Сборник с доклади от XXXIV Международен симпозиум | Page 271

Based on a combination of frequency bands B4( red), B3( green), and B1( aerosol) for both images, image segmentation was performed to extract the water area of the Batak Reservoir by using the Seeded Region Growing algorithm [ 3 ]. The obtained polygons are grouped into 255 clusters by classification without training − the " K-means clustering " method. For performing supervised classification after the band sets segmentation, 2 classes of test areas( training areas) were used –“ surface water body” and“ land” and Maximum Distance method were used.

To validate the results, 2858 control points were generated from Google Earth Engine. Many GIS functions were also used to process all the raster and vector data utilized, including intermediate and final results: georeferencing, intersecting, resampling, descriptive statistics, grid calculation, image correlation, visualization, and others.

3. RESULTS

Table 2 shows the Pearson correlation coefficient between NDWI, SWM, and WRI water indices. The WRI and SWM indices show the best correlation, followed by the correlation coefficient of the WRI and NDWI indices for the two satellite images used.

Table 2. Pearson’ s correlation coefficient between the water indices NDWI, Sentinel Water Mask и WRI

S2 L1C 2022-11-01 NDWI Sentinel WM NDWI 1.000

Sentinel WM 0,9689 1.000 WRI 0,9708 0,9995

S2 L1C 2024-07-28 NDWI Sentinel WM NDWI 1.000

Sentinel WM 0.9290 1.000 WRI 0.9643 0.9919

Applying the determined threshold values from the histograms of the calculated water indices from the two satellite images, the limits of the " Batak " Reservoir were identified( Fig. 2, Table 3).

271