Leveraging machine learning for hydrological drought prediction and mitigation

Abstract

Drought disasters have become a global issue, occurring more frequently due to climate change and increasing water usage patterns. Adaptation and mitigation efforts to reduce disaster vulnerability involve effective drought monitoring, such as drought predictions. This study aims to predict the hydrological drought index (HDI) for the next 5 years (20242028) in the Bendung Notog sub-watershed. The HDI prediction modeling is based on machine learning with an artificial neural network (ANN) algorithm using historical HDI values from the past 20 years (2004-2023). The historical HDI was calculated using the Threshold Level Method with discharge data transformed by the NRECA method. The drought prediction model demonstrates high accuracy with performance assessment values of MAE = 0.015, R = 0.91, R2 = 0.82, NSE = 0.82, and RMSE = 0.022. The HDI prediction results indicate that the Bendung Notog sub-watershed experiences dry conditions annually during the dry season, with the lowest HDI and longest drought duration occurring in 2024. Hydrological drought prediction is essential to minimize the negative impacts due to reduced water discharge, enabling strategic planning for future water needs.