<p>Accurate monitoring of agricultural land is a cornerstone of sustainable land management, particularly in arid regions like Saudi Arabia, where water resources are scarce. Traditional Land Use Land Cover (LULC) classification methods, dependent on manually engineered features, often lack robustness across diverse environmental conditions. While deep learning models like Convolutional Neural Networks (CNNs) automate feature extraction and enhance generalization, their computational complexity can be prohibitive. This research investigates a hybrid methodology to optimize this balance, integrating the powerful feature learning of DenseNet121 with the computational efficiency of advanced machine learning classifiers, specifically Decision Trees (DT) and XGBoost. The objective was to develop a precise and efficient tool for mapping key land covers—especially agricultural areas—in Najran City using 2020 Landsat 8 imagery. The proposed framework extracts complementary spatial and spectral features, which are then classified. Experimental results demonstrated that the DenseNet121-XGBoost hybrid model achieved superior performance, with an overall accuracy of 98.82% and a Kappa coefficient of 0.9638, significantly outperforming the standalone CNN. This study confirms the efficacy of hybrid deep learning for reliable agricultural land monitoring, providing a valuable decision-support tool for promoting sustainable practices in arid environments.</p>