Computational approaches for modelling soil water repellency: A comprehensive review

This is the authors' abstract. We don't add key points for this paper.

Soil water repellency (SWR) is a significant soil constraint in agricultural systems, impacting water infiltration, crop establishment, and productivity, especially in sandy soils. Accurate assessment and prediction of SWR are essential for enhancing soil and water management practices. Traditional measurement techniques, including molarity of ethanol droplet and water drop penetration time tests, provide reliable assessments on point-based samples. However, these methods require significant labour and time inputs to extend SWR assessments to farm and regional scales. Recent computational approaches integrating statistical analysis, remote sensing, machine learning, deep learning, and geospatial techniques have emerged as promising alternatives for SWR assessment and prediction. This study reviews existing computational approaches used for modelling SWR and related soil properties across agricultural systems. It synthesises commonly used data sources, including remote sensing imagery, vegetation indices, climate variables, topographic information, and soil properties, together with preprocessing techniques, modelling methods, and evaluation metrics. Furthermore, the review examines the strengths and limitations of existing approaches in terms of prediction accuracy, scalability, spatial–temporal generalisation, and interpretability. Particular attention is given to explainable artificial intelligence techniques for understanding the environmental drivers influencing SWR variability. Based on the reviewed literature, this study identifies existing research gaps and presents future research directions for developing scalable, interpretable, and data-driven computational frameworks for SWR modelling and precision soil management.