空氣汙染與氣象因子、其他汙染物之關係

Abstract

摘要 近年來大眾對於空氣品質要求越來越高，而PM₂.₅是影響空氣品質的關鍵因子，所以我們想探討的是氣象因子與汙染物對於PM₂.₅濃度有哪些的顯著影響，分析其顯著性與相關性；至於氣象因素方面，檢視氣溫、濕度、風速、風向與氣壓是否跟PM₂.₅有存在顯著的相關性。在污染物方面，檢視PM₂.₅與NO2、SO2與CO等污染物是否有存在顯著的相關性?此外在不同汙染物間的在特定氣象條件下具有交互作用是否有影響PM₂.₅的增加或減少，可能進一步影響PM₂.₅濃度變化。 我們採用SAS EG進行分析，主要運用多元線性回歸來觀測資料。過程包含透過散步圖進行初步觀察，接著進行參數估計檢定，為確保模型的嚴謹性，分析過程中執行了VIF、Durbin-Watson以及殘差分析，並針對偏離常態的資料進行剃除，以優化模型的完整性。 PM10(空氣懸浮粒子)是預測PM₂.₅的關鍵指標，而SO2、NO2及NOX亦展現正向影響，且風速與風向等氣象因子在線性模型中則呈現隨機分布。最終模型排除CO與離群值後，解釋力明顯提升，則證明該變數組合具備良好的環境預測。則未來可應用於環境監測預警甚至是引入其他分析以提升預測精度。 Abstract

In recent years, the public has placed increasingly higher demands on air quality, and PM₂.₅ is a key factor affecting air quality. Therefore, we aim to explore the significant impacts of meteorological factors and pollutants on PM₂.₅ concentrations, analyzing their significance and correlation. Regarding meteorological factors, we examine whether temperature, humidity, wind speed, wind direction, and air pressure have significant correlations with PM₂.₅. As for pollutants, we examine whether PM₂.₅ has significant correlations with NO₂, SO₂, and CO, and whether interactions between different pollutants under specific meteorological conditions affect the increase or decrease of PM₂.₅, potentially further influencing PM₂.₅ concentration changes.

We used SAS EG for analysis, primarily employing multiple linear regression to observe the data. The process included initial observation through scatter plots, followed by parameter estimation and testing. To ensure model rigor, VIF, Durbin-Watson, and residual analyses were performed during the analysis, and data deviating from normal patterns were removed to optimize model integrity.

PM10 (airborne particulate matter) is a key indicator for predicting PM₂.₅, while SO₂, NO₂, and NOx also show a positive impact. Furthermore, meteorological factors such as wind speed and direction exhibit a random distribution in the linear model. After excluding CO and outliers, the explanatory power of the final model significantly improves, demonstrating that this combination of variables possesses good environmental predictive capabilities. Therefore, it can be applied to environmental monitoring and early warning systems, and even incorporated into other analyses to enhance prediction accuracy in the future.