Citation
Hemming, S., de Zwart, F., Elings, A., Righini, I., & Petropoulou, A. (2019). Remote Control of Greenhouse Vegetable Production with Artificial Intelligence—Greenhouse Climate, Irrigation, and Crop Production. Sensors, 19(8), 1807. doi.org/10.3390/s19081807
Background
As the global population keeps growing, so does the demand for healthy, fresh, and accessible food. Greenhouse environments contribute an important function in providing fresh food at a high production rate while maximizing resource efficiency. The greenhouse industry has stumbled over the obstacle of finding enough skilled staff to manage these crop production systems. Modern high-tech greenhouses are equipped with process computers, and, to add more automated control, greenhouse climate and crop models have been developed. The use of Artificial Intelligence (AI) has reached breakthroughs in many areas and has not been used yet to control climate and irrigation and make crop management decisions for growing a greenhouse crop autonomously. An international challenge was conducted on autonomous greenhouses in 2018 at the high-tech research greenhouses of Wageningen University and Research in cooperation with five multi-disciplinary international teams to combine the use of modern AI algorithms and the greenhouse crop production of cucumbers cultivar “Hi-Power”. This article aims to describe the results of the teams in terms of optimizing crop yields and net profit using state-of-the-art AI algorithms for cucumber production.
Methodology
The experiments were conducted in six identical greenhouse compartments equipped with standard motor movers. Cucumber seedlings were grown in rockwool substrate cubes and placed on slabs on hanging gutters. The five participating teams were: Sonoma, iGrow, deep_greens, The Croperators, and AiCu. These teams were able to remotely control the motors’ activity in their greenhouse compartment by using their own original AI algorithm, varying in design and techniques. As part of the competition, a sixth team led by Dutch growers controlled a greenhouse compartment and served as a reference. The competing teams used their original AI algorithms to regulate the climate and irrigation setpoints through a central computer and operated in the greenhouse compartments accordingly. Standard sensors set up in the greenhouse compartments continuously measured data and calculations were made based on these and digitally sent back to the teams. Teams were also allowed to install additional sensors in their compartments at the start of the experiment. Three harvest quality parameters were established: A: no defects (>375g), B: defects in shape, color, or others (300-374g), and C: less than 300g per fruit. Harvest data was measured manually by greenhouse staff and digitally sent to the teams.
The teams were judged based on three criteria: Sustainability, based on resource use efficiency (20%), Net profit, based on the number of fruits harvested per price of the fruit and category (50%), and AI algorithm, based on originality and efficacy of the algorithm (30%). Two models were used to analyze and compare the different AI algorithm approaches since the operation resulted in differences in cropping, climate, irrigation strategies, harvest yields, and resource use efficiencies. However, the combined model did not represent the presence and effects of pests and disease. The combined model was carried out to compare the calculated output as the predicted fresh cucumber yield per greenhouse compartment versus the realized cucumber yield in the same greenhouse compartment to verify the models. Additionally, model calculations were carried out applying the cropping strategy, lighting strategy, and climate strategy of other teams for each greenhouse compartment to predict the changes in yield and compare these.
Figure 2 – Scheme of data exchange (Hemming et al., 2019)
Results and Discussion
Out of all the teams, including the Dutch growers, the Sonoma team resulted in the highest production of cucumbers and consistently did this throughout the experiment. They predicted that by having a high daily light integral, they would be able to achieve a greater harvest, as they did, focusing their algorithms on this. Other teams decided to increase daily light integral and maintain a low carbon dioxide concentration (The Croperators) or have lower daily light integrals in the beginning to minimize fruit abortion (Dutch growers) which resulted in lower yields but led to a greater understanding of important production factors. All teams began the experiment with a relatively low amount of carbon dioxide, where most teams increased their concentration throughout the experiment and later diminished their dosage towards the end, the Sonoma team decided to increase their carbon dioxide concentration continuously throughout the total cropping period.
The goal of maximizing crop management is finding the optimum combination of the climate, lighting, and cropping strategies. This can be achieved by attaining the greatest number of fruits per area without hindering photosynthesis. Light, for example, is the foundation for healthy growth. For this particular experiment, having higher light integrals resulted in higher yields for the teams. As predicted with the combined models, the teams iGrow, AiCu, and growers could have possibly achieved a higher yield if they would have used the lighting strategies of either team Sonoma or The Croperators. Overall, most of the teams were able to obtain a decent production using a low amount of resources while reaching a net profit close to or better than manual commercial growers.
