Watershed Scale Extreme Rainfall Analysis Under Climate Change

Abstract

Traditionally, to analyse the rainfall frequency at specific sites, individual rainfall stations are used that collect data on the annual maximum rainfalls. However, results from analysis such as this only capture the rainfall characteristics at each specific site and do not take the spatial covariation structure of extreme rainfalls of real rain storms into account. Thus, the picture that is created from this form of analysis is incomplete. Cheng's team is making use of covariance, so that they can understand how events in one place can influence on events in other places, rather than treating them entirely independently of each other. "By analysing historic typhoon rainfalls in Taiwan, we found that extreme rainfalls of true rainstorms exhibited strong spatial correlation. In particular, a single catastrophic storm could even result in annual maximum rainfalls of different durations at several rainfall stations in the same watershed," observes Cheng. "Traditional site-specific frequency analysis cannot not provide estimate of the return period of such catastrophic storms."Within the study, Cheng and his team developed a stochastic simulation approach which took into account the spatial correlation of extreme rainfalls, at the same time as preserving marginal probability distributions of extreme rainfalls at individual rainfall stations. This methodology enabled the team to estimate the return periods of real catastrophic storms. Importantly, where hydrologists tend to define extreme rainfalls as those with a rainfall depth of 24-hour duration and 100-year return period, that is based on the results of site-specific rainfall frequency analysis. However, rainfalls events such as those are only hypothetical and are this termed 'design rainfalls'. In reality, a catastrophic storm can result in heavy rainfalls over an extensive region and cause tremendous damages and casualties. Ultimately, the team has managed to develop a stochastic approach for the simulation of multi-site event-max rainfalls. By increasing the sample size and considering the spatial covariation of EMRs, the team was able to ascertain that the return periods of site-specific and multi-site rainfall extremes can be better estimated.