[Comparison of principal component analysis with VI-empirical approach for estimating severity of yellow rust of winter wheat]

The canopy reflectance of winter wheat infected by yellow rust with different severity was measured through artificial inoculation, and the disease index (DI) of the wheat corresponding to the spectra acquired in the field was obtained. Principal component analysis (PCA) was used to compute the first 5 principal components (PCs) of canopy spectra in the 350-1 350 nm range and the first 3 PCs of first-order derivative in blue edge (490-530 nm), yellow edge (550-582 nm) and red edge (630-673 nm), respectively. Step-wise regression was used to build models, the results of those models are compared with that of VI-empirical models, and the result shows that the model based on PCs of first-order derivative is particularly accurate compared to others, with the RMSE of 7.65 and relative error of 15.59%. Comparison was made between the estimated DI and the measured DI, indicating that the model based on SDr'/SDg' is suitable to monitoring early disease and the model based on PCs of first-order derivative is suitable to monitoring the more severe disease of yellow rust of winter wheat. The conclusion has great practical and application value to acquiring and evaluating wheat disease severity using hyperspectral remote sensing, and has an important meaning for increasing yields of crops and ensuring security of food supplies.