In this paper predictions of the normalized difference vegetation index (NDVI) are discussed. Time series of Earth observation based estimates of vegetation inform about changes in vegetation. NDVI is an important parameter for short-term vegetation forecasting and management of various problems, such as prediction of spread of forest fire and forest disease. Artificial neural networks (ANN's) are computational models and universal approximators, which are widely used for nonlinear, non-stationary and dynamical process modeling and forecasting. In this paper, first, a stepwise regression was used as feature selection method in order to reduce input data set dimensionality and improve predictability. Correlation in input data normally creates confusion over ANN's during the learning process and thus, degrades their generalization capability. The Principal component analysis (PCA) method was proposed for elimination of correlated information in data. A layer recurrent neural network (LRN) then was used to make short-term one-step-ahead prediction of the NDVI time series.