A Dynamic Graph Neural Network Approach for Forest Cover Change Detection and Carbon Sink Assessment Using Remote Sensing Imagery

Against the backdrop of accelerating global climate change, accurate assessment of forest cover dynamics and carbon sink capacity is critical to addressing the climate crisis and achieving the "dual carbon" goals.With the growing availability of high-resolution, multitemporal remote sensing data, efficiently processing complex spatial relationships and temporal dynamics has become a central challenge in forest change detection and carbon sink estimation.Traditional pixel-or object-based remote sensing classification methods often overlook spatial correlations and contextual information, leading to limited detection accuracy in regions with complex terrain.Similarly, carbon sink assessment methods based on statistical models or static data fail to capture the dynamic processes of forest change and their spatiotemporal coupling with carbon sequestration, resulting in insufficient accuracy and timeliness.Moreover, the low efficiency of large-scale data processing remains a pressing issue.To address these challenges, this study proposes a dynamic graph neural network-based approach for forest cover change detection and carbon sink assessment.On one hand, a dynamic graph model tailored for forest remote sensing imagery is constructed, leveraging the powerful spatiotemporal representation capabilities of graph neural networks to achieve precise detection of forest cover changes.On the other hand, based on the detection results, a carbon sink estimation model is developed that integrates forest type, growth stage, and climatic conditions to quantify carbon sink capacity and potential.The proposed method enhances both the accuracy and efficiency of forest change detection in complex environments and provides a theoretical and technical foundation for dynamically tracking forest carbon sink evolution, informing forest management policies, and guiding carbon trading strategies.