Farm Operations Management

Using vertical farm data: what to do before you buy a new sensor

A diagram showing the types of data recorded daily at a vertical farm (environment, cultivation, equipment)

A vendor proposes, “Let’s start by adding more sensors and collecting data,” and you are not sure that is really the right move. Or you keep careful records every day, yet they never seem to connect to a single management decision. Is that the kind of doubt you have been sitting with?

When people hear “using data,” they tend to think it begins with buying a new IoT or AI system. But the real starting point is much closer to hand.

The records pile up, but no one looks until after the accident

Picture your daily log. Many of you record temperature, humidity, and harvest volume on paper or in a spreadsheet, and have kept it up for a long time. The problem is what happens after you write it down. Whether anyone actually looks at it is another matter. When the harvest drops, that is the first time you go back, look, and realize after the fact, “the numbers really were off back then.” Does that sound familiar? You record the electricity bill every month too, but you just glance at it and stop at “that’s high.” You already have plenty of data on hand, yet you always look only after something has happened. What you are missing is not a new sensor. Who looks at what you already have, and when: that is what is absent.

There are two ways of looking. The “chasing after” kind, where you go back once an accident has happened, and the “regular check” kind, where you look at a fixed point during normal times when nothing is going on. Chasing after ends at “the numbers were off back then.” It is too late, so all it can do is confirm the cause in hindsight.

So before you buy a new system, the only thing to decide is who looks, and when. Here is one practice I have kept up on the floor of a leafy-greens PFAL: for fifteen minutes each morning, one person reads the previous day’s log out loud. Temperature, humidity, harvest volume, every morning, even when nothing is wrong. Then “that’s higher than usual” registers as a small sense that something is off, before it becomes an accident. With the electricity bill too, just a rough weekly look instead of once a month makes “this one week is oddly high” show up as the HVAC running too hard or a door left open. In a PFAL, a closed environment where lighting and HVAC run constantly, this was where things showed up earliest on my floor.

The same number is just a record if you look at it after the fact, but a yardstick for measuring “the gap from usual” if you look at it every day during normal times. Using data takes a big step forward on that switch alone. Adding a new sensor is fine once something comes up that you cannot catch even with that yardstick.

The same skew has been pointed out in the research world. According to a large-scale review of the smart-farming literature, there are plenty of reports that a technology works, but almost no studies that analyze how much it changed yield or return on investment once installed (see 1). That same review also points out that the most widely used technology is the sensor. It is understandable that the conversation leans toward “buy first,” but the effect that actually matters turns out to be surprisingly unmeasured.

The idea of making the electricity bill your normal-times yardstick holds up in the research too. In a case from a sunlight-type plant factory, one report finds that monthly electricity use per unit of floor area can serve as an energy-management indicator (see 2). That is a greenhouse case, where much of the power draw runs through a heat pump tied to outdoor temperature and so mirrors the seasons; a PFAL, by contrast, is dominated by lighting load, barely tracks outdoor temperature, and its consumption, on average, is nearly flat. So what you take from it is not the substance of “it mirrors outdoor temperature and the seasons” but the idea of “set power per unit area as one yardstick.” If you apply that yardstick in a PFAL, what shows up is not the seasons but operational drift, the HVAC running too hard or a door left open. I think that is how to read it — by substitution.

So what is the minimum set of records to keep? Line them up by whether they can serve as the basis for a management decision, and you get three: production (daily harvest volume, and if possible the yield rate), cost (the electricity bill, and how labor and materials are used), quality (the shipment grade, and how many out-of-spec items came up). These are probably scattered across separate ledgers and receipts. Not starting to measure something new, but lining up what already exists in the same place: that is where building the minimum set starts. If you are stuck from scratch on which items to line up and how, I hand out the recording format I have used on the floor as the 13 farm operations management templates, exactly as they are. Rather than something to apply wholesale, I think it is better to see it as a checklist that throws light on what is missing and what is excessive on your own floor.

Use the normal range as a yardstick, and don’t put the call on one person’s shoulders

Keep up the fifteen-minute morning read-aloud every day, and before long the reading becomes a ritual. You get used to it, and it stops registering. In my experience, this happens almost without fail. So you need two safeguards.

A single handwritten sheet of paper noting the normal range

The first is how to build the initial yardstick. At the start, no one knows what counts as “different from usual.” So accept that at first you will not judge, and just line up the numbers. Line up the temperature at the same time and the same place every morning, and a range of “ours is usually around here” naturally comes into view. Write that range down in one line on paper. For example, “the morning temperature is around here,” using the actual values from your own floor. Once you have put the normal range into words like this, the yardstick moves out of one person’s gut feel and reads the same for anyone. The trick is not to try to set the correct standard from the start.

