Monitoring desertification dynamics and transition patterns through spatial-frequency feature integration of remote sensing data: a case study of the Gurbantunggut Desert, China

One of the most severe types of land degradation in arid and semi-arid regions is desertification, which poses serious risks to ecological security, sustainable development, and the global climate system. However, long-term remote sensing monitoring is challenged by gradual transitions between different desertification levels and the ambiguity of spatial boundaries, leading to substantial uncertainty in fine-grained identification and quantitative assessment. To address these challenges, this study proposes a Deep Semantic Change Detection Network (DSCD-Net) that integrates spatial and frequency feature representations for fine-grained desertification monitoring using multi-temporal Landsat images acquired from 1990 to 2020. The framework takes bi-temporal multispectral images as inputs and directly predicts pixel-level desertification transition relationships, thereby accurately representing the location, direction, and intensity of changes. The proposed framework was applied to the Gurbantunggut Desert, China, to examine long-term desertification dynamics and transition patterns. Experimental results show that DSCD-Net achieved 92.11% OA, 78.18% mIoU, 74.36% Fscd, and 34.75% SeK, outperforming the comparison methods. Long-term analysis shows that the evolution of desertification in this region exhibits significant nonlinear characteristics and periodic fluctuations, with the moderate desertification state as a key transitional phase. Furthermore, precipitation during the snow accumulation and melting period is identified as the dominant climatic driver regulating desertification dynamics. By explicitly capturing transition pathways and intensity, the proposed framework provides a process-oriented ecological indicator framework for assessing land degradation and offers valuable support for ecological restoration planning and sustainable land management in arid environments.