flbeia_gadget_mse_age_4s_had_sam.R

# FLBEIA-Gadget MSE framework with the simple haddock Gadget example model (J. Begley) as operating model and SAM as assessment model
# updated 11-jun-2020

## conditioning the operating model

## install packages
#devtools::install_github("REDUS-IMR/FLBEIA", ref="FLBEIAgadget") # FLBEIA modified to work with a gadget model as OM
#devtools::install_github("REDUS-IMR/gadget", ref="gadgetr") 
#devtools::install_github("hafro/rgadget")

## load packages
library(dplyr)
library(FLCore) 
library(FLAssess)
library(FLash)
library(FLFleet)
library(FLa4a)
library(FLBEIA) # requires FLEBIAgadget version
library(ggplot2)
library(FLSAM)  # requires FLBIEAgadget version
library(gtools)
library(Rgadget)
library(gadgetr)

## a simple example with single stock, single fleet and one iteration. 
## change the working directory to the location of the gadget model
setwd(dirname(rstudioapi::getActiveDocumentContext()$path))
path_model <- paste0("models/had")
setwd(list.dirs(path = path_model, recursive = T)[1])
mypath <- getwd()

## To estimate the model parameters the suggested procedure is to use the iterative reweighting approach (gadget.iterative)
## fit <- gadget.fit()
paramsfile <- 'refinputfile'
fit <- gadget.fit(wgts = NULL, params.file = paramsfile, steps = "all")

## reset the directory
setwd("../..")
dir <- getwd()

## FLBEIA Conditioning
## biols: 
first.yr          <- 1978
proj.yr           <- 2000
last.yr           <- 2020
yrs <- c(first.yr = first.yr, proj.yr = proj.yr, last.yr = last.yr)
fls  <- c('fl1')
stks <- c('stk1')
fl1.mets       <- c('met1')
fl1.met1.stks  <- c('stk1')
ni           <- 1 # If gadget is the OM, it only works w/ ni=1
it           <- 1:ni 
ns           <- 4 
na <- 1 
nu <- 1 
stk1.age.min    <- 1
stk1.age.max    <- 10
stk1.unit       <- 1         

## main.ctrl
main.ctrl           <- list()
main.ctrl$sim.years <- c(initial = proj.yr, final = last.yr)

## Data: stk1_n.flq, m, spwn, fec, wt - use FLQuant
## stock stk1
## get the stock input data (derived from the Gadget model fitting output or assessments)
## read in gadget output
## stock number by age - stock size only for 'hindcasts'
data_n <- fit$stock.std %>% filter(year < proj.yr) %>% select(year, step, age, area, number) %>% rename(season = step, data = number) 

## stock weight by age
data_wt <- fit$stock.std %>% filter(year < proj.yr) %>% select(year, step, age, area, mean_weight) %>% rename(season = step, data = mean_weight)

## recruit number
data_rec <- fit$stock.recruitment %>% filter(year < proj.yr) 

## create FLQuant objects
stk1_n.flq <- iter(as.FLQuant(as.data.frame(data_n)), it) 
stk1_n.flq[1, ] <- data_rec$recruitment  # replace by recruit number if not in the Gadget 'immature' stock
stk1_wt.flq <- iter(as.FLQuant(as.data.frame(data_wt)), it) 
stk1_m.flq <- FLQuant(c(0.5/4, 0.2/4, 0.2/4, 0.2/4, 0.2/4, 0.2/4, 0.2/4, 0.2/4, 0.2/4, 0.2/4), # natural mortality per Q
                      dim=c(stk1.age.max, proj.yr-first.yr, na, ns), 
                      quant='age', dimnames=list(age = stk1.age.min:stk1.age.max, 
                                    year = first.yr:(proj.yr-1),
                                    season = 1:ns)) 
stk1_spwn.flq <- FLQuant(1, 
                        dim = c(stk1.age.max, proj.yr-first.yr, 1, ns), 
                        quant = 'age', 
                        dimnames=list(age = stk1.age.min:stk1.age.max, 
                                     year = first.yr:(proj.yr-1),
                                     season = 1:ns)) 
stk1_fec.flq <- FLQuant(c(0, 1, 1, 1, 1, 1, 1, 1, 1, 1), 
                       dim=c(stk1.age.max, proj.yr-first.yr,1, ns), 
                       quant='age', 
                       dimnames=list(age = stk1.age.min:stk1.age.max, 
                                     year = first.yr:(proj.yr-1),
                                     season = 1:ns)) 
stk1_mat.flq   <- stk1_fec.flq 
stk1_mat.flq[2:stk1.age.max] <- c(1, 1, 1, 1, 1, 1, 1, 1, 1) 
stk1_range.min       <- 1
stk1_range.max       <- 10
stk1_range.plusgroup <- 10
stk1_range.minyear   <- 1978
stk1_range.minfbar   <- 2
stk1_range.maxfbar   <- 8 

