Hazardmatch: an Application of Artificial Intelligence to Landslide Susceptibility Mapping, Howe Sound Area, British Columbia

HazardMatch is a computer system for the production of landslide susceptibility maps using semantics and semantic technology, a field of artificial intelligence. It provides a software framework within which a landslide expert can describe, using language as close to natural (specialist) language as possible, the properties of surface locations which are highly susceptible to landslides. It uses the same language to generate semantic descriptions of all sub-areas of the area of interest within which all landslide-relevant properties are effectively constant. By ranking all sub-area descriptions by their similarity to those around extant slope failures, maps of similarity scores and rankings are created. In this way, HazardMatch uses the same sort of knowledge that practitioners use to communicate between themselves in recognizing unstable slope environments. The results are readily explainable and justifiable to non-experts. Furthermore, if the results are demonstrably incorrect, the cause of error is usually traceable to an aspect of the expert’s landslide model that is at odds with the facts, or to the input data. These aspects can then be changed to produce more accurate maps. A preliminary test of this method was carried out in the Coast Mountains. Examination of soil slides on recently logged hill slopes yielded promising results.