Forecasting the Concentration of Atmospheric Pollutants: Skill Assessment of Autoregressive and Radial Basis Function Network Models

Sutapa Chaudhuri; Debanjana Das; Sayantika Goswami; Anirban Middey
The purpose of the present study is to develop a model to forecast the concentrations of some important atmospheric pollutants over Kolkata (22° 32′N; 88° 20′E), India during the period from 1st April 2009 to 30th November 2010 with considerable accuracy and adequate lead time. The pollutants considered in this study are respiratory suspended particulate matter (RSPM), nitrogen oxide (NOx), and sulphur oxide (SOx). The auto regressive (AR) models with different orders and radial basis function network (RBFN) model are developed to attain the objective. The skill of both the models is compared. The results of the study reveal that the 3rd order Auto-Regressive Model, AR (3) represents the best statistical model for the prediction of concentrations of all the three different pollutants over Kolkata. The study thus, depicts that the pollutants can be predicted with considerable accuracy and 3 days or 72 hours lead time using AR (3) model. The skill of the AR (3) model is compared with RBFN model. The result further reveals that the percentage error in forecast with 72 hours lead time is much less with RBFN model than AR model.
Concentration of pollutants; prediction; AR model; RBFN model
Download | Back to Issue| Archive