reports/ncChumDoc.md

In CamFreshwater/samSim: Closed-loop salmon dynamics simulator

North Coast Chum Case Study Documentation

Data available (data/northCoastDat)

Unless otherwise noted all North Coast chum data provided by Shaun Davies and Katie Beach (DFO)

area3OtolithResults.xlsx

1. Worksheet 1: Otolith results
   • Provides marked vs. unmarked catch breakdown 2011-2016
   • Data:
     • Year, gear, area, week, estimates of catch (unsure what this value represents precisely), proportion of readable and marked otoliths,
     • Depending on year ~69-91% of catch
2. Worksheet 2: weekly var
   • Week to week variability in proportion of catch allocated to hatcheries
   • Typically stable for 3-4 weeks before declining
3. Worksheet 6: Hatchery ppn (3)
   • Age specific catches from 2011-2016
   • Almost all age 4 except for 2016 when catches were dominated by age-5 (totals were low but not abnormally so)
4. Other worksheets have additional data that is likely not relevant – e.g. hatchery-specific estimates, data on individual fish
   • What is AlexData worksheet?

northCoastChumAgeStruc.xls

• Summary of results from otolith aging studies conducted on SA 2e, 2w, 3, 4, 5, 6, 7, 8 and 10 chum
• Data provided:
  • Management subarea, sample location, wild/enhanced, year, frequency at age, proportion at age, brood year at age, sample size
• For area 3 samples collected from two subareas and nine sample locations, collected from 2008-2012, with 2-151 fish per sampling location
• Generally age 4s are at least 75% but some years/locations ~50%

northCoastCommercialCatch_2011to2015.xls

• Provides daily, sub-area specific catch estimates for north coast statistical areas from 2005-2015
• Includes seine, gillnet and troll fisheries
• Data:
  • Estimate type (e.g. in-season), area opening descriptor, target species, fishing date, area/subarea, hours open, number of vessels, species specific catches retained and released
• Haven’t examined closely yet

area3ChumEnglishWB.xlsx

• Provides estimates of stream-specific survey effort (# years surveyed within decade) and returns (not age-specific) from 1950-2015
• Data:
  • CU, area, reviewer, survey assessment type, return estimates per year, and decadal averages
• 61 streams have data across all 4 CUs
• NOTE LARGE GAPS IN COVERAGE AMONG CUs
  • Lower Nass: several years with no indicator streams surveyed prior to 1992; 1993-2008 one stream surveyed in one year
  • Portland Canal-Observatory: some coverage in all years
  • Portland Inlet: some coverage in all years

summariesTRTC.xlsx

• Appears to provide data used to generate K. English reports on north coast salmon populations (i.e. all species totals as well as specifics for statistical areas)
• Basically all data is a function of common ERs at the SA level, applied to CU-specific estimates of escapements generated from indicator streams w/ expansions
• Used to generate catch data for our purposes (no estimates of FN catch repavailable)
• Worksheets include Pivot that combines species totals and subsets of SA data, TotalPlots which provides species specific totals, StatAreaPlots, and ConservationUnitPlots; last two most relevant
• Data for SAPlots and CUPlots worksheets runs from 1954-2014
  • ID variables
  • Q1-Q3 columns represent data quality
    • Q1 is categorical index of survey quality (e.g. visual surveys vs. mark-recapture vs. counting fence)
    • Q2 is survey execution reflecting EF1 (see below) and the proportion of indicators nsampled
    • Q3 is survey representation reflecting EF2 and strength of correlation between index streams and CU as a whole
  • Total index escapement + 2 expansion factors = observed escapement
    • EF1 corrects for missing indicators, EF2 corrects for non-indicator streams
  • Observed escapement + 1 expansion factor = total escapement
    • EF3 corrects for observer efficiency (assumed to always be 1.5)
  • Canadian harvest and exploitation rates + total escapement = TRTC (total return to Canada)
  • Total harvest and total ER included US harvest/ERs (which aren't shown); total harvest + total escapement = total run
  • Note all ERs are assumed to be static across CUs
• Some escapement estimates available for all years in Portland Inlet and Portland Canal-Observatory, but not in Lower Nass

outputAge.xlsx

• Provides age-specific recruitment estimates by brood year based on total returns to each statistical area and mean estimates of age composition
  • In SA3 8.9% age-3, 76% age-4, 14.4% age-5 and 0.06% age-6
• Not clear how these average values are generated, but presumably from recent otolith studies

chumSRDatPSF.csv

• Brood table for north/central coast chum stocks provided by PSF (E. Hertz) used to generate SR parameter estimates
  • Original data in NBBR/Age Tables/CM OUTPUT AGE StatArea.xlsx
• Assumes age structure stable through time
• Produced as output from NBBR model by K. English

psfMCMCPars.out

• Stock recruitment parameter estimates for 3 Skeena and 2 Nass CUs provided by E. Hertz
• Fit using jags model MultiStockHierPSF.txt which was replicated in our notation as MultiStockHierPSFRep.txt
• Priors for these CUs were also provided by E. Hertz; modified in script to represent beta (log(1/prSmax)) and tau (1/prCV)

