CGRG Bibliography of Canadian Geomorphology
Search Results
Author : Chung, C-J.; and Fabbri, A.G.
Date : 2005.
Title : Spatial predictions of the occurrences of future landslides and their validations.
Publication : Landslide Risk Management: Proceedings of the International Conference on Landslide Risk Management, Vancouver. Edited by: Hungr, O.; Fell, R.; Couture, R.; and Eberhardt, E. Leiden: A.A. Balkema.
Issue :
Page(s) : Paper on CD.
Abstract
A landslide hazard map, regardless of how or by whom it was produced, hopefully shows the locations of the future landslides. There are very many ways to construct prediction maps, from simple heuristic opinion-based procedures with little data to sophisticated mathematical models supported by complex databases and using advanced software and hardware technologies. From a review of existing mapping techniques, we have noticed the presence of one common unfortunate deficiency in the application of prediction models: lack of validation procedures of the prediction results. A systematic procedure is proposed in this contribution to eradicate the deficiency. It provides two analytical stages: one of "relative hazard mapping" and one of "vali-dation." The first stage needs formulating a mathematical model that generates a prediction map by dividing the study area into a number of "prediction" classes according to their relative likelihood of occurrence of the future landslides, conditional on the geomorphologic and topographic characteristics present at their location. The second stage requires empirically evaluating the prediction results for future landslide occurrence in each prediction class. Strictly speaking, the validation of the results of a prediction of future events is not possible, because the events are still to happen. However, a "wait-and-see" type of validation is often not acceptable as a decision alternative. A cross-validation procedure (it is often termed a blind test) consists of empirically measuring the validity of the prediction results. In addition, the statistics obtained from the cross-validation are used to estimate the probability of occurrence of the future landslides at each pixel. This estimated probability allows to further proceeding in risk analysis. The strategy of cross-validation is simple but effective. The procedure requires dividing the distribution of the past landslide occurrences into two groups, a "modeling group" and a "validation group." Using the landslides in the modeling group, a landslide hazard prediction map is generated that normally consists of a set of predicted hazard classes. The set of classes is spatially compared with the distribution of the landslide occurrences in the validation group. The statistics obtained from the comparison provide quantitative estimates of both the validity/reliability/robustness of the prediction results and the probability of the occurrences of future landslides. As in any prediction, the methods of how to predict do not have any scientific significance without the accompanying empirical measures of validity of the prediction results.
Bibliography of Canadian Geomorphology