prilagođeno pretraživanje po punom tekstu

ŠUMARSKI LIST 3-4/2022 str. 47     <-- 47 -->        PDF

to the planned alternatives, and route limitation was made by positioning the starting and destination points. First, road planning was carried out using the slope criterion used by the traditional approach. Second, the route was recalculated using ArcGIS Cost Path analysis, taking into account the landslide susceptibility obtained through the LR and RF methods. This study’s workflow is sum rized in Figure 3.
During the first phase, according to the importance of each factor as given by Sahin et al. (2020), chi-square, information gain, and random forest importance were applied from high to low importance as listed in Table 1. The table shows that each method and factor produced different feature weights and are in different rankings, according to the statistical method. There are differences between the first three factors in the chi-square ranking and in the factor rankings of information gain and random forest importance. Selections were based on the values obtained from the chi-square to determine the models and modeling that was carried out.
To determine the effects of the factors on the performance of the prediction model (Sahin et al. 2020), the factors’ importance values were ranked in ascending order. The factors that provided high performance (by choosing the best subset) were estimated in order to determine the highest AUC values. Various statistical tests (Wilcoxon signed-rank test, F-Test, Kolmogorov Smirnov test, and One Sample T-Test) were used by the LSM Tool Pack (Table 2). For this study, the Case 1 model-7 scenario was chosen. The best possible scenarios, using combinations of factors, are shown in Table 2.
Logistic Regression – Logistička regresija
The LR modeling approach has been frequently and widely used in landslide-susceptibility studies. The most successful combinations, using nine factors in total, are shown in Table 2. The AUC value (97.5522) of the Case 1 model-7 scenario, which was selected as the most successful LR approach, the estimated factors, std. Error, z-value, and Pr values are shown in Figure 4. The curvature factor correlated negatively with landslide formation, while the remaining eight factors (distance to fault, lithology, distance to road, slope, SPI, distance to stream, TPI, and TWI) correlated positively. Furthermore, slope, TPI, lithology, distance