Landslides triggered by rainfall are the cause of thousands of deaths worldwide every year. One possible approach to limit the socioeconomic consequences of such events is the development of climatic thresholds for landslide initiation. In this paper, we propose a method that incorporates antecedent rainfall and streamflow data to develop a landslide initiation threshold for the North Shore Mountains of Vancouver, British Columbia. Hydroclimatic data were gathered for 18 storms that triggered landslides and 18 storms that did not. Discriminant function analysis separated the landslide-triggering storms from those storms that did not trigger landslides and selected the most meaningful variables that allow this separation. Discriminant functions were also developed for the landslide-triggering and nonlandslide-triggering storms. The difference of the discriminant scores, DeltaCS, for both groups is a measure of landslide susceptibility during a storm. The variables identified that optimize the separation of the two storm groups are 4-week rainfall prior to a significant storm, 6-h rainfall during a storm, and the number of hours 1 in 3 /s discharge was exceeded at Mackay Creek during a storm. Three thresholds were identified. The Landslide Warning Threshold (LWT) is reached when DeltaCS is - 1. The Conditional Landslide Initiation Threshold (CTLl) is reached when DeltaCS is zero, and it implies that landslides are likely if 4 mm/h rainfall intensity is exceeded at which point the Imminent Landslide Initiation Threshold (ITLl) is reached. The LWT allows time for the issuance of a landslide advisory and to move personnel out of hazardous areas. The methodology proposed in this paper can be transferred to other regions worldwide where type and quality of data are appropriate for this type of analysis. (C) 2002 Elsevier Science B.V. All rights reserved.