## Projection biols
stk1_biol.proj.avg.yrs  <- c((proj.yr-3):(proj.yr-1)) 

## Create the object 
stks.data <- list(stk1=ls(pattern="^stk1")) 
biols    <- create.biols.data(yrs, ns, ni, stks.data)

## biols.ctrl: 
growth.model     <- c('gadgetGrowth')
biols.ctrl       <- create.biols.ctrl(stksnames = stks, growth.model = growth.model)

## SRs: 
## stock-recruitment model 
data_rec <- data_rec %>% select(year, area, recruitment) %>% rename(data = recruitment) 
data_rec$data <- data_rec$data/10000
data_rec$age <- 1 
stk1_rec.flq <- iter(as.FLQuant(as.data.frame(data_rec)), it)
stk1_ssb.flq <- ssb(biols[[1]][, 1:(proj.yr-first.yr),,]) 

## fit the stock-recruit model 
stk1_sr.model        <- 'rickerAR1'
stk1_params.n        <- 3
stk1_params.name     <- c('a','b','c')
sr.modelfit <- fmle(FLSR(model = stk1_sr.model,
                         ssb = stk1_ssb.flq[, 1:(proj.yr-first.yr-1), , 1],
                         rec = stk1_rec.flq[, 2:(proj.yr-first.yr), , 1]))
sr_params = as.data.frame(FLQuant((sr.modelfit@params@.Data[, 1]), 
                                  dim = c(stk1_params.n, last.yr-first.yr+1, 1, ns, 1, ni), 
                                  quant = 'param', 
                                  dimnames = list(param = c("a", "b", "c"), 
                                                  year = first.yr:last.yr, 
                                                  #area = na,
                                                  season = 1:ns,
                                                  iter = it)))
stk1_params.array    <- xtabs2(data ~ param + year + season + iter, 
                               data = sr_params,
                               exclude = NULL, 
                               na.action = na.pass)[, , , it, drop = F]   
stk1_uncertainty.flq <- FLQuant(c(1.5), 
                               dim = c(1.1, last.yr-first.yr+1, 1, ns, 1, ni), 
                               dimnames = list(year = first.yr:last.yr, season = 1:ns, iter = it))
stk1_proportion.flq <- FLQuant(0, 
                              dim = c(1, last.yr-first.yr+1, 1, ns, 1, ni), 
                              dimnames = list(year = first.yr:last.yr, season = 1:ns, iter = it))
stk1_proportion.flq[, , , 1][] <- 1 # all spawning occurs in season 1 
stk1_prop.avg.yrs    <- c((proj.yr-3):(proj.yr-1)) 
stk1_timelag.matrix  <- matrix(c(0, 1), nrow = 2, ncol = 1, dimnames = list(c('year', 'season'), 'all'))

## create SRs objects
stks.data <- list(stk1 = ls(pattern = "^stk1")) 
SRs     <- create.SRs.data(yrs, ns, ni, stks.data)

## Data per fleet - catch/landings from the gadget model output
data_catch <- fit$stock.prey %>% filter(year < proj.yr) %>% select(year, step, age, area, biomass_consumed) %>% rename(season = step, data = biomass_consumed)