  CU              prSmax  prCV        
  Skeena_Estuary  1118    1       
  Lower_Skeena    20378   10      
  Middle_Skeena   2544    1       
  Portland_C_Obs  42230   10      
  Portland_Inlet  24524   10
  

  • Analysis comparing estimates using their priors/model and our own is in ncStockRecModels.R; some details below in Apr. 30 update

northCoastDFOData_KBeach.xlsx

• Catch, ER, and escapement estimates that appear to account for otolith (i.e. estimate of AK hatchery chum contribution)
• Data from 2000-2017
• Estimates differ substantially from summariesTRTC.xlsx

Estimating stock-recruit parameters

First pass on fitting SR models - Feb. 13

• Received useable SR dataset
• Began working to estimate SR parameters using standard and hierarchcial Ricker models
  • Based on JAGS scripts and associated R functions written by Brooke Davis for retro SC chum project
  • Minor edits by CF with SR fitting script ncStockRecModels.R and stockRecFunctions.R
• Modifications due to small sample sizes for Lower Nass CU
  • Included 3 Skeena CUs in hierarchical model
  • Smax priors are a function of maximum observed abundance over entire time series, not just years with full recruit data
    • I.e. in stockRecFunctions.R escapement time series used to calc sMax, while paired escapement/recruitment used to estimate SR
  • Removing Lower Nass and running model with 5 CUs instead of 6 does not strongly influence parameter estimates; however estimates of alpha for Lower Nass do not show as much shrinkage as expected
• Stock Key: if(stk==1){ CU<-"Portland Inlet" }; if(stk==2){CU<-"Lower Nass" }; if(stk==3){ CU<-"Portland Canal-Observatory" }

Validate age structure (RESOLVED) - Mar. 8

• Double checked age-specific returns as follows; seem to be correct:

  for(i in 1:nrow(newChumDat)){
  age3Esc <- newChumDat$Age3[i+3]*newChumDat$Escape[i+3]
  age3Catch <- newChumDat$Age3[i+3]*(newChumDat$Total.ER[i+3]*(newChumDat$Escape[i+3]/(1-newChumDat$Total.ER[i+3])))
  newChumDat$rec3[i] <- age3Catch+age3Esc
  age4Esc <- newChumDat$Age4[i+4]*newChumDat$Escape[i+4]
  age4Catch <- newChumDat$Age4[i+4]*(newChumDat$Total.ER[i+4]*(newChumDat$Escape[i+4]/(1-newChumDat$Total.ER[i+4])))
   newChumDat$rec4[i] <- age4Catch+age4Esc
  age5Esc <- newChumDat$Age5[i+5]*newChumDat$Escape[i+5]
  age5Catch <- newChumDat$Age5[i+5]*(newChumDat$Total.ER[i+5]*(newChumDat$Escape[i+5]/(1-newChumDat$Total.ER[i+5])))
  newChumDat$rec5[i] <- age5Catch+age5Esc
  age6Esc <- newChumDat$Age6[i+6]*newChumDat$Escape[i+6]
  age6Catch <- newChumDat$Age6[i+6]*(newChumDat$Total.ER[i+6]*(newChumDat$Escape[i+6]/(1-newChumDat$Total.ER[i+6])))
  newChumDat$rec6[i] <- age6Catch+age6Esc
  }
  

Adjustments to stock recruit models - Mar. 16

• To make chum SR estimates consistent with Fraser River modeling strategy had to combine process and observation variance estimates into single value (i.e. sigma)
• Commented out tauv and trimmed code moderately; saved as MultiStockHierCF.txt
• These models fit more efficiently (neff much larger), gave reasonable precision estimates and behaved as predicted, i.e. data limited CU shrank more than others
• Given small sample size of Lower Nass CU (i.e. stk 2), didn't use fitted values
  • Instead that stock in nassChumMCMCPars.csv contains alpha values from the posterior hyperdistribution for alpha, beta0 value is equal to median from escapement time series and sigma is shared with Portland Canal-Observatory (larger sigma Portland Inlet)

Discrepancies with PSF estimates - Apr. 30 (RESOLVED MAY 31)

