Comparing Machine-Learning Models for Drought Forecasting in Vietnam's Cai River Basin

Drought occurs throughout the world, affecting people more than any other major natural hazards -especially in the agriculture industry. An effective and timely monitoring system is required to mitigate the impacts of drought. Meanwhile, extreme learning machine (ELM), online sequential extreme learning machine (OS-ELM), and self-adaptive evolutionary extreme learning machine (SADE-ELM) are rarely applied as the alternative drought-forecasting tools in the meantime. The present study aims to evaluate the ability of these models to predict drought and the quantitative value of drought indices, the standardized precipitation index (SPI), and the standardized precipitation evapotranspiration index (SPEI). For this purpose, the sea surface temperature anomalies (SSTA) events at NinoW and Nino4 zones were selected for input variables to forecast drought. The SPI/SPEI values may contain a one/three/six-month dry and a one/three/six-month wet period in short-term periods, and this causes instability. For this reason, 4 models for SPI/SPEI (12 months) were trained and tested by these methods, respectively. According to two statistical indices (RMSE and CORR) and stability of these methods, the SADE-ELM models perform the best, and the performance of the OS-ELM models are better than the ELM models.