The data layer AI in smallholder farming has to sit on

This is the part the current thesis rests on.

The platform held field location, field size and a field-mapping output, and the aircraft generated GPS and operational data on every job. The raw material was all there. The current work turns it into an object: a measured geometry, not a number typed into a booking form, with an identifier that persists and a boundary that can be checked against another.

It's a smaller step than it looks. The service already flies over the field, already knows where it flew, already records what it applied. Everything the parcel needs is being generated on every job.

Two things changed, and together they invert the problem.

Field boundary data can now be moved. Through the XAG partnership we reach boundary geometry from their platform, and we've built an extraction and mapping pipeline against it. Partnerships with local drone manufacturers mean this doesn't rest on one vendor.

This is the bridge between the two ventures. The earlier work proved the service, the operator network and the farmer relationship. The mapping pipeline turns the data that service was already generating into something that lasts. Which makes the real product question askable for the first time. What should the Digital Farm Parcel hold, and what should it let you do?

Farmers respond to yield. Show them what their land could produce and they engage. Nobody asks for a mapping platform, a farm record or a digital service. They ask what their land could grow.

Answering that takes more than interest. To tell one farmer what their specific field could yield, you need what that field did before, what was put on it, and what came of it. That's a history, and a history needs something stable to attach to.

That's the opening this venture takes. Not a gap in cadastral coverage, but a service we could be delivering to farmers who already want it.

The common framing, that APAC needs cadastral digitisation, misreads it, and the evidence is unambiguous about why. Three layers are evolving independently.

Vietnam has been building a national Multi-Purpose Land Information System as the platform for a National Land Database, financed in part through a multilateral development bank's VILG project across 30 provinces. Vietnam holds more than 100 million land parcels.

In parallel, FAO, under a donor-funded project with ITC, University of Twente, developed an AI workflow to trace crop field boundaries from free Sentinel-2 imagery, and tested it in Cambodia and Vietnam. The open dataset that came out of it, AI4SmallFarms, holds 439,001 hand-digitised field polygons across 62 tiles.

That's the strongest single piece of evidence here. A land database isn't a field layer, and the space between them is where this venture lives.

Indonesia's land registration programme has registered over 60 million parcels in eight years, created more than 38,000 jobs and grown the number of private surveying firms twelvefold. World Bank financing alone accounted for over 8 million parcels, 42 percent registered to women. A follow-on project targets the land rights of more than 11 million people.

The World Bank has separately identified fragmented and incomplete land information as a constraint on land governance in Indonesia.

Proposing cadastral mapping to Indonesia would be close to insulting. The opportunity there is making spatial infrastructure that already exists answer agricultural questions it was never built for.

The World Bank-financed SPLIT project splits collective land ownership awards into individual titles. As of July 2025, 163,176 individual electronic titles had been issued, 84 percent in female names. On completion around 750,000 agrarian reform beneficiaries are expected to gain individual rights over more than 1.3 million hectares.

Project reporting notes cadastral surveys were available for only 20 percent of collective CLOAs, with roughly a further million hectares still to survey.

Structurally different from Indonesia. Here an unfinished cadastral problem sits directly inside the agricultural system, which changes both the partner set and the order you do things in.

Of an estimated 3.5 million land parcels in Lao PDR, around 1.5 million had been registered and titled as of 2021, mostly in towns and cities, against a government target of a further 1.2 million. Development bank reporting shows titles or use rights in the Lao LandReg system rising from 482,233 at baseline to 1,203,331 by September 2023, with the system running in 32 districts against a target of 100.

As of October 2025, Lao LandReg was active in 8 provinces with 10 more planned by June 2026, over 430,000 paper land files had been scanned in Vientiane with 95,000 entries digitised, and a decision had been taken to centralise the system at the government data centre.

Laos is the cleanest illustration of the two-speed problem. Land digitisation is moving while farm parcel measurement is still being piloted. Small market, high development-partner interest, useful as a proof site.

The Digital India Land Records Modernization Programme, launched in 2016 and extended to 2025-26, gives each land parcel a 14-digit unique identification number based on the coordinates of the parcel's corners. As of November 2025, more than 360 million parcels across 29 states and union territories had received one, covering roughly 66 percent of agricultural land.

Land remains a state subject under the Indian constitution, and implementation varies a lot by state.

India doesn't have a missing-identifier problem. It has a spatial accuracy, interoperability, currency and state-variance problem. That makes it a platform and integration market and not a mapping market, and it's big enough that treating it as one country for go-to-market would be a mistake.

The ratings below are our synthesis. There's no published maturity index for this. Each cell is a judgement about what the venture runs into in that market.

Vietnam is the launch market. It has the technical gap, the venture history, the XAG and local manufacturer relationships, and, decisively, a national policy programme creating institutional demand for exactly this object. That last point is section 05.

Vietnamese family farms average around 0.4 hectares. In the Mekong Delta the average parcel has been measured at 0.14 hectares, with over 80 percent of farmers holding under one hectare. Nationally there are roughly 75 million parcels, seven to eight plots per farm household. Vietnam's farmland per person, around 0.3 hectares, is among the lowest in the world.

