Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Opih Forecast topic
No spam. Unsubscribe anytime.
Agency adopts updated Columbia River OPIH forecast methodology; staff report ~50% reduction in retrospective forecast error
Summary
Science staff presented a retooled Oregon Production Index Hatchery (OPIH) forecast for Columbia River coho that incorporates environmental covariates and ensemble modeling; the new approach cut retrospective mean absolute percent error roughly in half versus the historical model and has been peer reviewed and accepted by PFMC technical reviewers.
Science staff told the Fish Committee that they have implemented an updated Oregon Production Index Hatchery (OPIH) forecasting method for Columbia River coho that uses environmental indicators, ARIMA time‑series models and weighted ensemble forecasts to reduce historical overforecasting.
Shannon Conley, the Columbia River policy analyst who coordinates OPIH participation, said forecasting ‘‘is harder than predicting the weather’’ and emphasized the forecast’s central role in setting ocean and in‑season harvest opportunities. She said the OPIH team includes cross‑agency partners and that Washington staff co‑lead aspects of the effort.
Thomas Bierins, lead scientist for the Lower Columbia science unit, described the technical approach: identify candidate covariates (jacks/smolt metrics, North Pacific Gyre Oscillation, broader PDO measures and sea surface temperatures), fit ARIMA models across covariate subsets, and construct weighted ensemble forecasts that are evaluated via iterative one‑year‑ahead testing to mimic real forecasting tasks. He said the agency published code and methodology on the agency GitHub to enable technical review and reproducibility.
Bierins and colleague Mark Sorrell explained that the ensemble approach was tuned to weight more recent observations, which improved responsiveness to rapidly changing environmental signals. Sorrell said the team reduced retrospective forecast error from about a 60% mean absolute percent error with the older model to roughly 33% with the new ensemble approach — ‘‘about a 50% reduction in forecast error’’ — and that the method was presented to and accepted by the Pacific Fishery Management Council technical reviews.
Panelists described one diagnostic challenge: the Pacific Decadal Oscillation (PDO) behaved differently in recent years than during the long historical record used to train earlier models, which risked producing anomalously low forecasts if not handled. The team said they did not simply remove PDO but instead adjusted model weighting so the forecast could down‑weight older relationships and better track current ocean conditions. The method and code are published publicly for scrutiny.
Why it matters: more accurate preseason forecasts help managers set seasons and quotas that better align with conservation goals and harvest sharing while reducing last‑minute, disruptive changes. Commissioners praised the scientific team for responsive, transparent model development and asked follow‑up questions about PDO non‑stationarity and ongoing model validation.
Next steps include continued run reconstruction improvements, monitoring model performance in real time and engaging co‑managers and PFMC processes as seasons proceed.