## landings.n, discards.n,landings.wt, discards.wt, landings, discards, landings.sel, discards.sel, price
fl1.met1.stk1_landings.n.flq <- iter(as.FLQuant(as.data.frame(data_catch)), it) 
fl1.met1.stk1_discards.n.flq <- fl1.met1.stk1_landings.n.flq 
fl1.met1.stk1_discards.n.flq[] <- 0 

## economic parameter values are arbitrarily set for this example
fl1_effort.flq <- FLQuant(1, 
                         dim = c(1, proj.yr-first.yr, 1, ns), 
                         dimnames = list(year = first.yr:(proj.yr-1), season = 1:ns, iter = it))
fl1_capacity.flq <- FLQuant(1, 
                           dim = c(1, proj.yr-first.yr, 1, ns), 
                           dimnames = list(year = first.yr:(proj.yr-1), season = 1:ns, iter = it))
fl1_fcost.flq <- FLQuant(1, 
                        dim = c(1, proj.yr-first.yr, 1, ns), 
                        dimnames = list(year = first.yr:(proj.yr-1), season = 1:ns, iter = it))
fl1_crewshare.flq <- FLQuant(1, 
                            dim = c(1, proj.yr-first.yr, 1, ns),
                            dimnames = list(year = first.yr:(proj.yr-1), season = 1:ns, iter = it))
fl1.met1_effshare.flq <- FLQuant(1, 
                                dim = c(1, proj.yr-first.yr, 1, ns),
                                dimnames = list(year = first.yr:(proj.yr-1), season = 1:ns, iter = it))

## fleets: fl1
fl1_proj.avg.yrs            <- c((proj.yr-3):(proj.yr-1)) 
fl1.met1_proj.avg.yrs       <- c((proj.yr-3):(proj.yr-1))  
fl1.met1.stk1_proj.avg.yrs  <- c((proj.yr-3):(proj.yr-1))  

## create fleets object
fls.data <- list(fl1 = ls(pattern = "^fl1")) 
fleets   <- create.fleets.data(yrs, ns, ni, fls.data, stks.data)

## fleets.ctrl: 
n.fls.stks      <- 1
fls.stksnames   <- 'stk1'
effort.models    <- 'fixedEffort' 
effort.restr.fl1 <- 'stk1'
restriction.fl1  <- 'catch'
catch.models     <- 'gadgetCatch' 
capital.models   <- 'fixedCapital'       
flq.stk1 <- FLQuant(dimnames = list(age = 'all', 
                                   year = first.yr:last.yr, 
                                   unit = stk1.unit, 
                                   season = 1:ns, 
                                   iter = 1:ni)) 
fleets.ctrl      <- create.fleets.ctrl(fls = fls, 
                                       n.fls.stks = n.fls.stks, 
                                       fls.stksnames = fls.stksnames,
                                       effort.models = effort.models, 
                                       catch.models = catch.models,
                                       capital.models = capital.models, 
                                       flq = flq.stk1,
                                       effort.restr.fl1 = effort.restr.fl1, 
                                       restriction.fl1 = restriction.fl1)
fleets.ctrl$fl1$stk1$discard.TAC.OS  <- FALSE
fleets.ctrl$fl1$restriction <- "landings"

## BDs/covars/covars.ctrl/ NULL objects
BDs         <- NULL
covars      <- NULL
covars.ctrl <- NULL

## Gadget parameters placeholders
oneGDGT <- list()
oneGDGT$gadget.inputDir <- paste0(getwd(), "/models/had")
oneGDGT$gadget.mainFile <- "main"
oneGDGT$gadget.paramFile <- "refinputfile"
oneGDGT$runNow <- FALSE
oneGDGT$fleetMode <- "exclusive" # "exclusive" = gadget fleets only model, "mixed" = FLBEIA & gadget fleets mode

## setting up gadget an OM
## If Gadget model names and FLBEIA names are not the same
convertStockName <- list(stk1 = "had")
convertFleetName <- list(fl1 = "future")
stockList <- c("had")

