Predicting the Normalized Difference Vegetation Index (NDVI) by training a crop growth model with historical data

Abstract

Normalized Difference Vegetation Index (NDVI) is an important remote measurement in agriculture because it has a high correlation with crop growth and yield result. In this paper, we present a methodology to predict the NDVI by training a crop growth model with historical data. Although we use a very simple soybean growth model, the methodology could be extended to other crops and more complex models. The training process is an optimization problem, that is solved using the spectral projected gradient method. The quality of the prediction is measured by computing the Root-Mean-Square Error (RMSE) between predicted and true values, obtaining an error lower than 9%, which improves the results obtained by simple forecast techniques used as baseline estimators.