2025/09/04 by Gholipour, Mehranoosh, Kiani Sadr, Maryam, Lorestani, Bahareh +2
#Dust sources #Kermanshah #Multiple linear regression #Random forest model #Remote sensing technique
paper · doi:10.57647/j.jap.2025.0901.07
Identifying dust sources is essential for effective management. This study employs remote sensing and machine learning techniques, specifically Random Forest Model (RFM) and Multiple Linear Regression (MLR), to identify dust production areas in Kermanshah province from 2000 to 2020. The results indicate that the Random Forest Model (RFM) outperforms Multiple Linear Regression (MLR). Specifically, the accuracy of the RFM, with a correlation coefficient of 0.900, is significantly higher than that of the MLR, which has a correlation coefficient of 0.840. Furthermore, the Root Mean Square Error (RMSE) for the RFM was 5.59, whereas for the MLR it was 7.05, confirming the superiority of the RFM’s accuracy. Key factors influencing dust production include elevation, erosion, soil moisture, land cover, geology, daytime land surface temperature, and NDVI. The western and southwestern regions have been identified as hotspots for dust production. These findings can assist policymakers in effectively allocating resources and managing dust-related issues. Hazard zoning maps will facilitate the identification of areas requiring immediate intervention.