LAUSR

Abstract This study aims to predict land susceptibility (a term that indicates the degree of sensitivity of land to detachment of soil particles by wind) to dust emissions in Yazd province, central Iran, by combining a new integrated data mining (DM) model and the RegCM4 climatic model. The study further determines the relative importance of key factors controlling dust emissions by applying 12 individual DM models. The integrated model is based on the individual models returning Nash Sutcliffe coefficient (NSC) values > 90% for the spatial modelling of land susceptibility to dust emissions and using the area under the curve (AUC) for validation. 13 key factors controlling dust emissions are mapped including soil characteristics, climatic variables, vegetation cover, a Digital Elevation Model (DEM), geology and land use. Based on Spearman clustering analysis and multi-collinearity tests (tolerance coefficient -TC and variance inflation factor -VIF), the effective factors for dust emissions are classified into nine clusters and no multi-collinearity is found among the effective factors. DEM, NDVI (normalized difference vegetation index), geology and calcium carbonate are identified as the most important factors controlling dust emissions. Seven individual models return NSC in the range of 90–98% and are used to generate the integrated model for the final mapping of land susceptibility to dust emissions. Among 851 pixels located in the dust sources, 30% (255 pixels) and 70% (596 pixels) are randomly selected as validation and training datasets, respectively for the new integrated model. Using this model, 9%, 17%, 7% and 67% of the study area correspond to low, moderate, high and very high susceptibility classes, while the validation results in AUC = 99.3%. Simulations with the RegCM4 model reveal high consistency regarding the spatial distribution of the most susceptible areas and dust emissions. Overall, combining DM approaches and physical models is useful in aeolian geomorphology studies.