摘要
为尝试联合应用光学与微波遥感数据反演小麦覆盖区土壤含水量的可行性,收集了2014年3月28日RADARSAT-2微波数据和2014年3月24日Landsat8光学数据,同时开展了地面同步试验,测量了49个点的地面数据。首先根据地面实测数据优化了光学遥感反演地表小麦含水量模型,然后利用MIMICS模型和AIEM模型模拟研究区后向散射系数生成训练数据集,再以Matlab为平台建立BP神经网络、SVM(Support Vector Machine)、MEA-BP(Mind Evolutionary Algorithm-Back Propagation)神经网络、LS-SVM(least squares support vector machine)方法模型,构建小麦覆盖区地表土壤含水量反演模型,最后利用地面实测数据对反演模型进行了精度验证。结果表明,以LS-SVM方法构建的小麦覆盖区地表土壤含水量反演模型的精度最好,其RSME为0.010,相对误差为6.57%。说明联合应用光学与微波遥感数据,并结合简化MIMICS模型构建小麦覆盖区地表土壤含水量反演模型,其反演精度较高且具有可行性。
Soil moisture is very important for crop growth. To get soil moisture of large area effective- lybecomes significant to agricultural production. This study combined optical and microwave remote sensing data to inverse soil moisture of wheat covered surface. The study collected RADARSAT-2 mi- crowave data on March 24,2014,and Landsat8 optical data on March 9,8,2014. At the same time,the synchronous experiment was carried out on the ground,and data of 49 locations was measured. First- ly, wheat moisture content retrieval model of optical remote sensing was optimized according to the measured data. Secondly,the training data set was generated by combining MIMICS and AIEM model to simulate the study area's backscattering coefficient. And then the BP neural network, SVM (Support Vector Machine), MEA-BP (Mind Evolutionary Algorithm-Back Propagation) neural network and LS- SVM (Least Squares Support Vector Machine) were constructed based on Matlab. The soil moisture inversion model of wheat covered area was cons tested by using the ground experimental data. T od of constructing soil moisture inversion mode tructed. Finally the accuracy of retrieval models were he results showed that the accuracy of LS-SVM meth- 1 of wheat covered area is the best, and the RSME is 0. 010,with the relative error of 6.57%. The study used optical and microwave remote sensing data,combining with the simplified MIMICS model to construct the soil moisture inversion models of wheat covered surface, which has high precision and is feasible.