Satellite-based tree detection for urban green space monitoring in arid environments

Abstract

Urban Green Space (UGS) plays a vital role in maintaining urban ecological balance by providing environmental, social, and economic benefits such as heat mitigation and improved public health. However, rapid urbanization and the limitations of traditional monitoring methods make large-scale and accurate assessment of green spaces challenging. This paper introduces a high-resolution satellite image dataset for urban green space analysis, with a focus on tree counting in arid regions of Saudi Arabia, specifically Al-Qassim and Al-Madinah Al-Munawwarah regions. The dataset, consisting of over 13,000 satellite image tiles, is used to evaluate tree density using advanced deep learning object detection models, including Faster R-CNN, RT-DETR, YOLOv10, and YOLOv12. Experimental results show that YOLOv12 achieves the best performance with an F1-score of 67%, while YOLOv10 offers a strong trade-off between accuracy and speed, making it suitable for real-time applications. These findings demonstrate the effectiveness of modern object detection frameworks for scalable urban ecological monitoring in arid environments. This study establishes a data-driven foundation for real-time urban ecological monitoring, directly supporting urban planning initiatives and international sustainability targets such as the United Nations Sustainable Development Goal.