AI for Vertical Farming
Vertical Farming with A.I.
Sustainable Food Production
Integrating Artificial Intelligence (A.I.) into Vertical Farming is the next big step in this form of sustainable agriculture. When this type of agriculture first began it was riddled with problems and was quite expensive. Bringing costs down was an essential early step in the process. Now A.I. will help us increase the savings even more by spotting potential failures long before human eyes can see them.
What is Vertical Farming?
Vertical Farming is the practice of growing crops indoors, under artificial lighting, with little to no soil, up to 95% less water than traditional outdoor growing, and on a year-round basis. It is a valuable contributor to the domestic food chain.
Vertical Farming is composed of multiple layers, and can be any height, limited only by the architecture of its building. By way of example, a building with one hectare (2.5 acres/330’ x 330’) could have 20 “layers”, thus turning it into 20x its base area in terms of cropping ability (e.g. 20 hectares/50 acres). Imagine having an entire farm in the equivalent of an industrial warehouse, right in the city.
This diminishes the carbon footprint by minimizing transportation cost and being proximate to the point of consumption. It allows the provisioning of farm fresh food, sometimes within minutes of harvesting, particularly to high quality restaurants and local customers who demand that high quality food.
The single largest cost at first was using artificial lighting, something which outdoor growers get for free from the Sun. Full spectrum incandescents waste a tremendous amount of power and cause excessive heat.
With the advent of LEDs, and knowledge that blue and red light were essential but green was not, this allowed a 90% reduction in these costs. This was further aided by installing solar power panels on rooftops so growers could harness that “free power” from the Sun again, like conventional farmers, and reduce their electrical bills even further.
Growing in trays can consist of either hydroponics or the NASA-developed aeroponics, originally designed to allow astronauts to grow food on long trips. The former means growing in nutrient rich liquids; the latter refers to growing with confined, but exposed roots that are sprayed with nutrient rich mist. Commercially this is more expensive than free “dirt”, but the nutrients can be idealized for the crop, making it far more efficient and tightly controlled and thus virtually assuring success.
Some simple crops, like bean sprouts and watercress, can be turned around in just two weeks. In any case, the process is continuous, without soil exhaustion. The reliability of the crops, and the sheer variety, are what make Vertical Farming such an attractive proposition. Premium prices can be charged for this carbon-footprint-sensitive, locally grown produce.
By providing only exactly what is needed for plant growth, costs can be significantly reduced in areas of expensive or low-availability water, too. This could encompass drought-challenged places like California or the Middle East.
Automation plays a large role in Vertical Farming, controlling lighting, nutrients, and water delivery. Adding Artificial Intelligence (AI), Machine Learning (ML), and Computer Vision (CV) enhances every aspect.
Vertical Farming benefits the crops by being able to constantly assess the crops. Computer vision has already revolutionized medical imaging (MRI, PET, CAT, X-ray, etc.) for humans, allowing the diagnosis of disease years earlier than previously possible, where cures are a matter of pills or diet change rather than radical surgery.
The same is true for these crops. CV can spot fungus invasions, pests, and other problems, and then isolate and treat the menace by quickly moving it out of the rotation and into a quarantine until the problem is resolved. Normally this calls for human assessment, but ML and CV can learn to do this not only faster than a human, but entirely on their own, meaning far less human intervention is required.
Much more efficient recycling makes the operation less costly, too. There is no need to discharge water (and nutrients!) at all. Proper filtering and treatment makes everything completely reusable, requiring only replenishment for the produce that goes out the door.
Machine Learning allows the system to recognize when a plant is lacking essential nutrients, or ready to be harvested. While we are currently using a lot of hand pollination, that can be automated and handled by the A.I. in the future.
Currently we are growing a lot of leafy crops, because they are fast and efficient, but we’re adding corn, okra, Brussel sprouts, sunflowers, and even experimental orchards to produce apples, peaches, and so on. The choices are only limited to how efficiently you can use the available volume of space. Even a well-trimmed and maintained apple tree could use 8-10 “layers” that could grow smaller, faster crops, but a single semi-dwarf apple tree can produce 500 apples per “season”, of which you might coax two such seasons out of it each year with proper care and encouragement… Just a handful of trees could produce thousands of apples.
Now, as a cautionary note, these sorts of operations are susceptible to significant loss even from relatively short power outages. You may find a location with power from two different parts of the electrical grid, if that is possible, but regardless, it will still be necessary to have diesel generators for electrical backup. Think of it as some of the cheapest insurance you can buy—but buy (or rent) it you must.
To make it even more profitable, many are now looking at aquaponics as an additional source of income that benefits the crops. Aquaponics is basically hydroponics with fish added.
This provides another high value crop (fish) that also happen to provide high value fertilizer for the plant crops. It seems like a natural outgrowth, and fish assessment by computer vision is already established in the aquaculture industry.
CV and ML would have to be adapted to assess tree crops, and optimized for fish aquaponics, but this is entirely possible. No matter which crop is being investigated, the A.I. can assess maximal growth and crop maturity, based on what can be controlled, such as temperature, periodicity of lighting, CO2 content of the air, nutrient supply, and final output.
This would allow assessment of net weight comparisons vs. all of the inputs to precisely calculate costs and profits. Water utilization, nutrient input, labor costs, power efficiency would all form part of the equation. While a lot of harvesting is currently manual, that can be automated, too. It’s only a small step to add packaging, order fulfilment, and logistics for delivery into the mix to create a complete system.
Once you integrate market demand, production costs, pricing, and disease resistance (or “hardiness”) with the A.I., it can also decide which crops to emphasize to maximize profits. Ultimately, such a farm could be a “set-and-forget” operation, managed by just one or two people—alongside an incredibly competent A.I. managing the bulk of the work. Once we get to this point of sophistication it wouldn’t take many of these operations to feed an entire city.