An Updated Stray Rate Calculation

Introduction

This vignette describes updates to the existing straying implementation in the SDM. The previous approach used a uniform straying distribution across all potential watersheds. Based on recent empirical work by Sturrock et al. (2019) examining eight decades of hatchery release data in California’s Central Valley, we have updated the straying allocations to reflect observed patterns specific to each source hatchery.

Background Analysis

Sturrock et al. (2019) developed a beta regression model examining factors influencing straying rates including transport distance, return year flows, fish age, release timing, and environmental conditions (PDO). The model found transport distance was strongly associated with straying rates, with bay releases showing 9-26 times higher straying than on-site releases.

Implementation in the model

Our implementation in the model focused on these findings, and converted them into operational data structures and functions that the model could handle as updated inputs and sub-models.

We developed a predictive function that estimates stray rates based on environmental conditions and release strategies:

compute_adult_stray_rates <- function(type = c("natural", "hatchery"), 
                                    sim_year,
                                    total_releases, 
                                    released_month,
                                    flows_oct_nov, 
                                    flows_apr_may, 
                                    mean_pdo_return) {
    # Creates predictions for each hatchery using beta regression model
    # Accounts for age structure and release location (bay vs river)
    # Returns matrix of predicted stray rates
}

This function uses the beta regression model developed by Sturrock et al. to predict straying rates based on the key environmental variables identified in their analysis. The predictions vary by age class and distinguish between bay and river releases for hatchery fish reflecting the strong predictive influence of release location found in the research.

The implementation improves on the previous uniform approach by instead using a more dynamic representation of straying patterns by incorporating both emperical patterns and environmental influences.