Lake Detection and Water Quality Estimation in Sentinel-2 Data
For environmental scientists and remote sensing practitioners, this work provides a practical comparison of ML methods for lake detection and addresses a visualization gap in water quality assessment, though the improvements are incremental.
This paper compares three machine learning architectures for water body detection in Sentinel-2 data, finding that the best model outperforms classical NDWI thresholding in accuracy and robustness. It also proposes new color schemes for water quality indices to improve interpretability.
With climate change and increasing human pressure on natural landscapes, inland water resources are becoming progressively scarcer, more vulnerable, and more difficult to manage sustainably. Reliable and automated methods for detecting, monitoring, and assessing surface water bodies are therefore of growing scientific and practical importance. In this paper, we investigate and compare three distinct machine learning architectures for water body identification and monitoring. Their performance is evaluated through quantitative metrics and real-world examples. Furthermore, a direct comparison with classical NDWI thresholding is conducted on a representative test image to highlight differences between data-driven and index-based approaches. This analysis allows us to identify the best-performing model in terms of accuracy, robustness, and practical applicability. Beyond detection, a major challenge for meaningful water quality assessment lies in the consistent and interpretable visualization of spectral water indices. Standard color mapping techniques are often inadequate or potentially misleading for environmental applications. To address this gap, we propose a suite of meaningful color schemes adapted for water quality indices, facilitating clearer interpretation, comparison, and decision-making for human users.