Forecasting air pollutant concentrations using gene expression programming and multiple linear regression in Tehran, Iran

The forecasting of air pollutant trends has received much attention in recent years. It is animportant and popular topic in environmental science, as concerns have been raised about thehealth impacts caused by unacceptable ambient air pollutant levels. Of greatest concern aremetropolitan cities like Tehran, Iran. The aim of the present work is to evaluate the potentialof gene expression programming (GEP) in comparison with multiple linear regression (MLR)methods to provide reliable predictions of PM10 and O3 hourly concentrations, a task that isknown to present certain difficulties. The modeling study involves five measurementlocations within the Greater Tehran Area named Azadi, Golhak, Hesar, Pardisan and Vilastations which face significant air pollution problems. The PM10 and O3 data used cover theyear of 2005. The results of MLR were rather satisfactory, with values of the correlationcoefficient (R) for independent test sets ranged between 0.868 and 0.927 for PM10 andbetween 0.756 and 0.927 for O3 for the five stations and values of the index of agreementbetween 0.877 and 0.927 for PM10 and between 0.751 and 0.939 for O3. The performance ofexamined MLR was superior in comparison with GEP models that were developed in parallel(R for GEP models ranged between 0.846 and 0.923 for PM10 and between 0.773 and 0.940for O3). The results of the study clearly demonstrate the proficiency of the multiple linearregression in comparison with gene expression programming in estimating air pollutantconcentrations.

Keywords:

Air quality modeling , Gene expression programming , Ozone level , Particulate matter , Prediction