Exploiting historical rainfall and landslide data in a spatial database for the derivation of critical rainfall thresholds

Critical rainfall thresholds for landslides arepowerful tools for preventing landslide hazard. Thethresholds are commonly estimated empirically startingfrom rainfall events that triggered landslides in the past.The creation of the appropriate rainfall–landslide databaseis one of the main efforts in this approach. In fact, anaccurate agreement between the landslide and rainfallinformation, in terms of location and timing, is essential inorder to correctly estimate the rainfall–landslide relationships.A further issue is taking into account the averagemoisture conditions prior the triggering event, which reasonablymay be crucial in determining the sufficientamount of precipitation. In this context, the aim of thispaper is exploiting historical landslide and rainfall data in aspatial database for the derivation of critical rainfallthresholds for landslide occurrence in Sicily, southernItaly. The hourly rainfall events that caused landslidesoccurred in the twentieth century were specifically identifiedand reconstructed. A procedure was proposed toautomatically convert rain guages charts recorded on papertape into digital format and then to provide the cumulativerainfall hyetograph in digital format. This procedure isbased on a segmentation followed by signal recognitiontechniques which allow to digitalize and to recognize thehyetograph automatically. The role of rainfall prior to thelandslide events was taken into account by including in the analysis the rainfall occurred 5, 15 and 30 days before eachlandslide. Finally, cumulated rainfall duration thresholdsfor different exceedance probability levels were determined.The obtained thresholds resulted in agreement withthe regional curves proposed by other authors for the samearea; antecedent rainfall turned out to be particularlyimportant in triggering landslides.