This research was part of the SWURVE project and has been published in:
Fowler, H.J. and Kilsby, C.G. 2003. A regional frequency analysis of United Kingdom extreme rainfall from 1961 to 2000. International Journal of Climatology, 23(11), 1313-1334. [Abstract]
Fowler, H.J. and Kilsby, C.G. 2003. Implications of changes in seasonal and annual extreme rainfall. Geophysical Research Letters, 30(13), 1720, doi:10.1029/2003GL017327. [Abstract]
Increasing flood risk is now recognised as the most important sectoral threat from climate change in most parts of the world, with recent repeated severe flooding in the UK and Europe causing major loss of property and life, and causing the insurance industry to threaten the withdrawal of flood insurance cover from millions of UK households. This has prompted public debate on the apparent increased frequency of extremes and focussed attention in particular on perceived increases in rainfall intensities. Climate model integrations predict increases in both the frequency and intensity of heavy rainfall in the high latitudes under enhanced greenhouse conditions. These projections are consistent with recent increases in rainfall intensity seen in the UK and worldwide.
We use two methods to examine changes to the frequency and intensity of extreme rainfall events: a regional frequency analysis (RFA) based on L-moments (Hosking and Wallis, 1997) and a peak-over-threshold (POT) analysis. The RFA uses regional pooling (Hosking and Wallis, 1997) of rainfall maxima, standardised by median (RMED), to fit Generalised Extreme Value distribution curves and allow the estimation of long return period rainfall events for 1-, 2-, 5- and 10-day durations. These include estimates of uncertainty, measured using a bootstrap method. For the POT analysis, we consider the POT event, defined to occur if the total daily rainfall exceeds two standard deviations above the long term (1961-2000) mean wet day at a specific location. The standardised POT data is regionally pooled and used to examine changes in the timing and frequency of extremes, specifically using the statistical measures of modal POT month and annual POT frequency (Bayliss and Jones, 1993). Temporal variability in extremes is analysed in fixed decades from 1961-1970, 1971-1980, 1981-1990 and 1991-2000, using the conventional meteorological definition of seasons: winter (December-February), spring (March-May), summer (June-August) and autumn (September-November). These techniques are used to analyse the changing probability of extreme rainfall events for observed data from 1961- 2000 for 204 daily rainfall records across the UK regionally pooled into the nine regions (Wigley et al., 1984), with each containing at least 20 stations (Figure 1). A discordancy analysis (Hosking and Wallis, 1997) was used to establish that the distributions of station maxima within a region were acceptably similar.
Figure 1 Location of 204 UK daily rainfall records used in the study and the 9 rainfall regions. The regions are: North Scotland (NS), South Scotland (SS), East Scotland (ES), Northern Ireland (NI), Northwest England (NWE), Northeast England (NEE), Central and Eastern England (CEE), Southeast England (SEE) and Southwest England (SWE).
We find that there have been significant, seasonally and regionally varying, changes in the intensity of extreme rainfall events across the UK. The largest changes are observed in autumn and spring, but considerable change is also seen in winter and summer extremes. These findings contradict those of Osborn et al. (2000), who suggest that the largest changes to rainfall intensity have been experienced in winter and summer, as do regional climate models projections for future climate scenarios (Jones and Reid, 2001). These analyses (Osborn et al., 2000; Jones and Reid, 2001) only examine change to daily rainfall intensities. In many seasons, however, the largest change is found for 5- and 10-day duration events.
We find that in summer there is a consistent downward trend in estimated rainfall for both short and long-duration return periods, ranging from 2 to 100 years. This is observed particularly at 1- and 2-day durations (see Figure 2a) and most prominently in the south and east of the country, where there are also significant decreases in estimated rainfall for 5- and 10-day duration events. This is a result of a significant decrease in the median seasonal maximum event (SMED) since the 1970s. In winter, changes are spatially variable across the UK (see Figure 2b). In Scotland, there is an increase in estimated rainfall for a specific return period event, particularly at 5- and 10-day durations but, conversely, little or no change in more southern regions.
