A review on machine learning models for drought monitoring and forecasting

Drought, driven by shifting climate patterns, increasingly threatens hydropower, agriculture, and water supply systems, necessitating robust early detection and forecasting frameworks. This review critically examines recent machine learning (ML) approaches to drought prediction, focusing on meteorological, hydrological, and agricultural drought types across diverse temporal and spatial scales. We analyze the influence of input variables, such as precipitation, streamflow, climate indices, and remote sensing data, on model performance, and identify key trends including the rise of hybrid and deep learning models for capturing nonlinear dependencies and long-term patterns. Unique contributions include a comparative evaluation of model interpretability, scalability, and data requirements, as well as the identification of persistent gaps such as limited regional transferability and underrepresentation of socio-environmental factors. By proposing a framework for selecting optimal models based on data availability, complexity, and operational constraints, this review offers actionable insights for researchers and policymakers seeking to develop adaptive, context-aware drought mitigation strategies that ensure long-term water sustainability. © 2025 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license. http://creativecommons.org/licenses/by-nc-nd/4.0/