Application of Artificial Neural Network (ANN) to Vegetation Water Content (VWC) Estimation Using Hyper-spectral Measurements

Abstract

Hyper-spectral measurements obtained using a GER 3700 spectra-radiometer including 584 narrow bands in spectral region of 400–2400 nm were applied to estimate vegetation water content (VWC). Developments in hyper spectral remote sensing have led to developing the new groups of spectral indices and statistical models for estimating biophysical and biochemical properties of vegetation. In this study a back - propagation neural network with three groups of input data set including all spectral bands, the first ten principal components and the best narrow band indices were applied separately as input to estimate the VWC. The suitability of the network efficiency was analyzed applying cross validation technique. The best ANN was obtained using simple linear regression between measured VWC and network outputs in terms of RMSECV and R2cv. The results of this study showed that ANN has a high potential for estimating VWC by hyper-spectral data (Rcv=0.88, RMSEcv=0.31).