Abstract

Abstract. The theory of record-breaking processes offers a framework for understanding extreme events which is nearly independent of the theory of extremes. The mathematical theory of record-breaking processes is applied to the problem of identifying nonstationarity in hydrological records. A record flood event is simply an event which exceeds all previous events. The probability distribution and first four moments of the number of record events in an n-year period, R, are derived for a serially independent process. The variance of estimates of the mean, standard deviation, and coefficient of variation R is also derived. In addition, approximate confidence intervals are derived for the mean number of record-breaking events in a region with spatially correlated flood series. Using these results, in combination with 1571 flood records in the United States, we document that the average number of record breaking flood events over n-year periods ranging from [10, 80] behaved as if the annual flood series were serially independent for all regions of the United States. However, when spatial correlation of the flood records is ignored, as is the case in many previous studies, it appears as if flood records are not serially independent in the western and Midwestern regions of the United States. These results emphasize the importance of accounting for the spatial correlation structure of hydrologic records when performing regional hypothesis tests. 1.