Reconciling national land cover figures with FAOSTAT land use data

If you've ever sat down to fill in the SDG 15.3.1 reporting template and found your ministry's cropland figure sitting a few percentage points away from what FAOSTAT shows for the same year, you already know the problem. That gap is two different measurement systems answering the same question, and nobody upstream flagged that they measure different things.

Why the two numbers never started out equal

FAOSTAT's land use domain is built mostly from what countries report to FAO through questionnaires, census returns and administrative records. It's a use-based classification: agricultural land, arable land, permanent meadows and pasture, forest land. Those categories describe what the land is used for, as declared by an institution.

A national land cover layer built from satellite imagery is a cover-based classification. It describes what's physically on the ground at the time of the scene: tree canopy, bare soil, built surface, grass. Two parcels can carry the exact same land use declaration and look completely different from orbit, and the reverse happens constantly too: fallow cropland reads as grassland in a cover classification even though the cadastre still lists it as arable.

Add to that the IPCC's six land use categories (Forest Land, Cropland, Grassland, Wetlands, Settlements, Other Land), which most national greenhouse gas inventories and a lot of SDG reporting now leans on, and you've got three vocabularies in the same room: FAOSTAT's use categories, your cover classification's legend, and the IPCC's inventory classes. None of them were designed to nest inside each other cleanly.

Building a crosswalk you can actually defend to an auditor

The fix is building and documenting a concordance table that states, explicitly, which cover classes roll up into which use category, and where the mapping is many-to-one or genuinely ambiguous.

A few things that make this hold up under review rather than fall apart on the first question:

• Freeze the reference year before you compare anything. FAOSTAT figures often lag by a year or two depending on when a country submitted its questionnaire. Comparing a 2023 cover classification against a 2021 FAOSTAT entry and calling the gap "error" is a documentation failure: the reference years were never aligned in the first place.
• Write down every many-to-one decision once, not case by case. If your "shrubland" class sometimes becomes FAOSTAT grassland and sometimes becomes forest land depending on canopy threshold, that split needs a stated rule, not an analyst's judgment call repeated differently each cycle.
• Keep the crosswalk itself versioned. When your classification legend changes (and it will, as sensors and classifiers get refreshed), the crosswalk has to be re-validated against the new legend, not inherited silently from the old one.
• Separate definitional drift from actual land-cover change. A jump in "cropland" between two reporting cycles can mean real conversion, or it can mean your classifier caught more smallholder plots than the previous one did. If you can't tell which from the data alone, that's the footnote, not a number to quietly smooth over.

A classification that gets rebuilt from scratch every cycle, by whoever happens to hold the contract that year, turns this into a moving target. If the legend, the thresholds and the sensor source shift between commissions, your crosswalk has to be rebuilt from scratch too, and the SDG 15.3.1 baseline you reported three cycles ago stops being comparable to this year's figure for reasons that have nothing to do with the land. A land-cover change layer that stays on one consistent classification year over year lets you reuse the same crosswalk and attribute any divergence from FAOSTAT to the land use side rather than to shifts in your own pipeline.

When the numbers still won't line up

Sometimes they won't, and that's fine to say outright. FAO's own guidance on land use/land cover harmonization (used by UNCCD and the Trends.Earth toolset for SDG 15.3.1) accepts that Tier 1 default data and nationally produced Tier 2 or 3 figures can diverge, and expects countries to document which tier they're reporting under and why. Put that statement in your metadata, cite the crosswalk version, and move on. An auditable gap beats a reconciled number nobody can explain.

If your ministry is due for a refresh of the national land-cover baseline, it's worth building the next one around a classification that doesn't need re-crosswalking from zero every reporting cycle.