behaviour of SSI and FSI as this is useful for setting annual insurance contracts.Figure 5 shows the normalised seasonal occurrence of SSI and FSI (all normalised to the maximum seasonal values from the historical period).Most years in the historical period are either prone to flooding (exceeding 90th percentile of normalised seasonal FSI) or prone to wind damage (exceeding 90th percentile of normalised seasonal SSI) supporting the results of [7].However, there are 3 years, marked with red dots which exceed the seasonal 90th percentile of both SSI and FSI (note 2.8 years are expected by random chance, so this is indistinguishable from expectation).In the two largest cases these are single very large multi-day event at the start of February during which the majority of that season's losses occur (see table A1).
In the future period there are now 9 seasons which exceed the historical 90th percentiles of both SSI and FSI.In these seasons it is now the accumulation of multiple separate wind and flood events leading to the largest seasonal totals, rather than temporally compounding, multi-day events.There are also notably more FSI events than in the historical period.
Figure 5 also shows normalised seasonal SSI and FSI values greater than 1, which means the seasons are more extreme than in the historical period.This is particularly interesting for SSI, as overall we see a 9% reduction in SSI days, here we see it is possible to have much more extreme seasonal SSIs than in the historical period.This may be due to the increased likelihood of sting jets forming in future wind storms [33] which lead to localised intense wind gusts or due to changes in convective storm dynamics [48].
Figure A6 shows the accumulated weather pattern totals for each type of seasonal extreme.Although in a future period a significant number of FSI and compound seasonal totals are seen, when these are normalised by the number of days experiencing an event per season then there are not notable differences between the type of weather patterns seen for each of the SSI, FSI or compound events.

Conclusions
This study has used a set of 30 UK-centred weather pattern classifications defined in [38] to understand the occurrence of, and meteorological drivers of daily and seasonal GB compound wind-flood events in a present and future climate.To define the compound events the SSI index used in [17,34] and FSI index developed in [9] are used with the previously validated UKCP18 regional 12 km dataset.Key results are given below: • Throughout the study period January and February have the highest risk of daily extreme compound events occurring (exceeding 99th percentile of both SSI and FSI).These events are generally associated with cyclonic weather patterns, often from the NAO+ and Northwesterly pattern classifications (see figure 1).• In a future climate, the risk of extreme compound events increases almost four-fold, with 14 events in the historical period, and 55 in the future period.It is also more likely for there to be multi-day compound events (up to 3 days long) in a future climate, and the risk is considerably larger in February than other months.(see figure 2).However, neither in the current or future climate models runs do we find more than one extreme compound event in a season.• Extreme compound events happen in a larger range of weather patterns in a future climate, extending the range of patterns identified in [23] for those interesting in potential fluvial flooding (see tables A1-A3).• The most extreme SSI, FSI and compound events are all associated with storms, but the synoptic situation is subtly different for each type of event.In a future climate the region exposed to extreme precipitation shifts from Northern-GB to Southern-GB.Suggesting in future events that these events may be spatially as well as temporally compounding (see figure 3).• In both a present and future climate the most extreme compound events are associated with anomalously high wind gusts and precipitation for around 5 days before the event, but anomalously high river flows for approximately 2 weeks before the event.This could lead to the catchment saturation needed for flooding (see figure 4).• At seasonal timescales years tend to be either floodprone or wind-damage prone.However in a future climate there is a larger proportion of years experiencing both extreme seasonal SSI and FSI totals (see figure 5).
This study presents insights into the meteorological conditions present during daily compound wind and fluvial flood risk days over GB which extend the recent literature which predominantly uses precipitation as a proxy for flooding [3,4,7].Early warning systems of this type of event are critical to limit the total loss and damage experienced.This type of weather pattern framework is useful when thinking about predictability of compound windflood events at medium-range to seasonal prediction timescales [38].We note that this work uses only one possible future climate scenario.Future work could extend this analysis over a range of future climate models, and climate change scenarios.Different percentiles could also be used for the future SSI and FSI calculations to account for potential future adaptation measures to extreme weather.GB has been used here as a case study due to the interest from the insurance community, but this work could easily be extended to any country where SSI and FSI indices could be calculated and a similar weather pattern classification applied.A final limitation is the treatment of only daily compounds, when we know that floods could occur within several days of a storm first occurring (and therefore well after the time of the extreme winds), particularly in large river catchments.The strongest correlation between wind gusts and river flow over GB have previously been shown to occur at a lag of approximately 2 weeks [9], where persistent synoptic situations could impact local catchment saturation [6,47].Understanding the meteorological drivers of compound events over these timescales is also of future interest.
The results also add to the ongoing discussion between meteorological and insurance sectors about the appropriateness of current generation of climate data for modelling and understanding the impacts of extreme meteorological hazards in a present and future climate.

Figure 2 .
Figure 2. Frequency of SSI (left) FSI (middle) and compound SSI and FSI events (right) in the future UKCP18 simulation (2061-2080).Top: all events, middle: events exceeding the 95th percentile of the given metrics bottom: events exceeding the 99th percentile of the given metrics.Colors indicate the weather pattern on the day of the event, characterized using the met office DECIDER tool (see methods for details).The axis in each subplot are fixed in figures 1 and 2 for visual clarity.

Figure 4 .
Figure 4. Left: Daily-mean climatology's from October-March for the key variables used to produce FSI and SSI for the historical (solid) and future (dashed) periods.Right: Three weeks of GB-aggregate antecedent weather conditions for the historical extreme compound events (>P99 SSI and FSI, solid lines) and future (dashed lines).Lighter colour lines show the individual events for the historical period.

Figure 5 .
Figure 5. Normalised seasonal totals of SSI vs FSI for the historical period (left, HIST) and future period (right, FUT).Black dashed lines show the historical 90th percentile of SSI and FSI.Red dots show seasons that exceed both of these percentiles.

Figure A3 .
Figure A3.Composites of (top) maximum daily-total precipitation (bottom) daily-maximum wind gusts for the largest (exceeding the 99th percentile) events for ERA5-derived SSI (left) Grid-to-Grid-derived FSI (right) in the historical period (1981-2000) with data taken from the ERA5 reanalysis.Axis are the same as figures 3 and A4 for direct comparison with the UKCP18 events.Contours show MSLP composites.

Figure A5 .
Figure A5.Mean sea level pressure (MSLP) composites for the largest (>P99) events for SSI (left) FSI (middle) and compound events (right).Top: historical UKCP18 period, 1981-2000, bottom: future UKCP18 period 2061-2080.Crosses show the minimum of MSLP during the event (note some are outside of the plot window so number of X's may not match) the number of events shown in tables A1-A3.

Figure A6 .A. 6 .
Figure A6.Top: Daily pattern occurrences during the extreme quadrants of figure 5 (points represented by red dots).Bottom: As top, but normalised by the total number of days present in the extreme seasons.

Table A3 .
Details of compound events (exceeding 99th percentile in both SSI and FSI) in the UKCP future simulations (2061-2080) ensemble members 11 and 12.