4.1. Interseasonal (Seasonal exceedances)
The first objective is to assess the characteristics of clustering around the UK coast on interseasonal timescales.To do this we assess the number of exceedances per season from 1979/80 to 2021/22 above the 1 in 1- and 1 in 5-year thresholds in each of the three measured and modelled time-series andexamine how this varies in time and spatially among sites. The number of exceedances above the 1 in 1-and 1 in-5-year thresholds are plotted in Figures 4, 5 and 6 for surge, wave and still sea level, respectively. In each of these three figures the: 1 in 1-year exceedances are plotted in panels (a) and (b) and 1 in 5-year exceedances in panels (c) and (d); and measured data is plotted in figure panels (a) and (c) and the modelled data in panels (b) and (d). In these three figures, time (i.e., seasons 1979/90 to 2021/22) is represented on the x-axis and each site gauge is represented as a block on the y-axis, with the number of exceedances represented by a colour shading. On the y-axis the sites are plotted clockwise around the coast from Lerwick (at the top) for surges and still sea levels, and from Newbiggin for wave heights.The horizontal lines (blue: surges, red: wave heights and still sea levels) indicate the five regions shown in the inserted maps on the left side and in Figure 1. The total number of exceedances, and the number of seasons containing an exceedance for each site, is listed in Tables 1 and 2 for tide gauges and Table 2 for wave sites. Below we present the results, first for surges, then waves, and then still sea levels.
Storm Surge:First, we consider the measured storm surge component, shown in Figures4a and 4c. Without clustering, one might expect the 1 in 1-year exceedance to be reached each year at every site. However, it is evident in Figure 4a that some seasons have multiple 1 in 1-year exceedances across multiple sites, and some seasons have no exceedances at certain sites and in certain regions. The 2013/14 season is the clear standout with the majority of UK gauges seeing elevated counts of threshold exceedances. Spatial coherence among sites in each of the five regions is apparent; i.e., if one site as a high number of exceedances, this is likely also apparent at the other sites within that region. Furthermore, in most seasons, exceedance counts are high across two or more of the five regions during the same season; at the 1 in 1-year level there is no year when exceedance counts are higher for just one of the fiveregions alone, implying that seasonal clustering impacts on large spatial scales across the UK. The number of surge exceedances at or above the 1 in 1-year level are listed in Table 1. At all sites, the number of seasons when there was an exceedance is far less than the total number of seasons of data available, showing again that seasonal clustering is strongly prevalent in surge time-series. There were 161 storm surges exceedances in the 2013/14 season at or above the 1 in 1-year return level with 40 of the 42 operational tide gaugesexperiencing an exceedance this season. 33 (29) gauges experienced 2 (3) or more exceedances in 2013/14. The seasons at the 1 in 1-year return level threshold that havesimilar high counts of exceedances across a large spatial coverage are the 1992/93 and 2006/07 seasons.The 2006/07 season saw 114 storm surge exceedances across 41 out of 44 operational gauges for the 1 in 1-year level. In 1992/93, there was a similar number of storm surge exceedances (115)at the 1 in 1-year level, but these exceedances were seen across 32 out of 40 operational UK gauges. The 1992/93 season was the beginning of a 9-season period (1992/93-2001/02)which contained 7 out of the top 10 seasons on record in terms of total number of UK-wide storm surge exceedances at or above the 1 in 1-year return level threshold.Other periods of elevated exceedance counts occur with smaller spatial footprints. The 2011/12 season saw the second largest number of1 in 1-year surge exceedances in North Sea gauges. This was despite 5of the 12 operational gauges seeing no exceedances.The other UK regions during this season,except the Irish Sea,saw relatively low counts of exceedances relative to the rest of thetime period (1979-2021) and the number of operational gauges.
