Deep Learning Based Time-Series Forecasting Framework for Olive Precision Farming

In this paper, we present a time series forecasting framework that uses deep learning to predict the environmental attributes that affect the olive fruit farming. The proposed framework is composed of two phases: a data preprocessing phase and a prediction phase. In the prediction phase, we use both the Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) deep learning approaches to develop two models for predicting the environmental attributes. We evaluate the performance of the framework using real-life agriculture data collected for twenty years from a Spanish olive grove. Our results show that proposed LSTM and GRU models achieve remarkable accuracy, measured through different metrics, in predicting the temperature and relative humidity for the upcoming year based on historical data. We further use the predicted temperature in calculating the degree-day metric used to define the development phases of the olive fruit fly. This allows for foreseeing the best times to apply the counter measures to prevent the outbreak of such a fatal pest that affects a major Mediterranean crop: olive.