以時間序列法推估大氣中臭氧濃度可行性之研究

Abstract

預測空氣污染物濃度一直是各國學者所追求的目標，所使用的方法亦各有 不同，其大致上可歸為兩類，一種為以物理或化學反應現象為學理基礎之 擴散模式，另一種為根據資料本身之統計特性為主之統計模式。本研究即 以統計模式中之時間序列法探討當變數本身或變數與數間呈高度之相關情 形之預測模式，包括ARIMA、 ARIMA-Regression及 ARIMA - Transfer Function 三種模式，並以板橋、永和測站之臭氧為研究對象，比較上述 三種模式對臭氧四季逐時濃度之預測能力。結果發現ARIMA-Regression對 任何季節之臭氧濃度皆有不錯之預測能力，而加入解釋變數之ARIMA- Transfer Function雖然可使預估結果較好，但卻礙於模式本身之限制， 而可能出現尋求不出適當模式之情形。然而即使是傳統之ARIMA模式 ，其 一天24小時之逐時預測值與實測值之相關係數亦高達0.8上，為10:00 至17:00之高濃度臭氧預測較差，故以時間序列模式推估大氣中之臭氧濃 度，應不失為一可行方案。 There are two approaches to predict the concentrations of atm- ospheric air pollutants. One is the dispersion modeling based on the physical or chemical phenomena of air pollutants, the other is statistical analysis which simply incorporates a group of monitoring data to find out forecasting rules. The purpose of this study is to analyze and forecast ambient hourly ozone concentrations by statistical time series analysis. Three time series models , ARIMA, ARIMA-Regression and ARIMA-Transfer Function,were evaluated and compared in this study. The results indicated that ARIMA-Transfer Function model provides the est forecasting ability. However, sometimes it is not possible to find resonable forecasting rules by ARIMA - Transfer Function due to the model's limitation. The ARIMA- Regression model, on the other hand, provides a good prediction ability in all cases. Results from the simplest model, ARIMA, is not so good as the other two models. However, the correlation coefficients between measured and predicted data made by ARIMA were above 0.8 in most of the cases. Therefore , time series analysis should be a good approach for analyzing and fore- casting the ozone levels in the atmosphere.