粗蛋白、 粗纤维、 粗脂肪是评价牧草品质和饲用价值的重要指标。 针对目前已有的牧草品质检测方法存在费时费力、 容易产生化学废物等问题, 提出了一种利用牧草冠层高光谱数据来实现牧草品质实时、 无损监测的方法。 通过ASD FieldSpec 3地物光谱仪采集了青海湖环湖地区19种天然牧草的冠层光谱反射率, 并采样分析了牧草品质参数——粗蛋白、 粗脂肪和粗纤维的相对含量(%)。 光谱经去噪处理后, 分别选择原始光谱、 一阶导数、 波段比值以及小波系数与牧草品质参数进行相关性分析。 结果表明: 在所有高光谱参量中, 牧草品质参数含量与424, 1 668, 918 nm波段处的光谱一阶反射率以及低尺度(scale=2,4)的Morlet, Coiflets和Gassian小波系数之间的相关性较强。 在此基础上, 运用单变量线性、 指数和多项函数分别建立牧草品质的高光谱遥感估算模型, 分析结果显示, 以Coiflets小波系数(scale=4, wavelength=1 209 nm)为自变量的二次多项式模型、 以1 668 nm波段光谱一阶导数为自变量的二次多项式模型、 以918 nm波段光谱一阶导数为自变量的指数模型分别为估算牧草粗蛋白、 粗脂肪、 粗纤维含量的最佳回归模型, 模型检验均达到了极显著水平(0.762≥R2≥0.646), 说明在冠层尺度利用高光谱技术结合光谱一阶导数或小波分析的方法来估测牧草品质参数是可行的, 它将为牧草品质遥感监测提供依据。

Crude protein (CP), crude fat (CFA) and crude fiber (CFI) are key indicators for evaluation of the quality and feeding value of pasture. Hence, identification of these biological contents is an essential practice for animal husbandry. As current approaches to pasture quality estimation are time-consuming and costly, and even generate hazardous waste, a real-time and non-destructive method is therefore developed in this study using pasture canopy hyperspectral data. A field campaign was carried out in August 2013 around Qinghai Lake in order to obtain field spectral properties of 19 types of natural pasture using the ASD Field Spec 3, a field spectrometer that works in the optical region (350～2 500 nm) of the electromagnetic spectrum. In additional to the spectral data, pasture samples were also collected from the field and examined in laboratory to measure the relative concentration of CP (%), CFA (%) and CFI (%). After spectral denoising and smoothing, the relationship of pasture quality parameters with the reflectance spectrum, the first derivatives of reflectance (FDR), band ratio and the wavelet coefficients (WCs) was analyzed respectively. The concentration of CP, CFA and CFI of pasture was found closely correlated with FDR with wavebands centered at 424, 1 668, and 918 nm as well as with the low-scale (scale=2, 4) Morlet, Coiflets and Gassian WCs. Accordingly, the linear, exponential, and polynomial equations between each pasture variable and FDR or WCs were developed. Validation of the developed equations indicated that the polynomial model with an independent variable of Coiflets WCs (scale=4, wavelength=1 209 nm), the polynomial model with an independent variable of FDR, and the exponential model with an independent variable of FDR were the optimal model for prediction of concentration of CP, CFA and CFI of pasture, respectively. The R2 of the pasture quality estimation models was between 0.646 and 0.762 at the 0.01 significance level. Results suggest that the first derivatives or the wavelet coefficients of hyperspectral reflectance in visible and near-infrared regions can be used for pasture quality estimation, and that it will provide a basis for real-time prediction of pasture quality using remote sensing techniques.