基于地面光谱测量和主动微波遥感的反演土壤水分研究  

Inversion of soil moisture by surface spectral measurement combined with active microwave remote sensing

作　　者：尹承深 刘全明[1] 王春娟 王福强 YIN Cheng-shen;LIU Quan-ming;WANG Chun-juan;WANG Fu-qiang(College of Water Conservancy and Civil Engineering,Inner Mongolia Agricultural University,Hohhot 010018,China)

机构地区：[1]内蒙古农业大学水利与土木建筑工程学院,呼和浩特010018

出　　处：《西南农业学报》2022年第11期2595-2602,共8页Southwest China Journal of Agricultural Sciences

基　　金：国家自然科学基金项目(52069020);内蒙古农业大学“双一流”学科创新团队建设人才培育项目(NDSC2018-10)。

摘　　要：【目的】融合多源遥感数据的神经网络构建土壤水分反演模型,估测土壤含水率,更加快速地精准监测土壤水分分布,为灌区土壤墒情监测与节水农业提供基础信息指导。【方法】以河套灌区解放闸灌域为研究区,利用地面实测地表粗糙度与光谱数据,联合C波段微波雷达SAR四极化后向散射数据,通过多元逐步回归(MSR)、主成分回归(PCR)和偏最小二乘回归(PLSR)选择水分特征波段,并构建经验模型和BP人工神经网络模型评价土壤墒情。【结果】将光谱反射率及其一阶与二阶导数、对数形式4种光谱数据与土壤水分做相关性分析,经数学变换的光谱反射率与土壤含水率存在较强的相关性。发现光谱的一、二阶导数相关性更好,尤其是二阶导数变换的4个特征波段450~454、1412~1416、1421~1425、2208~2212 nm相关性最高,分别为0.50、-0.49、0.55和-0.59;二阶导数变换的模型拟合度远高于一阶导数变换的,MSR模拟土壤含水量的效果较好,其判定系数R^(2)和均方根误差RMSE分别为0.482和0.027,PCR次之,PLSR最差。在对比前述二阶倒数变换的PCR、MSR和PLSR 3个土壤水分模型基础上,确定联合光谱特征波段中心反射率二阶导数与地表粗糙度、雷达后向散射特性的BP人工神经网络(BP ANN)模型是最佳预测模型,模型R^(2)为0.792,预测精度及稳定性都优于前述经验回归模型。【结论】采用地面光谱联合主动微波遥感能够精确快速地预测河套灌区解放闸灌域盐渍化土壤水分,为微波遥感监测西北寒旱地区土壤墒情提供重要基础数据。【Objective】The study aimed to construct a soil moisture inversion model by fusing neural network with multi-source remote sensing data to estimate soil moisture content, to monitor soil moisture distribution more rapidly and accurately, and to provide basic information guidance for soil moisture monitoring and water-saving agriculture in the irrigation area.【Method】Jiefangzha zone of Hetao Irrigation District, Inner Mongolia, was selected as the study area, based on the measured ground spectra, surface roughness and four polarization scattering data of C-band microwave synthetic aperture radar(radar SAR),respectively by using the method of principal component regress(PCR),multiple stepwise regress(MSR) and partial least square regress(PLSR) to select feature band, and soil moisture distribution modeling was built and evaluated.【Result】Correlation analysis was done between the spectral reflectance and its first-and second-order derivatives, four spectral data in logarithmic form and soil moisture, and there was a strong correlation between the mathematically transformed spectral reflectance and soil moisture content. The correlation between the first and second order derivatives of the spectra was found to be better, especially for the four characteristic bands of 450-454, 1412-1416, 1421-1425 and 2208-2212 nm with the highest correlation of 0.50,-0.49, 0.55 and-0.59, respectively;The model fit of the second order derivative transform was much higher than that of the first order derivative transform.The MSR simulated soil water content better, with the coefficient of determination R^(2) and root mean square error RMSE of 0.482 and 0.027, respectively, while the PCR was the next best and the PLSR was the worst. Based on comparing the three soil moisture models of PCR, MSR and PLSR with the aforementioned second-order inverse transform, it was determined that the BP artificial neural network(BP ANN) model combining the second-order derivatives of the center reflectance of the spectral characteristic band

关 键 词：土壤水分 多源遥感 协同反演 神经网络 

分 类 号：S152.7[农业科学—土壤学]