Machine Learning and Data-Driven Methods for Estimating Near-Surface Hydrological Parameters: Benchmarking, Uncertainty, and Practical Application

Near-surface hydrological parameters — including soil infiltration rate, hydraulic conductivity, groundwater recharge rate, Manning’s roughness coefficient, and stage-discharge relationships — are fundamental inputs to water resource planning, irrigation management, flood modelling, and aquifer assessment. Yet they remain among the most difficult quantities to determine reliably. Traditional field and laboratory measurement methods are labor-intensive, site-specific, and often impractical at the spatial and temporal scales required for operational use. Classical empirical models (Horton, Philip, Kostiakov, Manning) offer tractable approximations but struggle to generalize across varying soil textures, land covers, channel geometries, and climate regimes.Over the past two decades, machine learning and artificial intelligence approaches — including artificial neural networks (ANN), support vector machines (SVM), adaptive neuro-fuzzy inference systems (ANFIS), gene-expression programming (GEP), random forest, and gradient boosting — have shown strong potential for estimating these parameters from readily available soil, topographic, and climatic inputs. More recently, physics-informed neural networks and hybrid process-ML architectures have opened further possibilities for physically consistent, data-driven parameter prediction. Despite a rapidly growing body of individual case studies, the field lacks systematic benchmarking across methods, consistent uncertainty quantification, and robust guidance on when and where each approach performs best. Reproducibility and transferability across regions and parameter types remain underexplored.This Research Topic aims to consolidate and advance this fragmented evidence base. It brings together original research, methods papers, and reviews focused specifically on the estimation of near-surface hydrological and hydraulic parameters using machine learning and data-driven methods — from laboratory column experiments and field measurement campaigns through to regional-scale prediction frameworks.We welcome contributions addressing, but not limited to, the following themes:- Development and evaluation of machine learning and AI-based models for estimating infiltration rate, hydraulic conductivity, groundwater recharge, Manning’s roughness, stage-discharge relationships, and related near-surface parameters.- Systematic benchmarking of multiple ML approaches against empirical models and against each other, using common datasets and performance metrics.- Uncertainty quantification in ML-based parameter estimation, including sensitivity analysis and model interpretability.- Physics-informed, hybrid, and knowledge-guided machine learning approaches for hydraulic and hydrological parameter estimation.- Field and laboratory validation of ML-derived parameters across contrasting soil types, land uses, and hydroclimatic settings.- Transfer learning and generalization: assessing model performance across geographic regions and data-scarce environments.- Review and meta-analyses synthesizing the state of the art in ML-based estimation of specific parameter types. Topic editors: Barkha Chaplot, Ozgur Kisi, Wan Zurina Source: https://www.frontiersin.org/research-topics/84368/machine-learning-and-data-driven-methods-for-estimating-near-surface-hydrological-parameters-benchmarking-uncertainty-and-practical-application