Challenges of modeling rainfall triggered landslides in a data-sparse region: A case study from the Western Ghats, India

Abstract

Accurate rainfall estimates are required to forecast the spatio-temporal distribution of rain-triggered landslides. In this study, a comparison between rain gauge and satellite rainfall data for assessing landslide distribution in a data-sparse region, the mountainous district of Idukki, along the Western Ghats of southwestern India, is carried out. Global Precipitation Mission Integrated Multi-satellitE Retrievals for GPM-Late (GPM IMERG-L) rainfall products were compared with rain gauge measurements, and it was found that the satellite rainfall observations were underpredicting the actual rainfall. A conditional merging algorithm was applied to develop a product that combines the accuracy of rain gauges and the spatial variability of satellite precipitation data. Correlation Coefficient (CC) and Root Mean Squared Error (RMSE) were used to check the performance of the conditional merging process. An example from a station with the least favorable statistics shows the CC increasing from 0.589 to 0.974 and the RMSE decreasing from 65.22 to 20.01. A case scenario was considered that evaluated the performance of a landslide prediction model by relying solely on a sparse rain gauge network. Rainfall thresholds computed from both the conditionally merged GPM IMERG-L and the rain gauge data were compared and the differences indicated that relying solely on a discrete, sparse rain gauge network would create false predictions. A total of 18.7% of landslide predictions only were identified as true positives, while 60.7% was the overall false-negative rate, and the remaining were false-positives. This pointed towards the need of having a continuous data that is both accurate in measurement and efficient in capturing spatial variability of rainfall.

Keywords

• Landslides
• GPM
• IMERG
• Conditional merging
• Remote sensing
• Satellite precipitation