Document Type : Research Paper

Authors

• Sina Nabizadeh ¹
• Ali asghar Naghipour ²
• Ataollah Ebrahimi ³
• Hamidreza Keshtkar ⁴
• Elham Ghehsareh ⁵

¹ Ph.D. Student, Department of Nature Engineering, Faculty of Natural Resources and Earth Sciences, Shahrekord University, Iran.

² Assistant Professor, Department of Nature Engineering, Faculty of Natural Resources and Earth Sciences, Shahrekord University, Iran

³ Associate Professor, Department of Nature Engineering, Faculty of Natural Resources and Earth Sciences, Shahrekord University, Iran.

⁴ Assistant Professor, Department of Arid and Mountainous Regions Reclamation, Faculty of Natural Resources, University of Tehran, Iran.

⁵ Associate Professor, Department of Nature Engineering, Faculty of Natural Resources and Earth Sciences, Shahrekord University, Iran

Abstract

Land use/land cover (LULC) maps play a key role in natural resource management and sustainable land-use planning. Recent advances in remote sensing, machine learning, and cloud-based platforms such as Google Earth Engine (GEE) have enabled efficient large-scale spatiotemporal analyses. In this study, LULC changes in the large and heterogeneous Karun-1 watershed were assessed using Landsat 7 ETM+ (2002) and Landsat 8 OLI (2024) imagery. Cloud-free composite images derived from nine scenes during the peak growing season were generated using a median filter. A total of 1,920 training samples for seven LULC classes were extracted based on field data, aerial images, and Google Earth. Vegetation and auxiliary indices (NDVI, NDBI, NDWI, and DSM) were integrated with spectral bands. Supervised classifications were performed using CART, Random Forest (RF), and Support Vector Machine (SVM) algorithms in GEE. The results indicated that all three algorithms produced highly accurate LULC maps, with SVM achieving the highest overall accuracy and kappa coefficient (above 92% in both years). Rangelands (≈40%) and forests (≈27%) dominated the watershed area, while a declining trend was observed in rangelands, forests, and especially water bodies over time. The findings confirm the effectiveness of integrating machine-learning algorithms with GEE for large-scale environmental monitoring. However, limitations related to the spatial resolution of Landsat imagery remain a challenge. Therefore, the use of higher-resolution data such as Sentinel imagery is recommended for future studies.

Keywords

• Remote sensing
• Support Vector Machine (SVM)
• Random Forest (RF)
• Classification and Regression Tree (CART)
• Auxiliary data