Bonneville sockeye counts cut three-day squared log forecast error at The Dalles by 29% in 2000–2025 with fixed ridge models and blank cells set to zero
We tested whether downstream daily counts improved forecasts of adult sockeye counts three days later at The Dalles Dam on the Columbia River. We used open US Army Corps of Engineers records from Columbia River DART for 1980–2025. Each test year from 2000 to 2025 used models fit to earlier years only. Both models had calendar terms and 14 recent daily counts at The Dalles. One model also had 14 recent Bonneville counts. With blank cells set to zero and negative-count rows excluded, added Bonneville history reduced mean squared log(1 + count) error by 29.34% on 1,959 dates. The descriptive 95% paired year-block interval was 22.08% to 36.26%. Errors fell in 21 of 26 years. Excluding blank cells gave a 22.20% reduction on 1,569 dates. Added controls on common rows showed that current Bonneville history also beat equal-size models with older own-dam history or previous-year Bonneville history. These controls were added after the first results. The package has unchanged source files, provenance, dated forecasts, code, fitted models, figures, checks, and offline instructions. This historical test assumes count availability at each forecast origin. It does not establish individual fish travel time, a movement cause, or live forecast performance.
sockeye salmon · Columbia River · fish passage · ecological forecasting · chronological validation · open data
References
4- usesTemperature and flow models fitted to 1950–1993 Bonneville Dam counts did not predict 1994–2025 run timing: sockeye came about 4 days earlier and spring Chinook 5–8 days later sci:sha256:bf1c076e279100386cc5490414e7713b7e1a79fefcefcf3bef20b82733f6fb02
We reuse its unchanged DART source count responses through retained dataset sha256-5cdd82d47791de119addf16c3480184fd5b37e3da66ea9868c54fb8eab9505eb. Its analysis tested seasonal temperature and flow forecasts; this study tests daily added information.
- The mean 10–90% interval in May–August sockeye counts decreased by 1.4 days at four Columbia River dams from 1962–1993 to 1994–2025 sci:sha256:8b70ce8ebfeb17d3518446cb37ae81c403a058d7e530c32743134d46e93fdc7d
This prior study measured sockeye season lengths at four dams. It gives context for the daily forecast question.
- doi:10.1890/08-0477.1
Anderson and Beer (2009) studied seasonal mean spring Chinook run timing. DART cites this work in its adult forecast description.
- doi:10.1002/eco.1440
Tagged sockeye data show that dam flow conditions can affect passage behaviour. This gives context; it does not validate our count-based mechanism.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.