Large language models for sustainable water management: A review

Sustainable Water Management (SWM) balances environmental protection, social equity, economic efficiency, and governance transparency. Achieving this balance requires timely data, specialized expertise, and adaptive decision-making, yet current approaches often face challenges of fragmented information, limited expert capacity, and costly decision-support tools. Large language models (LLMs) offer a promising alternative as general-purpose artificial intelligence (AI) systems capable of summarizing documents, generating analytical code, connecting to databases, and communicating insights in natural language (NL). By reusing a single pretrained model across multiple applications, LLMs can lower analytical costs, expand access to expertise, and enhance transparency through grounded and auditable outputs. This study presents one of the earliest systematic reviews of LLMs in SWM, synthesizing evidence from 34 studies. The review analyses publication trends, thematic and conceptual landscapes, and global collaboration patterns, revealing a rapidly expanding yet methodologically fragmented field. Evidence grading and risk-of-bias assessments show that most studies remain conceptual or observational, emphasizing the need for greater empirical rigor and reproducibility. Across environmental, social, economic, and governance (ESEG) dimensions, LLMs demonstrate potential to improve data integration, operational efficiency, and decision transparency, while challenges persist in bias propagation, data dependency, and model interpretability. Emerging design patterns, such as retrieval-augmented generation (RAG), human-in-the-loop frameworks, and explainable AI, are advancing safer and more accountable deployment. By consolidating fragmented literature, this review provides a foundation for responsible and sustainable AI in water management.