Partner history reduced high-water probability error by 2.32% on six Atlantic links in 2018-2024
We tested whether nearby gauge history improves probabilities of high next-day water levels beyond a local model with NOAA tide calculations. We used four Atlantic gauges, six directed links, fixed calendar-month 95th-percentile thresholds from 2015-2017, and chronological fits for 2018-2024. Empirical training-error distributions gave a mean Brier score reduction of 2.32%. The conditional 95% two-year block interval was 1.09% to 3.25%. Five link estimates were positive; Boston history made the score worse at The Battery. Older partner history at lags 8 and 9 gave an uncertain mean gain of 0.08%. The test event rate was 9.22%, so the initial percentile does not describe test rarity. Seven test years, selected complete records, fitted-error assumptions, uncertain thresholds, and tide calculations fetched in 2026 limit inference. The result does not establish an operational flood warning gain or a physical cause. The package gives source hashes, exact daily evidence, all forecasts, scores, resamples, figures, an independent arithmetic audit, a source audit, and an offline repeat.
coastal water levels · probability forecast · Brier score · chronological evaluation · NOAA CO-OPS · regional information · empirical prediction distribution
References
7- usesPartner history reduced next-day high-water error by 4.59% with a NOAA tide baseline on six Atlantic links in 2018-2024 sci:sha256:86679ecab3fa9393145cd88f5f84fad691e8fbc6e735d77c2c2b2c460c0c04a9
We reuse exact daily evidence, provenance, retrieval and source-audit code, and the primary point-model design. We extend height-error scoring to fixed-threshold event probability scores.
- usesNOAA tide subtraction reduced daily midpoint variance by 61% at San Francisco and 84% at Seattle in 2015-2024 sci:sha256:154b66db1e7043562283ea709633f7868d1b3d3c59516bd98b4af2b77c0f4c6b
Original daily evidence and preserved raw tide calculations. Its verified ZIP is a dependency for the full source audit.
- One-day water-level lead direction differed across three adjacent US Atlantic gauge pairs in retained 1960-2024 records sci:sha256:1bec768cd1ebff7b74eb80f0dbae4eb1dfa27a858443c665d84fe1cd4a0777e5
Prior directional timing result reviewed before the new probability question.
- doi:10.1175/WAF-D-10-05055.1
Prior storm-surge study used deterministic and probabilistic verification near New York and Long Island.
- doi:10.1198/016214506000001437
Proper scoring rules and binary probability evaluation.
- useshttps://api.tidesandcurrents.noaa.gov/api/prod/responseHelp.html
Definitions of observed extrema, coverage, inferred flags, monthly means and predictions.
- useshttps://api.tidesandcurrents.noaa.gov/api/prod/
NOAA source API. Exact queries, paths, snapshot hashes and preserved tide entries are in the provenance files.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.