Landslides are one of the most pervasive hazards provoking casualties and significant economic losses at the global scale. In order to substantially reduce the impact of landslides, it is essential to understand the factors controlling when and where landslides occur (hazard). Landslide hazard investigations in Central Africa are notably deficient due to the data-scarce context, rather than to the lack of landslide occurrence. Motivated by the observed discrepancy, this thesis aims at developing tools for regional hazard assessment adapted to data-scarce contexts, with a focus on the central section of the western branch of the East African Rift (WEAR).

As a fundamental requirement for assessing landslide hazard, this thesis research starts with a systematic collection of spatiotemporal data on landslide events, for the specific context of a political unstable region where a robust communication network is lacking and which covers large extents of remote areas where field accessibility is low. Information on 143 landslide events was collected, presenting the first regional event inventory in Central Africa. The event inventory directly contributes to an increased representation of Central Africa in global landslide inventories, and constitutes indispensable information for the subsequent analyses in this thesis.

Also crucial for hazard assessment is the availability of accurate rainfall data, as precipitation is the most common trigger for landslides. We compiled an unprecedented record of rain gauge data for the validation of satellite-based rainfall estimates (SREs) from the Tropical Rainfall Measurement Mission Multi-satellite Precipitation Analysis (TMPA). Despite the uncertainties in TMPA, we illustrate that satellite-based area-averaged rainfall estimates are indispensable and most suitable to provide the regional rainfall information required in hazard assessment, compared to data from scarcely distributed gauges with limited representativeness in the context of high rainfall variability.

The collected information on landslide events and validated SRE data from TMPA, serve to calibrate the first regional rainfall thresholds for Central Africa, that is, a principal tool for characterizing landslide hazard. We propose a novel statistical threshold approach based on the relation between antecedent rainfall (AR) and landslide susceptibility (S) for the definition of spatially varying rainfall thresholds (referred to as the AR-S approach). Susceptibility data were obtained from a continental-scale susceptibility model and antecedent rainfall was calculated using an improved index that accounts for the different lasting effect of large versus small rainfall. Subsequently, the availability of a more accurate regional-scale susceptibility model allowed us to test and improve the transferability of the AR-S approach, therefore contributing to the development of a robust rainfall threshold technique that is adapted to data-scarce contexts.