[M-AG33] Satellite Land Physical Processes Monitoring at Medium/High/Very High Resolution

[MAG33-P01] Utilizing Time-Series Satellite Imagery for Monitoring SDG 15.3.1 on Land Degradation Indicator: A Case Study in Taiwan

*Tee-Ann Teo¹, Yi-Fang Wu¹ (1.NYCU)

Keywords:Time-Series Satellite Imagery, Land Degradation, Time-Series Transformer, SDG 15.3.1 Indicator

The Sustainable Development Goal (SDG) 15.3.1 indicator aims to monitor and address land degradation, a critical global issue exacerbated by climate change and human activities. This study explores the application of time-series satellite imagery to monitor land degradation in Taiwan, focusing on its integration into the evaluation and reporting of the SDG 15.3.1 indicator. Utilizing Sentinel-2 imagery fused with digital elevation model data, we developed a deep learning-based Time-Series Transformer (TST) classification model. This model incorporates spectral and topographic features to improve the detection and quantification of land degradation. The research specifically targeted landslide-prone areas in the upper reaches of the Dajia River watershed, a region with a history of persistent slope failures. Through multiresolution segmentation and object-based analysis, we identified and classified landslide areas with an F1-score of 0.94, demonstrating high precision and accuracy. The analysis revealed critical land degradation patterns linked to slope instability, providing a robust method for calculating the land degradation ratio. The results for the SDG 15.3.1 indicator reveal significant progress in monitoring and predicting land degradation using time-series satellite imagery and advanced modeling techniques. The total study area of 93,045.917 ha showed a decrease in degraded land area, with the ratio dropping from 1.485% in 2019 to 1.321% in 2020 based on ground truth data and a predicted result of 1.219% for 2020. The predicted values align closely with actual measurements, demonstrating a minimal difference of 0.102%, indicating the reliability of the predictive model. These findings reflect effective land restoration efforts, particularly in addressing landslides, and the utility of satellite-based monitoring in supporting sustainable land management and achieving the goals of SDG 15.3.1.