Comparison of Bias Correction Methods for Summertime Daily Rainfall in South Korea Using Quantile Mapping and Machine Learning Model

This study compares the bias correction techniques of empirical quantile mapping (QM) and Long Short-Term Memory (LSTM) machine learning model for summertime daily rainfall simulation focusing on precipitation-dependent temporal variation. Numerical experiments using Weather Research Forecasting (WRF) were conducted over South Korea with lateral boundary conditions ERA5 reanalysis data. For spatial distribution mean rainfall, bias-uncorrected WRF (WRF_RAW) showed dry most region Korea. The results corrected by QM LSTM (WRF_QM WRF_LSTM, respectively) improved summer root square error values 0.17 0.69, respectively, which smaller than those WRF_RAW (1.10). Although WRF_QM performed better WRF_LSTM in terms monthly precipitation, presented a closer interannual variation to observation WRF_QM. coefficient determination calendar-day was highest following order: (0.451), (0.230), (0.201). However, had limitation reproducing extreme exceeding 50 mm/day due few cases precipitation training Nevertheless, simulated observed light-to-moderate (10–50 mm/day) others.