The topic of monitoring the state of vegetation using satellite technologies is covered in this paper. The forming of fields' NDVI images is considered. It is proposed to supplement satellite images with new images which formed on predicted values of the vegetation index. It can be helpful for timely detection of heterogeneous and defective areas of vegetation cover. The paper discusses methods for forecasting the NDVI values using Volgograd region as a study area. The results of training a recurrent neural network with the LSTM mechanism, as well as the results of training the XGBoost algorithm, are obtained. Based on the results of the training, the most important weather parameters affecting NDVI were identified. The performance of the trained models was evaluated using the RMSE metric.
Keywords: precision farming, vegetation indices, NDVI, forecasting, time series, LSTM, random forest