AI based drought assessment amid climate change: A systematic review

Artificial intelligence (AI) is rapidly gaining traction in drought assessment as conventional methods falter under climate change, yet its effectiveness and rigour remain untested. We address this gap through a systematic review of 103 studies spanning the drought assessment workflow. Evidence shows that performance depends on the entire modelling pipeline, modulated by study area characteristics and drought regime, so rather than seeking a single best model, model selection should follow purpose, resources, and architectural fit to the problem. Feature engineering, hybridisation, ensembling, and optimisation still improve accuracy but offer limited value unless paired with uncertainty estimates. Recurring pitfalls include data leakage, circularity, reverse-causal prediction, and architecture-task mismatch, while validation remains a broader challenge. Although impact-based modelling offers a more actionable framework, it requires separating hazard, exposure, and vulnerability. Advancing the field requires physically grounded models, validation with uncertainty estimates, and a shift from hazard prediction toward climate-resilient drought risk assessment.