Comparative analysis of sensors and classification algorithms for land cover classification in Islamabad, Pakistan

Land use and land cover (LULC) classification is essential for environmental monitoring and sustainable land management. The selection of satellite sensors and classification algorithms influences the accuracy of LULC classification. This study evaluates the performance of three satellite sensors, GF-6 (GF-6), S2 (S2), and L9(L9), and three machine learning classifiers, Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost), in classifying LULC in Islamabad, Pakistan. The satellite data with high-to-course spatial resolution data was utilized, and a comprehensive pre-processing workflow ensured high-quality imagery. The results indicate that XGBoost, paired with GF-6, achieved the highest overall classification accuracy (94.24%) and kappa coefficient (0.9279), outperforming RF and SVM. S2 combined with XGBoost also showed superior performance (92.89%) compared to other sensor-algorithm combinations. The study reveals that high spatial resolution (GF-6) significantly improves LULC classification, particularly in detecting forest and urban areas. Feature importance analysis identified GF-6 Red and NIR bands as the most significant predictors, especially for vegetation-related classes. The findings underscore the importance of selecting the appropriate sensor and classifier for specific LULC tasks, with XGBoost and high-resolution sensors like GF-6 providing the most accurate results. This study contributes to the growing body of research on LULC classification and offers valuable insights for urban planning and environmental monitoring.