探討以類神經網路取代克利金法中半變異元模式的可行性—以土壤重金屬資料來驗證

The Feasibility of Using Artificial Neural Network as an Alternative to Semivariogram Model in Kriging Method – Verified with Soil Heavy Metal Data

何信霖(Shin-Li HO)

指導教授 ： 張迪惠

朝陽科技大學/理工學院/環境工程與管理系/碩士(2009年)

摘要

空間內插法已成為研究地表科學的必要工具之一。而空間內插法中的克利金法由於具有無偏差預測與誤差變異數最小化兩大優點，因而最被廣泛應用。然而克利金法中的半變異元模式具有參數性質，使用者必需嘗試不同的半變異元模式並加以驗證，才能得到理想的預測結果，這過程造成使用者的不便。另一方面，由於類神經網路具有非參數性的特色，即無需針對不同的樣本進行模式的選擇與參數的設定，本研究探討以類神經網路取代克利金法中半變異元模式的可行性。以彰化市和美鎮鄰近的八種土壤重金屬砷(As)、鎘(Cd)、鉻(Cr)、銅(Cu)、汞(Hg)、鎳(Ni)、鉛(Pb)、鋅(Zn)為例，比較本方法與反距離權重法、以及克利金法中常見的半變異元模式的預測。研究結果顯示，反距離權重法與克利金法在不同的土壤重金屬的預測誤差上互有優劣，而在克利金法中最佳的預測方法並不侷限於某特定的半變異元模式。然而本方法的預測誤差又比克利金法與法反距離權重法低，如此證實類神經網路取代克利金法中半變異元模式的潛力。

關鍵字

克利金法 ； 類神經網路 ； 土壤重金屬 ； 反距離權重法

並列摘要

Spatial interpolation has become one of the requisite tools in Earth''s surface science. Due to the feature of unbias and minimized variance of predicted error, Kriging method was widely used among various interpolation methods. However, because the semivariogram model in Kriging methods is a parametric model, user need to try and error to find the best semivariogram model to yield sound results. Such process brings inconvenience for Kriging user. On the other hand, artificial neural network is a non-parametric model, that is independent to the structure of data and no need to tune parameters and select models. In this research, we explore the feasibility of using artificial neural network as an alternative to semivariogram model in kriging method. We use the data of soil heavy metal collected around Changhua and Hemei to verify and compare the results from proposed method and Kriging method using various semivariogram model as well as the results from IDW (Inversed distance weighted method). The soil heavy metal includes As, Cd, Cr, Cu, Hg, Ni, Pb and Zn. The result shows that the performance of Kriging method and IDW depends on cases. Sometimes Kriging is better but in other times IDW wins. Also, it is difficult to conclude that which semivariogram model is the best model based on this data set. Nevertheless, the proposed method yields better result comparing to IDW and Kriging based on predicted error, which shows the potential of using artificial neural network as an alternative to semivariogram model in kriging method.。

並列關鍵字

Inversed Distance Weight ； Kriging Method ； Soil Heavy Metal ； Artificial Neural Network

參考文獻

Burrough, P.A., 1991. Principles of Geographical Information Systems for Land Resources Assessment. Monographs on Soil and Resources Survey, 12. Oxford Press, Oxford, 194 pp.

Dalthorp, D., J. Nyrop, and M. Villani. 1999. Estimation of local mean population densities of Japanese beetle grubs (Scarabaeidae: Coleoptera). Environ. Entomol. 28:255–265.

Hosseini, E., J. Gallichand, and D. Marcotte. 1994. Theoretical and experimental performance of spatial interpolation methods for soil salinity analysis. Trans. ASAE 37:1799–1807.

Kravchenko, A., and D.G. Bullock. 1999. A comparative study of interpolation methods for mapping soil properties. Agron. J. 91:393–400.

Lari-Najafi Hossein, Mohammed Nasiruddin, and Tariq Samad, “Effect of Initial Weights on Back-Propagation and Its Variations.”, Systems, Man and Cybernetics, IEEE International Conference on , 1989 Vol. 1, P218 -219.