Estimation of Return Level of Daily Rainfall Using Peak Over Threshold Theory

Abstract

Return levels of daily rainfall at 21 observatories all over Japan were estimated by applying Peak-Over-Threshold (POT) theory, which is statistics theory of extreme value dealing with data exceeding some specified value, and characteristics of POT estimates were examined by comparing with estimates from Annual Maximum Series (AMS) by applying conventional extreme value theory. Threshold value required to apply POT theory was standardized by using average and standard deviation of annual maximum daily rainfall for summarizing relationship between threshold values and return level estimates at all observatories. Results show that POT and AMS estimates of return level were close for short return period for all observatories; however, significantly different for long return period for some observatories where the largest daily rainfall is far from the second largest.