The second is that when only one person looks, the read gets skewed. Have one person do the reading, but never let one person make the call. The reader can be fixed every day. Only when they think “this feels different from usual” do they turn to one other person right there and ask, “what do you think of this?” This is, in fact, also what keeps you from glazing over once you are used to it. Perfect concentration every day is impossible, so make it a rule that only when something catches do you say it out loud and pull in a second person. The idea is that even with a skim, it is enough if the setup lets you stop the moment one thing catches.

Getting used to it is a given. Build it so you can still catch things once you are used to it. A way of running things that leans on perfect concentration will not last.

Deciding in advance to separate who looks from who decides is plain, but it is where the leverage is. In research, the proposal that an information platform linking sensor data and control can raise crop management and decision-making to a higher level comes up repeatedly within the same UECS-family efforts (see 3, 4, 5, 6). But what is proposed there is, at bottom, the implementation story that you can build such a system; most of it does not step into the operational question of who looks at that screen, and when, to decide the next move. That is exactly why deciding who looks and who decides, up front, remains a point you have to build by hand in the end, whether before or after you bring in a system.

Add a measurement only after you hit the wall of isolating a cause

Stand on the management side and the view changes again. A vendor brings a proposal, “install this and yield goes up,” and next week’s board meeting decides whether to adopt it. Have you sat in on a scene like that? At that point, how does “first decide who looks, and when” mesh with the call to invest or not?

A single electricity invoice as the entry point to using data

What matters here is not whether to add or not, but the view of “with the yardstick you already have, can that gap be isolated all the way to its cause?” Once the morning read-aloud and the weekly electricity bill let you catch “different from usual,” the next wall always comes. The gap is visible, but the records on hand cannot separate why it happened. Say the harvest dropped and the electricity bill rose too. But the daily-log temperature is within the normal range. Here you cannot pin down “is it the HVAC, the water, or something else entirely?” With the records on hand, isolating the cause stops. That is the first sign to “add a measurement.” Conversely, add a sensor before you hit that wall and, with no habit of looking, you end up just glancing at it. The order is backwards.

This order works cleanly in a closed environment like a leafy-greens PFAL. The variables are relatively few, and the daily log and electricity bill catch much of it. Conversely, in a greenhouse jerked around by outside air and sunlight, or with fruiting-vegetable crops and their many variables, there are situations where you cannot even begin without several measurements from the start. Take what I am describing as a closed-environment story.

So when a vendor proposal comes at the board meeting, the first thing I want to confirm is not “yield goes up.” It is whether there is, first, “a gap we cannot currently explain and are struggling with.” If the line is that the trouble comes first and this measurement is needed to isolate it, it is worth the money. If the trouble has not even been put into words yet and the pitch comes as just “install it and it goes up,” no one will look once it is in, so I usually pass on adopting it.

One more thing I look at in an investment decision is whether “who reads the added measurement, and when” can be put on the rails of how you run things. A measurement small enough to add one line to the fifteen-minute morning slot will last. Anything that you can only see by opening a separate dedicated screen every day, however high-functioning, will eventually go unwatched. This too I have seen many times on the floor. The decision to add a measurement and the work of building the habit of looking are one continuous thing. A new measurement, too, only comes alive once it fits inside the frame of “who looks, and when.”

Trouble comes first, then adding a measurement works. There are examples with clear numbers for this. In irrigation, there is a report of a commercial-farm trial in fruiting vegetables that switched from watering on a fixed timer to watering only when a sensor reading of substrate moisture showed it was needed. It was a Korean case: coir-substrate hydroponics in a commercial greenhouse, growing tomato and strawberry. For tomato, water use was cut to roughly one-third to one-sixth of the timer method with almost no change in yield (see 8). For strawberry the reduction was not as large, cut to roughly sixty percent of the timer method (see 7). Fertilizer cost was cut by roughly forty to sixty percent in both. The crops and facilities differ from your vertical farm, but what to take is less the numbers themselves than the order: “it worked because they added a single point to measure the trouble of overwatering they could not see.” That order does not change for leafy greens.

Carry the floor’s sense of unease up to management, and swap out what you read

Then again, before the question of whether to add a measurement, the “trouble” that decision rests on usually starts from a small sense of unease on the floor. Yet something catches as “this is a bit odd” on the floor, and before it can become material for a management decision, it vanishes inside the floor as “well, let’s wait and see.” In my experience, this is where the most slips away. So build just one mechanism to keep that unease from vanishing. When something catches, you do not have to reach a conclusion on the spot; just leave one line in the margin of the daily log: “this bothered me.” Do not judge, just leave a mark. It vanishes because you try to decide that morning whether it is wait-and-see or an accident. Leave it without deciding, and when you look at the electricity bill weekly, “come to think of it, it caught a few times this week” turns dots into a line. Once is fine to wait and see. Repeat it many times and it has already become a problem worth bringing up to management. The floor’s unease does not reach the board meeting as is; it becomes words only once the marks pile up. You put one buffer in between.

