Reconciling sub-national land-cover change figures into one national total

Every reporting cycle starts the same way. Each province or district sends in a land-cover change figure, usually tied to whatever study or contractor covered that area that year, and someone in the statistics unit has to make those numbers add up to a single national total for the SDG indicator narrative, or the forest reference level submission, or the national communication to UNFCCC. They almost never add up cleanly on the first pass.
The mismatch is a methods problem, and it shows up the same way across most national statistical offices we've talked to.
Why the provincial numbers won't just sum
The most common cause is legend mismatch. One province's classification scheme calls a mixed shrub-and-grass transition "degraded forest," another calls the same ground cover "other wooded land," and a third doesn't have a category for it at all, so it gets folded into "other land." Sum the three provincial totals and you've either double-counted that transition zone or dropped it entirely, depending on which direction the misclassification ran.
Reference dates are the second problem. If one province's baseline imagery is from a 2019 dry-season acquisition and the neighboring province used 2021 wet-season imagery, the change figures aren't measuring the same interval, even though both get reported as "this cycle's change." Mixed sensor years compound this: a province mapped from one satellite constellation at 10 m resolution next to one mapped at 30 m from an older archive will show systematically different minimum mapping units, which shows up in the totals as noise that looks like real change.
Then there's the boundary seam itself. Adjacent provincial studies rarely use identical administrative boundary files, so a strip of land along the border gets counted in both totals, or in neither, depending on which shapefile each contractor pulled from the gazetteer that year.
None of these are errors in the individual provincial studies. Each one can be internally sound and still produce a national total that won't hold up when a reviewer traces it back to source.
A rollup method that survives the audit trail
The fix works upstream of the aggregation stage: removing the conditions that cause the mismatch before you ever sum anything.
Start with a single classification legend applied to the whole territory, not a crosswalk table mapping twelve provincial legends onto one national scheme after the fact. Crosswalks are defensible in a methods annex, but every crosswalk step is a place where a reviewer can ask "how was this reclassified, and by whom" and you need an answer on hand.
Second, pin every province to the same reference dates for the baseline and the comparison year. If the national total covers 2015 to 2025, every sub-national figure feeding it needs to be measuring that same ten-year interval, not whatever interval each original provincial study happened to use.
Third, run the classification over the administrative boundaries as a single continuous surface, then cut the result into provinces for reporting, rather than classifying province by province and stitching the edges afterward. That's the only way to guarantee the border strip gets counted exactly once.
This is the logic behind a consistent, annually-refreshed layer run over the full national territory from wide-swath imagery: one classification pass that gets cut into sub-national figures instead of sub-national studies that get added up. The provincial totals still exist, since districts and ministries of agriculture need them for their own purposes, but they're drawn from the same source as the national number rather than feeding into it as separate inputs.
What this means for the next reporting cycle
If your national total currently comes from reconciling commissioned studies after the fact, the honest fix sits upstream of the spreadsheet: get every sub-national figure from the same legend, the same reference dates, and the same continuous classification surface, so the rollup is addition rather than negotiation.
That's a different commissioning decision than most ministries are used to making, and it's worth asking whether your next cycle's land-cover change layer is built this way before the reconciliation problem shows up again.
If your ministry is weighing that decision for the next reporting cycle, Habitat Loss Mapping is built around exactly this problem.