基于高光谱遥感数据的水稻叶温反演

Rice leaf temperature inversion based on hyperspectral remote sensing data

【目的】研究水稻叶温与冠层反射光谱间的关系,为水稻叶温的模拟与监测提供理论依据。【方法】利用FieldSpec Pro FR光谱仪和Raynger ST红外温度探测仪测量水稻抽穗期冠层的反射光谱和叶片温度,分析原始反射光谱、一阶微分光谱、归一化植被指数(NDVI)、比值植被指数(DVI)、再归一化差值植被指数(RDVI)和转换型土壤调整指数(TSAVI)与叶温的关系。【结果】叶温的变化直接影响水稻冠层光谱的反射率,影响水稻红边特征。一阶微分光谱与叶温存在极显著相关性(P<0.01,下同),990 nm处相关系数(0.889)最高,885 nm处相关系数(-0.893)最低。选取叶温敏感波段光谱组合计算植被指数,发现RDVI和TSAVI与叶温的关系呈极显著相关,相关系数分别为0.724和0.733。由RDVI和TSAVI建立经验模型,结果显示由TSAVI建立的叶温估算模型效果更好,其验证样本的决定系数为0.610,相对误差为1.97%,均方根误差为2.546。【建议】综合考虑多种预处理方法,最大程度还原光谱信息;优化特征波长的提取,提高建立模型的精度;基于高光谱技术,实现冠层叶温的无损监测。

【Objective】The canopy reflectance spectra of rice at different leaf temperatures were measured to study the relationship between leaf temperature and canopy reflectance spectra,which provided a theoretical basis for the simulation and monitoring of rice leaf temperature.【Method】The reflectance spectra and leaf temperature of canopy in rice heading stage were measured by FieldSpec Pro FR spectrometer and Raynger ST infrared temperature detector.Original reflection spectrum,first-order differential spectrum,normalized vegetation index(NDVI),difference vegetation index(DVI),renormalized difference vegetation index(RDVI),and converted soil adjustment index(TSAVI)and leaf temperature relationship were analyzed.【Result】The change of leaf temperature directly affected the reflectance of rice canopy spectrum and affected the red edge characteristics of rice.There was highly significant correlation between the first-order differential spectroscopy and leaf temperature(P<0.01,the same below).The correlation coefficient at 990 nm was the highest(0.889),and the correlation coefficient at 885 nm was the lowest(-0.893).Vegetation index were calculated by spectral combination of leaf temperature sensitive bands.The relationship between RDVI and TSAVI and leaf temperature was highly significantly correlated,and the correlation coefficients were 0.724 and 0.733,respectively.The empirical model was established by RDVI and TSAVI.The results showed that the model established by TSAVI had better effects.Its determinant coefficient for sample detection was 0.610,relative error was 1.97%and root mean square error was2.546.【Suggestion】Comprehensive consideration of multiple pre-processing methods to maximize spectral information;optimize the extraction of characteristic wavelengths and improve the accuracy of model building;and the non-destructive monitoring of the canopy leaf temperature should be realized based on the hyperspectral technique.