Accuracy comparison of Elamn and Jordan artificial neural networks for air particular matter concentration (PM 10) prediction using MODIS satellite images, a case study of Ahvaz

Due to the complexity of air pollution action, artificial intelligence models specifically, neural networks are utilized to simulate air pollution. So far, numerous artificial neural network models have been used to estimate the concentration of atmospheric PMs. These models have had different accuracies that scholars are constantly exceed their efficiency using numerous parameters. The current research aims to compare Elman and Jordan recurrent networks for error distribution and validation to estimate atmospheric particular matters concentration in Ahvaz city. The used parameters are relative humidity, air pressure, temperature and aerosol optical depth. The latter one is extracted from MODIS sensor images and air pollution monitoring stations. The results show that Jordan model with RMSE of 219.9 milligram per cubic meter has more accuracy rather than Elman model with RMSE of 228.5. The value of R2 index that shows the linear relation between the estimated from the model and observed values for Jordan is equal to 0.5 that implies 50% estimation accuracy. The value is because of MODIS spatial resolution, inadequacy in numbers as well as spatial distribution of meteorological station inside the study area. According to the results of the current research, it seems that air pollution monitoring stations have to be increased in terms of numbers and suitable spatial distribution. Also, other ancillary data like volunteer geographic air pollution data entry using mobile connected cheap sensors as portable stations may be used to implement more accurate simulation for air pollution.