Resumen
Dust formation is one of the most seriously damaging environmental issues in arid and semiarid areas. In this study, the decision tree-based Chi-square Automatic Interaction Detector (CHAID) algorithm was used to determine the non-linear relationships of soil physical and chemical properties with seasonal and annual dust deposition rate (DDR) in Gavkhouni swamp sub-basin, Central Iran. The results were compared with those obtained by the multiple linear regression (MLR) method. A set of 124 atmospheric dust samples was seasonally taken from 31 sites. A set of 96 surface soil samples was also collected. DDR and dust particle size distribution, as well as the physical and chemical properties of soil samples were investigated. The results showed that the highest and lowest DDR belonged to summer and autumn, respectively. Based on the CHAID algorithm results, the most important soil properties affecting DDR in autumn, winter, spring and summer, as well as annual DDR were soil organic matter content (importance coefficient [IC] = 0.34), gypsum (IC = 0.42), sand (IC = 0.39), silt (IC = 0.31), and sand (IC = 0.23), respectively. Based on the CHAID algorithm results, it appears that particle size distribution of surface soil, especially sand content is a determinant factor affecting seasonal and annual DDR in the study area. In this study, the MLR model had unacceptable accuracy as compared with the non-linear CHAID algorithm method. Therefore, it seems that in areas with high ecological complexity and complex nonlinear relationships among input and output data, the nonlinear methods such as CHAID are superior to linear methods such as MLR.