Air quality forecasting through integrated model using air dispersion model and neural network

摘要

The study presents the development of an integrated model based on air dispersion model and neural network (NN) that optimises the performance of each of the methodology. The Air dispersion model and NN models are two important techniques for evaluating air pollution concentration. The performance of the present integrated model is tested for Respirable Suspended Particulate Matter (RSPM) in Delhi, one of the most polluted cities of the world, in the month of December, representing the winter season, which has worst meteorological scenario. The air dispersion models are analytical solution of atmospheric diffusion equation that take account the wind speed as a power law profile of vertical height and vertical eddy diffusivity as an explicit function of downwind distance from the source and vertical height. Weather Research and Forecast (WRF) model is used for simulating the hourly meteorological parameters using NCEP/NCAR FNL data at resolution of l°x 1°. The concentrations of dispersion models with meteorological variables are used as input parameters to the neural network for forecasting daily air quality. Results show a marked improvement for all simulations when neural network is added downstream of the air dispersion models. A comparative study of integrated model, air dispersion model and observed data demonstrates that the use of NN in order to correct the dispersion model for air quality forecasting could be the reasonable model combination when the air dispersion model gives some systematic error with respect to observed data.