At the 1 in 5-year return level threshold (Figure 4c), the rarity of exceedances of this magnitude is apparent, as expected given the higher threshold, with long periods of few exceedances for large parts of the UK coastline and a much smaller overall count of exceedances (Table 2). Like with the results from the 1 in1-year threshold, some spatial coherence among sites in each of the fiveregions is apparent. However, with the exception of the 2013/14 season, exceedance counts are typically only higher across one or two of the five regions. This implies thatmeteorological forcing generallyresults in the seasonal clustering of smaller magnitude exceedances on a national scale, whereas the seasonal clustering of higher magnitude exceedances occur on a more local scale and are likely more reliant on storm paths.At the 1 in 5-year return level threshold, the 2013/14 season remains the most significant season on record with 52 storm surge exceedances. 33 of the 42 operational tide gauges saw at least 1 surge exceedance. The 1992/93 season sees the second highest seasonal total of 1 in 5-year surge exceedances (32), but the spatial coverage is reduced. All but 3 of the surge exceedances occurred across 16 North Sea and North Atlantic tide gauges. Interestingly, the 2006/07 season that had a high number of 1 in 1-year return level surge exceedances does not see a matching signal at the 1 in 5-year return level. Each region has numerousmulti-season periods where zero, or very few, exceedances were seen. For example, between 2003/04-2018/19 in the English Channel, 1 in 5-year surge exceedances were only recorded in the 2013/14 season. There are also periods where it is evident that few exceedances occurred across the entire UK tide gauge network. The seasons ofthe early-mid 1980s saw many seasons containing few exceedances, but there were 18 or more tide gauges that were not yet operational in this period. In contrast, some recent seasons where nearly all tide gaugeswere operational recorded few surge exceedances at the national scale.The 2003/04 season recorded only 2 surge exceedances at only 1 gauge, the seasons of 2016/17 and 2019/20 only saw 1 exceedance each, and the 2009/10 seasonrecorded no surge exceedances across all operational UK gauges.
The seasonal counts of the corresponding modelled storm surge extremes from the CODEC hindcast, at or above the 1 in 1-year and 1 in 5-year return level are shown in Figures 4b and 4d, respectively. Interestingly, although the CODEC hindcast has been extensively validated against measured still sea level data (with small root mean square errors and mean absolute errors in the high-frequency time-series; see Muiset al. 2020), the counts in exceedances above higher thresholds do not appear to closely agree. This would suggest that while the hindcast accurately captures the mean characteristics of measured still sea level well, the more extreme levels are not as accurately predicted at the gauge locations. For the coincident time period (1979/80-2018/19), it is apparent that CODEC underestimates the total number of exceedances across all gauges at the 1 in 1-year return level (1771 to 1502 exceedances), and the 1 in 5-year return level (382 to 179 exceedances), relative to the measured data.Differences in measured versus modelled total countsare alsolikely to be greater than stated, asthere is a significant proportion of measured data missing due to non-operational tide gauges,whilst the modelled hindcast provides continuous gap-free data throughout the time period.Although there issome coherence between the measuredseasonal storm surge exceedances and CODEC in relation to some ofthe periods of low, or no countsat the 1 in 1-year return level, the hindcast tends to overestimate the spatial footprint. The most evident difference between the modelled hindcast and the measured surge exceedances is the lessened signal of the 2013/14 season. It remains as one of the most significant seasons but does not dominate the record like in the measured surge record. Instead, the most significant season in the modelled hindcast records is 1989/90. At the 1 in 1 (5)-year return level, 158 (37) surge exceedances were recorded across 40 (27) sites, whereas in 2013/14 there were 147 (31) exceedances across 44 (29) sites.
Wave:Next, we consider wave heights. The seasonal counts of measured high significant wave heights at each of the 45 waves site, at or above the 1 in 1-year and 1 in 5-year return level are shown in Figures 5a and 5c, respectively.Due to the relatively short duration of measured wave data, it is more difficult to characterise interseasonal change. As with surge, the 2013/13 season stands out, with spatial coherence among sites being apparent in each of the five regions. The general pattern of wave exceedances is comparable to surge exceedances from the corresponding English tide gauges, but there are a greater number of neighbouring wave sites recording exceedances with the higher spatial density of wave sites.At the 1 in 1-year return level (Figure 5a), numbers of exceedances are fairly consistent across all regions throughout the short record, with only the 2012/13 season having a high number of sites not experiencing an exceedancewith most sites being operational. At the 1 in 5-year return level (Figure 5c), the North Sea sites appear to more consistently experience seasons with exceedances, but this is difficult to characterise relative to the particularly short data lengths. For example, Chapel Point has 4 seasons with a wave height exceedance above the 1 in 1-year level out of only 9 seasons in its record (Table 3). Sites in the English Channel have been operational for longer and show that periods of elevated counts of 1 in 5-year wave height exceedances can be preceded and/or followed by periods of no exceedances. There were only 8 1 in 5-year exceedances recorded between 2003/04-2012/13 across all operational sites in the English Channel and just one exceedance occurred between 2018/19-2020/21. In contrast, the extreme season of 2013/14 produced significantly higher exceedance counts in the English Channel. At both 1 in 1- and 1 in 5-year return levels, the 2013/14 season has the highest number of exceedances across all sites and most instances of a site experiencing >1 exceedances in a season. This is primarily seen in English Channel wave sites. Across all sites and seasons of measured wave height data, ~25% of all 1 in 5-year exceedances happened at English Channel sites in the 2013/14 season.
