Definition of rain-triggered landslide thresholds for early warning systems at different geographical scales in Colombia

Abstract

ABSTRACT : Due to its geographic and hydro-meteorological conditions, Colombia has a long history of mass movement disasters. According to Gómez, García, and Aristizábal (2021), most landslides have been triggered by rainfall and have caused 34,248 deaths over 99 years (1921-2019). For this reason, the definition of the rainfall conditions that trigger landslides that could be implemented in early warning systems (EWS) is one of the prospective intervention actions that challenge Colombia’s disaster risk management policy. This thesis focuses on determining a tool that contributes to the solution of the country’s landslide problem and was developed in research stages presented in seven chapters in this document: The first chapter presents the objectives outline and research background. The second chapter focuses on the presentation of the target areas (different geographic scales), landslide conditioning factors and rainfall databases used for the study. The third chapter focuses on a spatial and temporal analysis of landslides in Colombia and the world, which shows a local and global overview of the problem. The fourth chapter conducts a rainfall-landslide analysis and evaluates the performance of the satellite rainfall estimations (SRES) in the study area. The fifth chapter defines landslide thresholds that can be used for landslide predictions, and the sixth chapter articulates the results obtained in previous chapters by presenting a methodology proposed as a forecasting component of Landslides Early Warning Systems (LEWS) and its validation. Finally, the seventh chapter presents general conclusions and recommendations for future work on the scope of this research. The methodological framework proposed is based on empirical and statistical landslide prediction approaches. It is focused on the same three geographic levels in which the risk management structure in Colombia are aligned: the Andean Region at the national level, the Antioquia department and the Medellín municipality at the regional and local levels, respectively. In addition, this methodology proposes to work on these scales in an articulated and non-articulated way, allowing them to be applied separately in each region or locality. Although it is focused on Antioquia and Medell´ın, it could be replicated in other areas of the country and the world with a complex topography and significant climatic heterogeneity. This methodology proposes working with ground-based rain information and satellite rain estimations (SREs: CHIRPSv2 and MSWEPv2.6) due to the scarcity of groundbased rain or its drawbacks in the recorded data. Moreover, the results indicated that CHIRPSv2 is the best-performing SRE for landslide prediction in the Colombian Andean Region. Finally, the methodological framework validation pointed out that this tool can be used as a forecasting component of LEWS, generating hazard warnings in the areas evaluated, with results at the pixel-resolution level when using CHIRPSv2 (0.05°) and at the station area when using ground-rainfall gauges data. Although overall, the results using groundbased rainfall as input data outperformed those using CHIRPSv2, CHIRPSv2 results were satisfactory, with accurate probabilities of occurrence in statistical models and high warnings using the empirical thresholds, and corroborating that the use of this SRE would yield good results in areas with a scarcity of ground-rainfall data.