The Performance Analysis of Different Water Indices and Algorithms Using Sentinel-2 and Landsat-8 Images in Determining Water Surface: Demirkopru Dam Case Study


Yilmaz O. S., Gülgen F., Balık Şanlı F., Ates A. M.

Arabian Journal for Science and Engineering, cilt.48, sa.6, ss.7883-7903, 2023 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 48 Sayı: 6
  • Basım Tarihi: 2023
  • Doi Numarası: 10.1007/s13369-022-07583-x
  • Dergi Adı: Arabian Journal for Science and Engineering
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Communication Abstracts, Metadex, Pollution Abstracts, zbMATH, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.7883-7903
  • Anahtar Kelimeler: Water surface, NDWI, MNDWI, Maximum likelihood, Random forest, K-means
  • Yıldız Teknik Üniversitesi Adresli: Evet

Özet

© 2023, King Fahd University of Petroleum & Minerals.In this study, the most appropriate algorithm and water index to determine the boundaries of the dam water surface using remote sensing (RS) techniques were investigated. Water surface boundaries of Demirkopru Dam were determined using Sentinel-2 L2A (MSI) and Landsat-8 (OLI) satellite images. Demirkopru Dam was chosen as the study area as it is suitable for floating photovoltaic (FPV) solar power plant installation. Normalized difference water index (NDWI) and modified NDWI indices were used to determine the water surface boundaries of the dam. Thirty-six classification results were obtained using K-means, maximum likelihood classification (MLC), and random forest (RF) algorithms. The best classification accuracies of the produced maps have been calculated as 80.3%, 73.1%, and 73.2% by RF, MLC, and K-means, respectively. In addition, the water coastlines determined by classifications were compared with the continuously operating reference station (CORS-TR) data in a local area by calculating the root-mean-square error (RMSE). Compared with the CORS-TR measurements of the dam coastline obtained from the images classified by the RF algorithm, the minimum RMSE values were calculated as 13.8 m and 10.1 m for Landsat and Sentinel images, respectively. While the minimum RMSE value for coastlines obtained with various layer stacks of Landsat images classified by the MLC algorithm is 36.7 m, it could not be calculated in Sentinel images due to poorer classification results. For the coastlines obtained from the images classified by the K-means algorithm, the minimum RMSE values were calculated as 14.5 m and 9.6 m for Landsat and Sentinel images, respectively. According to the comparisons based on classification accuracy and CORS-TR measurements, it is concluded that the RF algorithm performs better than others for the dam water surface. Moreover, it was determined that the NDWI presented better results when the water level was the lowest for Demirkopru Dam. Also, in this study, the MLC algorithm has better results in detecting water surfaces using Landsat images. It was concluded that the K-means algorithm is also very effective in water surface detection. In this study, various water extraction indices, algorithms and free Landsat and Sentinel images were used to extract the water surface in a selected reservoir for the FPV installation. This study guides a series of algorithms and indexes used to detect water surfaces. In addition, it has been shown that the use of RS techniques, which are more practical than classical approaches in determining water boundaries, will be more effective in planning and design in terms of engineers, investors and various organizations who will realize the FPV installation.