Evaluating Erosion Rates Through Advanced DEM Differencing and Co-Registration Techniques Using Underutilized Satellite Data: A Case Study from Southern Taiwan

Taiwan's geological setting, characterized by rapid tectonic uplift and among the world's most intense precipitation patterns and recurring extreme rainfall events, offers a natural laboratory for studying sediment flux and erosion rates in mountain river basins. The availability of open-source satellite-derived digital elevation models (DEMs) provides an invaluable opportunity to evaluate their suitability for constraining sediment flux in these dynamic environments. The Laonong River, one of Taiwan's prominent and vulnerable watersheds, has been selected as a representative study area due to its history of past and ongoing landslides, making it ideal for understanding erosion processes and sediment transport dynamics. This study assesses erosion rates in the Laonong River Basin over the past two decades using satellite-derived DEMs from diverse optical and radar sources. By evaluating the suitability of underutilized global DEMs, including ASTER GDEM, NASADEM, SRTM, ALOS World 3D DEM (AW3D30), Copernicus DEM, FAB DEM, and TanDEM-X EDEM, and benchmarking them against a high-accuracy LiDAR DEM, we aim to enhance our understanding of the erosional processes. Accuracy assessments are conducted in stable areas through spatial domain analysis, utilizing comprehensive metrics, including RMSE, bias, and standard deviation, to quantify discrepancies and ensure rigorous error evaluation. Additionally, metadata analyses identify voids and artifacts filled from external sources, while Fourier analysis is applied to detect and mitigate vertical biases, enabling a robust examination of DEM suitability in this complex terrain.

Our findings revealed that while Copernicus DEM, FAB DEM, and TanDEM-X EDEM exhibited good vertical accuracy in the spatial domain, their reliance on external DEMs for void-filling rendered them unsuitable for multitemporal analysis. Similarly, ASTER GDEM was excluded due to its high standard deviation, significant negative bias, and prolonged acquisition period, averaging over 13 years. As confirmed through Fourier analysis and in the spatial domain against LiDAR DEM, AW3D30 demonstrated excellent vertical accuracy and minimal vertical bias. NASADEM, being the successor of SRTM, was preferred over its predecessor due to lower vertical bias and minimal external void-filling. Consequently, NASADEM and AW3D30 were identified as the most reliable DEMs for capturing topographic changes across different decades in the Laonong River Basin. Horizontal co-registration was refined to sub-pixel accuracy using the Nuth and Kääb method, while Fourier analysis was employed for vertical alignment, effectively minimizing biases across DEMs acquired at different time points. Spectral analysis identified long-wavelength topographic features crucial for correcting offsets and enhancing the accuracy of DEM differencing. Our results estimate that about 119 Mm³ of sediment volume has been transported out of the system over 20 years, as calculated from NASADEM and LiDAR DEM. We documented the spatial pattern of erosion and deposition across the whole Laonong River basin in the DEMs of Differences (DoD) maps, and the results were validated from the Google Earth imageries. These findings highlight the capability of underutilized satellite-derived DEMs in capturing sediment erosion rates over multiple decades, demonstrating their utility in environments where erosional signals are dominant over the inherent noise in the dataset.