Trend and correlation analysis of k-day extreme precipitation over Belgium

k-day extreme precipitation depths (k = 1,2,3,...30) for the climatological network of Belgium (165 stations) are analysed to detect a possible evolution in the occurrence of extreme rainfall events during the 1951-1995 reference period. The calendar year and the hydrological summer and winter are considered separately. Spearman's rank correlation coefficient shows a strong spatial correlation between extreme k-day precipitation events, depending on the time of the year (lower during summer than during winter) and increasing with k. In some cases the distances of decorrelation exceed 200 km which is comparable to the size of the country. Due to this correlation, tests for trends have been carried out on the leading principal components (PC) derived from the covariance matrix. Various PC selection rules have been applied to identify the number of components to analyse. The number of components needed to reproduce a given proportion of the total variance varies, with larger values for summer than fur winter and a decrease with growing k. The Fisher test is used as a global test. It combines the individual Mann-Kendall trend tests carried out on the selected PC scores. Significant trends have been found in extreme winter k-day precipitation for all the values of k and none in extreme summer precipitation. The results fur the annual k-day precipitation depths are between those for the two seasons: no trend for small k because summer events dominate and a significant trend fur k larger than 7 due to the winter events. Analysis of a few stations with long-term series shows no significant trend for the period 1910-1995, these series also reproduce almost the same trends as those found for the shorter 1951-1995 period.