The seasonal exceedance counts of the modelled high significant wave heights from the closest ERA5 hindcast grid node to the corresponding wave site, at or above the 1 in 1-year and 1 in 5-year return level are shown in Figures 5b and 5d, respectively. There are fewer ERA5 grid nodes than CCO wave sites due to the coarse resolution of the ERA5 wave hindcast.It is important to note that spatial coherencebetween sites in the modelled datawill be strengthened by the fact that 13 sites share the same ERA5 grid nodes.Due to short wave records, the modelled data can only be contrasted to the measured data for the last ~10-15 years. The 2013/14 season is seen to have a high number of counts and a large spatial footprint at both return levels. However, like with the modelled surge component, the 1989/90 season has a higher total of exceedance counts than 2013/14 (but it cannot be validatedwhether this is unusual from the measured data like it was with surge levels, as measured wave data is not available this far back in time at most sites).After the 2013/14 season, the modelunderestimates the number of exceedances by 173 at the in 1-year level and 46 at the 1 in 5-year level when compared to the measured data.Expectedly, there are some clear similarities in the general patterns seen between the ERA5 hindcast wave heights and the ERA5-forced CODEC surge exceedances as both are primarily driven by the meteorology of the ERA5 reanalysis. At the 1 in 1-year return level, the ERA5 seasonal exceedances closely resemble the CODEC seasonal exceedances at the English tide gauges.One of the only noteworthy differences is the strong wave signal in 1979/80 which is not seen at most English tide gauges in the modelled surges. Interestingly, when discounting the shared wave grid nodes and focusing on the 30distinctEnglish wave grid nodesanalysed in the ERA5 hindcast, compared to 46 UK grid nodes analysed in the CODEC hindcast, the modelled wave data has 58 more exceedances than the surge hindcast at the higher 1 in 5-year return level threshold. The modelled seasonal exceedances of wave heights also a higher level of spatial coherence, with 4 more seasons having 6 or more neighbouring sites all experiencing at least one exceedance, despite the lower overall number of sites. This further highlights the smaller spatial scales associated with the seasonal clustering of higher magnitude exceedances.
Still Sea Level:Finally, we consider high stillsea levels (e.g., tide plus surge, offset by MSL rise). The number of still sea level exceedances at or above the 1 in 1-year and 1 in 5-year per season, is shown in Figure 6a and 6c for the measured data at each of the 46tide gauge sites.As with storm surges and waves, the 2013/14 season is the most significant season on record for the total number of UK-wide measured high still sea level exceedances at both the 1 in 1- and 5-year return level threshold.There were 175 (68) high still sea level exceedances across 39 (34) of the 42 operational tide gauges at the 1 in 1- (5-) year return level. It is rare for a site to experience more than one 1 in 5-year still sea level exceedance in a season yet in 2013/14 this was the case for 21gauges. The 2006/07 season had the second highest count of 1 in 1-year still sea level exceedances, experiencing 149 across 40 of 44 operational gauges. However, as with storm surges, this season does not see a matching signal of 1 in 5-year return level exceedances. Only 15 1 in 5-year high still sea level exceedances were recorded across 12 gauges.Numerous1 in 1-year return level exceedances that are clustered in the 1993/94 to 2001/02 period (Figure 6a) are also not represented to asimilar extent at the 1 in 5-year return level. There are clear differences in the measured still sea levels compared to the measured seasonal storm surge exceedances (Figure 4a and 4c). Seasons that contain storm surge exceedances on local and regional scales do not necessarily see a corresponding signal of still sea level exceedances.At the lower 1 in 1-year return level,there are254 more measured still sea level exceedances than storm surge exceedances (Table 1). All regions except the North Sea recorded more still sea level exceedances than storm surge exceedances in the record. It is evident from Figures 6a and 4a that the higher number ofstill sea level exceedances are clustered spatially and temporally to a larger extent than the storm surge exceedances, as there is a greater number of seasons with nostill sea level exceedances at regional and national scales.The 9-season period (1992/93-2001/02) which contains 7 out of the top 10 seasons on record for 1 in 1-year return level surge exceedances, however, does also contain 7 of the top 10 seasons for high still sea level exceedances (6 matching seasons). Conversely, at the higher 1 in 5-year return level threshold (Figure 6c), there were 44more storm surge exceedances than high still sea level exceedances in the record (Table 2). This is despite both the North Atlantic and Bristol Channel regions recording more highstill sea level exceedances than storm surge exceedances. The North Sea saw the highest number of storm surge exceedances (121) and the greatest difference between the number of still sea level exceedances (44 more surge exceedances, 26 more than the next largest regional difference in theEnglish Channel) at the 1 in 5-year return level.
