Predictive Analysis of Strawberry Harvest in an Agrivoltaic Field Using Machine Learning

Agrivoltaic systems (AVS) represent a promising solution for sustainable land use by integrating photovoltaic panels with crop cultivation. This study examines an AVS in Basilicata, Italy, focusing on the impact of photovoltaic panels on microclimate conditions and strawberry yield in a hydroponic setup. Environmental parameters were monitored at 15 minute intervals, while strawberry production data were collected as daily aggregates. A key challenge emerged due to label sparsity, the misalignment between high-frequency sensor data and low-frequency yield measurements, which complicates traditional supervised learning approaches. Data exploration revealed microclimatic variations influenced by panel proximity, with temperature increasing by up to 3°C and humidity decreasing by 5% in the sector closest to the panels. Principal component analysis and K-means clustering highlighted distinct environmental patterns, while yield comparisons showed significant reductions under shaded conditions. Varietal performance varied, suggesting potential for shade-tolerant cultivar selection. The study underscores the need for advanced machine learning techniques to address label sparsity in agrivoltaic yield prediction. It also demonstrates the value of integrated monitoring systems for optimizing AVS design and crop management. Future research should explore time-series modeling and multisensor data fusion to enhance predictive accuracy in such complex agricultural-energy systems.