start rec covariate WP

WP_RecCovariates_Gaichas.Rmd

@@ -0,0 +1,89 @@
+---
+title: "Working Paper: Recruitment covariate testing in WHAM"
+author: "Sarah Gaichas and Jon Deroba"
+date: "`r Sys.Date()`"
+output:
+  bookdown::html_document2: 
+    toc: true
+    toc_float: true
+    code_fold: hide
+  bookdown::pdf_document2:
+    includes: 
+       in_header: latex/header1.tex
+    keep_tex: true
+  bookdown::word_document2:
+    toc: true
+link-citations: yes
+csl: "canadian-journal-of-fisheries-and-aquatic-sciences.csl"
+bibliography: zoopindex.bib
+urlcolor: blue
+---
+
+```{r setup, include=FALSE}
+knitr::opts_chunk$set(echo = TRUE,
+                      message = FALSE,
+                      warning = FALSE)
+
+library(tidyverse)
+library(here)
+library(DT)
+library(patchwork)
+library(wham)
+
+```
+
+# Introduction
+
+The Atlantic herring research track assessment working group (WG) prioritized investigation of recruitment drivers as potential stock assessment model covariates, because low recruitment in recent years is an important issue for the stock and for fishery management.
+
+The WG used a boosted regression tree analysis (Molina 2024) to identify zooplankton indices that best explained patterns in herring recruitment.  These indices included large copepods in spring (influencing growth of herring postlarvae and juveniles), small copeopods in fall (influencing survival of herring larvae over the winter), haddock egg predation (influencing egg mortality), and temperature (influencing larval and juvenile survival).
+
+In this working paper, we evaluate each of these indices as potential recruitment covariates in the Atlantic herring assessment implemented during the research track in the Woods Hole Assessment Model (WHAM, @stock_woods_2021). 
+
+
+# Methods
+
+We are using the `devel` version of WHAM: https://github.com/timjmiller/wham/tree/devel
+
+Model [mm192](https://drive.google.com/drive/folders/1sQdDsfdnVbiiY4X7Rgr-fvegwT7Fa1Az?usp=drive_link) is our starting point.
+
+Haddock egg predation, Zooplankton, and temperature indices were explored as covariates on herring recruitment.
+
+Recruitment is modeled as deviations from the "recruitment scaling parameter", leaving one option for modeling effects of covariates on recruitment: "controlling".
+
+A "controlling" recruitment covariate results in a time-varying recruitment scaling parameter.
+
+[Jon's text on the haddock egg predation testing here... including trying to fit a stock recruitment function that didnt work]
+
+We explored indices with different zooplankton groups, seasons, and regions according to herring life history and results from the boosted regression tree:
+
+*  Jan-Jun (Spring) large copepods in spring herring BTS strata with lag-0 to represent food for pre-recruit juveniles
+*  Jul-Dec (Fall) small copepods in fall herring BTS strata with lag-1 to represent food for larvae in general
+*  Sep-Feb small copepods in herring larval area with lag-1 to represent food for larvae more specifically
+*  Combinations of large and small copepod covariates above 
+
+We evaluated 
+
+*  Options for covariate input (millions of cells vs. log(cells), VAST estimated SE vs. WHAM estimated SE)  
+*  Options for covariate observation model ("rw" vs. "ar1")  
+*  Options for recruitment link ("none" vs. "controlling-linear" with lag-0 for large copepods and lag-1 for small)  
+
+Short story:
+
+*  Models with covariates input on the log scale generally converged
+*  Models with WHAM estimated covariate SE ("est_1") generally converged
+*  Under the above conditions, most models with and without the recruitment link converged for all covariates
+
+*  Models with the Jan-Jun (Spring) large copepods covariate also converged with input as millions of cells and VAST estimated SE
+
+*  I'm still figuring out where to find all the diagnostics in WHAM, so "converged" may not be "a good model"
+
+[Way too much detail including false starts](https://noaa-edab.github.io/zooplanktonindex/WHAMcovariate_tests.html)
+
+
+# Results
+
+# Discussion
+
+# References
+

zoopindex.bib

@@ -122,6 +122,25 @@ @book{collette_bigelow_2002
 	keywords = {Nature / Animals / Fish, Science / Life Sciences / Zoology / Ichthyology \& Herpetology},
 }
 
