基于Landsat的石漠化信息提取方法研究——以辰溪县为例

Research on Extraction Method of Rocky Desertification Information Based on Landsat--Take Chenxi County as an Example

【目的】对湖南省怀化市辰溪县的石漠化提取方式进行研究分析,为南方丘陵的生态修复提供可靠依据。【方法】基于Landsat-9 OLI影像,结合基岩裸露率和植被覆盖度,构建石漠化分级指标,并利用专家经验决策树和基于向量机的面向对象模型进行石漠化提取,将石漠化程度分为重度、中度、轻度、潜在、无石漠化5个等级。通过真彩色遥感影像、谷歌地球历史影像和野外实地核查进行精度对比。【结果】专家经验的决策树法验证总精度(89%)高于基于向量机的面向对象法的验证总精度(64%),Kappa系数分别为0.852052和0.530209。研究区石漠化总体呈轻度,主要分布在东北部。【结论】建议在南方丘陵地区广泛应用专家经验的决策树提取方法,以提高石漠化监测的准确性和实用性。地势陡峭和坡耕地等特征区域应重点关注。

[Purposes]The author aims to study and analyze the extraction method of rocky desertification in Chenxi County,Huaihua City,Hunan Province,in order to provide a reliable basis for ecological resto⁃ration of southern hills.[Methods]Based on Landsat-9 OLI image,combined with the exposure rate of bedrock and vegetation coverage,the rocky desertification classification index was constructed,and the expert experience decision tree and object-oriented model based on vector machine were used to extract rocky desertification,and the degree of rocky desertification was divided into five levels:severe,moder⁃ate,mild,potential and no rocky desertification.The accuracy was compared by true color remote sens⁃ing image,Google Earth historical image and field check.[Findings]The total validation accuracy of ex⁃pert experience decision tree method(89%)was higher than that of vector machine-based objectoriented method(64%),with Kappa coefficients of 0.852052 and 0.530209,respectively.The rocky de⁃sertification was generally mild and mainly distributed in the northeastern part of the study area.[Con⁃clusions]The study suggests that the decision tree extraction method based on expert experience should be widely applied in the hilly areas of South China to improve the accuracy and practicability of rocky de⁃sertification monitoring, and the characteristic areas such as steep terrain and sloping farmland should bepaid more attention.