Improving the SM2RAIN-derived rainfall estimation using Bayesian optimization

Abstract

The rainfall product derived from the SM2RAIN (Soil Moisture to Rain) algorithm has been widely used. However, there is still a large uncertainty partly due to the soil moisture input and parameters estimation of the SM2RAIN algorithm, which limits the application of the model in alpine regions. Here, the SM2RAIN-BayesOpt algorithm was developed by integrating the SM2RAIN algorithm and Bayesian optimization to improve the estimation of parameters (Z, a, b, Tbase, Tpot), subsequently incorporating SMAP Level-3 soil moisture products for rainfall estimation. The performance of the SM2RAIN-BayesOpt algorithm was evaluated based on observed rainfall data under different environmental conditions in three typical alpine regions, namely Tibetan Plateau, Heihe River Basin, and Shandian River Basin. Moreover, SM2RAIN-BayesOpt, IMERG-V06B, and ERA5 reanalysis rainfall estimates were also compared with in-situ rainfall observations. The results showed that the proposed SM2RAIN-BayesOpt algorithm can obtain more accurate rainfall estimates in all studied areas in terms of different evaluation metrics. It was also found that our proposed SM2RAIN-BayesOpt algorithm performs better in alpine meadows and grassland than in desert and forestland. SM2RAIN-BayesOpt algorithm can considerably improve the accuracy of rainfall estimation, and it is of significant value for rainfall monitoring in alpine regions where observational data are scarce.