DeepPaSTL: Spatio-Temporal Deep Learning Methods for Predicting Long-Term Pasture Terrains Using Synthetic Datasets

Effective management of dairy farms requires an accurate prediction pasture biomass. Generally, estimation biomass site-specific data, or often perfect world assumptions to model systems when field measurements other sensory inputs are unavailable. However, for small enterprises, regular data inconceivable. In this study, we approach the by predicting sward heights across field. A convolution based sequential architecture is proposed height predictions using deep learning. We develop a process create synthetic datasets that simulate evolution growth over period 30 years. The learning (DeepPaSTL) trained on dataset while spatiotemporal characteristics growth. purely learns from trends in through available spatial and agnostic any climatic conditions, such as temperature, precipitation, soil condition. Our performs within 12% error margin even during periods with largest dynamics. study demonstrates potential scalability predict size quantization prediction. Results suggest DeepPaSTL represents useful tool both short long horizon predictions, missing irregular historical measurements.