Comment on "Estimating Total Dissolved Solids in Groundwater Using Machine Learning Models" by Sumita Gulati, Anshul Bansal and Ashok Pal, in Natural Resources Research DOI: 10.1007/s11053-025-10480-3

In this paper, specific technical aspects of the study conducted by Gulati (Nat Res Res 34:1623–1644, 2025) are critically examined. In their original study, Gulati (Nat Res Res 34:1623–1644, 2025) utilized several machine learning (ML) techniques to estimate the total solids in groundwater systems. While the study demonstrates the increasing role of data-driven approaches in hydrogeological applications, certain methodological choices, including performance metric formulation, train-test strategy, model comparability, and statistical validation, require further technical clarification. The present commentary aims to provide a focused methodological evaluation rather than a comprehensive review. In doing so, it highlights quantitative inconsistencies in reported evaluation metrics, differences in model partitioning strategies that affect result comparability, and ambiguity regarding feature attribution and statistical testing procedures. These issues have direct implications for reproducibility and the scientific validity of performance claims. The novelty of this commentary lies in its targeted assessment of methodological reproducibility in ML-based groundwater modeling studies, which has not been explicitly addressed in prior critiques. By discussing these aspects, the commentary intends to support the development of more transparent, statistically robust, and reproducible ML practices in future hydrogeological research.