+@article{stock_woods_2021,
+	title = {The {Woods} {Hole} {Assessment} {Model} ({WHAM}): {A} general state-space assessment framework that incorporates time- and age-varying processes via random effects and links to environmental covariates},
+	volume = {240},
+	issn = {0165-7836},
+	shorttitle = {The {Woods} {Hole} {Assessment} {Model} ({WHAM})},
+	url = {https://www.sciencedirect.com/science/article/pii/S0165783621000953},
+	doi = {10.1016/j.fishres.2021.105967},
+	abstract = {The rapid changes observed in many marine ecosystems that support fisheries pose a challenge to stock assessment and management predicated on time-invariant productivity and considering species in isolation. In single-species assessments, two main approaches have been used to account for productivity changes: allowing biological parameters to vary stochastically over time (empirical), or explicitly linking population processes such as recruitment (R) or natural mortality (M) to environmental covariates (mechanistic). Here, we describe the Woods Hole Assessment Model (WHAM) framework and software package, which combines these two approaches. WHAM can estimate time- and age-varying random effects on annual transitions in numbers at age (NAA), M, and selectivity, as well as fit environmental time-series with process and observation errors, missing data, and nonlinear links to R and M. WHAM can also be configured as a traditional statistical catch-at-age (SCAA) model in order to easily bridge from status quo models and test them against models with state-space and environmental effects, all within a single framework. We fit models with and without (independent or autocorrelated) random effects on NAA, M, and selectivity to data from five stocks with a broad range of life history, fishing pressure, number of ages, and time-series length. Models that included random effects performed well across stocks and processes, especially random effects models with a two dimensional (2D) first-order autoregressive, AR(1), covariance structure over age and year. We conducted simulation tests and found negligible or no bias in estimation of important assessment outputs (SSB, F, stock status, and catch) when the operating and estimation models matched. However, bias in SSB and F was often non-trivial when the estimation model was less complex than the operating model, especially when models without random effects were fit to data simulated from models with random effects. Bias of the variance and correlation parameters controlling random effects was also negligible or slightly negative as expected. Our results suggest that WHAM can be a useful tool for stock assessment when environmental effects on R or M, or stochastic variation in NAA transitions, M, or selectivity are of interest. In the U.S. Northeast, where the productivity of several groundfish stocks has declined, conducting assessments in WHAM with time-varying processes via random effects or environment-productivity links may account for these trends and potentially reduce retrospective bias.},
+	language = {en},
+	urldate = {2021-05-26},
+	journal = {Fisheries Research},
+	author = {Stock, Brian C. and Miller, Timothy J.},
+	month = aug,
+	year = {2021},
+	keywords = {Environmental effects, Natural mortality, Random effects, Recruitment, State-space, Stock assessment, Template Model Builder (TMB), Time-varying},
+	pages = {105967},
+	file = {ScienceDirect Full Text PDF:/Users/sarah.gaichas/Zotero/storage/UQJRN3K3/Stock and Miller - 2021 - The Woods Hole Assessment Model (WHAM) A general .pdf:application/pdf;ScienceDirect Snapshot:/Users/sarah.gaichas/Zotero/storage/3FFX3ISS/S0165783621000953.html:text/html;ScienceDirect Snapshot:/Users/sarah.gaichas/Zotero/storage/SGFEARBH/S0165783621000953.html:text/html},
+}
+
 @article{ng_predator_2021,
 	title = {Predator stomach contents can provide accurate indices of prey biomass},
 	volume = {78},