ASHS 2025 Conference

Greenhouse tomato production with high-wire system and indeterminate tomato cultivars facilitates year-round production with high quality and productivity. However, maintaining optimal climate conditions in greenhouse is expensive due to high operational costs. Optimizing climate control strategies requires in-depth understanding of controlling systems, outdoor climate, and plant physiology. But skilled and experienced growers may not always be available. Artificial intelligent-driven climate control (AI) has been emerged as a potential solution. Yet, few trials have conducted, which may not be at an equivalent scale as the industry and following the industry standard. To address this gap, we compared AI and conventional climate control strategies (human decision-based; CV) for greenhouse tomato production in two identical high-tech greenhouse compartments (namely, AI and CV each with 481.7 m²) over 145 days after the final transplanting with management practices established by commercial growers. Each compartment had 420 plants of the indeterminant cultivar Maxxiany at a planting density of 3 plants m⁻². The AI algorithms were developed using datasets from commercial growers and a digital twin via physiology-informed neural network (photosynthesis and transpiration modules). Leaf pruning in AI was determined based on weekly light integral below canopy (Kim and Kubota, 2025), while CV followed conventional pruning based on harvesting trusses. To evaluate the performance of AI, parameters for crop development, yield, and fruit quality were collected in addition to environmental conditions and resource usage for lighting, cooling, heating, and fertigation. AI maintained relatively higher day and night temperature with high heating pipes temperature and keeping windows closed. AI also resulted in more leaves within canopy from fewer leaf pruning compared to CV. Those contributed to increase in cumulative irrigation volume (936 vs. 785 l m⁻² for AI and CV) and thus total fertilizer use (878 vs. 639 g m⁻²). AI used more natural gas for heating (190 vs.79 MJ m⁻²) and more electricity for supplemental lighting (91.4 vs. 80.4 kWh m⁻²). However, AI had higher cumulative yield (9.3 ± 0.3 vs. 8.5 ± 0.3 kg m⁻²) and greater PAR-based productivity (grams of fruits per PAR mol; 4.1 vs 3.6 g mol⁻¹). These findings suggest that AI increased resources use (water, fertilizer, natural gas, and electricity) but also resulted in higher yields as a trade-off. Further optimization of AI’s algorithms regarding fertigation and heating strategies may improve economic feasibility of AI application in greenhouse tomato production.