A Review of Surface Water Body Extraction from Remote Sensing Images: From Visual Interpretation to Transformer-Based Approaches

Abstract

The correct delineation of surface water bodies from remote sensing imagery (RSI) is essential for surface water monitoring, water conservation, and environmental change assessment. Various methods have been developed over the last three decades to improve the accuracy of surface water body extraction. This review provides a detailed overview of remote sensing-based surface water body extraction methods. Freely available datasets, pre-processing techniques, loss functions, and evaluation metrics, along with their advantages and limitations, are also discussed in this article. The review highlights a clear transition from traditional methods, such as visual interpretation and thresholding, toward machine learning (ML) methods and deep learning (DL) approaches, which include conventional deep learning models and hybrid transformer-based methods due to their superior ability to represent complex spatial and spectral properties of surface water bodies. Detection of small and fragmented water bodies, class imbalance, computational costs, model generalization, and the small size of annotated datasets are still significant challenges. This article summarizes recent advancements, current challenges, and future research directions. This article helps beginners and researchers understand key remote sensing concepts and their applications in surface water body extraction (SWBE).