Landslide Susceptibility Model Validation: A Routine Starting from Landslide Inventory to Susceptibility

The main purpose of the present study is to evaluate the model validation stage of a routine landslide susceptibility mapping. For the purpose, model validation is assessed in three stages; (1) during model data production, (2) during model construction, and (3) during the production of model consequences; landslide susceptibility maps. As the results of these evaluations, it is revealed that training and testing data sets should be separated considering an appropriate separation ratio which is about 80 % and 20 % of the presence (1) data after completion of inventory studies. Correct classification percentages, error matrices, and the Kappa index are suggested to be considered for the training data sets during model construction. Additionally, again the correct classification percentage and the Root Mean Square Error (RMSE) should be considered during this stage for the testing data sets as well. In order to evaluate the spatial performance of the produced landslide susceptibility maps, the use of the Receiver Operating Characteristic (ROC) curves and the Area Under Curve (AUC) statistics is recommended. In the present study, the maximum Kappa index (k) value was calculated to be 0.459 for both the random sampling 1 (Rnd1) in the model 1 and for the random sampling 2 (Rnd2) in the model 2 during the model construction stage. The AUC values were calculated for these random samplings to be 0.781 and 0.790 respectively during the production of the model consequence stage in the study.