評估模糊集群分析結合克利金法在界定農地土壤管理區的適用性

Abstract

模糊集群分析 (fuzzy clustering analysis, FCA) 是目前最常被應用於界定土壤管理區 (management zone) 的方法，係將土壤性質的觀測資料轉成隸屬度函數 (membership function) 做為分類依據。依據已歸類分組之土壤在空間上的分布，界定不同管理區在空間位置上的邊界。實際上，當土壤的採樣密度無法涵蓋所有空間位置的土壤時，則無法精確的界定出不同管理區的邊界。由於克利金法是目前最常被用來分析處理土壤空間變異與推估空間分佈的技術，故已有文獻提出以模糊集群分析結合克利金法 (kriging) 來解決界定管理區邊界的問題，然而卻多未對其適用性做詳細的探討比較。本研究以台灣地區耕地土壤的管理區界定為例，探討模糊集群分析結合克利金的兩種方法 (一是將土壤性質經模糊集群分析後轉成的隸屬度值作克利金推估，另一是將土壤性質的克利金推估值代入模糊集群分類模式) 界定管理區的適用性。本研究區位於彰化縣埤頭鄉台糖元埔農場，為面積約20公頃的農地。本研究以表土之土壤酸鹼度值 (pH)、電導度 (EC) 與質地做為界定土壤管理區的性質，經多次模糊集群分析的分組模擬與克利金法的空間推估，藉以比較不同性質與不同分組模擬和空間推估步驟的影響。結果顯示，以質地所界定之管理區，其各區內之變異較未分組前小，而以pH和EC所界定之管理區其各區內之變異則無顯著減小。再者，從分組錯判率的評估結果發現，先以土壤性質建立模糊集群分類模式，再將研究區內各土壤性質的克利金推估值依分類模式歸類分組，其錯判區塊數較少；若將各土壤性質之先依分類模式轉換成不同組別的隸屬度值(也就是屬於不同組別的可能性)，再以克利金推估未採樣點之土壤性質屬於各組別的隸屬度值，並依其大小來決定該未採樣點之土壤性質所屬之組別，則歸類分組的錯判區塊數會較多，故以前者所界定的管理區是較為可靠的。

Currently, fuzzy clustering analysis (FCA) is increasingly used to delineate zones for site-specific management. Sampled observations of soil properties can be classified into groups by using FCA given a membership function. The boundaries of management zones are thus determined on the spatial distribution of grouped soils. In practice, the sampling density will not be intensive enough to determine the management zone boundaries precisely. Thus, a scheme of fuzzy clustering analysis combined with kriging technique, which is frequently used in spatial interpolation of soil properties, has been proposed. However, there are few detailed discussions for assessing the feasibility of fuzzy clustering analysis combined with kriging to delineate management zones of agricultural soils in Taiwan. In this study, a comparison of two approaches of fuzzy clustering analysis combined with kriging, kriging membership values of fuzzy classification and putting kriged soil properties into fuzzy classification, was carried out. The study site was 20 ha in area in Changhua County, Taiwan. The observed data of pH, electronic conductivity (EC), and soil texture were used for illustration of the effects of fuzzy classification, kriging interpolation and data of soil properties used on determination of management zones. The results of classification showed that the soil texture-based management zones were with lower within-group variation but were not the pH- and EC-based management zones. The soil texture-based management zones were more robust than those pH- and EC-based. Moreover, compared with one of kriging membership values of fuzzy classification to determine management zones, the other of putting kriged soil properties into fuzzy classification got a lower misclassification rate. The approach of putting kriged soil properties into fuzzy classification was thus recommended in this study.