** Introduction**

**Particulate matter is any liquid or solid component (except pure water) that is dispersed in the Earth's atmosphere and is microscopic or sub-microscopic but larger than the molecular size. These particles play an important role in the Earth's climate. Suspended particles are created by various natural or anthropogenic processes and are among the deadliest types of air pollution, especially smaller particles less than 10 micrometers in diameter. Since the number of pollution station is very low, Satellite measurements have been widely used to estimate particulate matters (PMs) on the ground and their effects on human health.**

**Methodology**

**In this research, we tried to estimate PM10 using the regression model based on the Aerosols optical depth. Because the AOD value recorded by satellite sensors is affected by the weather conditions, to increase the accuracy of the PM10 estimation, meteorological parameters were also used in the AOD to PM10 conversion model. The used meteorological parameters include surface wind speed, surface temperature, relative humidity, visibility, and planetary boundary layer height. **

**Since the data used were extracted from four different sources with different temporal and spatial resolution, it is necessary to apply a method for integration and synchronization in space and time. To generate AOD and PM10 data from the Aqua satellite, pollution stations were mapped onto satellite images and AOD values for the nearest neighbor pixels as well as values for a 3 by 3 window were extracted. So, two pairs of AOD and PM10 were formed, one with the nearest neighbor values and the other with the weighted average AOD values in a 3 by 3 window. Because the PM10 values were more closely related to the AOD values than the nearest neighbor, the pair of the weighted average was excluded from the calculations. There were extracted 100 samples in warm season (June, July, August and September) for model development (Shokoufa station data) and 65 samples for validation (Cheshmeh and Atisaz station data) and 140 samples in cold season (November, December, February and March) for model development, and 50 samples for validation. In the last step, the accuracy of the model was evaluated using indices such as coefficient of determination, mean error deviation (Bias), and mean square error (RMSE). **

**Results and Discussion**

**The results showed that AOD and PM10 have a better relationship with each other in the warm season than in the cold season. Only two variables of AOD and wind speed were included as independent variables in the best model presented for the warm season; both of which have a direct relationship with PM10, that is, with increasing of both variables the value of PM10 increases. The results showed that the regression model of warm season can only predict 16% of the PM10 variations correctly, which is not a satisfactory result.**

**In the multivariate cold season regression model, only the visibility remained, and other variables that had no significant effect on model improvement were excluded from the regression model. Multivariate correlation coefficient of this model was estimated to be 0.59. Therefore, the cold season regression model, at best can predict 35% of the PM10 variations correctly. By deleting the visibility variable, it was attempted to measure the impact of other variables such as AOD on PM10 estimation. In this model, the boundary layer height, AOD and temperature variables were retained. The boundary layer height variable has a negative relationship and the other two variables have a positive relationship with PM10. The maximum effect of temperature on PM10 is justified by the increase in boundary layer and the relationship of these two dependent variables which decreases PM10 density, but since this role of temperature element is represented by the same boundary layer height variable, what remains is the secondary role. Temperature is in the PM10 particle production. However, the latter model is weaker than the previous model and its multivariate correlation coefficient is 0.45 and accounts for 20% of the PM10 variations.**

**In the evaluation of the model in the warm season, the root mean square error at the Cheshmeh station was 31.76 µg / m3 and at the Atisaz station was 33.56 µg / m3. In the cold season, the root mean square error was estimated to be 47.10 µg / m3 at the Cheshmeh station and 49.81 µg / m3 at the Atisaz station, respectively. However, using the model with independent variables AOD, boundary layer height and temperature, the root mean square error was estimated to be 38.42 μg / m3 at the Cheshmeh station and 39.11 μg / m3 at the Atisaz station. The former shows a decrease of approximately 10 micrograms per cubic meter. Therefore, although the latter model with independent variables of boundary layer height, AOD and temperature had less multivariate correlation coefficients and determination coefficients than the model with independent observational variables in cold season, it yielded better results based on evaluation of the model for different locations and days from modeling location and days.**

**Conclusion **

**Generally, based on the results, it can be stated that regression models of warm and cold seasons are statistically acceptable at a confidence level of 99%. Therefore, the amount of PM10 fluctuations that is justifiable by the model is not accidental, although the modifications justified by models are low. The calculated errors in the evaluation section showed that the proposed models are not very accurate. It was also found that in the warm season, the wind speed can improve the results of the regression model of the relationship between AOD and PM10, and in the cold season, the variables of the boundary layer height and temperature in the regression model are statistically acceptable and improve the results of the model.**