EVALUATION OF WATER INDICES AND THRESHOLDING METHODS FOR DETECTING WATER SURFACES OVER LARGE AREAS USING SENTINEL-2 IMAGERY

ELİF KARABAL

EVALUATION OF WATER INDICES AND THRESHOLDING METHODS FOR DETECTING WATER SURFACES OVER LARGE AREAS USING SENTINEL-2 IMAGERY

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Hacettepe Üniversitesi, Fen Bilimleri Enstitüsü, Geomatik Mühendisliği A.B.D., Türkiye

Tezin Onay Tarihi: 2026

Tezin Dili: İngilizce

Öğrenci: ELİF KARABAL

Danışman: Mustafa Türker

Özet:

Inland water systems on Earth are of great importance for ecosystem sustainability, biodiversity conservation, and water resource monitoring. Therefore, the rapid, repeatable, and reliable detection of water surfaces using remote sensing technologies has become a significant research topic. The aim of this study is to evaluate the performance of different water indices for the detection of surface waters over large areas from Sentinel-2 satellite images and to identify the indices that provide the best performance. An area of approximately 245 km × 185 km from Turkey's Lake District was selected as the study area. The most commonly used Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI-1, MNDWI-2), Automated Water Extraction Index (AWEI_sh, AWEI_nsh), and Water Ratio Index (WRI) were selected. Sentinel-2 images from 2023 were used for the evaluation. To determine the optimum threshold value, different global thresholding methods were used, includingK-Means clustering, Otsu, Kapur, percentage slicing (P-tile), and zero threshold methods. The achieved results were evaluated using the error matrix-based accuracy analyses. The results showed that not only the spectral index used but also the thresholding method has a significant impact on the results of water surface extraction. In this study, the AWEI_nsh – K-Means thresholding pair provided the best performance with an overall accuracy (OA) of 97% and a Kappa coefficient of 0.939. On the other hand, the fixed threshold method did not produce suitable results for all indices. Furthermore, it was observed that the P-tile threshold method produced more balanced and consistent results among different indices. The findings of this study revealed the necessity of evaluating indices and thresholding methods together in water surface extraction studies from Sentinel-2 satellite images.