Accurate and scalable crop classification is important for food security and sustainable resources management. The temporal development of crops, i.e., their phenology, is a continuous phenomena that if properly captured, can help to discern them. The novel model, Neural Controlled Differential Equations (NCDE), extends the disruptive Neural Ordinary Differential Equations (NODEs) for processing time series, and allows their hidden state to be dynamically controlled by a continuous representation of the data. This thesis’ objective was to explore if a model with a fully continuous hidden state, such as NCDEs, can achieve improved performance in crop classification. The experimental results obtained showed that their performance is still inferior to the state-of-the-art, however a careful study of their weaknesses suggested potential opportunities for further improving them. In particular it was found that the continuous data representation method chosen for the data may severely affect the performance of the model. Therefore, evaluating improved interpolation and regression methods with respect to those evaluated here are a promising direction to boost their performance.