• Attempted to get PSF and our own stock recruit estimates to converge
• Ran four model combinations to try to identify the discrepancy
  1. Normal model - our standard priors, all six CUs and MultiStockHierSimple.txt model
  2. Trimmed model - our standard priors, five CUs and MultiStockHierSimple.txt model
  3. PSF replicate model - PSF priors, five CUs and MultiStockHierPSFRep.txt model
  4. PSF model - PSF priors, five CUs and MultiStockHierPSF.txt model (basically a copy of B. Connors model with very minor changes
• We couldn't easily compare model 4 to our outputs because of differences in variable names, however 3 and 4 provided nearly identical summary statistics
• Model 3 and 4 outputs were NOT consistent w/ parameter distributions (i.e. raw output data in data/northCoastDat/psfMCMCPars.out
• They were close to model 2 and slightly less similar to model 1, suggesting our model structure is acceptable
• For now using MultiStockHierSimple.txt model with our own priors
• Resolved May 31 divergence due to error by CF in calculating total number of recruits; etimates now similar

Issues w/ data limited productivity - June 1

• When running low productivity scenarios based on 10th percentile of alpha, noticed that L Nass CU (data limited) had negative productivity values, i.e. well below replacement
• Pattern due to using muAlpha as proxy for this CU's alpha; former has a highly uncertain distribution with a substantial amount of distribution below 0
• Since there is no evidence to suggest that the Lower Nass has disproportionately lower productivity than other CUs suggest
  • a) using informative priors to constrain hyper-distribution to more realistic values (e.g. change prior on muAlpha's sigma from dunif(0, 100) to dunif(0, 10)
  • b) sample other two CUs distributions with replacement to produce a mean area 3 value (currently selected this option for both alpha and sigma)

Concerns around forward simulated variability - June 5

• During meeting with S. Davies, K. Beach, AM Huang and K Holt several commented that forward simulated variability appeared reduced relative to historic
• Appears to largely be due to taking medians (individual trials still quite variable)
• Although sigma values are low relative to Fraser, this seems unlikely to be driving the pattern as they are consistent with values in other chum SR studies (e.g. Collie et al. 2012 also varied between 0.3 and 0.7)

• May also be influenced by:

  1. How data are generated, i.e. as model outputs (indeed individual samples of simulations show expected levels of variability)
  2. Less interannual variability in exploitation rates
  3. Unrealistically low levels of outcome uncertainty

Catch comparison and outcome uncertainty - June 7

• Used two sources of catch data to help parameterize catch obs error (difference between method 1 and method 2) and outcome uncertainty (deviations between estimated ER and target ER)
• Catch methods
  • 1) NBSSR model outputs (mod) - primary source of catch data for north coast chum
    • Distinct American and Canadian harvest rates produced using expanded escapement and adjusted exploitation rates from other fisheries (either pink or sockeye depending on time period)
  • 2) Hail data (otoHigh and otoLow) - available for subset of years (2000 - 2017)
    • Consists of commercial catch data + est. release mortality + Nisgaa marine catch
    • Available at area level and majority assumed to be Alaskan hatchery origin
    • To calculate potential SA3 origin harvest multiply by the proportion of thermally marked fish observed in catch as well as a dummy variable representing the ppn that may be migrating into SA3 (rather than 4 or 6); assume high scenario (0.9; OtoHigh) and low (0.1; OtoLow)
• otoHigh and mod catches track each other fairly closely, but begin to diverge ~2010; mean relative difference is 1.76 +/- 3.14 (i.e. otolith catches are 2x higher on average)
  • otoLow exploitation is much lower than either
  • Suggest using 1.5 as sigma for catch
• Outcome uncertainty is scaled relative to target ER of 10%
  • modER typically ~10% lower, otoHighER ~25% high; both sigmas ~0.75
  • Suggest using 0.75 as sigma for mixedOU
• Summary figures are in miscSensistivity/catchComparison/catchTrends.pdf and analysis in misc/ncCatchComparison.R
• Notes on simulations conducted examining different levels of obsSig and OUSig are in simulationRunNotes.md

Update stock recruitment models - July 23 - Given evidence of autocorrelation in residuals, decided to re-estimate SR parameters using an AR Ricker model (MultiStockHierAR.txt) - Prior for rho ~ dnorm(0, 1) - Resulted in realistic estimates for all parameters - Increasing sd in prior to 10 resulted AR effects completely disappearing - Limited examples to compare to, but generally analysts seem to use uninformative prior w/ broad uniform distribution - At same time changed prior on hyperdistribution for alpha to be set on TauA_global ~ dgamma(1, 1) - I.e. sigA_global no longer defined - Data limited CU's parameters estimated by drawing 1000 samples from dataset that contains the other 5; beta based on 1/median observed escapement

CamFreshwater/samSim documentation built on Sept. 25, 2023, 10:22 a.m.