In the heart of Saudi Arabia’s Najran city, a groundbreaking study is reshaping how we understand and manage land use, offering promising avenues for sustainable agriculture. Published in *Scientific Reports*, the research introduces a hybrid deep learning and optimization-based approach to land use and land cover (LULC) classification, providing critical insights for farmers, policymakers, and agritech innovators.
The study, led by Aisha M. Mashraqi from the Department of Computer Science at Najran University, leverages a combination of Convolutional Neural Networks (CNN) and Random Forest (RF) algorithms to analyze Landsat-8 satellite images. The goal? To create a robust system that can accurately classify land cover types and guide sustainable agricultural practices.
“Our hybrid model not only improves the accuracy of land classification but also enhances the interpretability of the results,” Mashraqi explains. “This is crucial for making informed decisions that can support sustainable agriculture in arid regions.”
The research tested ten different CNN-RF variants, with the top performers achieving impressive overall accuracies: VGG19-RF at 97.56%, GoogleNet-RF at 96.15%, DenseNet121-RF at 92.39%, and ResNet152-RF at 92.26%. These models demonstrated high precision, recall, and F1 scores, ensuring reliable separation of agronomic classes.
The classified land cover data reveals significant insights. Built-up areas account for approximately 29–33% of the region, while vegetation areas range from 14–25%. Bare ground and water areas each make up about 9–22% of the landscape. These statistics highlight the pressures of urban growth on agricultural lands and the need for strategic planning.
For the agriculture sector, the implications are substantial. The study’s findings can be translated into operational sustainability indicators, such as vegetation and bare-soil area metrics to inform crop rotation and soil management practices. Additionally, the delineation of built-up areas can help safeguard agricultural buffers, and water-body identification can prioritize irrigation efficiency and groundwater recharge areas.
“This research provides a framework that can be easily applied to other semi-arid regions,” Mashraqi notes. “Accurate and periodically updated spatial information is essential for sustainable production, and our model offers a reliable tool for achieving this.”
The hybrid model’s success over single-architecture baselines opens new avenues for commonplace LULC monitoring, agricultural risk screening, and policy tracking. As Saudi Arabia advances towards its Vision 2030, such innovative approaches will be instrumental in balancing urban development with sustainable agricultural practices.
In the broader context, this research underscores the potential of integrating advanced machine learning techniques with remote sensing data. As agritech continues to evolve, the fusion of these technologies could revolutionize how we manage land resources, ensuring food security and environmental sustainability in arid and semi-arid regions worldwide.