Repository navigation
Description
What
The oisst column of external/ocean-daily.csv is empty for every day (0 of 4,840 rows) and has been since v0.1.0. ocean() in scripts/fetch-external.mjs asks coastwatch.pfeg.noaa.gov/erddap for ncdcOisst21Agg_LonPM180. Today that returns HTTP 503. erddapSeries logs a warning and continues on a failed chunk, so the weekly refresh succeeds with no OISST at all.
Every consumer falls back to Coral Reef Watch SST: sets.sst is oisst ?? crw_sst in the importer, and /ocean uses r.oisst ?? r.crwSst. So all SST on the site is CRW CoralTemp (5 km, PacIOOS dhw_5km), while the credits and labels say OISST:
- the footer (
+layout.svelte: "SST: NOAA OISST v2.1"); - /about (sea-surface temperature source);
- /ocean (weekly chart subtitle and source);
- /seasons and /seasons/[season];
- /water-quality ("satellite SST (OISST)").
Found while reviewing #149, whose SST chart read ocean.oisst directly and was empty for every month.
Why
The labels name a product the numbers don't come from: different resolution and cell (OISST 0.25° south-west of Mo'orea vs CRW 5 km to the north). Anyone reproducing a result from the stated source gets different values. The silent continue also means a dead source can't fail the job.
Acceptance
- Decide: fetch OISST from a working ERDDAP (NCEI or another CoastWatch node; check the dataset id and the
zlevdimension), or drop OISST and label SST as Coral Reef Watch CoralTemp everywhere. -
fetch-external.mjsthrows when a series returns no values at all, or when a non-final chunk fails (as the chlorophyll job does since Environment: chlorophyll-a / ocean colour around Moorea #23). - Labels, credits and the About page match the series actually used; one shared label helps.
- If the SST series changes, bump
IMPORTER_VERSIONonly if import logic changes (the fingerprint already coversexternal/), and re-check the SST findings: the Analysis: negative-binomial GLM/GAM of catch with effort offset #54 catch model, /water-quality's nursery-vs-satellite offset and /ocean.
Part of #69.
Activity
Review fixes for sampling conditions (#124)
Charts: compare sampling conditions within nursery-months (#149)
Luen commentedon Nov 4, 2026
Confirmed against commit 5a1476f and live endpoints on 4 November 2026:
Recommendation: restore OISST through that endpoint, validate the full historical series, then replace the snapshot. Fail on non-final chunk errors and require finite values (a nonempty Map can still hold only nulls). Tolerate final-chunk failure only for expected publication lag. Failure must occur before writeCsv so the previous snapshot survives.
Make fallback provenance visible too: if CRW fallback stays, unconditional OISST labels can remain misleading during OISST publication lag. Use shared source copy with coverage/fallback counts, or choose one explicit series per analysis. Credit CRW heat-stress data independently.
Restoring OISST changes numerical inputs. Recompute the catch-model SST term, nursery/satellite offset, ocean and season summaries, and compare coverage before/after; do not assume the conclusions persist. Changing external/ already invalidates the cache; bump IMPORTER_VERSION only if import logic changes. No code or snapshots changed in this investigation.