Estimation of SPEI Meteorological Drought Using Machine Learning Algorithms

Accurate estimation of drought events is vital for the mitigation their adverse consequences on water resources, agriculture and ecosystems. Machine learning algorithms are promising methods prediction as they require less time, minimal inputs, relatively complex than dynamic or physical models. In this study, a combination machine with Standardized Precipitation Evapotranspiration Index (SPEI) proposed analysis within representative case study in Tibetan Plateau, China, period 1980–2019. Two timescales 3 months (SPEI-3) 6 (SPEI-6) aggregation were considered. Four models Random Forest (RF), Extreme Gradient Boost (XGB), Convolutional neural network (CNN) Long-term short memory (LSTM) developed SPEIs. Seven scenarios various combinations climate variables input adopted to build The best XGB scenario 5 (precipitation, average temperature, minimum maximum wind speed relative humidity) RF speed, humidity sunshine) estimating SPEI-3. LSTM 4 speed) was better SPEI-6 estimation. model 7 (all variables, i.e., + solar radiation). Based NSE index, performances classified good fits both timescales. produced satisfactory results could be used rapid tool decision making by water-managers.