大丰麋鹿保护区狼尾草粗蛋白含量的高光谱遥感估算

Estimation of Pennisetum alopecuroides crude protein content in the Jiangsu Dafeng David’s Deer Reserve using hyperspectral remote sensing technology

粗蛋白是评价牧草质量的重要指标,利用高光谱遥感技术可以充分了解粗蛋白的含量及其动态变化,为牧草品质遥感监测提供科学依据。利用光谱仪测定大丰麋鹿自然保护区半散养区主要优势植物狼尾草(Pennisetum alopecuroides)的冠层反射率,并分析了其粗蛋白的相对含量。选择狼尾草原始反射率光谱及其构建的植被指数与其粗蛋白含量进行相关性分析,筛选显著相关的光谱特征;运用偏最小二乘回归(partial least squares regression,PLSR)和多元逐步回归(stepwise multiple linear regression,SMLR)建立狼尾草粗蛋白含量高光谱预测模型,通过比较模型精度确定最优预测模型。结果表明,基于显著性波段的PLSR模型精度最高(验证集R^(2)=0.897,RMSE=1.303,RPD=3.110);基于显著性波段的SMLR模型反演效果最差(验证集R^(2)=0.170,RMSE=3.691,RPD=1.098);基于植被指数的PLSR和SMLR模型精度较为接近,验证集RPD约为2;基于显著性波段的PLSR模型反演效果最好。本研究结果可为保护区大区域尺度狼尾草粗蛋白含量的定量反演提供参考依据。

Crude protein(CP)is an important index for evaluating forage quality.Rapid determination of the crude protein content of forage grass may provide a scientific basis for monitoring forage quality using remote sensing.The canopy reflectance of the dominant species Pennisetum alopecuroides in the semi-captive area of the Jiangsu Dafeng David’s Deer Reserve was measuerd using a back-held Field Spec Pro FR^(2)500 hyperspectrograph,and the crude protein content was determined.Pre-processed spectral reflectance and spectral indices correlating significantly with crude protein content were selected to establish the partial least squares regression(PLSR)and stepwise multiple linear regression(SMLR)models.The PLSR model based on significant bands showed excellent accuracy for predicting crude protein content(R^(2)=0.897,RMSE=1.303,RPD=3.11 in the validation set),whereas the SMLR model had the lowest degree of accuracy in the validation set(R^(2)=0.17,RMSE=3.691,RPD=1.098).The predictive ability was similar between PLSR and SMLR models based on significant spectral indices,with RPD reaching approximately 2.The PLSR model based on significant bands yielded the most accurate model,which can quantitatively predict the crude protein content of P.alopecuroides in a large area in the reserve.