Nonstationary rainfall frequency analysis considering climate variability

Abstract

Recently, flood damage frequently occurs due to severe rain storms. One of primary causes that affect the recently increasing intensity of rain storm is global warming. In Korea, water engineers need to estimate design rainfalls in design target year to establish comprehensive flood mitigation plans reflecting an increasing trend observed in annual maximum rainfall data series. The increasing trend in rainfall series intensifies extreme hydrologic events as part of climate change. This study investigates the effect of climate factors on the trend in rainfall observations and the design rainfalls in Korea. This study also provides practical methodology to estimate design rainfalls in design target year considering trend in rainfall observations and climate variability. We collect rainfall data at 56 gauging stations in Korea which have observed data more than 30 years managed by the Korean Meteorological Administration. To estimate design rainfalls in design target year, parameters of probability distribution of rainfall data in design target year are estimated based on regression equations between rainfall observations and climate factors through downscaling methods. Various statistical methods are used in downscaling climate factors including a standard linear regression and non-linear neural networks. This study applies the backpropagation algorithm of artificial neural networks to the prediction of parameters of a distribution in design target year. Finally this study constructs the I-D-F curve in design target year and compares with the I-D-F curve established by the conventional rainfall frequency analysis.