Trends in UK ¯ood behaviour are examined using extensive peaks-over-threshold and annual maxima data from 890 gauging stations. The analysis focuses on changes that have occurred at the national scale; data were therefore pooled across all available sites. Two annual series are examined, representing ¯ood size and frequency of ¯ood occurrence. Results are presented for the 40-year period 1941±1980, for the 50-year period 1941±1990 (annual maxima only), and for some limited data for 1870±1995. Three main tests for trend (linear regression, normal scores regression and Spearman's correlation) were applied using (i) conventional theoretical signi®cance levels, and (ii) a permutation approach. Test results were generally signi®cant under the conventional approach but were not signi®cant under the permutation approach. This difference occurs because some of the independence assumptions required for the conventional approach are not satis®ed. There is a high yearto-year variation inherent in the data and, relative to this variation, any underlying trends are not signi®cant. Regional and seasonal effects are considered but no signi®cant trends emerge.
Although no trends in national ¯ood behaviour are seen, the year-to-year variations in UK ¯oods appear to show some systematic behaviour (¯uctuations) when smoothed using locally weighted regression. Further investigation suggests that these ¯uctuations are driven climatically; similar ¯uctuations are evident in UK annual rainfall values. The in¯uence of climatic variation on the pattern of ¯ood occurrences also means that, at the national scale, progressive changes in the UK ¯ood regime, e.g. arising from land-use change, are hidden. Over the period 1940 to 1980 smoothed curves suggest increased ¯uctuation, with emergence of some possible cyclic tendencies. However, in the context of the limited data going back to 1870, this increased ¯uctuation is not atypical. Overall, no evidence for climate change emerges for the periods for which data are currently available, although climate clearly has a marked effect on ¯oods. The observed climate-linked ¯uctuations indicate that inferences from short ¯ood records may differ appreciably from those obtained from longer series. This could have potentially important implications for ¯ood-risk studies and trend detection.