Detecting water in agricultural landscapes from remote sensing imagery: Methodological choices, sensor constraints and performance metrics

Monitoring surface water in agricultural landscapes is a key requirement for irrigation management, leak detection, and sustainable water use. Although remote sensing literature extensively addresses water detection, most studies focus on large-scale Surface Water Mapping (SWM) in heterogeneous landscapes, where extensive water bodies such as lakes, rivers, or coastal zones occupy a substantial portion of the scene. In contrast, fine-scale water detection in agricultural environments typically involves small, fragmented, and highly imbalanced targets embedded within predominantly vegetated or cultivated areas. As a result, methods and performance metrics developed for large-scale mapping cannot be directly transferred to fine-scale agricultural scenarios without adaptation. This paper analyses 49 peer-reviewed studies published between 2020 and 2025 that address water detection in agricultural and rural contexts using multispectral, thermal, and Synthetic Aperture Radar (SAR) imagery from satellite and Unmanned Aerial Vehicles (UAV) platforms. Rather than providing a purely descriptive review, the work examines how methodological choices — ranging from spectral indices and decision trees to machine learning, deep learning, and foundation models — interact with sensor characteristics, processing levels, and evaluation metrics. The analysis highlights systematic trade-offs among model complexity, data availability, and robustness, identifies recurrent limitations in multiple accuracy metrics in scenarios where land pixels vastly outnumber water pixels, and synthesizes the practical implications of spectral band selection (VNIR, SWIR, TIR) and platform resolution. A central contribution of this review is the demonstration that, in agricultural water detection, preprocessing choices, sensor characteristics, and the use of appropriate evaluation metrics often have a greater influence on reported performance than the complexity of the detection algorithm itself. Based on these findings, the paper offers comparative insights and methodological recommendations to guide the selection and validation of water-detection approaches in agricultural remote sensing applications. • Review (2020–2025) of fine-scale water detection using remote sensing imagery. • Comparative synthesis of spectral, machine and deep learning models under real-world constraints. • F1-Score and IoU outperform Overall Accuracy for robust evaluation under class im- balance. • Analysis of preprocessing, sharpening, augmentation, and multi-sensor fusion impacts. • Identifies gaps in spatial refinement and hybrid segmentation–decision frameworks.