水资源与水工程学报

陈鲁皖，韩玲，秦小宝，王文娟.基于区域特征相似度的微波土壤水分反演结果可信度评价水资源与水工程学报[J].,2018,29(1):242-248

基于区域特征相似度的微波土壤水分反演结果可信度评价

Reliability evaluation of SAR-retrieved soil moisture based on regional feature similarity

DOI：10.11705/j.issn.1672-643X.2018.01.42

英文关键词:SAR-retrieved soil moisture reliability principal component analysis regional feature similarity watershed algorithm

基金项目:国家重大高分专项军事测绘专业处理与服务系统地理空间信息融合分系统（GFZX04040202-07）；中央高校基本科研业务费专业资金项目（310826175031）

作者	单位
陈鲁皖，韩玲，秦小宝，王文娟	(长安大学地质工程与测绘学院，陕西西安 710064)

摘要点击次数: 1713

全文下载次数: 865

中文摘要:

微波遥感反演土壤水分的精度评价，一般通过选取有限数量采样点的地表实测土壤水分数据与反演值进行比较，实际上只能反映局部采样区的反演精度。本文提出了微波反演土壤水分中一种可以评价整个研究区反演土壤水分可信程度的方法。首先选择采样区的实测土壤水分数据和地表粗糙度数据，通过多元回归统计拟合得到土壤水分反演经验方程，对整个研究区进行土壤水分反演；然后从TM和SAR数据中通过反演和提取，选择影响土壤水分的10个因子（土壤湿度、地表温度、NDVI、土壤质地指数、地形指数、雷达入射角以及Landsat TM的b₃、b₄、b₅、b₇共4个波段），采用主成分分析法（PCA）筛选提取主成分，将前3个主成分合成为RGB影像；再使用分水岭算法分割包含前3个主成分的RGB影像，得到一幅分割区域图；最后计算各分割区域的6维特征向量（土壤湿度、地表温度、NDVI、土壤质地指数、地形指数、雷达入射角）与反演时选择采样区的特征向量间的马氏距离，得到区域特征相似度数据集，基于该数据集计算反演结果可信度。以此为基础，利用2008年甘肃黑河地区的ENVISAT ASAR双极化数据（VV、VH）和实测土壤水分数据进行土壤水分反演并评价反演结果的可信度，同时使用反演区域中多个采样区土壤水分实测数据和确定系数（R²）评价反演精度，对比可信度和R²，表明提出的反演可信度可以有效反映土壤水分反演精度。

英文摘要:

The accuracy evaluation of SAR-retrieved soil moisture is generally carried out by comparing a limited number of measured soil moisture data with inversion values. In fact, the evaluation results could only reflect inversion accuracy of finite sampling area. This paper presented a method for evaluating reliability of soil moisture retrieved from the whole study area. First, measured soil moisture data and surface roughness data were selected, and then empirical equation of soil moisture retrieval was obtained by multiple regression statistical fitting. Soil moisture inversion was carried out in the whole study area. Then, principal component analysis (PCA) was used to extract principal component from 10 factors (soil moisture, surface temperature, NDVI, soil texture index, topography index, radar incidence angle and b₃, b₄, b₅ and b₇ in Landsat TM) affecting soil moisture by choosing from TM and SAR data, and the first three principal components were combined into RGB images. The RGB image containing the first three principal component was segmented using watershed algorithm. A graph of a split target area could be get. At last a data set about regional feature similarity was obtained by calculating mahalanobis distance between 6 dimensional feature vectors of each segmentation region and feature vectors of each quadrat area. These components of region feature vector included soil moisture, land surface temperature, NDVI, soil texture index, surface roughness and radar incidence angle. The reliability of the inversion results was computed based on the data set. The results were validated by using Envisat ASAR (advanced synthetic aperture radar) C-band dual polarization (VV, HH) data and the observed values of ground truth measurements synchronizing with Envisat ASAR. Soil moisture was retrieved, and reliability of inversion results was evaluated. By using measured data of soil moisture in several sampling areas, the accuracy of inversion results was evaluated. By comparing reliability and R², the results show that reliability of inversion results can effectively reflect the accuracy of soil moisture retrieval.

查看全文查看/发表评论下载PDF阅读器

关闭