The number of modelled 1 in 1- and 5-year high stillsea level exceedances(again offset for MSL rise)from the CODEC hindcast is shown in Figure 6b and 6d for the grid nodes closestto each tide gauge.For the total count of exceedances across all gauges, the CODEC hindcast appears to better represents still sea level exceedances than surge exceedances at both 1 in 1- and 5-year thresholds for the coincident timeperiod. At the 1 in 1-year return level (Figure 6b), CODEC slightly underrepresents the measured data by a difference of 94 still sea level exceedances, whilst at the 1 in 5-year return level (Figure 6d) CODEC slightly overestimates with 118 more modelled stillsea level exceedances.The modelled still sea level data is consistent with the some of the patterns seen across the modelled data for surges and waves. When exceedances occur in the modelled data, they tend to affect multiple neighbouring sites. The modelled still sea level data has over double the number of seasons where 10 or more neighbouring tide gauges experience at least one 1 in 1-year exceedance. The 2013/14 season has a particularly reduced signal in the modelled still sea levels, despite having the third highest total count for a season; there were 87 (38) fewer exceedances compared to 1989/90. Like with surge levels, there is some coherence between measured and modelled still sea levels in relation to period of low, or no counts.
4.2. Intraseasonal: Days-betweenexceedances
The second objective is to examine the characteristics of clustering on intraseasonal timescales. To achieve this, we analyse the number of days between consecutive exceedances at or above the 1 in 1- and 5-year return levelsfor each of the threeparameters, at each gauge site.We do this only for the measured datasets and the entirety of the measured data records available. As highlighted in Section 4.1, the modelled hindcast datasets see key differenceswhen compared to the measured datasets in theseasonal count patterns of storm surge, wave height, and high still sea level exceedances.
The number of days between consecutive measured exceedances that occur within 365 days of the previous exceedance, at or above the 1 in 1- and 5-year thresholds, are shown in 7a and 7b for surge levels, 7c and 7d for wave heights, and 7e and 7f for still sea levels, for all gauge sites available. Consecutive events occurring within 365 days of each other are evident at both return level thresholds. However, it is clear that therearemany more examples at the lower 1 in 1-year return level. Certain periods can be seen where many exceedancesappear to cluster over a similar time period, with numerousexceedances occurring within 365 days of the last.In each of the three parameters, there are a number ofexceedances that occurred within very small time periods (i.e., less than three days apart) for the 1 in 1-year threshold. Interestingly, there are many exceedances of still sea level at or less than 1 day, and between 6 and ~10 days (Figure 7e); a feature not seen in the surge (Figure 7a) and wave (Figure 7c) results. This is likely a result of the spring neap tidal cycle and confirms the findings of Haigh et al. (2016); when storms are separated by 4–8 days, one will always occur during neap tide, and the combined still sea level, even with a large storm surge, is unlikely to be high enough to lead to a high still sea level. As expected, given the higher threshold, there are much fewer occurrences of exceedances occurring within shorter periods at the 1 in 5-year level, particularly in the surge and wave records.
The total number of exceedances and the minimum, maximum, median, and average number of days between consecutive exceedances for each site is listed in Tables 1 and 2 for tide gauges and Table 3 for wave sites. At the 1 in 1-year return level, the clustering of exceedances is prevalent.For example, there are 1793still sea level exceedancesthat occur across all 45-gauge sites within 365 days of the last exceedance. However, at the 1 in 5-year return level, the clustering of exceedances becomes far less apparent. Only ~7% of the 1793 still sea level exceedances at or above the 1 in 1-year are at or above the 1 in 5-year level.A similar proportion of 1 in 5-year exceedances is seen across the other parameters analysed when compared to their respective 1 in 1-year return level exceedance totals (~7% for surges,~8% for waves).Nevertheless, the proportions of 1 in 5-yearexceedances occurring in quick succession is still of significance given the rarity of exceedances of this magnitude.Importantly, all parameters see the highest number of the exceedances occurring on short temporal scalesin the periods of highest data availability. Therefore, it is likely that due to the drop off in data availability before the 1990’s for tide gauge data, and the late 2000s for wave data, that a significant number of exceedances have been missed. The gap between many consecutive exceedancesmay also consequently be overestimated. Despite this, it is clear that significant numbers of exceedances are clustering over short periods.
