Landslides susceptibility mapping using Geographically Weighted Regression with Gaussian distance weighting kernel and comparison with Logistic regression algorithm : a case study in Semirom, Isfahan, Iran

Huge economic detriment and victims are the results of Landslide. Since there are no physical sensors that would detect landslides directly, detection of landslides is a substantial challenge. Hence, the purpose of this article is to enhance the prediction results of landslide occurrence in Semirom region in Isfahan Province, Iran using a new methodology. Accordingly, the slope, aspect, lithology, land use, distance from faults, distance from river, distance from main roads, precipitation, plan curvature, profile curvature and the distance from residential regions are considered as the effective factors in landslide occurrence. By means of Geographically Weighted Regression (GWR) using the Gaussian distance weighting kernel with landslide inventory map, the Landslide susceptibility map was created. However, before combining the generated factors mentioned above, their independence among each other has been determined. For this reason, the correlation matrix was calculated and it showed that most of the correlations between factors were very minor, proposing that all landslide factors are adequately independent. Coefficient of Determination for GWR with Gaussian distance weighting kernel was 0.8557. The root mean squared error (RMSE) was 0.0804 and the mean root mean squared error (NRMSE) was 0.4468.

Keywords:

Landslide , Landslide susceptibility mapping , Geographically Weighted Regression(GWR)