Sea level rise: 3.90M vs assessment 642,610 ac — resolve which water-level mask defines 'already water'

[cboettig]

Both hypotheses we could test ourselves are now eliminated. This does need TNC's method.

The gap

	CA acres
2025 Assessment slr5ft	642,610
ours, as published	3,901,527

Our ingest is faithful — ogrinfo against the NOAA source confirms we took CA_*_slr_5_0ft (ocean-connected, correctly excluding _low_), straight per-feature copy plus ST_MakeValid, 3,909,250 ac re-verified independently via the res-10 hex. What we publish is NOAA's connected inundation surface, which includes the existing ocean and bay by construction (NOAA's slr_0_0ft layer is literally the current water extent). The Assessment measures newly-inundated land.

Ruled out (1): any choice of inundation level

Measured all 42 layers per region — connected and low-lying, 0.0–10.0 ft in half-foot steps — across all 7 California regions, EPSG:3310. Job + full output: validation/slr-mask-sweep/ (sweep-results.tsv).

level	connected	low-lying	conn(5) − conn(level)	+ low-lying(5)
0.0	3,471,723	442,899	429,804	694,828
0.5	3,489,428	456,408	412,100	677,124
1.0	3,536,170	444,291	365,357	630,382
3.0	3,674,605	404,369	226,922	491,946
5.0	3,901,527	265,024	0	265,024

642,610 falls between subtracting the 0.5 ft and 1.0 ft connected surfaces — not a datum anyone would choose. The only sub-2% fits pair mismatched levels (conn(5) − conn(2.5) + low(2.5) = 650,090), which is numerology, not a definition.

Ruled out (2): a spatial land clip

Tested whether existing water was removed by clipping to the state land boundary instead of subtracting the 0 ft surface. Clipped against the CA-Nature ecoregion extent — the app's own definition of California, behind the pinned 101,498,000-ac denominator (the mask validates on load at 101,501,429 ac). Output: clip-results.tsv.

definition	unclipped	∩ CA land	vs 642,610
connected 5 ft	3,901,527	889,511	+38.4%
connected 5 − 0	429,804	427,260	−33.5%
connected 5 − 0 + low-lying 5	694,828	692,275	+7.7%

The land clip is nearly a no-op — 427,260 vs 429,804 — because subtracting the 0 ft surface has already removed the ocean. It cannot be what closes the gap.

Per-region, best candidate (conn(5−0) + low-lying 5 ft, CA-clipped):

region	conn 5−0	low-lying 5	total
Delta	244,895	239,280	484,175
SFBay	130,990	6,620	137,610
South	21,378	8,600	29,978
North1	19,228	9,271	28,499
Central	9,681	1,126	10,807
North2	1,004	115	1,119
Catalina	83	4	87
total			692,275

The ask

Our closest reconstruction of newly-inundated land at 5 ft is 692,275 ac (NOAA ocean-connected 5 ft minus connected 0 ft, plus low-lying 5 ft, clipped to California) against your 642,610. We have ruled out both a different inundation level and a land/ocean clip. What produced slr5ft — which NOAA layers, which surface treated as already-water, and any additional mask (tidal wetland, land cover, county/region subset)? Delta contributes 484,175 ac of our total, so if our Delta handling differs from yours that alone would explain the 52,000-ac difference.

Recommended default if no answer arrives: adopt connected(5−0) + low-lying 5 ft as the published definition (boettiger-lab/data-workflows#363 has the rebuild spec for all 28 NOAA regions) and document the +7.7% residual against the assessment.

Also blocked by this: report Q21 — "what percent of the 30x30 network is at risk from sea-level rise?" (reported ~0.5%).

Done when: the definition is known and data-workflows#363 rebuilds to it, or the fallback above ships with the residual documented.