Research on field boundary tracing describes the Vietnam and Cambodia landscapes as highly fragmented, with sub-hectare paddies separated by narrow earth dykes, and notes that frequent cloud forces reliance on rare clear-sky days.

The AI4SmallFarms work notes that Sentinel-2's 10-metre resolution is a problem given how small the fields are, and its published results show higher-resolution imagery catching small fields Sentinel-2 misses.

Here's the crux, plainly. The question isn't whether satellites can see agriculture. Obviously they can. The question is whether pixels can be turned into reliable individual field shapes at the scale of fragmented smallholder farming, and then whether those shapes can be attached to a farmer, a crop, a history and an economic activity.

A drone flying a treatment mission is physically present at ten to thirty centimetre resolution with satellite-corrected positioning. It resolves what a 10-metre pixel can't.

So the argument for service-generated boundaries is technical before it's commercial. The service is on the field, and being on the field is what produces geometry good enough to be worth keeping.

Satellite imagery earns its place in three specific roles.

DJI's SmartFarm platform automatically identifies farmland borders using AI to get the area of each parcel, and records the operation information of each parcel to build farmland management archives. It also does field scouting, growth detection, prescription maps and 3D route planning for orchard terrain.

DJI's own user guide states that field management is currently independent of field planning, that at present it's for area measurement and yield analysis, and that in future field management will be integrated with operation data, yield data, seedling data, agricultural notes and pesticide records.

That second citation is worth reading twice. The largest agricultural drone manufacturer in the world documents that its boundary object and its operation history aren't yet joined. The join is the product. And it's stated as roadmap by the incumbent whose interest is a single-vendor stack.

We have a partnership with XAG and access to boundary data from their platform, with an extraction pipeline and internal data mapping built against it. We also hold partnerships with local drone manufacturers in the region.

The multi-vendor position is what makes this defensible. A venture whose only asset is a pipeline into one manufacturer's system is a feature that manufacturer can switch off. A venture that normalises boundaries and history across XAG, local manufacturers and, where farmers permit, DJI is doing something no manufacturer will do for a competitor's customer.

The clearest way to explain vendor-neutrality to a non-technical audience is the cooperative that uses two brands of drone. Half its members' fields are described in one platform, half in another, and neither platform will ever hold the whole picture. So the cooperative can't answer a question about its own land, even though every individual field is mapped somewhere.

Four consequences, and they compound.

The earlier venture had to manufacture demand through demo events and loyalty caps. This programme generates it. The venture's job is to be the infrastructure the programme is missing.

Whether cooperatives currently feel traceability as a painful unmet need, or whether commune-level verification works fine on paper and spreadsheets, is the first thing to test. Policy requiring something isn't the same as an operator feeling the cost of it. That test is in section 11.

A Digital Farm Parcel is a lasting, uniquely identified record of a piece of cultivated land, with a few defining properties.

Three other candidates were on the table.

The parcel is the smallest thing that's stable over time, meaningful to a farmer, addressable by a machine, and the natural key for every product downstream.

Identity rules, chosen so the record survives the messiness of real land.

That last rule is the one people argue about, so here's the reason. A cultivated field and a legal parcel routinely disagree. Farmers cultivate across boundaries, rent adjacent land, leave margins uncultivated, subdivide informally. Force them into equivalence and you corrupt both.

Six states, with the moves between them defined.

Every geometry enters the system on a source tier. The tier travels with it permanently and governs what it may be used for. Accuracy figures below are typical ranges from published instrument specifications.

The critical distinction is P3 against P4. A spray footprint records where the drone flew, which isn't where the field is. It leaves out headlands, margins, obstacles and anything the operator skipped. Treat them as interchangeable and the system quietly fills up with fields that are all about a fifth too small.

When two geometries claim the same land, the system doesn't pick a winner and throw the other away. It records the disagreement.

The reason there's a queue and not automatic resolution is the standard irony of automation. Resolve the easy conflicts automatically and the leftover ones get rarer and stranger, so they have to surface somewhere a person is looking.

Every parcel carries a confidence score built from source tier, how old the last verification is, how many independent confirmations there are, and how much the sources disagree. Confidence shows up anywhere the geometry is used, and it gates the products. A finance or emissions claim needs a floor that a service-planning query doesn't.

These two get collapsed into each other constantly. Walking backwards separates them. One is a question about a place, which the parcel solves. The other is a question about a photograph, which the parcel does nothing for.

The four buildable applications all depend on the same two things: the parcel, and the operation record attached to it. Neither is an intelligence product. So the build order puts the substrate first and the models second.

Whether an application needing seasons of history can survive the wait gets answered in the validation plan.

Farmers own their boundary data. We discuss boundaries with them directly and hold an anonymised version in the system for aggregate questions like crop and yield distribution.

That instinct is right, and it's made structural here.

A vendor partnership grants us access to a vendor's system. It can't grant us a farmer's consent, because the farmer's consent went to the vendor. Moving a farmer's geometry into a new platform with new purposes needs its own basis, from the farmer.

