Publications

Liou, YA; Lin, JJ (2025). Characterizing the antecedent rainfall and ATI-MODIS-derived soil moisture content of shallow landslides in Taiwan. LANDSLIDES.

Abstract
Landslide prediction remains a critical challenge due to limitations in in-situ monitoring and the uncertainty of where to position equipment in landslide-prone areas. This study aims to improve understanding of the antecedent conditions of slopes, particularly focusing on the role of soil moisture content (SMC). While many models and remote sensing approaches have been applied to landslide prediction, significant uncertainties persist. To address this, we investigated the relationship between antecedent SMC and rainfall intensity in characterizing shallow landslides in Taiwan. Using the apparent thermal inertia (ATI) method from MODIS, this study facilitated detailed spatial and temporal monitoring of surface moisture. Data were collected from ten in-situ monitoring stations located at diverse elevations. The Pearson correlation (r) between in-situ SMC measurements and ATI-MODIS-derived SMC ranged from 0.38 to 0.87, with RMSE values from 0.03 to 0.14 and NSE values from 0 to 0.71, highlighting the model's reliability. This approach was further applied to estimate SMC in remoted areas near the stations. Between 2009 and 2013, eleven major disaster rainfall events occurred, including four typhoons and seven strong frontal activities. We selected ten landslides triggered near the targeted in-situ monitoring stations. In total, 162 SMC fluctuation periods were delineated, and 248 single and continuous rainfall events were analyzed. The analysis revealed a critical threshold: landslide events were more likely to occur the combined value of the rate of change in SMC during ascent and descent periods and rainfall intensity exceeded 100 mm per day, with an accuracy of 86.4%, validated against landslide inventories. Regions with higher coefficients of variation (Cv) in SMC, particularly during the warm season in central and southern Taiwan, exhibited greater landslide susceptibility. Despite the effectiveness of the threshold-based model, missing rainfall data during certain periods, such as Typhoon Morakot in 2009, impacted predictive accuracy in certain areas. This study underscores the importance of continuous and accurate rainfall data for improving threshold-based landslide prediction models. Future research should address challenges in regions lacking rainfall, soil moisture data, and soil physical properties and explore refining thresholds based solely on satellite-retrieved SMC and rainfall intensity for specific landslide events.

DOI:
10.1007/s10346-025-02495-x

ISSN:
1612-5118