A novel bias correction method for extreme rainfall events based on L-moments

The effectiveness of bias correction (BC) of global-scale future climate projections is crucial in climate change studies. The magnitude of error in the BC affect climate change adaptation decisions. The existing BC methods vary in their complexity, and exhibit limitations on data length, degrees of freedom etc. This study proposes a new method, L-moments Scaling (L-mS), which is both parsimonious and efficient in bias-correcting extreme rainfall events. The L-mS method applies corrections to the first three L-moments of the data to bias correct the entire distribution. The proposed method's efficiency was demonstrated at two stations in India, Chennai and Hyderabad, for 1-day Annual Maximum (AM) precipitation simulations from EC-EARTH and MIROC5 models. A comparison was performed with five widely used BC methods using two validation procedures: Strict Split-Sample (SSS) and Bootstrapped Split-Sample (BSS). The results revealed that the L-mS method could outperform all the five BC Methods with increased accuracy (0%–18% in SSS and 3%–21% in BSS), and with minimal variability among the bootstrapped samples in terms of Normalized-Root-Mean-Square-Error (NRMSE). The method was also applied on 1° gridded data over India for 1-day, 2-day, 3-day, 7-day AM, and Annual Totals, and the future (2021–2050) projections were bias-corrected using L-mS method. The L-mS method produced at least 2.5 and 3 times lesser error in mean and standard deviation, respectively, compared to observed extremes, across all the grids. The L-mS method was able to utilize the inherent nature of frequency domain analysis to outperform similar advanced methods by correcting the entire AM data with few key statistics and could serve as an efficient tool in the BC of extreme climate variables. Also, the bias-corrected future projections indicated the magnitudes of extreme rainfall events are expected to decrease in 35%–40% and increase in 60%–65% of the grids.