精确估算吸收性光合有效辐射分量(EPAR)对于检测植被水分、 能量及碳循环平衡具有重要意义。 应用ASD地物光谱仪与SUNSCAN冠层分析仪对冬小麦整个生育期内冠层光谱反射和光合有效辐射进行监测, 并利用冠层反射率数据构建了24个高光谱特征参量, 通过分析不同光谱特征参量与冬小麦FPAR的相关关系, 建立冬小麦FPAR光谱参量估算模型。 结果表明: 除蓝边幅值Db外其余高光谱参量均与冬小麦FPAR呈极显著相关(p＜0.01)。红边面积SDr与蓝边面积SDb的比值(VI4)与FPAR的相关系数最高, 达到0.836。 提取相关性较高的7个光谱参量分别与冬小麦FPAR建立最优线性与非线性估算模型, 通过精度检验分析, 优选了冬小麦FPAR最合适的模型。 对于线性模型, 绿峰位置λg与FPAR的反演模型最好, 其预测模型的R2, RMSE和RRMSE分别为0.679, 0.111和20.82%; 对于非线性模型, 绿峰反射率Rg与红谷反射率Rr的归一化比值(VI2)与FPAR的反演模型最好, 其预测模型的R2, RMSE和RRMSE分别为0.724, 0.088和21.84%。 为进一步提高模型精度, 分别运用多元线性逐步回归与BP神经网络建立多个高光谱参量同时参与的模型, 与单参量模型相比, BP神经网络模型的反演精度明显提高(R2=0.906, RMSE=0.08, RRMSE=16.57%)。 利用高光谱特征参量测定冬小麦FAPR具有可行性, 这为实时、 有效、 准确监测冬小麦生长过程中FPAR的动态变化提供了一种新的方法和理论依据。

Estimating fraction of absorbed photosynthetically active radiation (FPAR) precisely has great importance for detecting vegetation water content, energy and carbon cycle balance. Based on this, ASD FieldSpec 3 and SunScan canopy analyzer were applied to measure the canopy spectral reflectance and photosynthetically active radiation over whole growth stage of winter wheat. Canopy reflectance spectral data was used to build up 24 hyperspectral characteristic parameters and the correlation between FPAR and different spectral characteristic parameters were analyzed to establish the estimation model of FPAR for winter wheat. The results indicated that there were extremely significant correlations (p＜0.01) between FPAR and hyperspectral characteristic parameters except the slope of blue edge (Db). The correlation coefficient between FPAR and the ratio of red edge area to blue edge area (VI4) was the highest, reaching at 0.836. Seven spectral parameters with higher correlation coefficient were selected to establish optimal linear and nonlinear estimation models of FPAR, and the best estimating models of FPAR were obtained by accuracy analysis. For the linear model, the inversion model between green edge and FPAR was the best, with R2, RMSE and RRMSE of predicted model reaching 0.679, 0.111 and 20.82% respectively. For the nonlinear model, the inversion model between VI2 (normalized ratio of green peak to red valley of reflectivity) and FPAR was the best, with R2, RMSE and RRMSE of predicted model reaching 0.724, 0.088 and 21.84% for. In order to further improve the precision of the model, the multiple linear regression and BP neural network methods were used to establish models with multiple high spectral parameters BP neural network model (R2=0.906, RMSE=0.08, RRMSE=16.57%) could significantly improve the inversion precision compared with the single variable model. The results show that using hyperspectral characteristic parameters to estimate FPAR of winter wheat is feasible. It provides a new method and theoretical basis for monitoring the dynamic change of FPAR in real time, effectively and accurately during the growth stage of winter wheat.