基于多时相SAR的IEM和Oh模型地表土壤含水率反演与融合  

Soil Moisture Inversion and Fusion Based on IEM and Oh Model Using Multi-temporal Synthetic Aperture Radar Images

作　　者：黄对[1] 刘九夫[1] 张建云[1] 王文 崔巍 HUANG Dui;LIU Jiu-fu;ZHANG Jian-yun;WANG Wen;CUI Wei(State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering,Nanjing Hydraulic Research Institute,Nanjing 210029;State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering,Hohai University,Nanjing 210098,China)

机构地区：[1]南京水利科学研究院水文水资源与水利工程科学国家重点实验室,江苏南京210029 [2]河海大学水文水资源与水利工程科学国家重点实验室,江苏南京210098

出　　处：《地理与地理信息科学》2021年第6期24-31,共8页Geography and Geo-Information Science

基　　金：第二次青藏高原综合科学考察研究项目(2019QZKK0203);国家自然科学基金项目(52109026);水利部公益性行业科研专项经费资助项目(Ss520026JG)。

摘　　要：基于水云模型分析植被后向散射系数的影响,采用考虑组合粗糙度定标的积分方程模型(IEM)与Oh模型,在不利用实测土壤含水率的前提下,建立基于多时相多极化SAR影像的IEM和Oh模型地表土壤含水率反演方法,以2004年5期Walnut Gulch流域ENVISAT-ASAR影像为例,采用加权平均、算术平均两种融合方法对两种模型的3期反演结果进行融合。结果表明:1)流域植被覆盖度低,各地类在植被去除前后的土壤含水率差值多期均值多在0.002 cm^(3)/cm^(3)以下,植被对反演结果的作用不显著;2)基于IEM、Oh模型反演的0720期、0805期、0824期土壤含水率的时空分布较一致,基于IEM模型反演的0714期与0818期土壤含水率较低,基于Oh模型的反演方法受影像极化方式制约,0714期、0818期土壤含水率的空间连续性差;3)基于IEM模型反演结果的均方根误差(RMSE)与偏差(Bias)低于Oh模型,其中0805期土壤含水率存在低估,而Oh模型反演结果在各期存在不同程度高估;4)考虑数据权重的加权平均方法优于算术平均方法,两种方法融合后的RMSE值降低了0.003~0.065 cm3/cm3,0720期、0805期两种方法的融合结果均改进了IEM模型反演结果偏低、Oh模型反演结果偏高的不足。该研究可为基于多时相SAR的多模型/方法土壤含水率反演以及多时相高精度土壤含水率获取提供参考。Based on five-period Synthetic Aperture Radar(SAR)images of the Walnut Gulch watershed,the performances of integral equation model(IEM)and Oh model for estimating soil moisture(m_(v))were compared,then the weighted average method and arithmetic average method were employed to obtain the fused m_(v) data.The water-cloud model was applied to remove the influence of vegetation firstly.Then the relationship between surface roughness Z_(s)(h_(RMS)^(2)/l_(opt)),mv,and the better-simulated back-scattering coefficient was established for IEM,and mv empirical inversion methods were derived for VV polarization image at two incidence angles.For Oh model,since there were two periods of images without VH polarization data,the VH polarization images of two adjacent dates were separately normalized,then different equations of the Oh model were solved simultaneously to obtain mv.Results showed that the removal of vegetation or not has little effect on the estimation results due to the low vegetation coverage in the Walnut Gulch watershed.The temporal and spatial distributions of mv derived from IEM and Oh models for three periods(July 20,August 5 and August 24)were generally consistent.The IEM model underestimated mv for periods of July 14 and August 18,while Oh model exhibited poor spatial continuity of mv in the dates of July 14 and August 18 due to the restrictions of image polarization type.Comparison between the estimated mv and observations of 19 stations in the Walnut Gulch watershed showed that the RMSE and Bias of IEM were lower than that of the Oh model.For IEM,mv of August 5 was underestimated,while mv of all three dates was overestimated to some degrees for Oh model.The RMSE after fusion was reduced by 0.003 to 0.065 cm3/cm3,suggesting that the weighted average method considering the error of the actual site data performed better than the arithmetic average method.Fusion results of the two methods for July 20 and August 5 both improved the inversion effect of a single model.This study can provide a reference for soil

关 键 词：土壤含水率 积分方程模型 Oh模型 合成孔径雷达 融合 

分 类 号：S152.7[农业科学—土壤学] TP722.6[农业科学—农业基础科学]