Assessing green area changes in Najran City, Saudi Arabia (2013–2022) using hybrid deep learning techniques

Abstract Advances in remote sensing technology have facilitated the assessment of land vegetation and agricultural development using satellite images. This study’s normalized difference vegetation index (NDVI) is vital in assessing vegetation coverage and tracking land-use changes. Integrating NDVI with convolutional neural network (CNN) models and random forest (RF) improves the accuracy of cropland discovery and vegetation evaluation in Najran city. This study proposes a hybrid method for discovering Najran, Saudi Arabia's agricultural lands. The proposed method combines the strengths of CNNs, namely, MobileNetV2, GoogLeNet, and DenseNet121, with RF, NDVI, and Landsat 8 satellite images to classify and analyze terrestrial vegetation accurately. Study data from 2013 and 2022 were used to evaluate and compare cropland discovery. The study finds that green space increased from approximately 860,297,400 m² in 2013 to approximately 909,567,900 m² in 2022. Various crop categories contributed to this increase, with vegetation surging by 220,794,300 m² in 2013, followed by a decrease to 161,650,800 m² in 2022. The average green area shifted positively from 57.36% in 2010 to 58.85% in 2022, indicating an overall enhancement in green land discovery and utilization in Najran city over the past decade. The hybrid systems for classifying agricultural land areas produced a confusion matrix that reached an overall accuracy of 96.87% in 2013 to 98.89% in 2022. The Kappa coefficient, a measure of agreement, ranged from 0.945687 in 2013 to 0.973832 in 2022, demonstrating the strength and robustness of the proposed hybrid approach.