. . . . "2026-06-06"^^ . . "Open light+SST+bathymetry HMM reproduces juvenile white-shark daily geolocation at 148 km median to held-out Argos (GPE3: 54 km) - physically consistent and fully open" . "This replication PARTIALLY SUPPORTS the original claim that pop-up archival tag (PAT) light-level data yield usable daily geolocations of juvenile white sharks. An open, fully-reproducible Hidden Markov geolocation model — fusing an astronomical twilight emission, a satellite-vs-onboard temperature emission, and a bathymetric depth/land constraint, with a land-barrier movement model, on a baseline-free species-range grid — reproduces the broad daily tracks of the four co-deployed (PAT + SPOT) sharks, reaching a pooled median great-circle error of 148 km to the held-out SPOT Argos referee (per-tag 62–206 km). This is well below the 276 km temperature-only floor of the prior chain (more than halving it), and within the range typically reported for open light-level geolocation. It is larger than the proprietary Wildlife Computers GPE3 product (54 km) on the same held-out fixes — by a factor of about 2.7 on the pooled median, but the gap is tag-dependent rather than uniform: on the coastal-nursery tag 08_01 the open method comes within 62 vs 30 km of GPE3, while the wide-ranging tags drive most of the difference. The chain's falsifiable prediction — that adding the light signal brings the fitted Brownian movement sigma off its optimisation bound — is CONFIRMED on every referee tag. The open method is thus a valid but less-accurate reproduction of the manufacturer geolocation, with two compensating properties GPE3 lacks: it is physically consistent (it never places the animal on land, where GPE3 does on 10 referee days, and it respects bathymetry by construction) and it is fully open and recomputable, whereas GPE3 is proprietary and non-recomputable." . . "Per-tag median great-circle error to held-out SPOT Argos fixes (quality classes 1/2/3), open HMM vs GPE3: 07_05 = 161 vs 95 km; 08_01 = 62 vs 30 km; 08_02 = 206 vs 41 km; 08_09 = 136 vs 67 km. Pooled median-of-medians: 148 km (open HMM) vs 54 km (GPE3) vs 276 km (prior temperature-only chain). Fitted movement sigma = 0.0017–0.015 rad on all four referee tags, all far below the 0.094 rad search upper bound (the falsifiable sigma-off-bound prediction is confirmed). Physical consistency: 0 of ~580 open-HMM daily positions fall on land (GPE3 places the shark on land on 10 of the four tags' days); every open-HMM position respects bathymetry (cell seabed >= observed daily max dive depth) by construction. The state-space grid is baseline-free — sized from the documented NE-Pacific + Gulf-of-California species range (lon[-125,-106], lat[22,38]), so the GPE3 baseline touches neither the model fit nor the grid, and the Argos referee is held out throughout. Referee robustness: the raw Argos data contains corrupt fixes (an out-of-region Florida position; implied inter-fix speeds up to 1078 km/h) — exactly the outliers the paper's Technical Validation warns users to filter — but the median metric absorbs them (cleaning leaves both 54 km and 148 km unchanged). Fairness: the paper documents GPE3's inputs as light levels + satellite SST + deployment/pop-up anchors + a swim-speed prior, and NOT the SPOT Argos fixes — the same inputs the open method uses — so the comparison against the held-out Argos is fair.\n\nGithub repository: https://github.com/annefou/white-shark-geolocation-light" . "1. Small referee sample: only 4 of the 7 analysed tags carry a co-deployed SPOT Argos referee; the per-tag error spread is large (62–206 km).\n2. The open method is ~2.7x less accurate than GPE3 — a partial, not full, accuracy reproduction.\n3. Per-tag failure modes (which double as future-work levers): latitude error dominates 08_09 and 08_01 (the intrinsic day-length->latitude weakness of light geolocation, worse near the equinox); longitude error dominates 08_02 and grows away from the equinox, consistent with onboard-clock drift (a degradation the paper itself flags); 07_05 was held at the aquarium and released in Monterey Bay, north of its natural range (paper Usage Notes), and its baseline-free track retains 1 residual land-crossing segment of 133.\n4. GPE3 is a proprietary black box: its exact swim-speed prior, any light-curve filtering, and internal tuning are unknown — only its documented input list could be matched." . . . . . "Anne Fouilloux" . "2026-06-07T21:14:11.533Z"^^ . . . . . "Open light+SST+bathymetry HMM reproduces juvenile white-shark daily geolocation at 148 km median to held-out Argos (GPE3: 54 km) - physically consistent and fully open" . . "RSA" . "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" . "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" . . .