本研究提出以非平穩空間模型(non-stationary spatial model)應用於台灣地區之二氧化氮資料,此非平穩空間模型為由數個基底函數和數個平穩過程之線性組合。在模型的設定之下,欲估計之參數個數很多,遂以 Tibshirani(1991)提出的『最小絕對壓縮挑選機制』(Least absolute shrinkage and selection operator, Lasso)進行參數估計,此方法具備同時選模與估計參數的優點。透過使用Efron et al.(2004)所提出『最小角度迴歸法』(lars)套件,以R軟體解出最小絕對壓縮挑選機制之估計值將會非常有效及簡單。本研究將分析結果繪製成空間分佈圖,以觀察台灣地區二氧化氮濃度之分佈情形。研究結果顯示二氧化氮月平均濃度在秋季及冬季較高,其相關係數在空間上呈現非平穩性質。
In this research, a non-stationary spatial model is applied to Taiwan NO2 data. The proposed non-stationary model contains some basis functions and some stationary processes. This model is very flexible to characterize various non-stationary or stationary features. Under the model setting, the number of the parameters needed to be estimated is large. For solving this problem, the Lasso method (Tibshirani, 1996) is used to estimate parameters. Lasso can deal with model selection and parameter estimation simultaneously. The lars package (Efron et al., 2004) in R language is used to solve the Lasso estimate efficiently. The result of the analysis is displayed in plots to observe the NO2 distribution. The monthly mean of the NO2 concentration is higher in autumn and in winter. The feature of correlation reveals the non-stationaraity of this data.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。