Air quality in Hong Kong has deteriorated significantly over the past few years. The situation is much worse in the western part of Hong Kong, such as Tung Chung. The number of days air pollution index (API) over 100 at Tung Chung is increasing (11 days in year 2001, 16 days in year 2002, 21 days in year 2003 and 42 days in year 2004). The high API values recorded at Tung Chung are mostly due to high ozone concentration, sometimes with high sulfur dioxide concentration. On 14^th September 2004, an API of 201 was recorded at Tung Chung because of high ozone concentration. Therefore, it is desirable to predict air pollutant concentration more accurately. Generally, people have mostly used linear models to predict air pollutant concentration. However, these models are not accurate to predic...[ Read more ]

Air quality in Hong Kong has deteriorated significantly over the past few years. The situation is much worse in the western part of Hong Kong, such as Tung Chung. The number of days air pollution index (API) over 100 at Tung Chung is increasing (11 days in year 2001, 16 days in year 2002, 21 days in year 2003 and 42 days in year 2004). The high API values recorded at Tung Chung are mostly due to high ozone concentration, sometimes with high sulfur dioxide concentration. On 14^th September 2004, an API of 201 was recorded at Tung Chung because of high ozone concentration. Therefore, it is desirable to predict air pollutant concentration more accurately. Generally, people have mostly used linear models to predict air pollutant concentration. However, these models are not accurate to predict the level of secondary pollutants like ozone because ozone formation is complex and nonlinear. As polluted air poses a threat to public health, especially for sensitive subgroups in the community, therefore, the objective of the study is to predict air pollutant concentration more accurately and issue warning before air pollution episodes occur. Other statistical models, regression tree model and neural network model, are used to predict sulfur dioxide concentration and ozone concentration in the study. Using root average square error criterion, it is shown that neural network model is the most accurate to predict ozone concentration.