multivariate spline adaptive regression model (mars)

The phenomenon of dust is one of the most important environmental crises in arid and semi-arid regions of the world, which has harmful effects on human health and the environment. Dust or fine dust refers to very small and light particles of silt, clay or sand, which are moved and transported for a long distance in the earth's atmosphere as a result of wind erosion and desertification by the wind, and the horizontal visibility is between It is limited to 1 to 2 km or less. Due to being located in the dry and semi-arid belt of the world, Iran is exposed to numerous local and extra-regional dust systems. The dust storm in our country during the last few years has involved the provinces of the country as a serious crisis and has brought dangerous consequences for the residents of these areas in terms of environment, health and economy. Identifying the dust source area using remote sensing techniques is one of the most important methods in the world. The purpose of this research is to identify the centers of dust harvesting and its risk zoning using RF and MARS models. It is in Ilam province. 38 dust harvesting centers were identified in the study area. 70% of the identified foci were used for modeling and 30% for evaluation. Then 7 factors including soil, lithology, slope, vegetation cover index (NDVI), distance from the river, climatic units and land use were prepared as independent and effective variables on the center of dust collection. Then, using RF and MARS models, dust risk zoning maps were prepared.

The results of the identification of dust source area indicate that a total of 38 dust source area were identified in the entire region. The rule for detecting dust source in this research was based on the Gaussian plume diffusion model. In this way, when a dust emission cone is observed in a satellite image, the top of the cone represents the dust point. The results of the random forest model indicate that land use and soil science factors had the greatest role in the occurrence of dust. Using this model, the areas of sensitivity to dust extraction centers showed that about 18.1% of the studied area is in the category of very high sensitivity, and 63.2% of all dust source area are also in this range. Has taken. According to the results of MARS model, respectively, land use factors and climatic classes had the greatest effect on the creation of dust collection centers, and soil science had the least role. Examining the floor area of the dust sensitivity map using the MARS model showed that the area with high sensitivity has the lowest area (6.7) and the area of high, medium and low floors is 19, 42.8 and 31.5 respectively. It has been estimated. About 44.7% of collection centers are located in the area with very high sensitivity and there are 2.6% dust collection centers in the area with low sensitivity. The results of the random forest model showed that 16.6% of the area of Ilam province is in the area with low sensitivity, 42.3% in the area with medium sensitivity, 23% in the area with high sensitivity and 18.1% in the area with high sensitivity. It is located a lot. The results indicate that the highest concentration of dust collection (63.2%) is located in the area with high sensitivity. The results show that about 81.6% of the dust harvesting centers are located in the area with high and very high sensitivity. In both models, the highest percentage of dust collection centers is in the very high sensitivity class, which includes a small area of the region, especially in the MARS model.

The results of both models indicate that land use had the greatest impact on the creation of dust source area. Evaluation of the models using the ROC curve showed that in relation to the success rate, the RF and MARS models have accuracy of 0.91 and 0.86, respectively. Both models show a high correlation between sensitivity maps and distribution of dust centers. The results showed that the RF model has a higher accuracy in determining the areas sensitive to dust collection centers in the study area.