The Role of Conditioning Factors in Determining Rainfall Intensity Necessary for Triggering Shallow Landslides in Portugal

Shallow landslides are geological phenomena that affect soil of small thickness originated from the weathering of the bedrock and downslope transportation. The goal of this work is to analyze the correlation between the conditioning factors and the triggering factor (rainfall) of shallow landslides in the continental area of Portugal. The understanding of the correlation between these factors can be of great help for the development of early warning systems, since it enables near real time updates of susceptibility maps relying on the rainfall forecast and the  specific physical characteristics of different regions. We used the DISASTER landslide historical database and analyzed the following conditioning factors: elevation, slope, aspect, lithology, land use, distance to rivers and faults. The historical rainfall data were obtained from the gridded NetCDF file provided by the Copernicus climate services. An automatic script was created to filter in the database the landslides that can be considered a shallow landslide and the ones that were probably triggered by rainfall events. Another automatic script was created to extract from the NetCDF file the intensity of the rainfall event that triggered the landslide. Then, we used the Boruta algorithm for feature selection. The Boruta algorithm helps to reduce the number of features in a dataset by identifying features that do not influence the study variable. In our case, the algorithm analyses which conditioning factor influences the rainfall intensity necessary to cause the respective landslide. It was found that only the lithology, slope, elevation and aspect had a significant contribution to the definition of the necessary rainfall intensity. In order to analyze how the changes in the conditioning factors affect the rain intensity necessary to cause the landslide, we grouped the events by lithology.  Two-mica granites were the lithology with the widest range of rainfall intensities that triggered landslides, reaching the lowest and higher values. This result possibly demonstrates that regions dominated by two-mica granites have higher susceptibility to landslides. Next, the Pearson correlation was used to determine whether the correlation between the relevant conditioning factors and the triggering factor were positive or negative. As a preliminary result, we found that all the Pearson correlations were low and positive, showing that the increase of value of conditioning factors result in a small increase in rain intensity necessary to cause landslides. This correlation can be probably explained by analyzing the scatter plot “rainfall intensity/slope”. The plot shows that the slopes lower than 10 degrees and higher than 20 degrees show a minimum rainfall intensity higher than the ones within slopes between 10 and 20 degrees. This could be explained by the fact that shallow slopes have low gravitational potential energy demanding high rainfall intensities to trigger a landslide and steep slopes could not have enough material accumulated to generate a landslide with low rainfall intensity. The next step will be to run a statistical model to completely correlate the conditioning factors with the respective rainfall intensities.