@prefix this: . @prefix sub: . @prefix np: . @prefix rdf: . @prefix prov: . @prefix npx: . @prefix dc: . @prefix xsd: . sub:Head { this: a np:Nanopublication; np:hasAssertion sub:assertion; np:hasProvenance sub:provenance; np:hasPublicationInfo sub:pubinfo . } sub:assertion { sub:jws-geolocation-pangeo-fish-replication-outcome a ; "2026-06-04"^^xsd:date; "Replication outcome: pangeo-fish depth-temperature geolocation of juvenile white sharks vs published GPE3 tracks"; "An independent, fully open geolocation method (pangeo-fish: a HEALPix-NESTED hidden Markov model), run here in a deliberately minimal configuration — temperature-at-depth as the only emission signal, matched against the GLORYS12V1 ocean reanalysis, anchored only by the release/pop-up endpoints — does NOT reproduce the accuracy of the paper's proprietary GPE3 (light-plus-SST) geolocations when both are judged against co-deployed Argos SPOT fixes as the referee. (pangeo-fish supports richer multi-signal emissions and known reference-point/acoustic anchoring that this baseline did not use — see limitations; the figure below is a floor for this configuration, not the method's ceiling.) Across four juvenile white sharks carrying both a recovered archival PAT tag and a SPOT tag, the open method's median great-circle error to Argos was 276 km (median of per-tag medians; range 202 to 354 km per tag), versus 54 km for GPE3 (range 30 to 95 km) on the same fixes — roughly three to twelve times larger per tag. The paper's geolocation product is therefore reproducible in principle but not to comparable accuracy with open tooling: the result qualifies the reproducibility of the released geolocations rather than disputing their correctness."; ; """From results/summary.csv (read directly, not from memory). Tag selection: recovered PAT tags with full time-series (PAT_RECOVERY==YES & DATA_TS==YES). Of 7 candidates, 5 entered the HMM; 2 were excluded honestly — 02_01 (PAT2 records only internal/body-heat temperature, no external ambient sensor, so the emission is invalid) and 06_10 (its basin-scale GLORYS box repeatedly failed to download). Four of the analysed tags have a co-deployed SPOT tag (Argos referee); one (07_01) is PAT-only and compared to GPE3 only. Referee tags — median great-circle error to Argos (same fixes for both methods): tag n_argos pangeo-fish vs Argos GPE3 vs Argos fitted sigma (rad) 07_05 68 300.3 km 94.5 km 0.0070 (interior) 08_01 75 354.5 km 30.4 km 0.0937 (at bound) 08_02 62 201.8 km 41.3 km 0.0937 (at bound) 08_09 62 251.1 km 67.0 km 0.0937 (at bound) ---------------------------------------------------------------------------- aggregate (median of per-tag medians): pangeo-fish 275.7 km | GPE3 54.2 km PAT-only (no referee; NOT an accuracy validation): 07_01 pangeo-fish-vs-GPE3 median offset 169.9 km. pangeo-fish-vs-GPE3 track-agreement medians on the referee tags: 258.2 / 273.1 / 178.6 / 184.6 km (07_05/08_01/08_02/08_09). Key diagnostic: the fitted Brownian σ saturated at its upper bound (0.0937 of max 0.0942 rad) for 3 of the 4 referee tags — a signature that the temperature observations were weakly informative, so the model defaulted to maximum diffusion. Only 07_05 fit an interior σ. State space: HEALPix NESTED level 9 (~6.4 km) for every tag (recorded per-tag in summary.csv). Pipeline executed end-to-end via the repo notebooks; numbers verified 2026-06-05. GitHub repository: https://github.com/annefou/white-shark-geolocation-replication"""; """Caveats that bound this conclusion: 1. **Validity of temperature-matching.** White sharks are regionally endothermic, but the PAT *external* sensor records ambient water temperature, so matching tag temperature-at-depth against GLORYS thetao is physically valid. (Checked deliberately — not a confound.) 2. **Diagnosed cause of the accuracy gap — weak thermal constraint, not a code error.** For 3 of the 4 referee tags the fitted Brownian σ saturated at its upper bound (≈0.0937 of max 0.0942 rad). σ pinned at the ceiling is diagnostic that the temperature observations carried little positional information, so the posterior defaulted to maximum diffusion. Temperature fields are spatially smooth (many locations share near-identical profiles), giving a flat emission likelihood — the opposite of GPE3's light-based longitude, which is clock-sharp. 