Figure 2 50-year rainfall event return period estimates shown for nine regions and four decadal periods, 1961-1970, 1971-1980, 1981-1990 and 1991-2000. a, Summer, 1-day duration. b, Winter, 10-day duration.
Figures 3a and 3b show decadal rainfall estimates for a 25-year return period event for autumn and spring respectively. The most recent decade from 1991-2000 provides a dramatic departure from previous estimates in all regions and for all durations. In autumn, increases in rainfall estimated for a specific return period at 5- and 10-day durations are mainly seen in Scotland and the east of England, but increases at 1- and 2-day durations are observed across the country. This is due to increases in SMED, particularly at the shorter durations. In spring, the picture of change is similar, with increases in extreme rainfall most prominent at the 1- and 2-day durations but increases at 5- and 10-day durations in northern and western parts of the UK.
Figure 3 25-year rainfall event return period estimates shown for nine regions and four decadal periods, 1961-1970, 1971-1980, 1981-1990 and 1991-2000. a, Autumn, 10-day duration. b, Spring, 2-day duration.
It is clear from our research that there have been significant changes to both the timing and occurrence of multi-day intense rainfall events over the past decade. We estimate that the magnitude of multi-day extreme rainfall has increased two-fold over parts of the UK since the 1960s. Annual recurrence probabilities are quadrupled in some regions, with intensities previously experienced every 25 years now occurring at 6 year intervals. This is comparable to climate model projections for the end of the 21st century (Figure 4).
Figure 4 Comparison of estimates of 10-day duration, 25-year return period (or 4% chance) rainfall event for both observed and regional climate model data. a, observed 1961-1990. b, observed 1991-2000. c, HadRM3 control (1961-1990) scenario, ensemble mean d, HadRM3 future (2070-2100) scenario, ensemble mean.
In terms of flooding, a change to the frequency and timing of extreme rainfall events may be as important as changes to their magnitude and duration, as this determines the timing of the return of catchment soils to near field capacity and thus the timing of major flood events (Bayliss and Jones, 1993). Autumn has typically been the dominant season of flooding in northern and western parts of the UK. This may be due to the soils in these areas returning to near field capacity earlier in the year than those in the south and east, which do not usually reach these values until December or January (Bayliss and Jones, 1993). Further south however, until recently, the dominant season of flooding has been winter (Bayliss and Jones, 1993). However, recent flood events, both in autumn 2000 and autumn 2001, have predominantly affected the south and east of the UK.
Figure 5 The change in timing of extreme rainfall events over southern parts of the UK from 1961-2000, measured using peak-over-threshold probabilistic analysis. a, The annual frequency and modal month of 1-day POT events in the south east of England, overlain by 5-year moving average annual POT frequency. b, The decadal station frequency of 10-day POT events in central eastern England.
We find that there has been little change to the timing and frequency of extreme rainfall events in northern and western parts of the UK. However, in southern and eastern regions, there have been significant changes to both the timing and frequency of intense rainfall that may have severe implications for flooding. Figure 5a shows the annual frequency and modal month of POT events in south eastern England. Over the 1990s, the annual frequency of POT events has increased substantially, to 50% above the long-term average. Changes in timing are also apparent, with a much greater concentration now occurring in autumn and winter months. This is further demonstrated by Figure 5b, which demonstrates the increased frequency of POT events in central and eastern England during the 1990s, and the change in timing from predominantly summer to autumn. These changes are found throughout the southern UK during the 1990s, for all rainfall durations.
This research suggests that causal mechanisms such as the frequency, duration and timing of extreme rainfall events are changing. These seasonal changes may be caused by atmospheric circulation anomalies in the Scandinavia pattern or the North Atlantic Oscillation and help to explain recent severe flood events in the European region. However, further research is needed in this area to firmly establish links between flood generating mechanisms and large-scale circulation patterns if we are to fully understand the risk implications of the estimated changes in extreme rainfall occurrence for flooding.