Machine learning and predictive models for water management: a systematic review

Introduction Water resource management faces strategic challenges posed by climate change, urban expansion, and land-use transformations. Machine learning (ML) has emerged as a promising alternative, capable of learning patterns from large datasets, contributing to the design of forecasting models, and revolutionizing the sustainable management of water. Methods This systematic review followed PRISMA 2020 guidelines. The study identified 35 records, reviewed 18 full texts, and excluded 17 studies. Searches targeted Scopus, Web of Science, IEEE Xplore, ScienceDirect, and were supplemented by Google Scholar and manual reference screening. The equation combined water-related terms such as “water management” with machine learning terms such as “deep learning,” “artificial intelligence,” etc. Inclusion required peer-reviewed articles with sufficient methodological description and English or Spanish full text. Exclusions comprised narrative reviews, gray literature, and studies lacking algorithmic details. The period spanned 2010–2025 to capture ML growth. Results The results show that deep learning models (especially LSTM) offer significant improvements in time prediction, while assembly-based algorithms (Random Forest, XGBoost, CatBoost) stand out for their robustness in data-constrained situations. Hybrid ML + physical model approaches showed high efficacy in correcting bias and improving hydrological projections. Gaps in reproducibility, uncertainty analysis, and integration of anthropogenic factors were identified. Geographic focus favored Asia, Europe, and North America with 10–50 years series. Common metrics included RMSE, MAE, R 2 , NSE, and KGE. It is concluded that ML constitutes a strategic tool to strengthen water management in scenarios affected by climate variability and data scarcity. Discussion ML captures nonlinearities, adapts to noisy data, and integrates multi-source sensor and satellite data. Reproducibility remains limited, as few studies publish code or hyperparameters. Integration of anthropogenic factors (dams, irrigation, urbanization) remains insufficient. Future research must adopt reproducibility frameworks, incorporate explicit uncertainty analysis, and advance physically informed hybrid models. The evidence confirms ML's value for water management under climate variability and data scarcity, but consolidation requires addressing methodological weaknesses.