3. **Field resolution and habitat bound the achievable skill.** GLORYS12V1 (~8 km, daily-mean) smooths the mesoscale fronts/eddies that carry positional information; the ~0.1 °C tag sensor is *not* the limit, the field is. Movement is basin-scale through the relatively homogeneous offshore California Current, where the vertical thermal gradient — the thing that makes temperature-at-depth informative — is often weak (vs pangeo-fish's coastal sea-bass demo). Leading mechanistic hypothesis: **skill correlates with thermocline strength**. 4. **The referee.** GPE3 is itself an estimate, not truth — only the Argos SPOT fixes serve as the accuracy referee, and only where deployments overlap SPOT coverage (the 3–4 co-deployed tags). 5. **This is a MINIMAL pangeo-fish configuration — a baseline, not the method's ceiling.** The result tests pangeo-fish with a single emission variable (temperature-at-depth), only the two release/pop-up endpoints as anchors, and the GLORYS field — with Argos deliberately held out as the independent referee. pangeo-fish itself supports much more, and none of it was exploited here: (a) **richer multi-signal emissions** — open light-level geolocation (TwGeos/FLightR/SGAT/probGLS), satellite-SST matching (OSTIA/MUR/GHRSST), joint temperature–salinity matching (GLORYS `so`), and bathymetry as a hard exclusion (GEBCO); (b) **known reference-point anchoring along the track, not just endpoints** — including **acoustic-receiver detections** (the MBA project carried acoustic tags on ~21 sharks), which pin position to within ~1 km at known times. So the ≈276 km figure is a floor for this bare configuration, NOT a verdict on pangeo-fish as a method. Caveats on the acoustic route specifically: the detection records are likely **not** in the PAT/SPOT archive used here (they live in a separate acoustic-telemetry network — ATN/OTN / California arrays — and must be joined on shark ID); receivers are dense in the coastal nursery but absent offshore, so acoustics tighten the coastal portions, not the offshore basin-scale excursions where temperature struggles most; and any detection used as an anchor must NOT also be used as a referee (avoid the circularity we avoided with GPE3). 6. **Field/resolution future work.** A natural follow-up that could narrow the gap (a candidate `extends`/`qualifies` chain step): re-run the same HMM with a **finer, eddy-resolving, data-assimilative ocean field** — a regional California Current reanalysis (CCS-ROMS, ~1–4 km, which also covers the older 2001–2009 deployments) or a Destination Earth ocean digital twin for recent-enough tags. Falsifiable prediction: σ comes off its bound and median error drops. Caveats: (i) resolution only helps where thermal structure physically exists but is unresolved, not where the water is genuinely homogeneous; (ii) the field must be high-res AND correctly phased (assimilative) — a sharp-but-misplaced eddy field could add error, so forecast-style twins are not automatically better than GLORYS; (iii) most km-scale products do not reach back to the 2001–2009 windows, so a regional reanalysis may beat a global twin for these specific tags. None of the field improvements remove pangeo-fish's single-signal disadvantage vs GPE3's light+SST+endpoint fusion — which is what caveat 5's multi-signal fusion addresses."""; ; ; . } sub:provenance { sub:assertion prov:wasAttributedTo . } sub:pubinfo { "Anne Fouilloux" . this: dc:created "2026-06-06T09:01:50.741Z"^^xsd:dateTime; dc:creator ; dc:license ; npx:introduces sub:jws-geolocation-pangeo-fish-replication-outcome; npx:wasCreatedAt ; "Replication outcome: pangeo-fish depth-temperature geolocation of juvenile white sharks vs published GPE3 tracks"; . sub:sig npx:hasAlgorithm "RSA"; npx:hasPublicKey "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"; npx:hasSignature "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"; npx:hasSignatureTarget this:; npx:signedBy . }