Anonymising doesn't solve this. A field boundary plus a commune resolves to a household across most of the Mekong Delta, where holdings average 0.4 hectares and neighbours know each other's land. Boundary geometry identifies people by default. The anonymised aggregate layer is fine for crop and yield distribution questions.

The consent conversation sits at first field contact, run by the operator, in person, and it's the same conversation as the yield conversation. That's a design decision with a commercial reason behind it: the farmer is being told what their land could produce, and the reason we can tell them is that we'll keep a record of this field. The permission and the value proposition are the same sentence.

Consent is recorded per parcel, per purpose, and can be withdrawn. Withdrawing removes the farmer's identity link and takes the geometry out of active use, while keeping the operation record the service provider needs for their own accounts.

The architecture separates three things that commonly and dangerously get merged.

The commercial products sit almost entirely in the third band. Making that explicit early is what lets the first band be genuinely farmer-owned without hollowing out the business model.

Eleven entities. The fields listed are the ones carrying design weight.

Five rules that govern the whole system.

The system is a log with views built over it. The reasons are specific.

Aggregate queries return results only above a minimum cell size, set by parcel count and distinct-farmer count inside an administrative unit. Below that the query returns nothing instead of a rounded figure. Rounding leaks, suppression doesn't.

That's what the platform stores. What follows is what a farmer or cooperative gets, because a data model isn't a product.

A record of one field that a person can read on a phone in under a minute. Anything longer won't get read.

This is the surface that opens every relationship, so it gets designed first.

The comparison has to be real and local. A national average doesn't persuade someone who knows their own delta. The published cooperative results in section 05 are the starting comparison, and they're strong precisely because they're field-level and nearby.

Whether a farmer acts on a yield comparison or just finds it interesting gets tested in section 11. Interest is already observed.

What a cooperative director gets, which is a different product from what a farmer gets.

The second and third items are what make this a purchase. A cooperative in a programme with traceability requirements has an obligation it currently discharges on paper.

Derived. Production area, variety, season, water management, fertiliser, residue handling, all keyed to a parcel and drawn from operations already recorded. The farming-diary problem from the earlier venture, growers skipping a long manual task that buyers needed, is solved by generating the diary from operations instead of asking anyone to keep one.

Buyers interviewed during the earlier venture said they'd use a platform that automated diary entry for residue level and crop protection use. That's directly relevant now.

Reconstructed from the pilot and current operating knowledge, described as method.

The same job, with three additions that cost the operator almost nothing and change what the system knows.

The design constraint is that the operator's day must not get longer. Every extra step has to be automatic or under thirty seconds, because operators are paid by hectares covered and will abandon anything that costs them a job.

Drone operators won't take small isolated jobs, because transport and setup dominate the economics of a single sub-hectare field.

That's why the cooperative is the entry unit, not the individual farmer. A cooperative gathers adjacent parcels into a single visit. Where the earlier venture proposed a scheduling tool to group nearby bookings after the fact, the cooperative provides the grouping structurally.

Illustrative arithmetic on the pilot's own figures, to show the shape. At the pilot's rice spraying price of USD 7.9 per hectare, a single 1.3-hectare field yields around USD 10 of revenue against a trip. Twenty adjacent parcels averaging 1.3 hectares yield around USD 205 against one trip. The pilot noted a provider with five drones could cover up to 50 hectares a day. Density is the entire economic argument.

Seven components, each with an owner and the way it goes wrong.

Cooperatives enrolled in the One Million Hectare programme in the Mekong Delta, starting with the provinces the pilot already scouted: Dong Thap, An Giang, Kien Giang, Long An, Tay Ninh.

That last point is a targeting signal. What's observed so far is that farmers engage when shown yield potential.

Three parties, each of which gains something they currently lack.

The earlier venture had already developed a scheme where large crop protection orders could earn a drone, and planned to help those customers become service providers.

Four streams, in the order they can credibly be turned on.

Streams 3 and 4 justify the infrastructure investment. Stream 1 pays for being on the field. Anyone pitching this as a drone booking business has misread it.

Willingness to pay gets tested at every level in section 11.

Gate three does the most work in this domain, and it should. The whole premise of this venture is that the data doesn't exist yet, so anything assuming it does is the fastest way to fail.

Three bands, held apart. The most consequential single decision in the system.

The test: the platform stays correct and usable with the intelligence band switched off. Boundaries would come only from instruments, records would still assemble, compliance would still generate.

This matters commercially as much as technically. A traceability claim you can't explain is worth nothing to a verifier, and a yield figure a farmer can't interrogate is worth nothing to the farmer.

Seven surfaces. Each is named by the question it answers, because a surface you can't describe as a question doesn't need to exist.

S6 is the surface that makes the rest defensible, and the one most likely to be cut for looking unglamorous. It belongs in the first phase.

Three horizons, sequenced against the programme calendar and the growing season instead of engineering preference. Winter-spring is the natural pilot window, as it was in the pilot.

Running the models quietly in Horizon 1 is deliberate. Every model here can be checked against ground truth before anyone relies on it, because the drone is measuring the same thing the model is predicting. Few domains offer that, and it would be wasteful not to use it.