Estimating Leaf Biochemical Information from Leaf Reflectance Spectrum using Artificial Neural Network

Leaf being the basic component of almost all plants on the earth, its biochemical status controls many critical physiological and ecological processes including photosynthesis and primary production that are crucial to our living environment. Leaf reflectance spectrum, which is caused by the absorption of leaf biochemical substances to a great extent, becomes an effective and fast way for leaf biochemical estimation. In this paper, the influence of chlorophyll and leaf water on leaf reflectance spectrum was investigated at first, and two artificial neural networks (ANN) were established for chlorophyll and leaf water estimation. In the end, the accuracy of ANN was validated using measured data. The main problem of ANN modeling in this paper is that the great number of spectral bands of leaf reflectance spectrum and insufficient measured leaf samples hamper the gathering of training sample set to establish the neural network. Four bands and band combinations (spectral indices) sensitive to chlorophyll and leaf water respectively based on the result of sensitivity test were selected purposefully as independent variables for ANN modeling so as to reduce the dimension. And simulated leaf reflectance spectra coupled with biochemical parameters by a within leaf radiative transfer model were used as part of training sample set. Validation results of rice leaves showed that the accuracy of chlorophyll and leaf water estimation using neural network is satisfactory.