A New Clustering Method to Generate Training Samples for Supervised Monitoring of Long-Term Water Surface Dynamics Using Landsat Data through Google Earth Engine

Water resources are vital to the survival of living organisms and contribute substantially development various sectors. Climatic diversity, topographic conditions, uneven distribution surface water flows have made reservoirs one primary supply in Iran. This study used Landsat 5, 7, 8 data Google Earth Engine (GEE) for supervised monitoring dynamics reservoir eight Iranian dams (Karkheh, Karun-1, Karun-3, Karun-4, Dez, UpperGotvand, Zayanderud, Golpayegan). A novel automated method was proposed providing training samples based on an iterative K-means refinement procedure. The Function Mask (Fmask) initial map generate final samples. Then, Support Vector Machines (SVM) Random Forest (RF) models were trained with generated mapping. Results demonstrated satisfactory performance RF model procedure (with overall accuracies 95.13%) comparison direct Fmask 78.91%), indicating approach’s success producing three feature sets also evaluated. Tasseled-Cap (TC) achieved higher than Spectral Indices (SI) Principal Component Transformation Image Bands (PCA). However, simultaneous use all features (TC, SI, PCA) boosted classification accuracy. Moreover, long-term changes showed a downward trend five sites. Comparing latest year’s area (2021) maximum extent that sites experienced significant reduction (16–62%). Analysis climate factors’ impacts revealed precipitation (0.51 ? R2 0.79) more correlated temperature (0.22 0.39) changes.