Quantitative landslide risk analysis with limited dataset by leveraging geospatial approaches in Indonesia

Landslide risk analysis is a key process in mitigating adverse losses. However, in an area with a limited dataset, the assessment can be challenging. Accordingly, this study aims to analyze the landslide risk within a data scarcity environment by leveraging the open-access geospatial data with ground checks. This study estimated the landslide risk in the District of Sukajaya, Regency of Bogor, Indonesia, by overlaying the landslide hazard and vulnerabilities that were exemplified by the exposed elements. This study utilized the available dataset on historical landslides and conditional factors to generate the spatial (susceptibility) and temporal probabilities for the landslide hazard. The spatial probability evaluated two robust approaches, i.e., machine learning-based (i.e., Random Forest) and deep learning-based (i.e., TabNet) methods for the Landslide Susceptibility Mapping (LSM). The result revealed that the TabNet-based LSM outperformed the Random Forest-based LSM. Moreover, the landslide exposures were quantified from the open-access dataset for populations and buildings. The resulting landslide risk estimated that the Sukajaya District had a total building risk of IDR 1.20 trillion at 4.88 $$\text {km}^2$$ and a total population risk of 39,204 people, or around 58.58% of the total population. This study is beneficial for policymakers and authorities in mitigating the prevailing landslide disaster and thus, achieving better sustainable development.