Convention needed: disputed-territory attribution in admin (gaul0) zonal extraction — affects all admin-aggregated Atlas tables

[peetmate]

@bjyberg — surfaced via CR-115 (duplicate adm0_obs rows; producer-side tracking issue AdaptationAtlas/hazards_prototype#11) but it is cross-cutting, so flagging for a convention call rather than patching one pipeline.

Problem

At admin0 zonal extraction zone_id = gaul0_code, and disputed-territory slivers each have their own gaul0 code attributed to a claimant country. So a country ends up with multiple adm0 rows — its main polygon + each disputed sliver it claims. E.g. KEN → main 137 + Ilemi 135; both admin1/admin2 = NULL, so a naive WHERE iso3='KEN' returns 2 rows and consumers double-count/plot.

Disputed slivers (hazards_prototype/R/misc/calc_disputed_weights.R): Abyei 100 (SSD/SDN), Bīr Ṭawīl 110 (EGY/SDN), Hala'Ib 133 (EGY/SDN), Ilemi 135 (KEN/SSD). (W. Sahara excluded.)

Why it's a convention, not a fix

Every admin-extracted Atlas table — exposure, hazard, hazard_exposure, observational, … — inherits this. How we attribute disputed territory is politically sensitive and must be consistent Atlas-wide.

Options

1. Drop slivers — country = its main polygon only (disputed area excluded everywhere).
2. Count each sliver for every claimant — politically-neutral; each claimant's national stat = area-weighted (obs-grid cell count) pooled mean/sd over {main + claimed slivers}. Disputed area appears in both countries' stats. (Pete's preferred direction.)
3. Apportion — split the sliver between claimants by some weight (less neutral).

Ask

Which convention should all admin-extraction producers adopt? Once set, the pipeline implements it uniformly (the obs adm0_obs republish + the same collapse in hazard/exposure producers). Until then, consumers keep the client-side GROUP BY / AVG dedup.

Related: STAC/CDH cataloging #2 · producer issue AdaptationAtlas/hazards_prototype#11

[peetmate]

Folding in a related finding (CR-075) — the same disputed-territory slivers also diverge between pipelines, so the convention needs to cover cross-pipeline parity too:

• Observational (CHIRPS/CHIRTS-ERA5) ships the slivers as separate adm0 rows (EGY → 3 adm0 polygons: Bīr Ṭawīl 110 + Egypt 120 + Hala'ib 133; KEN → 2; SDN → 2).
• NEX-GDDP-CMIP6 instead collapses them to admin1 inside the main-country adm0 polygon, and double-attributes Bīr Ṭawīl + Hala'ib to both EGY and SDN admin1 lists.

So beyond "how to attribute," the convention should also fix:

1. adm0 row parity — obs and NEX-GDDP must produce the same number of adm0 rows per iso3.
2. double-attribution rule — Bīr Ṭawīl / Hala'ib are currently in EGY+SDN on one side, EGY-only on the other; pick one consistent rule.

Once the convention lands, both producers re-bake to match row-for-row at adm0.

[bjyberg]

Thought I already commented on this issue, but can't find it, so sorry if it is a duplicate. This has/is a know issue in the Atlas since we switched from the vector based extraction pipeline to the raster admin extraction pipeline. The current fixes are a bit messy and split across a few locations, however there is a function in the hazards_prototype repo here, which implements a fix. This function was run over most of the initial data as it was being uploaded to the s3 for use in the Atlas, it should be integrated into the pipeline function to be sure it is consistent across the whole pipeline - probably in the zonal extraction wrapper function & anywhere that zonal extraction occurs.

I tested using some DGGS (a5 and h3) to see if these could be helpful in reducing the number of extractions we have to do when a boundary changes, but results were not great so far. I think we stick with the zonal extraction and manage the duplicates within the pipeline.