Modeling Herring Populations

PROJECT

Herring Modeling Populations

Background

The Herring Research and Monitoring (HRM) program's goal is to improve the ability to predict the herring stocks through modeling. During the first phase of the HRM program (2012-2017) we focused on fitting the existing model used by Alaska Department of Fish and Game (ADF&G) into a Bayesian framework to allow us to include prior information on herring in general (probability distribution of mortality rates across herring populations) and estimate uncertainty around model estimates. We call the modified model the Bayesian age-structured assessment model for herring (BASA). We also modified the structure of BASA to fit to aerial surveys of age 1 herring to see if they can predict recruitment at age 3. We gathered information from herring stocks around the world to determine how the herring crash in Prince William Sound (PWS) compares to declines and recovery of other herring stocks. In the second phase of the program (2017-2021) we used the Bayesian model to estimate herring abundance in PWS and add several new pieces to the puzzle. These pieces include using antibody levels in herring to better estimate disease impacts, testing how fast herring may mature (and the consequences of getting it wrong), finding possible environmental linkages, and looking ahead for the testing of alternative management strategies.

Methods

We used the BASA model to predict Prince William Sound herring biomass using survey data from two aerial surveys (one on milt, and one on age-1 schools), the acoustic spawning survey, and the age-sex-length data, all collected by the monitoring projects in the HRM program. Environmental factors were gathered to determine which best predicted recruitment, mortality, and spawn timing in PWS herring. In addition, extensive modeling showed that it is possible to use disease antibody data to estimate the size of past outbreaks and the extent to which disease increase mortality in herring. Lastly, a new time series of disease antibody data were included in the stock assessment model to further improve estimates.

What we are learning

A key change in BASA was made in 2019, when we reconfigured the BASA model to use a more efficient Bayesian algorithm (called the No-U-Turn algorithm). This new approach cut the run time of the model by 60-80% which has allowed us to use the BASA to explore more relationships between recruitment, mortality, and the environment. For example, higher pink salmon returns were associated with higher mortality in herring, but this appeared to be inconsistent over time. There was weaker evidence for other factors such as humpback whale predation, indirect interactions with walleye pollock, and regional climate patterns (the North Pacific Gyre Oscillation).

The herring modeling project (PI Trevor Branch) includes work by University of Washington graduate researchers Melissa Muradian, John Trochta, and Joshua Zahner; and work presented separately by postdoctoral scholars David McGowan and Bia Dias.