基于高光谱技术剑湖湿地茭草磷含量估算模型研究

以云南剑湖湿地典型植被优势种茭草为研究对象，利用ASD Filed Spec 3光谱仪采集茭草反射光谱，并测定其磷含量，通过高光谱数据建立茭草磷含量估算模型。结果表明：茭草的反射光谱曲线与健康绿色植被的反射光谱曲线趋势一致，通过一阶导数变换可以更清楚地分析原始光谱的细节特征，作为植被光谱特有的“三边”参数，可以定量分析茭草光谱特征。分别利用光谱反射率、反射率一阶微分值和“三边”参数与磷含量作相关性分析，相关性最高的变量分别为平滑光谱1 085 nm波段、光谱一阶微分1 259 nm波段、红边面积，相关系数分别为0.528、0.619、0.526；基于Landsat 8 OLI波谱重采样的低维光谱数据，近红外波段对茭草磷含量较为敏感，相关系数为0.519；通过主成分分析提取的变量平滑光谱提取第1成分和反射率一阶微分提取第1成分与磷含量的相关性最高，相关系数分别为0.547和0.494。在建立的估算模型中，多元逐步回归模型估算效果较优，其次为主成分回归模型和单变量回归模型；茭草磷含量最佳估算模型为基于反射率一阶微分建立的多元逐步回归模型，R²为0.76、RMSE为13.69、P为91.08 %、MAE为10.75。

The Estimation Model for Phosphorus Content of Zizania Cuciflora in Jianhu Wetland Based on Hyperspectral Technology

Taking the dominant species of sturgeon of Jianhu wetland in Yunnan Province as the research object, the reflectance spectrum of Zizania Cuciflora was collected by ASD Filed Spec 3 spectrometer, and its phosphorus content was measured. The estimation model of Z. Cuciflora phosphorus content was established by hyperspectral data. The results showed that the reflectance spectrum curve direction of the Z. Cuciflora is consistent with the healthy green vegetation. The detailed characteristics of the original spectrum can be analyzed more clearly by the first derivative transformation. The spectral characteristics of the Z. Cuciflora can be quantitatively analyzed by" three edges”parameters as a unique for vegetation spectrum. The correlation was analyzed between the spectral reflectance, first-order differential of reflectance," three edges”parameters and phosphorus content, the highest correlative variables were respectively B1 085 (1 085 nm of smoothed spectrum), D1259 (1 259 nm of spectral first order differential) and SDr (red edge area), and the correlation coefficients were respectively 0.528, 0.619 and 0.526. The low-dimensional spectral data of resampling Based on Landsat 8 OLI spectrum, the near infrared band was more sensitive to the phosphorus content of Z. Cuciflora and the correlation coefficient was 0.519 on the low-dimensional spectral data of resampling Based on Landsat 8 OLI spectrum. The correlation coefficients of BF1 (first component extracted from smoothed spectrum) and DF1 (first component extracted from first order differential spectral) with phosphorus content were respectively 0.547 and 0.494. The multivariate stepwise regression models have better estimation effect, followed by the principal component regression models and the univariable regression models in all estimation models. The best phosphorus content estimation model of Z. Cuciflora was a multivariate stepwise regression model based on the first order differential of reflectance, R2, RMSE, P and MAE were respectively 0.76, 13.69, 91.08 % and 10.75.