The Application of Hyper-spectral Data and RBF Neural Network Method to Retrieval of Leaf Area Index of Grassland

Abstract

In accordance with the 120 sites of grassland canopy spectral reflectance and the leaf area index(LAI) data collected by Chinese Academy of Agricultural Science,the method of Radial Basis Function(RBF) neural network was developed for the prediction of LAI after the compression of spectral reflectance using principal component analysis(PCA).The PCA results show that the cumulative reliability of the first 9 PCs is up to 99.782%,covering the majority of original spectral information.The 120 sites of LAI and 9 PC samples were classified randomly for training dataset(90 sites) and predicting dataset(30 sites),and were used to establish the neural network and predict the LAI,respectively.The results show that the accuracy rate of training data is up to 100%(RMSE=0.009 6,R2=0.999).The root mean square error(RMSE) and correlation coefficient(R2) for the prediction dataset are 0.839 and 0.218 6 respectivdg,thus achieving more preferable results and improved the accuracy(RMSE=0.416 5,R2=0.570)of the traditional multiple linear regression method.