Combining spectral indices and machine learning for accurate surface water bodies extraction: An SVM-based application framework with spatial assessment (IoU, Dice)
Department of Remote Sensing and GIS, Earth Sciences Faculty, Shahid Chamran University of Ahvaz, Ahvaz, Khuzestan, Iran
10.22034/gp.2026.71865.3525
Abstract
Objective: In recent years, accurate extraction of surface water areas from satellite images has gained special importance due to their key role in water resources management and environmental monitoring. Despite development of numerous spectral indices for water identification, performance of these methods in vast mountainous and heterogeneous areas is often limited by spectral interference, shadow and dark surfaces.
Methods: In this study, an integrated framework based on the combination of common spectral indices including NDWI, MNDWI, AWEI_sh, AWEI_nsh and WRI along with Landsat-8 OLI multispectral raw bands is presented for extracting surface water areas. Then, the support vector machine (SVM) algorithm was used as a machine learning method to optimally integrate feature space and separate water and non-water classes. Reference data were prepared from a combination of field observations, visual interpretation of high-resolution images, and Open Street Map (OSM) vector data, and spatial dependence of the samples was reduced by applying quality filters and spatial compression.
Results: Model performance evaluation using quantitative criteria including Recall: 90%, IOU: 53%, and Dice: 70% showed that model provides stable performance in identifying water bodies. SVM model was able to obtain a Recall value of 0.90 for water class, which indicates high power of the model in correctly identifying most of real water pixels. Also, Dice value of 0.70 and IoU of 0.53 indicate moderate to acceptable spatial agreement between the predicted map and reference data.
Conclusions: Results indicate that enriching feature space by combining raw bands and spectral indices plays a key role in improving continuous and accurate identification of water body boundaries. This framework can be used as an efficient and reliable method for extracting surface water areas in complex and large-scale areas and support decision-making in water resources management.
Riahinia, M. , Zareie, S. and Rangzan, K. (2026). Combining spectral indices and machine learning for accurate surface water bodies extraction: An SVM-based application framework with spatial assessment (IoU, Dice). Journal of Geography and Planning, (), -. doi: 10.22034/gp.2026.71865.3525
MLA
Riahinia, M. , , Zareie, S. , and Rangzan, K. . "Combining spectral indices and machine learning for accurate surface water bodies extraction: An SVM-based application framework with spatial assessment (IoU, Dice)", Journal of Geography and Planning, , , 2026, -. doi: 10.22034/gp.2026.71865.3525
HARVARD
Riahinia, M., Zareie, S., Rangzan, K. (2026). 'Combining spectral indices and machine learning for accurate surface water bodies extraction: An SVM-based application framework with spatial assessment (IoU, Dice)', Journal of Geography and Planning, (), pp. -. doi: 10.22034/gp.2026.71865.3525
CHICAGO
M. Riahinia , S. Zareie and K. Rangzan, "Combining spectral indices and machine learning for accurate surface water bodies extraction: An SVM-based application framework with spatial assessment (IoU, Dice)," Journal of Geography and Planning, (2026): -, doi: 10.22034/gp.2026.71865.3525
VANCOUVER
Riahinia, M., Zareie, S., Rangzan, K. Combining spectral indices and machine learning for accurate surface water bodies extraction: An SVM-based application framework with spatial assessment (IoU, Dice). Journal of Geography and Planning, 2026; (): -. doi: 10.22034/gp.2026.71865.3525
)