請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/70907
標題: | 應用貝氏最大熵法推估土壤有效深度 Prediction of Soil Depth Using Bayesian Maximum Entropy Approach |
作者: | Shao-hua Chang 張少華 |
指導教授: | 廖國偉 |
關鍵字: | 土壤深度,山坡地土地可利用限度,地理資訊系統,貝氏最大熵法,克利金法,地形濕度指數,K-means分群法, Soil depth,Bayesian maximum entropy (BME),Geographic information system,Kriging method, |
出版年 : | 2018 |
學位: | 碩士 |
摘要: | 管理土地資源時,土壤深度是其中一項重要的參考因子,在難以現地勘查的位置,便需要進行土壤深度推估。然而,在過往的研究中,還無法達到很好的推估效果。因此本研究的目的,便是以土壤深度的空間分布關係與相關地文因子為依據,找出更有效的推估方法,並提高土壤深度推估的準確性。
為了建置土壤深度推估模式,首先考量土壤深度與地文因子之關聯性,整理前人研究文獻後,篩選出較具代表性的地文因子包括坡度、坡向、剖面曲率、平面曲率、地形濕度指數及地質等。接著,使用貝氏最大熵法推估土壤深度,輸入資料包括實測土壤深度資料及利用前述地文因子所建置的土壤深度模式,結合以上兩項資料進行區域內未知點的土壤深度推估。同時亦使用克利金法進行土壤深度推估,比較傳統線性空間推估方法與貝氏最大熵法的推估準確率。 依據山坡地土地可利用限度中對土壤深度之分級規範,分為甚淺層、淺層、深層、甚深層四個等級,本文中使用不同推估方式時,均依照上述四種類別計算準確率。首先,利用土壤深度的實測資料,檢測現今使用土壤深度圖之正確率,結果僅達27.81%,顯示土壤圖有更新的必要性。比較克利金法及貝氏最大熵法推估之準確率,分別為40.40%及82.94%。克利金法雖可考量資料空間分佈之特性,但無法將地文因子對土壤深度之影響納入考量,且克利金法之前提假設為常態與最佳線性不偏推估等,並非完全符合實際資料狀況。相對而言,貝氏最大熵法可同時納入常態、非常態之資料,並結合實測點資料與香 地文因子資料,進行非線性的推估,因此準確度最佳。 Soil depth is one of the most essential information for land resource management; however, the predictions of soil depth in previous studies still have certain differences compared with measured soil depth. The objective of this study is thus to propose an effective approach that is able to improve the accuracy of soil depth estimation, considering its spatial variability and using some spatial soil parameters derived from high-resolution digital elevation data. To establish the prediction model, considering the relation between soil depth and terrain attributes, six attributes are chosen to construct the soil depth model, including slope, aspect, plane curve, profile curve, geology and compound topographic index. Selection of the aforementioned attributes are based on earlier studies. Secondly, Bayesian maximum entropy (BME) approach is utilized to estimate the soil depth, which is a non-linear method based on a rigorous statistic theories and the processing of diverse information sources. The traditional linear geostatistic method, Kriging, is also presented; based on the analyzed results, BME processes a better prediction performance compared to that of the Kriging approach. Classification of soil depth in Taiwan are fairly shallow, shallow, deep, fairly deep. Based on the above classification, the accuracy of Kriging and BME approach through the classification, it is 40.40% and 82.94% respectively. To discuss the difference, Kriging method is based on a least-squares function and remain linear estimators. In contrast, the BME approach is completely non-linear and consider the terrain attributes simultaneously, which makes it a better tool to process soil depth estimation. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/70907 |
DOI: | 10.6342/NTU201802476 |
全文授權: | 有償授權 |
顯示於系所單位: | 生物環境系統工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-107-1.pdf 目前未授權公開取用 | 4.84 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。