透過您的圖書館登入
IP:18.218.209.8
  • 學位論文

以隨機模擬建立之解析尺度協助插值方法中經驗參數之選擇

Stochastically established resolution analysis helps to determine empirical tuning parameters in general interpolation schemes

指導教授 : 喬凌雲

摘要


在地球物理及環境研究問題中,解析度評估對空間分析具有重要的意義,但經常受到學界所忽略。通過解析度的評估與分析能夠正確的解釋與判斷不同的空間分析方法所產生之結果。對於以層析成像為例的地球物理逆推問題而言,解析度分析的限制主要在於計算上無法取得完整的逆矩陣,以與理論斐謝核(Frechet kernel)組成解析矩陣。隨機模擬(stochastic simulation)是解決此一困境,用以估計大型逆推問題中經驗解析長度(empirical resolution length)的潛勢方法。尤其有趣的延伸則是針對傳統上最廣為應用的若干空間分析方法,如克力金法(Kriging)、最小曲率法(minimum curvature)之屬,在其方法的架構上本來就沒有解析的外顯(explicit)資料算核(data kernel),傳統意義的解析度分析因此無法進行,也從未受到應有的重視。我們針對這些普遍使用,但從不討論解析能力的空間分析方法,推廣隨機模擬方法來求得解析長度,並以實際地球物理資料的插值問題來凸顯這種新解析分析方法的應用。值得注意的是在克利金法以及最小曲率法中,存在一些僅憑經驗或者任意決定的控制參數,例如克力金法中的影響範圍(influence range)以及最小曲率法的張力參數(tension parameter)等,本研究嘗試憑藉經驗解析長度來瞭解不同的插值方法以及採用不同的經驗控制參數設定的推估模型在解析尺度中的實際規範為何,而經由解析度的分析亦能幫助我們挑選推估模型或是經驗控制參數的決定。

並列摘要


Resolution analysis has been a crucial appraisal procedure in general estimation problems to help with the correct interpretation. However, complete resolution information is usually inaccessible due to the sizeable matrix inversion involved with the construction of the resolution matrix. Furthermore, there are not explicit forward kernels embedded within formulations for popular interpolation algorithms such as the Kriging and the minimum curvature gridding schemes. Stochastic simulation has been proposed to make the resolution evaluation for sizeable inverse problems tractable. We generalize the method of getting resolution information for the popular interpolation schemes. Furthermore, there are usually certain empirically determined tuning parameters involved in these interpolation schemes, for example, the ideal function and influence range for fitting the semi-variogram in the Kriging method and the tension parameter in the minimum curvature gridding scheme. In this study, we will show that our proposed resolution analysis not only provide the crucial spatial resolution variation, more importantly, it helps to determine those critical tuning parameters that have been determined empirically and arbitrarily.

參考文獻


An, M. (2012), A simple method for determining the spatial resolution of a general inverse problem, Geophys. J. Int., 191, 849–864.
Backus, G. E., and J. F. Gilbert (1968), The resolving power of gross Earth data, Geophys. J. R. Astron. Soc., 16, 169–205.
Briggs, I.C. (1974), Machine contouring using minimum curvature, Geophysics, 39, 39-48.
Cressie, N. A. C. (1990), The origins of kriging, Math. Geol., 22, 239–252.
Cressie, N.A.C. (1993), Statistics for Spatial Data. John Wiley and Sons, Inc., New York.

延伸閱讀