Building EUDR traceability: plots, batches and due diligence data

The legal side of the EU Deforestation Regulation fits on a page. The engineering does not. To file a due diligence statement for a shipment you need to answer, with evidence, a question most supply chains were never built to answer: which plots of land did this exact batch come from?

The core data model

Four entities carry most of the weight:

The links between them matter more than the entities. A batch of roasted coffee is made from green-coffee batches, which were made from lots, which came from plots. If any link is missing, the trace breaks and so does the statement.

Plots as geometry

Store boundaries as real geometry, not as text. In Postgres, PostGIS gives you validation, area calculation, overlap checks and spatial indexes.

create extension if not exists postgis;

create table plots (
  id bigserial primary key,
  producer_id bigint not null references producers(id),
  country_code char(2) not null,
  boundary geometry(MultiPolygon, 4326),  -- for plots over 4 ha
  centroid geometry(Point, 4326),          -- enough for small plots
  area_ha numeric(10, 4),
  valid_from timestamptz not null default now(),
  check (boundary is not null or centroid is not null)
);

create index plots_boundary_idx on plots using gist (boundary);

Validate on the way in

Field-collected polygons are messy: self-intersections, duplicate points, swapped latitude and longitude, plots in the sea. Reject or repair them at upload, never at filing time.

-- Problems worth catching before they reach a due diligence statement.
select id,
  st_isvalid(boundary)                                   as valid_shape,
  st_area(boundary::geography) / 10000                   as area_ha,
  st_y(st_centroid(boundary)) between -90 and 90         as lat_ok
from plots
where boundary is not null;

-- Two producers claiming overlapping land is a red flag worth reviewing.
select a.id, b.id
from plots a join plots b on a.id < b.id
where st_overlaps(a.boundary, b.boundary);

Keep history. When a boundary is corrected, add a new version rather than overwriting the old one, so you can always show what you knew when a statement was filed.

Deforestation checks

To show a plot was not deforested after 31 December 2020, intersect it with forest-cover data for 2020 and with forest-loss data since. Public sources include the EU Joint Research Centre's global forest cover map for 2020 and the tree-cover loss data published through Global Forest Watch. Treat satellite results as evidence for a risk assessment, not as a verdict: cloud cover, plantations mapped as forest and small plots near the data's resolution all produce false alarms that need a person to review.

Following batches through processing

Mills, crushers and roasters blend inputs. Record every transformation with its inputs and outputs, and decide your model up front:

Data collection in the field

The data starts with a field officer at a farm. Build for that reality: an offline-first mobile app, GPS capture by walking the boundary, photos of documents, local languages, and sync when the phone finds a signal. Data quality at this step decides everything downstream.

Producing the statement

For each shipment, the system should assemble the statement automatically: product codes, quantities, the country of production and the geolocation of every contributing plot, then submit it to the EU information system and store the reference number against the shipment. Your EU customers will ask for that reference, or for the underlying data, often through an API.

If a batch cannot be traced to its plots in a query, it cannot be traced in an audit either.

Build checklist