Multi-Task Learning for Budbreak Prediction

• DOI: 10.48550/arxiv.2301.01815
• Publication Date: 2023-01-01
• Authors: Aseem Saxena; Paola Pesántez-Ca...; Rohan Ballapragada; Markus Keller; Alan Fern
• Keywords: Frost (temperature)

Grapevine budbreak is a key phenological stage of seasonal development, which serves as signal for the onset active growth. This also when grape plants are most vulnerable to damage from freezing temperatures. Hence, it important winegrowers anticipate day occurrence protect their vineyards late spring frost events. work investigates deep learning prediction using data collected multiple cultivars. While some cultivars have over 30 seasons others little 4 seasons, can adversely impact accuracy. To address this issue, we investigate multi-task learning, combines across all make predictions individual Our main result shows that several variants able significantly improve accuracy compared each cultivar independently.