Next, we consider the percentage of consecutive exceedances around the UK occurring at six different chosen timescales, namely: <1-3 days, 3-14 days, 14-50 days, 50-100 days, 100-365 days, and ≥365 days. Results are shown in Figure 8 for the three parameters, for both the 1 in 1- and 1 in 5-year thresholds. For all parameters, the majority (>65%) of 1 in 1-year exceedances occur within 365 days of the previous exceedance. Importantly, significant proportions occur on far shorter timescales. ~24% (~44%) of still sea level exceedances occur within 3 (50) days of the previous exceedance. ~10% of storm surgeexceedances occur within 3 days and~26% (~39%) occur within 14 (50) days of the previous exceedance.~20% (~35%) of wave height exceedances occur within 14 (50) days. Of thestill sea levelexceedances occurring within 2 weeks of the last, ~75%occur in less than 2 days(Supplementary Figure S4). At the 1 in 5-year return level, exceedances occurring within 365 days of the previous exceedance account for ~26% of surge exceedances and ~34%of still sea levelexceedances. Although there are 33 examples of quick succession still sea level exceedances occurring within 14 days of the last exceedance, all but one occurred within 4 days of the previous exceedance.
The percentage of consecutive exceedances at the six specified timescales are shown in Figure 9a-e for surge and still sea level, but this time averaged across the sites within the specific fiveUKregions (shown in Figure 1a) and again for all sites (Figure 9f). Spatially, clustered periods of 1 in 1-year exceedances account for similar proportions of consecutive exceedances across all UK regions; there are negligible differences between regions. The regions that see the smallest proportions of successive 1 in 1-year still sea level exceedances occurring within 365 days of the previous exceedance are the North Sea and English Channel.The Bristol Channel and North Atlantic see the highest proportions occurring within 365 days of the prior exceedance. Both regions see over half of still sea level exceedances occurring within 50 days of the last exceedance. At the higher 1 in 5-year return level, the North Sea sees the smallest proportions of exceedances occurring within 365 days of the previous exceedance for both still sea level and surge parameters.Only ~21% of still sea level exceedances and~17% of surge exceedancesoccur within 365 days of the prior exceedance. The Bristol Channel has the highest proportions for the two parameters. Despite no 1 in 5-year still sea level exceedances occurring in the Bristol Channel within 14 days of each other, ~60% occur within 365 days of the last. Conversely, the Bristol Channel has a high proportion of 1 in 5-year surge exceedances occurring within 14 days of each other (~17%).North Atlantic gauges have no still sea levelexceedances occurring within3-14 or 50-100 days of the last. Nevertheless, the North Atlantic still has the second highest proportion of exceedances (~33%) occurring within 365 days of the prior exceedance. The North Atlantic has a higher percentage (~21%) of 1 in 5-year surge exceedances occurring within 14 days of the previous exceedance than the North Sea hasexceedances occurring within 365 days of the last (~17%), yetNorth Atlantic gauges record no instances of a 1 in 5-year surge exceedance occurring within either 14-50 days or 50-100 days of the previous exceedance.
The percentage of consecutive exceedances at the six specified timescales are shown in Figure 10 for significant wave height; as the CCO data is only for England, the data is averaged across the sites in five main English regions (Figures 10a-e, the regions are shown in Figure 1b) and again for all sites (Figure 10f). The results must be approached with some caution, as some regions have a low number of sites, and the overall wave record length is short. However, for the measured exceedances, the North Sea has by far the highest percentage (~20%) of 1 in 1-year exceedances occurring <1-3 days after the previous exceedance,when compared to the other English regions. The other regionssee ~14-21% of consecutive exceedances occurring within 14 days of the previous – except for the Irish Sea which only experiences ~4% of exceedances in<1-3 days of the prior exceedance and no exceedances within 3-14 days. At the 1 in 5-year return level, the differences become starker as the greater rarity of these exceedances combine with the short records. The Irish Sea sees no exceedances within 14 days of the previous exceedance, the Celtic Sea has no exceedances within 50 days of the prior, and the Bristol Channel has no exceedances within 100 days of the prior.~5 (~11%) of North Sea 1 in 5-year return level exceedances occur within 3 (14) days of the previous exceedance, whereas the English Channel sees no exceedances within 3 days of the previous exceedance but has ~18% occurring within 14 days.