利用高光谱植被指数反演植被水分含量时, 快速、 准确的找到实测光谱数据与植被水分相关性最高的植被指数是研究的重点。 在农田尺度上, 以春小麦野外光谱数据与叶片含水量的定量关系为基础, 通过灰色关联度分析, 筛选出与叶片含水量灰色关联度较高的5种典型的水分植被指数, 并建立了估算春小麦叶片含水量(LWC)的偏最小二乘回归(PLSR)模型和BP神经网络(back propagation artificial neural networks, BP ANN)模型。 结果表明: (1)光谱一阶导数可以有效去除噪声影响并突出光谱特征信息, 尤其是在750～830, 1 000～1 060和2 056～2 155 nm等区间明显提高了与LWC的相关性。 (2)灰色关联法能够较好的表征各水分植被指数与叶片含水量间的关联性, 其中基于原始光谱建立的前5个水分植被指数都是两波段比值植被指数, 基于光谱一阶导数建立的水分植被指数基本上都是两波段归一化差值植被指数。 (3)所建立的两种模型中, 基于光谱一阶导数建立的PLSR和BP神经网络模型R2分别为0.80和0.81, 稳定性基本相同且都较好; 两种模型RMSE都是0.55, RPD分别为2.01和1.41, 说明PLSR模型的预测精度比BP神经网络模型高。 从模型的验证效果来看, PLSR模型在估算春小麦叶片含水量方面有一定的优势, 为高光谱定量反演春小麦叶片含水量提供一定的参考。

When vegetation index is used to retrieve water content, it is important to find the vegetation index which has the highest correlation between measured spectral data and vegetation water content. In this paper, Fukang science experimental base of Xinjiang University was selected as the study area. Based on spring wheat field spectral data and leaf water content data, this paper selected 5 typical water vegetation indices that have higher grey correlation degree with leaf water content through grey correlation analysis. And used 2 kinds of methods including the partial least squares regression (PLSR) and back propagation artificial neural network (BP ANN) to establish the quantitative inversion models of soil water content. The results showed that: (1) The first derivative of spectrum can effectively remove the noise influence and highlight the spectral characteristic information, especially in the range of 750～830, 1 000～1 060, 2 056～2 155 nm, which significantly improves the correlation with LWC. (2) The grey correlation method can better characterize the relationship between water vegetation indices and leaf water content, and the first 5 water vegetation indices based on original spectrum were two band ratio vegetation index, and the water vegetation indices based on the first derivative spectra were basically two band normalized difference vegetation index. (3) Among the two established models, R2 of PLSR and BP neural network model established by the first derivative reflectance were 0.80 and 0.81 respectively, which showed that the two models have good stability in inversion of leaf water content; the RMSE of the two models were 0.55 and RPD were 2.01 and 1.41 respectively, which indicated that the prediction accuracy of PLSR model was higher than that of BP neural network model. From the validation of the model, the PLSR model has some advantages in estimating leaf water content of spring wheat, which provides a reference for hyperspectral quantitative inversion of it.