Passage to Spawner Submodel

Overview

The Passage to Spawner (P2S) is a Bayesian model that produces estimates of spawning adults for Battle Creek, Clear Creek, Deer Creek, and Mill Creek with uncertainty. The model can be used to get spawning adults for a tributary as part of any stock-recruit-based SRJPE model alternative.

Submodel Objective

The purpose of the P2S submodel is to take in observed upstream passage, or escapement, estimates from a tributary and model the relationship with spawner counts, either from redd surveys or holding surveys. The P2S produces estimated spawner counts with uncertainty for use in creating the stock portion of a stock-recruit model in the SRJPE.

The model also estimates a conversion rate, or proportion of adults that pass though the video counting systems that become spawning adults (i.e. prespawn mortality). The P2S can also be used for forecasting; it takes in a standardized environmental covariate and uses this variable to predict spawners from upstream passage counts. In this way, the P2S provides resiliency for years where data may be missing (i.e. years where upstream passage data were collected but spawner surveys were not conducted).

Conceptual model

P2S estimates adult abundance needed for the stock recruit model. See figure 1 below to see how it fits into the full JPE model system.

Figure 1: Conceptual Diagram of SRJPE model. P2S sub-model highlighted in yellow

Submodel Architecture

The submodel takes observed data (in orange) and predicts spawner counts (in pink). Parameters estimated by the model are light pink; for more information, see documentation.

Model inputs

The model takes in two datasets:

Observed adult counts or estimates from all methods (upstream passage, holding surveys, redd surveys, and carcass surveys) aggregated by year and stream. This is in the SRJPEdata::observed_adult_input data object.
Observed environmental covariates aggregated by year within streams, and standardized (scaled to center on 0). This is in the SRJPEdata::adult_model_covariates_standard object.

Running the P2S submodel

For a given stream and covariate

The model is run for a given stream and environmental covariate.

Options for streams are battle creek and clear creek.
Environmental covariates are water year type, growing degree days, maximum flow, median passage timing, and passage index.

# run the model for battle creek and water year type
battle_inputs <- prepare_P2S_inputs("battle creek", "wy_type")
battle_P2S_model <- run_passage_to_spawner_model(battle_inputs)

For all covariates within a stream

This truncates the dataset to only those years where all covariate data is available. You can look at parameters that help diagnose the performance of the different covariates like R2_data, R2_fixed, mean_redds_per_spawner, and b1_survival.

# run the model for battle creek and water year type
battle_covariate_comparison <- compare_P2S_model_covariates("battle creek")

battle_covariate_comparison |> 
  filter(parameter %in% c("R2_data", "R2_fixed", "mean_redds_per_spawner",
                          "sigma_redds_per_spawner", "b1_survival",
                          "spawner_abundance_forecast"))

Model Output

The P2S model produces a STANfit object when run. You can extract parameter estimates from the model object with the function extract_P2S_estimates(). This produces a table.

battle_parameter_estimates <- extract_P2S_estimates(battle_P2S_model)

knitr::kable(head(battle_parameter_estimates, 10))

parameter	mean	se_mean	sd	2.5	25	50	75	97.5	n_eff	Rhat	year
log_mean_redds_per_spawner	-0.7756127	0.0009767	0.1662082	-1.1088123	-0.8826713	-0.7743096	-0.6676304	-0.4474190	28957.38	1.0001819	NA
sigma_redds_per_spawner	0.6176322	0.0005426	0.1185665	0.4352698	0.5341672	0.6007244	0.6824753	0.8992626	47743.00	1.0001319	NA
b_survival	0.2207787	0.0029467	0.3314244	-0.4351591	0.0046350	0.2197987	0.4356461	0.8865241	12650.26	1.0004669	2001
log_redds_per_spawner	-1.2754301	0.0005788	0.1728325	-1.6260166	-1.3910668	-1.2711890	-1.1559771	-0.9504445	89168.74	1.0000197	2001
log_redds_per_spawner	-1.1327236	0.0003877	0.1164060	-1.3684956	-1.2099801	-1.1305666	-1.0535601	-0.9122176	90168.53	0.9999576	2002
log_redds_per_spawner	-0.5759191	0.0029505	0.3364033	-1.2524814	-0.7937494	-0.5754964	-0.3554868	0.0866867	12999.47	1.0004778	2003
log_redds_per_spawner	-0.9447890	0.0005517	0.1621853	-1.2723532	-1.0527563	-0.9414259	-0.8333916	-0.6380928	86427.66	0.9999703	2004
log_redds_per_spawner	-0.6980111	0.0029671	0.3493897	-1.3982250	-0.9257723	-0.6980626	-0.4669272	-0.0134806	13866.49	1.0004266	2005
log_redds_per_spawner	-0.8264333	0.0029588	0.3382511	-1.5006455	-1.0461285	-0.8252512	-0.6038720	-0.1573202	13068.74	1.0004673	2006
log_redds_per_spawner	-0.7938967	0.0002782	0.0860919	-0.9653668	-0.8509358	-0.7926342	-0.7352431	-0.6292074	95740.56	0.9999721	2007

Extract predicted spawners by year, with uncertainty

We can also see forecasted spawner count for a wet or dry year with uncertainty by looking at the abundance_forecast parameters, which use the average upstream passage as a predictor variable and water year type as the predictive covariate. Deer and Mill Creek model fits are limited in utility due to a lack of available data, so the confidence intervals for those tributaries are much greater than for Battle and Clear Creeks.

Resources

See the Passage to Spawner Full Model Description for full documentation of development, testing, and results of the submodel. Codebase for the analysis in the Full Model Description can be found here. See adult data report for full documentation on adult Chinook salmon data collection.

Liz Stebbins, Ashley Vizek, Erin Cain, Josh Korman