spatial modeling

This study modeled sensitive areas to dust storms in Isfahan province, which is sensitive to successive droughts, and dust storms because of its climatic condition, and proximity to the desert, using meteorological codes related to dust, AOD values, and Maximum Entropy model (MaxEnt).

200 occurrence points of dust were determined using dust meteorological codes and AOD values of MODIS sensor, Terra satellite, (2011-2022). Ten parameters including temperature, rainfall, albedo, altitude, slope, land use, enhanced vegetation index (EVI), normalized difference moisture index (NDMI), normalized difference salinity index (NDSI), and frequency percentage of erosive wind seed were considered dust-predictive factors. Finally, the MaxEnt model was utilized for modeling dust susceptibility. The performance of the model was specified using the AUC value and the importance of each influential factor was identified utilizing the Jackknife test.

  The findings indicated that areas susceptible to dust are mainly bare lands, salt lands, and poor rangeland located mostly in the north, northeast to parts of the east and southeast of the Province, and also the central parts towards the southwest of Isfahan Province. According to the results, the MaxEnt model, with AUC=0.72, had a good efficiency in modeling susceptible areas to dust storms in Isfahan Province.

The major conclusion of this study is that the MaxEnt model had good performance in mapping susceptible areas to dust in Isfahan Province. The results of this research can be useful for decision-makers in identifying the areas prone to dust storms.

In the present study, random forest (RF) and support vector machine (SVM) were used to assess the applicability of ensemble modeling in landslide susceptibility assessment across the Kolijan Rostaq Watershed in Mazandaran Province, Iran.

Both models were used in two modeling modes: 1) A solitary use (i.e., SVM and RF) and 2) Their ensemble with a bivariate statistical model named the weights of evidence (WofE) which then generated two more models, namely SVM-WofE and RF-WofE. Further, the resulting maps of each stage were dually coupled using the weighted arithmetic mean operation and an intermodal blending of the previous stages.

Accuracy of the models was assessed via the receiver operating characteristic (ROC) curves based on which the goodness-of-fit of the SVM and the SVM-WofE models were 0.817 and 0.841, respectively, while their respective prediction accuracy values were found to be 0.848 and 0.825. The goodness-of-fit of the RF and the RF-WofE models respectively was 0.9 and 0.823, while their respective prediction accuracy values were found to be 0.886 and 0.823. The goodness-of-fit and prediction power of SVM and SVM-WofE ensemble were respectively 0.859 and 0.873. The same increasing pattern was evident for the ensemble of RF and RF-WofE where their goodness-of-fit and prediction power increased, respectively, up to 0.928 and 0.873. Moreover, the goodness-of-fit and prediction power of RF-SVM ensemble were increased up to 0.932 and 0.899, respectively. The results of the averaged Kappa values throughout a 10-fold cross-validation test as an auxiliary accuracy assessment attested to the same results obtained from the ROC curves.

Successive intermodal ensembling approach is a simple and self-explanatory method so far as the context of many data mining techniques with a highly complex structure has been simply benefitted from the weighted averaging technique.