最小二乘法联合光学与雷达遥感数据估算玉米叶面积指数

针对单源数据经验模型估算精度较低等问题,提出采用最小二乘法联合光学和雷达遥感数据构建联合估算模型,以中国科学院河北怀来遥感综合实验站为研究区,以夏季玉米为研究对象,利用Landsat8和Radarsat2影像实现研究区叶面积指数估算:首先分别建立了多光谱数据和雷达数据与实测叶面积指数之间的回归模型,然后利用最小二乘算法联合不同数据间的回归模型构建估算模型,最后利用迭代法估算叶面积指数并通过验证数据对估算结果进行评价分析,同时与单源数据经验模型、多源数据加权平均模型和基于物理模型查找表估算结果进行对比。通过对研究区59个样本点数据分析表明:基于最小二乘算法联合光学与雷达遥感数据能够提高叶面积指数的估算精度(R2=0.5442,RMSE=0.81),优于单源遥感数据拟合经验模型(DVI经验模型:(R2=0.485,RMSE=1.27))、基于权重的光学微波联合模型(R2=0.447,RMSE=1.36)和物理模型查找表法(R2=0.333,RMSE=1.36),并当叶面积指数大于3时,对其由于信息饱和或误差引起的低估或高估现象具有一定的抑制作用。

Estimation of Corn LAI by Synergy Multi-spectral and SAR Remote Sensing Data based on Least Squares Method

As a result of different kinds of RS data containing varied information about green plants,to avoid the problem of low precision,the joint inversion model that constructed by the least squares method combined optical and radar remote sensing data such as Landsat8/OLI and Radarsat2 data was put forward to estimate LAI.And this research area was based on Remote Sensing Synthetic Experiment Station of Chinese Academy of Sciences in Huailai,Hebei Province and the research objects were maize.First of all,conventional method was used for remote sensing image preprocessing and then measured LAI was considered to build the empirical expressions between the extracted information from multi\|spectral data and radar data.Secondly,the least squares method that combined with Regression Model from different data was used to build the joint inversion model.At last,the joint inversion model was used to estimate the LAI based on iteration method and assess the result by the verification data.For comparison,the empirical model using vegetation index or backscattering coefficient as predicted variable,the weighted averaging model using multi\|source data and the Look\|up table method from physical model were also considered for LAI estimation.The result shows the better fit result was found between the predicted LAI from Partial Least Squares method and measured LAI (R²=0.5442,RMSE=0.81).Moreover,partial least squares method also couldimprove the overestimated and underestimated phenomenon from empirical method or weight fusion model due to the data quality,system error or saturation of remote sensing data.
