TY - JOUR

T1 - Autologistic modelling of susceptibility to landsliding in the Central Apennines, Italy

AU - Atkinson, Peter M.

AU - Massari, R.

N1 - M1 - 1-2

PY - 2011/7

Y1 - 2011/7

N2 - In previous research, a logistic regression of landslide occurrence on several explanatory variables was fitted and used to map landslide susceptibility for a small area in the central Apennines, Italy. Here, the spatial dependence or spatial correlation in the residuals from the fitted regression model was accounted for by inserting an autocovariate into the model. The autocovariate was estimated by applying a Gibbs sampler to the susceptibilities for neighbouring pixels. As in any landslide susceptibility analysis, accuracy was difficult to assess because of the requirement for data on future landslides. However, by comparing the ordinary logistic model to the autologistic model obtained on the same set of data, it was possible to assess the influence of the autocovariate. The autocovariate rendered the model simpler because several variables lost their significance and were, therefore, omitted from the model. Further, areas of high landslide susceptibility estimated from the autologistic model were geographically clustered, as one would expect, and this may be advantageous in terms of (i) interpreting the model and (ii) displaying the results to non-experts.

AB - In previous research, a logistic regression of landslide occurrence on several explanatory variables was fitted and used to map landslide susceptibility for a small area in the central Apennines, Italy. Here, the spatial dependence or spatial correlation in the residuals from the fitted regression model was accounted for by inserting an autocovariate into the model. The autocovariate was estimated by applying a Gibbs sampler to the susceptibilities for neighbouring pixels. As in any landslide susceptibility analysis, accuracy was difficult to assess because of the requirement for data on future landslides. However, by comparing the ordinary logistic model to the autologistic model obtained on the same set of data, it was possible to assess the influence of the autocovariate. The autocovariate rendered the model simpler because several variables lost their significance and were, therefore, omitted from the model. Further, areas of high landslide susceptibility estimated from the autologistic model were geographically clustered, as one would expect, and this may be advantageous in terms of (i) interpreting the model and (ii) displaying the results to non-experts.

KW - Landslide susceptibility

KW - Logistic regression

KW - Autologistic regression

KW - Spatial scale

U2 - 10.1016/j.geomorph.2011.02.001

DO - 10.1016/j.geomorph.2011.02.001

M3 - Journal article

VL - 130

SP - 55

EP - 64

JO - Geomorphology

JF - Geomorphology

SN - 0169-555X

IS - 1-2

ER -