Pick a place and a pair of survey cycles: read the ledger of what its farmland became — acre by acre, class by class — and score what the next project would cost.
Since 1982, state law has required the Department of Conservation to report farmland conversion biennially. Twenty vintages, 1984 to 2022, are published openly — plus 716,776 polygons that each record the cycle they became urban and what they were before.
The state's own front end to that record makes you draw a polygon, submit a geoprocessing job and wait — and it answers "what is inside this shape now". Not what was here, and not what it became.
Meanwhile the digital-agriculture platforms serve the grower, in season, on land they already farm. The nearest thing in the GIS market serves the assessor, valuing a parcel. Neither does conversion accounting.
The statewide crop survey gives every field a stable id that survives across vintages. So a true per-field transition matrix is not a spatial overlay at all — it is two attribute queries and a join. A whole county resolves in 7.6 seconds.
The state's agricultural significance model is a fixed 100-point weighted sum written in 1993 and published in 1997. Every input is now open: soil capability and the Storie Index from the federal soil service — keyless, and readable straight from a browser in 0.7 s.
Both calculations are performed today — by hand, per project, in a Word document, with a screenshot pasted in. Nobody had put either one on an open map.
From-class × to-class, in acres, between any two survey cycles. Click a cell and the map, the table and the balance narrow to exactly those acres.
The full vintage axis with the methodology breaks marked — so you can see which comparisons the record actually supports.
Out of active agriculture, into it, and the net — the line a staff report actually needs, with the method printed beside it.
The statutory 100-point model computed live for any field, every factor printing its source and its branch.
A county planner, a groundwater agency, a land trust and an environmental consultant can all read the same screen — and each one can trace every number back to the survey that produced it.
Madera County, 2018 → 2023. 10,229 and 10,389 fields, 9,437 matched across both surveys, in 7.6 seconds.
4,022 acres of vineyard replanted to orchard. 5,198 acres of 2018's young plantings now bearing. 4,431 acres of orchard pulled to idle — and 3,506 coming back the other way.
New plantings in the same county fell from 7,200 acres to 1,315. That is a permanent-crop cycle turning, visible five years before it shows up in a yield statistic.
| Orch | Vine | Young | Field | Past | Idle | |
|---|---|---|---|---|---|---|
| Orchard | · | 272 | 606 | 460 | 115 | 4,431 |
| Vineyard | 4,022 | · | 367 | — | 389 | 2,338 |
| Young per. | 5,198 | 621 | · | — | 15 | 121 |
| Field crops | 2,653 | — | — | · | 3,050 | 278 |
| Pasture | 829 | 195 | — | 4,865 | · | 618 |
| Idle | 3,506 | 192 | — | 4 | 1,016 | · |
The idle class did not exist before 2021. Measured statewide: 7 acres in 2016, zero in 2019 and 2020 — then 813,945. Charted naively that is a 137,000× increase in idle farmland that never happened.
The conversion layer is duplicated. One county returns 5,390 rows carrying 1,057 distinct acreages — a naive sum reports 1,532,533 acres converted in a county with 473,000 acres of land.
The obvious class field is a placeholder. It reads '**' on 93.8% of 446,914 records. Style a map on it and the state renders one flat grey.
A protection code is spelled two ways. Filtering one spelling silently drops 173,699 acres of contracted farmland.
The same letter means two things. In the farmland record D is urban; in the crop survey D is deciduous fruit and nut. Share one legend and orchards render as cities.
The comparability ribbon greys out what cannot be compared, and the method note says why — in the app, in the export, and in the record.
Click any field and the statutory model runs live: soil capability and the Storie Index acreage-weighted from the federal soil service, project size, water availability, and the surrounding agricultural and protected land inside a quarter-mile zone of influence.
The score alone is never the answer — the model's own sub-tests decide, and a tool that prints only the total is not implementing the model.
So we tested it against a study actually filed with a California county in 2023.
Band 40–59 → one sub-score below 20 → not significant. Reproduced by the app's test suite on every run.
| Land capability classification | 25% |
| Storie Index | 25% |
| Project size | 15% |
| Water resources availability | 15% |
| Surrounding agricultural land | 15% |
| Surrounding protected resource land | 5% |
Every factor in the app prints its measured value, its weight, its source and which branch it took — because a number without provenance cannot be defended at a hearing.
24 external feeds, 20 curl-checked live on the build date — every field name, count and acreage in the design measured from a real response. A fabricated endpoint is worse than a blank.
The demo ships a regression suite that re-measures all five traps, the model's lookup tables and the filed worked example against the real services on every run.
Keyless and sovereign: no vendor account, no token, no data leaving your estate. Deployable behind your own sign-on, and re-pointing it to another state or country is configuration, not development.
| The question | What this is not | What it is |
|---|---|---|
| Significance under environmental review | A determination of significance | A screening that computes the statutory model and shows its work — the lead agency decides |
| The authoritative record | A system of record | An accounting surface over the state's published record, which stays the record |
| What is happening in the field | An observation, a yield or a condition product | A reading of official surveys — the fastest input here is biennial, and the app says so |
| Where coverage runs out | Silence presented as "no change" | An explicit "not surveyed here" — including counties where urban land is not delineated at all |
Coexistence, not replacement. The state's viewer stays the authority; this is the surface that makes its record legible, comparable and quotable — and hands the user back to it.
The design started from the decisions four public-sector roles actually have to make — then every endpoint, field name and acreage behind it was checked live before it entered the build.
A GIS admin describes the app; the platform authors the map spec, the layout and the interactivity. No code.
On Strata or on an existing ArcGIS estate. Open basemaps, open standards, sovereign deployment.
Every figure carries its vintage, its method and its source — into the export and into the record.
See it live → tabaqat.net · info@tabaqat.net