Tree-based ensemble model prediction for hydrological drought in a tropical river basin of India

Conventional hydrological modeling is highly tedious for a data-sparse tropical river basin heavily loaded by hydraulic structures. The application of promising machine learning algorithms in this context is not explored much. This study investigates the suitability of two machine learning approaches—random forest regression model (RF) and gradient boosting regression model (GBM) in predicting the hydrological drought indices—standardized streamflow index (SSI) and streamflow drought index (SDI) based on the predicted meteorological drought indices—standardized precipitation index (SPI) and standardized precipitation evapotranspiration index under the future climate Representative Concentration Pathway (RCP) scenarios RCP 4.5 and RCP 8.5. The study predicts long-term hydrological drought indices SSI and SDI from 2006 to 2099 based on global climate models projections. The future drought years were extracted using the run theory. The study reveals that the river basin was prone to hydrological drought occurrences 50–65% of the time during the historic period (1988–2005) and 32–70% during the future period (2006–2099). It is evident from the results that both RF and GBM models follow a similar pattern of dry events in the future, which implies the prediction to be accurate. The study recommends the RF model for the study area, considering the model performances.