PERFORMANCE EVALUATION OF AIR POLLUTION MODELS

Abstract

In a rapidly developing country like India, the transport sector is growing rapidly and number of vehicles on Indian roads has increased from 0.3 million in 1951 to 37.2 million in 1997 i.e. a increase of almost 124 times. This has lead to overcrowded roads and a polluted environment. These alarming increases in the pollution in our metropolitan cities cause various health hazards. To predict the vehicular pollutants various models have been developed abroad and the most popular among them are the CALINE models and the Finite Line Source Models. However, the suitability of these models for Indian conditions must be thoroughly investigated before they are applied for prediction of pollutants concentration in India. In this dissertation, an effort has been made to study the various air quality models and to evaluate the performance of CALINE-4 and General Finite Line Source Model for eight locations in Delhi in terms of carbon monoxide concentration. The General Finite Line Source Model was developed for Indian conditions and CALINE-4 was developed for American conditions. Prediction of carbon monoxide concentration has been done for all the eight selected locations of Delhi. The Comparison between model predicted concentration and observed concentration was performed using statistical methods, like regression analysis, significance test and Index of agreement, to evaluate model performance. After doing statistical analysis both models gives satisfactory results. The t-test shows that tcaiculated value is always less than t,ab„iates value for degree of freedom 15 and level of significance 0.05 for all the eight locations for both models which imply that difference between observed and predicted value is insignificant. Regression analysis shows good correlation between observed and predicted values by both models with r2 value ranged from 0.575 to 0.9398 for CALINE-4 and 0.7006 to 0.8751 for GFLSM. The minimum value of Index of agreement for CALINE-4 is 0.51 and 0.61 for GFLSM, which implies that GFLSM predictions are more error free than CALINE-4 predictions. Hence evaluation of the performance of both models has been satisfactory in terms of statistical analysis. The application of CAL[NE-4 and GFLSM for prediction of CO concentrations shows that both models generally under predicts in most cases. This means the predicted values will generally be less than the observed values and therefore the modeled values can be safely adopted for decision-making purpose.