4.1 Attribution of weather patterns
To assess whether there has been an anthropogenic influence on the patterns seen over the UK, seasonal frequencies of weather patterns are compared for HadGEM3-A, which contains both 15 members ensembles with Natural Forcings only and with All Forcings (Fig. 3). The figure shows that for all seasons between 1980-2010 there are very few noticeable differences in pattern frequencies between ALL and NAT with there being no significant range differences for 28/26/23/25 of the 30 weather patterns for djf/mam/jja/son respectively. The same analysis was carried out for MRI-ESM2-0 for the 5 NAT and 5 ALL ensemble members between 1980-2010. This showed a few extra noticeable differences with no significant range differences for 22/27/28/19 out of the 30 patterns for djf/mam/jja/son respectively. The patterns out of the 30 that showed significant range differences were compared for both models and there was very little agreement between the two models on these weather pattern changes with no overarching trends (Supplementary Table 2). The only weather patterns that showed the same change in both models were pattern 20 (increasing in frequency in SON), pattern 5 (decreasing in frequency in SON), and pattern 29 (increasing in frequency in MAM). Overall, based on these models, no clear signal has appeared suggesting that anthropogenic climate change has not influenced the frequency of weather patterns significantly before 2010.
4.2 Future climate signal in weather regimes
To assess the influence of anthropogenic climate change on weather patterns in the future, the UKCP Global is examined for a past period (1960-2010), as well as a future period (2071-2100) for both a low (RCP 2.6) and high (RCP 8.5) emissions scenario. Given very few significant range differences were found between ALL and NAT forcings from 1980-2010, the historic period is used as a baseline for weather pattern frequencies when looking at the influence of anthropogenic climate change on the ACS.
The results show that there are significant range differences for the majority of the 30 weather patterns within all seasons between UKCP Global historic and future RCP 8.5. Figure 4 shows the differences for Autumn with the boxplots for future RCP 2.6 falling between the past and future RCP 8.5 time slices, indicating that anthropogenic influence is behind these changes in weather pattern frequencies. The main pattern seen in this figure for Autumn, is that there are significant increases in the lowest numbered patterns (summer-types) and a decrease in the highest numbered patterns (winter-types). This is also seen for the season of Summer, but not Winter or Spring in the UKCP projections (Supp. Fig. 4-6). These differences seen between the lowest and highest emission scenarios in the future are much greater than the differences seen between the past and the future low emissions scenario.
The CMIP6 models also show the same trends in Autumn but to a lesser extent, with a large increase in the lower numbered summer-type weather patterns and a decrease in the higher-numbered winter-type ones (Fig. 5). This change can also be seen in the Summer in MRI-ESM2-0 and HadGEM3-GC31 (Supplementary Table 3), however such a trend is not seen in Spring or Winter. This extension of the summer seen in the models could have big impacts on a number of weather variables including rainfall- which is examined in detail in the next section. It is also important to note that these changes have not appeared to occur in weather patterns seen when comparing ALL and NAT forcings before 2010 in the attribution section.
To examine when this signal emerges, the combined frequency of the predominantly summer type patterns (1-6) is plotted against the combined frequency of the predominantly winter type patterns (25-30) for the lowest and highest emissions scenarios for UKCP Global RCP 2.6 and RCP 8.5 respectively (Fig. 6). This shows that the signal could emerge as early as 2025 under these scenarios with the two emissions scenarios deviating to outside the 1900-2000 range from around 2035 onwards. By Autumn 2095 the frequency of winter-type patterns in SON could decrease by over a third under the highest emissions scenarios and summer-types increasing by a quarter. Under low emissions this change in seasonality in Autumn is much reduced as can be seen in figure 6.
4.3 Impacts of weather pattern changes on UK rainfall
To investigate whether the increase in summer-type patterns impacts rainfall, the ACS is isolated and compared to the overall change. The main focus of this analysis is for Autumn, given the strong signal seen in section 4.2. To assess the relative impact of the ACS against the overall signal, the method used in section 2.3 is applied for a past time period (1971-2000) and the future time period (2071-2100) under different emissions scenarios for Autumn, using the three regions chosen (Fig. 7). The climatology profile for each weather pattern for both the probability of an extreme rainfall day in Autumn (SON) (a day exceeding 99.5th percentile) and mean SON daily rainfall is calculated from the UKCP Global dataset over a baseline period running from 1900-1950.
Future circulation changes in the models when isolated show a 9-12% decrease in mean rainfall in SON from 1985 to 2085 over the three English regions (Table 1). The signal of this change is true for all 3 models examined under high emissions scenarios, with the strongest signals seen in the UCKP Global RCP 8.5 and HadGEM3-GC31 ssp585, with the MRI-ESM2-0 ssp370 showing a weaker ACS. The overall signal for Autumn mean rainfall, a decrease of 9-11% using the UKCP Global RCP 8.5 simulations is almost identical to the ACS, a decrease of 9-12%. This suggests that the overall mean rainfall signal leading to drier Autumns in the future is primarily driven by atmospheric circulation changes. Future ACS changes in the models when isolated also shows a 21-23% decrease in extreme daily rainfall events in Autumn (Table 1). The overall signal over the same time interval, however, shows a 71-103 % increase in extreme rainfall days in SON using the UKCP Global RCP 8.5 simulations. This would likely be even higher if it wasn't moderated by the 21-23% decrease due to the ACS.
Table 1: The change in rainfall indices between 1985 and 2085 for both the ACS and overall signal in Autumn. The two indices are the percentage change in mean precipitation (Δ mean pr) and the percentage change in the number of extreme rainfall days exceeding the 99.5th percentile (Δ 99.5th percentile daily pr). The uncertainty in the ACS comes from the model range (MR) as multiple models are examined, and for the overall signal using the UKCP Global 15 ensemble members, the 95% confidence intervals (CI) are used.
There are also big differences for rainfall indices between emission scenarios RCP 2.6 and RCP 8.5 for UKCP Global, both for the ACS and overall signal (Table 2). Overall, under a low emissions scenario the increase in Autumn extreme rainfall days is around 21-47 % depending on the region, compared to 71-103 % under the highest emissions scenario at the end of this century. The difference is even more pronounced for mean rainfall where there is only a 5% decrease in mean rainfall compared to a 10% decrease, when comparing RCP 2.6 and RCP 8.5 respectively. The ACS is also significantly reduced for both indices under a lower emissions scenario, with the signal between 2-3 times greater for the higher emissions scenario.
Table 2: The change in rainfall indices between 1985 and 2085 for both the ACS and the overall signal for different UKCP Global emissions scenarios. The two indices are the percentage change in mean precipitation (Δ mean pr) and the percentage change in the number of extreme rainfall days exceeding the 99.5th percentile (Δ 99.5th percentile daily pr).