| Abstrakt: |
Understanding the spatial distribution pattern of dust-sensitive regions (DSR) and prioritizing the factors affecting them are important steps in controlling and alleviating the dust events hazards in arid and semi-arid regions of the world. In this research, in addition to spatial modelling of DSR, the effectiveness of its controlling agents in Iran has been investigated using machine-learning algorithms and game theory. Input variables were chosen using the variance inflation factor, least absolute shrinkage, and selection operator (Lasso) and Boruta methods. The Lasso and random forest (RF) algorithms were employed for modelling the spatial variability of DSR across Iran. The error evaluation metrics and Shapely values were respectively applied to measure the performance of both algorithms and the drivers' contributions in spatial variations of DSR. The R-square (R2), root mean square error (RMSE), mean absolute error (MAE), and Nash–Sutcliffe efficiency coefficient (NSEC) were calculated to measure the performance of both algorithms. The Lasso resulted in R2 = 55, NSEC = 0.60, RMSE = 0.38, and MAE = 0.23, meaning that the Lasso is less efficient in spatial modelling of DSR compared to the RF model, which resulted in R2 = 0.91, NSEC = 0.89, RMSE = 0.16, and MAE = 0.12. Topographic diversity, slope, and vegetation cover followed by soil sand content and precipitation were identified as the influential environmental agents in the spatial variability of DSR over Iran. However, bulk density and land use were identified as less important drivers in these regions of Iran. These results can provide a platform for alleviating dust pollution hazards in the DSR of Southwest Asia, especially in Iran. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink