Analysis and forecasting of the life cycle of vegetation are essential in planning agricultural work as well as monitoring of agricultural crops and forecasting their productivity. In practice, vegetation indices are often used that are calculated from the values of satellite image pixels like normalized difference vegetation index (NDVI). Forecasting of this index in precision agriculture allows indicating problems, which are related to agricultural crop growth in time and make timely decisions about necessary measures to fix these problems. In recent years, a series of signal processing methods such as signal decomposition techniques based on spectral analysis is widely used in time series forecasting tasks. These methods are used in time series forecasting tasks for data preprocessing, obtaining multiple derived time series or sub-signals from original time series. Variational mode decomposition (VMD) is a modern decomposition method that can be used for time series decomposition. However, it is a non-causal method, and not suitable for time series forecasting, if all historical observations of time series are fed into VMD method. In this paper, the sub-signal approximation approach is developing. First of all NDVI time series is fed into a modified VMD method and sub-signal is obtained. Then a developed approach is used in order to get approximately calculated sub-signal values. NDVI time series is forecasted using as input data historical observations of the NDVI time series and an approximated sub-signal and then using only historical observations of the NDVI time series.