Investigation of Landslide Debris Migration in the Wushe Reservoir Catchment Area, Taiwan

Following the Chi-Chi earthquake in Taiwan, the surface soil and rocks were fractured and loosened, making the geology more unstable. As typhoons continued to batter the mountainous regions, the resulting heavy rainfall—in terms of total accumulation and intensity—exacerbated the situation. This caused an increase in the size of the collapsed and exposed upstream catchment areas, leading to more damage from rainwater erosion. The migration of soil and sand into the reservoir, triggering landslides of varying scales or other soil-related disasters, has severely worsened reservoir siltation. Traditional dredging methods are ineffective in solving this issue. Therefore, addressing the migration of soil and sand has become crucial in mitigating and slowing down the process of reservoir siltation.

This study focuses on the Wushe Reservoir catchment area, which is experiencing significant siltation. According to the Water Resources Administration of the Ministry of Economic Affairs, the current reservoir capacity is less than 25% of its original design. The study primarily analyzes historical data and ongoing monitoring of the area. By observing the impact of rainfall on the slopes, the study aims to gain a deeper understanding of soil and sand migration in the catchment area.

Additionally, the study employs Synthetic Aperture Radar (SAR) images for Small Baseline Subset (SBAS) analysis to detect potential slope sliding within the study area. Landslide potential is identified through SAR data, and GNSS (Global Navigation Satellite System) is used to confirm whether slope sliding is occurring. Data from existing on-site single-frequency and dual-frequency GNSS monitoring equipment are also analyzed for verification.

The study uses NDVI (Normalized Difference Vegetation Index) and GNDVI (Green Normalized Difference Vegetation Index) to identify bare land affected by landslides and river channels. The accuracy of these interpretations is evaluated using precision analysis metrics. The average accuracy for bare land identification is 73.91%, with an average Kappa coefficient of 69.94%. Rainfall events are categorized to map landslides caused by different rainfall conditions and a landslide mapping model is established. The amount of debris is estimated based on the collapsed area, and soil loss is calculated using the Universal Soil Loss Equation (USLE). These results are cross-verified with changes in reservoir capacity over the years to validate the study's findings.