Thursday, November 15, 2018

Logistics for distributed, diverse production

Farming operations don't exist in a vacuum. Farm products feed into a processing and distribution network, and, if nutrients are to be recycled, waste materials need to come back to farms to maintain and improve soils. Both flows, outward and inward, are served by keeping the supply lines (the need for transportation) as short as can reasonably be managed, given other constraints like economies of scale in processing, storage, distribution, and marketing.

This post will focus on the outward flow, from farm to processing and storage to distributor to vendor to customer, which, I'd argue, applies even to urban farms with direct-to-customer relationships, albeit in compressed form.

The usual model is that farms specialize, sometimes producing only a single commodity and typically only a few, with production profiles that closely resemble those of neighboring farms, feeding what they produce into bulk collection points (grain elevators, for example) and from those into processing, storage, and distribution networks that are optimized for certain measures of efficiency, notably the profitability of the companies involved.

But specialization of this sort does not lend itself well to high biological diversity, which, at the microbial level, is an important measure of soil health. This is less true of grazing operations, since pasture can be biologically very diverse, but even ranchers may suppress the growth of broadleaf plants to increase the proportion of grass in the mix, and, in any case, in temperate climates grazing may not be available all year around. For part of the year livestock may need to be fed hay, silage, grain, soy, etc., grown for this purpose or purchased from distribution chains similar to those described above.

So, even if we were already in possession of technology adequate to perform the detailed, plant-by-plant, leaf-by-leaf manipulations necessary to conduct diverse perennial polyculture at scale, we would still be faced with the challenge of pairing a production system with extended, overlapping harvests of perhaps dozens of different commodities from any given locale, with a processing and distribution system accustomed to large volumes of single commodities arriving more-or-less all at once.

This is both an informational challenge and also one of physical logistics.

The informational component begins with estimations of how much of what is likely to available from whom, how soon, with what expectation of quality. This can be combined over multiple farms to aggregate marketable quantities of each crop. By the time harvest arrives how much of each farmer's production should be combined with that from which other farms to go where should already be worked out.

The physical component is an exercise in materials handling, beginning with harvesting particular plants or plant parts from the midst of dense, mixed stands, without damage either to the crop being harvested or to what is left behind after harvest. The next step will likely be some form of containerization, so each farm's output can be identified, even after being combined with that from other farms for transport.

The informational component isn't so different from other applications of information technology, and, given the ability to provide reliable harvest predictions, developing that will be relatively straightforward. Most of what has yet to be worked out relates to the physical, materials handling component, the details of how to perform delicate harvesting operations and how to handle the crops once they're harvested.

Since efforts to develop such capabilities can't be expected to turn a profit next year, perhaps not even within the next ten years, this is an appropriate focus for government-funded research, and we should all be insisting that the relevant government agencies make this type of research a priority, while some soil remains to be saved and regenerated.

Friday, February 23, 2018

The point isn't to replace people; it's to go beyond what people can do at scale.

Imagine a horticulturist at work, repotting plants, snipping off withered leaves, moving plants farther apart as they grow, setting them out in gardens and landscapes, diagnosing nutrient deficiencies, diseases, infestations, and so forth, maybe even engaging in a little selective cross-pollination to produce new varieties, as time allows. A horticulturist needs to be a soil scientist, a botanist, a plant pathologist, a designer, and, in many cases, an entrepreneur.

Now imagine automating all of the most tedious physical tasks.

Even that is no small feat; skilled horticulturists probably have among the most secure jobs in existence. Nevertheless, it can be done and will almost certainly eventually be done.

So now imagine that you're running a company that is closing in on that goal and looking forward to what to take on next. Do you attempt to automate those aspects of the work requiring design and/or business sense, to entirely replace horticulturists with robots? Or do you press on with tackling more delicate tasks, faster, based on increasingly comprehensive sensory data backed by increasingly sophisticated processing and accumulated knowledge, with the intention of gradually moving this approach to plant cultivation out onto broad acres as the cost of doing so falls far enough to be competitive with other methods?

I suppose the answer depends on who you perceive the customer to be and what you perceive to be marketable, and whether you're only interested in business-to-business relationships or also want to serve the end consumer. (Note: people may quite willingly buy their food from a vending machine, but there's likely to be much more resistance to the idea of buying potted plants or flower arrangements from one. The interaction with another human being is integral to such transactions, and for good reason.)

I see the role of robots as coming in stages, beginning with augmenting what skilled humans can do by relieving them of the most time-consuming, tedious tasks, then taking the performance of those tasks deeper, by improving precision, data collection, and the conversion of that data into knowledge, then moving the capabilities developed into a broader context, such as by applying horticultural methods at the scale of agriculture.

The goal shouldn't be to replace humans, particularly not skilled, knowledgable humans. It should be to displace crude methods by bringing skill and knowledge to bear at scale.