Ephemeral relationships in salmon forecasting: A cautionary tale

Abstract

The influence of climate on marine populations is important for predicting stock abundance of marine fishes, and has led to increasing interest in environment-based forecasts (EBFs) for harvest management. While some climate indices have proven useful for explaining fluctuations in Pacific salmon stock abundance, there have also been sudden failures of EBF models. I analyzed temporal patterns in prediction skill for a variety of climate and ecosystem indicators as predictors of marine survival for a coastal coho salmon stock by computing prediction skill for 29 climate and ecosystem indices across multiple time scales to explore patterns of skill across time. Results demonstrate that predictive skill of EBF models is often ephemeral, arising and falling suddenly across time. This behavior can be explained both on a statistical basis and as a consequence of complex interactions between climate, ecosystems, and populations involving both climate regime shifts and ecosystem phase transitions. Forecast failures are problematic for traditional forecast-dependent harvest management approaches. Solutions for this problem may include improved forecast models and improved climate and ecosystem indicators, but developing management systems that are robust to forecast uncertainty would provide a more reliable response to expected rapid ecosystem changes in response to climate.