The goal of this experiment was to see how AI algorithms would become better at making crop management decisions and their effects on the greenhouse climate and production. At the end of the cucumber crop’s growth period, all the algorithms developed for this experiment resulted in establishing a successful greenhouse environment. As this was the first successful experiment on remotely controlling a greenhouse for cucumber production using artificial intelligence algorithms, the authors demonstrated that AI algorithms can compete, and even outperform, experienced manual growers. Artificially assisted or managed greenhouse systems could be a useful tool to grow crops where there is limited knowledge. In the future, more developments and studies in this area will be needed to make AI an alternative for trained and skilled greenhouse workers and growers, which to this day, greenhouses cannot function without.
It’s interesting to see as time progresses, with more developments in state-of-the-art technology and intelligence, how we have incorporated these advancements in several areas including the agricultural sciences. In my view, further research will be definitely needed to keep implementing this technology in controlled environment agriculture and could be very beneficial in improving crop production and quality. However, with more autonomous greenhouses around the globe, less skilled personnel may be needed to operate them. What are your thoughts on AI-assisted or -managed greenhouses serving as a potential alternative to skilled growers?
I would make the analogy that AI-assisted greenhouses are to horticulture as calculators are to mathematics. Once a problem is identified, these tools help to find the answer and correct it more quickly. However, if you cannot identify the problem in the first place, you wouldn’t know how to look for a solution. While some greenhouse crops have diseases with easily-recognizable symptoms (e.g. tip burn of strawberry), identifying all the stressors negatively impacting crop yield is too complex for computers to handle. The plant could have a nutrient disorder originating from the rhizosphere, physiological disorder due to environmental stressors, or a disease that is pathogenic in nature. As long as there is an intricate and complex combination of factors that can decrease crop yield, AI-assisted greenhouses aren’t capable of crop cultivation entirely independent from humans. Instead, there needs to be human intervention during the cultivation process, which would allow for the identification of variables that could be optimized. Instead of serving as an alternative to skilled growers, AI-assisted greenhouses will require growers with a different skill set which includes the establishment and maintenance of these AI technologies.
This type of research conducted in high-tech greenhouses has led to a better understanding of the climate, lighting, and cropping strategies needed for optimum crop growth. Although, few countries possess this type of technology to be able to do these kinds of studies. I can’t help but wonder if having more autonomous greenhouses in the future would lead to more accessibility between fresh produce and their consumers, or further distance them. Does someone have more insight into how this may benefit low and middle-income countries in accessing fresh food?
A recent article in hortidaily referenced an article that mentioned the abandonment of greenhouses for vegetable production in Kenya (likely referring to soil based high tunnels not the more technically advanced greenhouses equipped with sensors and automatic controls). Apparently, in certain coastal regions, up to 70% of greenhouses are abandoned, often due to a lack of technical support and extension information on proper management. Based on this, it would seem that future AI may be able to help farmers make management (venting, irrigation, pest control….) decisions based on weather predictions or pest/disease forecasting. However, when the great cost of AI technology, and the need for professional instillation and service is considered it seems unlikely that small scale farmers and institutions like schools, which make up a large portion of greenhouse users in parts of Kenya, will be able to afford these systems when even non-AI greenhouse control systems can cost thousands of dollars.
https://www.hortidaily.com/article/9327722/greenhouses-don-t-work-in-kenya/
Status of Greenhouse Farming in the Coastal Humid Climatic Region of Kenya: https://www.hrpub.org/download/20180830/UJAR4-10411904.pdf
These are such exciting developments for crop production. I think this competition was a great way to determine which AI approaches to plant growth (being high focus on light intensity, CO2 levels, etc.) were most effective. I found the most beneficial part of this paper being its highlights on how each AI system, even the best performing one, could improve based on the successes of other AI systems. I think this will be a great foundation for deciding how to improve AI algorithms to best manage greenhouses. I feel that this opens the opportunity to greatly expand the size greenhouse production systems; greenhouse staff could then focus on problem-solving in specific areas or for specific issues rather than manage every detail throughout the growth space. I think this is a better use for the time and skills of greenhouse growers and can help reduce the cost needed to feed more consumers.