Modeling evapotranspiration using Encoder-Decoder Model

In recent years, deep learning algorithms successfully applied to develop decision support systems in fields such as medical, bibliometric, and many more. The field of agriculture is one of the most important where developing decision support systems using deep learning needs to be explored due to its direct impact on human well-being. Agriculture is the largest customer of freshwater. Hence, optimizing irrigation and developing an efficient irrigation decision support systems is essential. The amount of evapotranspiration is the most critical factor in scheduling irrigation. In this work, the ability of an Encoder-Decoder Long Short-Term Memory model to model the daily evapotranspiration was investigated for three locations in Portugal. The model predicts evapotranspiration three days into the future, given six days of historical data. Two models were trained; one is used the teacher forcing method and the other one without the teacher forcing method. Our results show that by using the teacher forcing method, the model's performance could be improved. The performance of the Encoder-Decoder Long Short-Term Memory was compared to a simple Long Short-Term Memory model, Convolutional Neural Network, a Multi-Layer Perceptions model, and Random Forest regression. The Encoder-Decoder Long-short term memory performed better on the test set.