Insight, Measurement & Agriculture

The Cheaper It Gets to Verify, the Less Anyone Visits

Satellite radar can now confirm whether a rice paddy was flooded, plot by plot, across millions of fields. That is a real advance. It also removes the last budgeted reason anybody had to go and look at the farm.

Line-art diagram: a satellite in orbit at the top left projects a cone of radar signal down onto a grid of green rice paddies below; a dashed arrow leads to a laptop in the centre showing a plot map with a green tick and a rising line chart; a further dashed arrow leads to a faded figure of a farmer with a clipboard standing on a dashed-outline plot on the right, rendered semi-transparent to indicate absence.
The satellite confirms the practice. The dashboard bills against it. The person on the right, whose visit produced almost everything anyone knew about the programme, is being drawn in dashed lines.

What falling measurement costs quietly take away

Mitti Labs has raised around US$9.5 million to expand from India into Indonesia and the Philippines. Its platform combines satellite radar, field observations and physical models to build plot-level digital twins of rice fields, supporting water management and methane reduction interventions alongside farmers.

The technical proposition is sound and the problem it addresses is real. Synthetic aperture radar can detect standing water on a paddy through cloud cover, at night, repeatedly, which is exactly the observation that alternate wetting and drying requires somebody to make. Doing it plot by plot across millions of smallholdings, rather than by sampling a few hundred, collapses a cost that has blocked smallholder carbon programmes for two decades.

Nobody should be dismissive about this. Verification cost has been the binding constraint on paying smallholders for practice change, and this class of tool genuinely removes it.

Two-panel line-art diagram titled What a satellite records, and what it does not. Left panel titled What the radar sees: an overhead view of a single rice paddy with regular rows of shoots on a sky-blue flooded ground, and a small legend chip labelled flooded / not flooded with a coral tick beside it. Right panel titled What the radar does not see: a paddy edge with three figures standing at the bank, a farmer holding a labelled container of pesticide, a woman with a wage slip in hand and a small child, each with a small grey question mark hovering above their head. Footer: The satellite answers the funder's question. It does not answer the household's.
The left panel is what a payment can be tied to. The right panel is what a household lives with, and the two panels are the reason a satellite alone is not enough.

What follows from removing it is less obvious, and it is the reason to write about this at all.


What the satellite sees, and what it cannot

Be precise about the observation. Radar confirms that water was on the field, or was not, on a given date.

That is verification of a practice, not of an outcome, and of a field, not of a household.

Observed from orbit, and not observed at all

Observed. Water present on the field. Date and duration of flooding. Field boundary and area. Cropping calendar. Change against last season. Cheap, repeatable, at continental scale, and increasingly accurate.

Not observed. Methane, and nitrous oxide at all. Who controlled the water. Whose labour did the extra work. Whether the farmer had a choice. What was given up to comply. Why a farmer stopped. Only obtainable by asking, on the plot.

The left column is what the credit is issued against. The right column is what determines whether it keeps being earned.

The emissions figure itself is modelled, not measured. Radar sees flooding; an emissions factor converts flooding into methane; and the nitrous oxide increase that partially offsets the methane reduction is not observable from orbit under any circumstances. A digital twin is a model wearing the vocabulary of observation, and the assumptions inside it are doing a great deal of the work.


Legibility, and what it costs

James Scott's account of legibility is the right frame here. Administering anything at scale requires simplifying it into categories a distant office can process, and those simplifications then become the reality decisions are made on. What is not in the register does not exist administratively. The local practical knowledge that does not fit the categories, what Scott called métis, is not merely omitted; it becomes invisible to the institution, and its absence is not noticed because there is no slot in the form where it would have gone.

A plot-level digital twin is an unusually clean example. It is a register of rice fields legible to a carbon buyer in another jurisdiction, and it is a considerable achievement. Everything in the right-hand column above is métis.


The mechanism worth naming

Here is the specific dynamic, and it is not a criticism of anybody's technology.

Field visits in smallholder programmes have historically been funded because verification required them. Somebody had to go and check. That was the budget line, the justification, and the reason the visit happened.

Those visits also produced, incidentally, almost everything anyone knew about how a programme was actually landing. The disputes with neighbours over water. The household that dropped out and why. The fact that the extra monitoring labour fell to women. The season when the canal ran dry and the practice was abandoned quietly. None of that was the purpose of the visit. All of it came back with the enumerator.

Remove the verification rationale and the incidental function goes with it, unless somebody funds it deliberately and on its own terms. Nobody usually does, because it was never a line item to begin with. It was a by-product.

A programme can therefore become perfectly verified and completely misunderstood. Every hectare confirmed, every credit issued, an audit trail nobody can fault, and no one in the chain able to say whether the practice is sustainable for the household, who is bearing its cost, or what will happen in the first bad year.

This is the same failure of instrument we described in The Benefits Nobody Was Looking For, arriving from the opposite direction. There, a real effect went unmeasured because no indicator asked for it. Here, an excellent indicator arrives and removes the reason anybody would have found the effect by accident.


The counter-argument, which is strong

The obvious response is that cheap remote verification frees resources rather than destroying them. If confirming the practice costs a hundredth of what it did, the saved budget can fund better social research than was ever affordable before.

That is entirely possible and it is the outcome to aim for. It requires only one thing: that somebody decides to spend the saving that way rather than banking it as a lower cost per tonne.

The competitive pressure runs the other way. In a carbon market, cost per verified tonne is the number buyers compare, and a programme that spends its verification saving on household research will lose on that metric to one that does not. This is not a prediction of bad behaviour. It is what the metric selects for.


What to do about it, specifically

Decouple the field budget from the verification budget.

The field research allocation should be a fixed proportion of programme value, not a residual of assurance need. If it is a residual, it falls automatically as remote sensing improves, which is precisely backwards: the social evidence becomes more valuable as the technical evidence gets cheaper, because it is the only remaining source of everything the satellite cannot produce.

Buy a panel, not a survey.

The questions that matter here are longitudinal. Does the household still practise this in season five, after the payment structure changes. A standing panel of a few hundred farms, visited repeatedly, costs a fraction of the programme and answers the only question anybody will care about in 2032.

Publish the discontinuation rate.

Farmers leaving is the single most informative statistic a programme of this kind produces and the one least likely to be reported. It is also, incidentally, detectable in the satellite data, which is a genuine opportunity: the same tool that removes the reason to visit can identify exactly which plots are worth visiting.

That last point is the constructive version of this whole argument. Remote sensing should not replace field research. It should target it, by finding the anomalies, the dropouts and the places where the model and the practice diverge, and sending somebody there.

The Lab works on this in regenerative agriculture and through field research designed to sit alongside technical measurement rather than under it. Our work in Lombok runs on the same question of what a programme means to the household it lands on.

If you are building or funding an MRV platform and want the field component designed as a complement rather than a casualty, tell us what you need to know.


Sources

  • Mitti Labs, Series A funding to expand rice-MRV platform into Indonesia and the Philippines, August 2026 (round size and geographic scope are as reported by the company).
  • Scott, J. C. (1998), Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed, Yale University Press.

This is an independent insight piece by Transitions Lab. For the Lab's applied work, see Regenerative Agriculture & Land Systems and Field Research. See also When the Agent Pays on the same disappearing-visit dynamic when an agent transacts on the user's behalf. To discuss a study, see Contact.

Read more Articles & insights See all articles → See it in the field Case studies See all case studies →