The impact of natural factors and human activities on the dust-vulnerable regions in central Iran

Abstract

Identifying dust-vulnerable regions (DVRs) and their key drivers are essential for preventing land degradation and reducing dust pollution in arid regions. This research aimed to develop a dust-vulnerability map for Kerman Province in central Iran using the support vector machine (SVM) model. The model, utilizing four kernel functions, was used for modeling and assessing based on sensitivity, 1-specificity, and the area under the curve (AUC) of the receiver operating characteristic (ROC). The best DVR prediction map was validated using the dust frequency index (DFI) map. Subsequently, the significant components were prioritized through the learning vector quantization (LVQ) algorithm. All functions effectively delineated vulnerable areas, with the radial basis function (RBF) (sensitivity = 0.967; 1-specificity = 0.014; AUC = 0.958) surpassing others (sensitivity = 0.956; 1-specificity = 0.04; AUC = 0.977). The study area was classified into low (32%), moderate (11.6%), high (13.6%), and very high (42.8%) vulnerability zones using the optimal model. The classification's accuracy, particularly for the low and very high-risk classes, was validated by the reclassified DFI map, achieving balanced accuracies of 82% and 77%, respectively, indicating the reliability of the SVM model with the RBF function in identifying these classes. The LVQ analysis indicated that slope angle, geology, soil moisture, soil carbon density, and proximity to mines were critical factors influencing vulnerability to dust storms in central Iran. Land-related factors and human activities were identified as having a more substantial impact on dust vulnerability compared to climatic factors. These results can aid policymakers in prioritizing strategies to combat land degradation, manage dust storms, and alleviate the negative effects on air quality and human health, particularly in highly vulnerable regions.

Keywords: Arid regions; Dust storms; Environmental factors; Machine learning; Remote sensing.