頁籤選單縮合
題名 | 應用支持向量機探討山坡地土壤沖蝕程度之研究=A Study of Soil Erosion Degree on Slopeland by Using Support Vector Machine |
---|---|
作者 | 林俐玲; 王兆文; 沈哲緯; 陳品岡; 翁志成; Lin, Li-ling; Wang, Chao-wen; Shen, Che-wei; Chen, Pin-gang; Weng, Chih-cheng; |
期刊 | 水土保持學報 |
出版日期 | 20110300 |
卷期 | 43:1 2011.03[民100.03] |
頁次 | 頁49-68 |
分類號 | 441.2 |
語文 | chi |
關鍵詞 | 山坡地土地可利用限度; 土壤沖蝕程度; 支持向量機; Slopeland utilization limitation; Soil erosion degree; Support vector machine; |
中文摘要 | 早期台灣山坡地土地利用由於缺乏通盤規劃,且農地資源隨經濟結構改變之影響,爰於 1976年公布山坡地保育利用條例以規範山坡地範疇,為管理山坡地農地資源狀況,乃訂定「山 坡地土地可利用限度分類標準」,將山坡地依其平均坡度、土壤有效深度、土壤沖蝕程度及母 岩性質等因子,區分宜農牧地、宜林地及加強保育地。並依四項因子條件全面查定山坡地土地 利用分類使用,其中土壤沖蝕程度因子之人為判定不易,為減少人為判定之誤判,故本研究以 南投縣竹山鎮大坑段、大鞍段、筍子林、鯉魚尾及雲林縣古坑鄉草嶺段等5個地段之地籍單元 共計7,622筆,亂數選取輕微沖蝕、中等沖蝕及嚴重沖蝕各850筆土地為訓練資料,餘為測試資 料之使用。 本研究方法採用支持向量機,選取坡度、地形曲率、降雨沖蝕指數、土壤沖蝕指數、土地 覆蓋與管理指數等因子進行山坡地土壤沖蝕程度分析,透過分類矩陣得知訓練模式之總體準確 度77.18%,而驗證成果以嚴重沖蝕驗證正確性最高,達九成以上,輕微沖蝕次之,亦達到七成, 最後是中等沖蝕,準確率有六成以上,故本研究建立之土壤沖蝕程度分類模式,應可輔助查定 人員現場判定參考,提升土壤沖蝕程度等級判定之效率。 |
英文摘要 | Due to the lack of a comprehensive planning on the utilization of slopeland in Taiwan in the early years, as well as the effect of economic structure changes on farmland resources, “Slopeland Conservation and Utilization Act” was promulgated in 1976 to regulate slopeland use. For the management of farmland resources on slopeland, “The classification standard of slopeland utilization limitation” was established to classify lands as suitable for agriculture, animal husbandry, forestry purposes or as land subject to strengthened conservation. The classification was based on four factors which are average slope, soil effective depth, soil erosion degree, and parent rock. Comprehensive classification of slopeland utilization was determined based on these four factors. In view of difficulties in manual verification of soil erosion degree, the study employed a total of 7,622 cadastre entries in five cadastral sections: Da-Keng Section, Da-An Section, Sun-Zih-Lin Section and Li-Yu-Wei Section in Jhushan Township of Nantou County; and Tsao-Ling Section in Gukeng Township of Yunlin County. For training data, 850 pieces of land were randomly selected from slight erosion, medium erosion and severe erosion, respectively, while the rest of the data were used as test data. The study analyzed soil erosion degree of the slopeland by using Support Vector Machine and selecting factors including slope, terrain curvature, erosivity index, erodibility index, land cover and management index. Via classification array, the general accuracy of the training results is 77.18%. The highest ranking of accuracy in the testing results lies in the severe erosion group, which is above 90%; followed by the slight erosion group, which achieves around 70%; and the last is the medium erosion group, which has an accuracy of at least 60%. Therefore the predictive model of soil erosion degree established by the study can be used as reference for future verifications. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。