One more thing: the lines grow and you end up back to skimming. Does that happen? Treat this as unavoidable. Whenever you add a measurement, always ask whether you can drop one of the old lines. The read-aloud slot is fixed at fifteen minutes, so the line count does not grow either. Needing a new number must mean one of the ones you have been reading every morning has finished its job as a yardstick. If the temperature you lined up early on is now “around here,” stable and long unbroken, there is no need to say it out loud every morning. You can drop it down to an occasional weekly check. Don’t grow the slot; swap the contents. Otherwise it grows one line at a time and turns into a long read that no one looks at.

So the call to add and the call to drop come as a set. As much as whether a measurement is worth adding, ask each time what you can stop reading. Precisely because the slot is finite, only what is truly worth seeing every morning survives.

Up to here the story has been “measurement for isolating a cause comes after you hit the wall,” but on a separate track, there are measurements you keep running from the start. Abnormalities that are too late to fix and beyond recovery by the time you notice, like a failure or a disease, will not be caught in time by the normal-times read-aloud. Measurements that monitor continuously and raise an alert are kept in place without waiting for trouble. And then there is calibrating the yardstick itself. If a sensor returns a skewed value, even the normal range you took the trouble to write down gets distorted, and you end up manufacturing the very “chasing after” this article dislikes, on the data side. That the equipment is putting out correct values should be re-confirmed periodically, as a prerequisite of running things. Along with that: update and hand off the sheet with the range written on it each time the staff changes, and at night or when short-staffed, when no human eye reaches, do not try to fill it with manual operation but leave it to continuous monitoring. These two are also classic places where person-dependent operation falls apart, so it is easier to decide them from the start.

The stance of choosing what to read and what to drop to fit your own constraints is echoed in the research on choosing technology. Agricultural IoT is widely used for many purposes, management, monitoring, control, but no single technology or configuration is optimal for every situation. So the user has no choice but to select to fit their own constraints, the review concludes (see 9). There is no all-purpose, correct set that drops in from outside.

It starts not with a sensor but with a sheet of paper

“We haven’t put in sensors or IoT in the first place, so using data is still a long way off.” Do you feel that and stall at the entrance? But it starts not with a sensor but with a sheet of paper. If you are recording nothing yet, before you go buy IoT, pick just one number you already care about and try writing it down at the same time every day. For most floors, the thermometer reading and harvest volume, or even just the electricity bill that arrives every month, is fine. The easiest to start with is the electricity bill. It arrives as an invoice every month, so all you do is change “glance at it and file it away” into “write this month’s number next to last month’s, lined up in one place.” That is already proper use of data. The question “who looks, and when” begins the moment you have two numbers to compare. There is no need to wait for the system to come together.

So those stalled at the entrance are not, in fact, before the starting line. The electricity bill is surely on hand, and you may have delivery slips and a thermometer on the wall too. What there is to do is not to obtain data but to line up what you are throwing away and decide to look at it on a set day. Start with one number, one person in charge, one set time. Add the second number once the first one has become a yardstick.

The ability to read the numbers you line up grows even when you narrow the items. For instance, what I often watched on the PFAL floor was the share of waste from trimming. End it at “lots of waste today” and nothing remains; but from that same number you can read several facets, the way the waste came out, the care taken in that day’s work, a flaw somewhere in the cultivation process, whether the plants are packed too tightly. Even with one number, just lining it up against the normal range and asking “why is it high today” changes the range of moves you can make. Before pricier equipment, this way of reading works.

Last, an honest word about where to draw the line. What running the records on hand can cover is the range of “notice the difference and act on it yourself.” The temperature is different from usual, the door was open, the water supply was left running, cases where the cause is inside your own daily operation and, once the difference is visible, you can reach out and fix it. That is where this method works best, and it is wider than most people think.

There are two places where you let go. One is the wall of isolating a cause. The difference is visible, but the records on hand simply cannot separate the cause, and guessing wrong genuinely costs money. That is the first point to add a measurement, and, if possible, to bring in an expert who reads such data for a living. At that stage, the question has moved from operation to technology. The other is anything touching safety or the equipment itself, electrical capacity, wiring, chemicals, structure. There you run no experiments with your own yardstick at all. Leave it to an expert, or route it to an investment decision about equipment. Because the price of failure there is not “the harvest drops” but something beyond recovery. The rule of thumb is this. If reading the numbers tells you “what to do,” keep that inside your own frame. If reading the numbers tells you only “someone who knows more than me is needed,” that is the boundary. And noticing that boundary early is itself one of the things the habit of looking every day teaches you.

That you don’t have to buy a pricey system right away is not just a matter of floor instinct. Commercial smart-farming systems hit a high-cost wall to adoption, and in their place, low-cost equipment and home-built setups that farmers assemble themselves are being examined as alternative means of spreading them, one report finds (see 4, 10). The way in does not have to be a large investment — and that is genuinely treated as a real option.

For a deeper compilation of monetization know-how, there is also 172 hints to raise the profitability of a vertical farm. Once you can turn the records on hand into a yardstick, read it as your next step.