Utilizing Multi-Source Abundance Estimation And Climate Variability To Forecast Pacific Salmon Populations

Abstract

Data limitation is a common property of many fisheries. Some Pacific salmon populations are a typical example of this situation because the monitoring of numerous tributaries within an area becomes logistically intractable. Fishery management often responds to this scenario with qualitative stock assessments in the form of harvest projections In some cases, fishery data, although limited, exists in a variety of sources and may be integrated to develop quantitative population estimates. The first objective of this investigation is to generate a modeling process that combines multiple data sources to estimate abundance and escapement estimates for data-limited salmon populations. Second, we consider the reliability of these estimates by testing for robustness to various simulated levels of measurement error in the data. Finally, we perform rigorous development and selection on an age structured spawner-recruit model that incorporates abundance and escapement estimates and identifies potential environment-recruit relationships. We demonstrate our technique with a case study on summer chum salmon from the Kuskokwim and Yukon Rivers, Alaska. Recent declines of summer chum returns to this salmon-dependent region have created hardships for the local area residents. We developed a maximum likelihood statistical framework that synchronously combined all available data sources from this management region to estimate abundance and escapement. Successful estimation was dependent on an independent estimate of abundance for a least a few years. We provide error estimates of the modeling process through bootstrap methods. Simulations showed that measurement error had negligible effect on abundance estimates, whereas performance for escapement estimation was tied to the sequence of abundance years. High explanatory power was attained by including environmental variables in the spawner-recruit relationship developed from these population estimates. We used a three-stage modeling process to maintain biological realism in the predictor variables. Recent changes in variables chosen for the best model were consistent with poor environmental conditions and estimates of forecasting error were much lower than models using no environmental information. Based on our findings, we recommend that managers consider the utility of multiple source estimation and environmental variability with our modeling approach for future regulatory decisions of Pacific salmon fisheries in data-limited regions.