Satellite-based Grapevine Phenological Stage Detection through a Deep Supervised Machine Learning Approach

The accurate detection of grapevine phenological phases is crucial for optimal vineyard management. This paper presents the development of a machine-learning model for grapevine phenology stage detection using satellite data. It leverages multi-temporal satellite imagery and employs the GRU (Gated Recurrent Unit) neural architecture to capture temporal dependencies and timely identify the occurrence of key phenological stages. We compare the performance of proposed model with a traditional baseline approach based on the concept of Growing Degree Days, demonstrating significant improvements in detection accuracy. Experimental results highlight the model’s ability to effectively capture vegetation dynamics and growth patterns, offering accurate and near real-time information for vineyard management. By providing a reliable and scalable solution, the developed model could enable growers to make informed decisions to optimize grape production and quality.