## specify stocks and fleets for gadget input and simulations 
had.fleets <- c("comm", "survey", "future")
had.stocks <- c("had")
had.stocks.mature <- c("had")
had.surveys <- c("survey")
had.forecasts <- c("future")
had.forecasts.tac.proportion <- c(0.232, 0.351, 0.298, 0.119) ## this needs to be the same as FLBEIA
#had.forecasts.tac.proportion <- c(1, 1, 1, 1)

## specify fleets and metiers
fleetList <- c("fl1")

## Below is using the same information as the FLBEIA conditioning
#fl1.mets       <- c('met1')
#fl1.met1.stks  <- c('stk1')

## parameterize mortality
## m2=NULL means we calculate m2 from gadget result, m2=0 means we use only residual mortality (m1). 
## NOTE: m1 can be a vector or scalar.
## stockStep is the step number where the stock is going to be calculated
stockstep = 4
stk1_ind1_range.startf <- 0.12 # the initial time step 
stk1_ind1_range.endf <- 1 - 0.12 # the last time step
had.params <- list(stockStep = stockstep, 
                   minage = stk1.age.min, 
                   maxage = stk1.age.max, 
                   minfbar = stk1_range.minfbar, 
                   maxfbar = stk1_range.maxfbar, 
                   startf = stk1_ind1_range.startf, 
                   endf = stk1_ind1_range.endf, 
                   m1 = c(0.5, 0.35, 0.2, 0.2, 0.2, 0.2, 0.2, 0.3, 0.4, 0.7),
                   m2 = NULL)

## gadget simulation parameters (should be the same as FLBEIA)
firstYear <- first.yr
projYear <- proj.yr
finalYear <- last.yr

## reset the directory
dir <- getwd()

## Load helper functions
source(paste0(dir, "/gadget-fls.R"), local = T)
#----------------
#updateFLFleet("fl1", 0 , 0 , 0)
#stop()


## management procedure (MP)

## indices:
flq <- biols[["stk1"]]@n[ , , , 1] 
#dimnames(flq)$season <- "all"
unc <- id <- q <- flq 
unc[] <- rlnorm(prod(dim(flq)), 0, 0.3)
q[] <- rep(runif(dim(flq)[1], 1e-05/2, 1e-05*5), dim(flq)[2])
id <- biols[["stk1"]]@n[ , , , 1]*unc*q 
#dimnames(id)$season <- "all"
stk1_indices <- c('ind1') 
stk1_ind1_index.flq <- id 
stk1_ind1_index.q.flq <- q 
stk1_ind1_index.var.flq <- unc
stk1_ind1_range.startf <- 0.12  
stk1_ind1_range.endf <- 1 - 0.12 
stk1_ind1_range.min <- stk1.age.min+1
stk1_ind1_range.max <- stk1.age.max 
##  YFT_cpue_range.plusgroup <- 0
stk1_ind1_range.minyear <- first.yr
stk1_ind1_range.maxyear <- proj.yr-1
stk1_ind1_type <- "FLIndex"

## create indices objects
stks.data <- list(stk1 = ls(pattern = "^stk1")) 
oneIndAge <- create.indices.data(yrs, 1, ni, stks.data)

### obs.ctrl: 
## age-structured observation model (age2ageDat) 
stkObs.models <- 'age2ageDat'
flq.stk1 <- FLQuant(dimnames = list(age = 'all', 
                                    year = first.yr:last.yr, 
                                    unit = stk1.unit, 
                                    #season = "all",
                                    iter = 1:ni)) 
obs.ctrl.age <- create.obs.ctrl(stksnames = stks, 
                                stkObs.models = stkObs.models, 
                                flq.stk1 = flq.stk1)
obs.ctrl.age[['stk1']][['indObs']] <- vector('list', 1)
names(obs.ctrl.age[['stk1']][['indObs']]) <- c("ind1")
obs.ctrl.age[['stk1']][['indObs']][['ind1']] <- list()
obs.ctrl.age[['stk1']][['indObs']][['ind1']][['indObs.model']]  <- 'ageInd'
nage <- stk1.age.max
ny <- length(first.yr:last.yr)
ages.error <- array(0, dim = c(nage, nage, ny, ni))
## generate errors using the Dirichlet distribution (n, alpha)
for(a in 1:nage){
  for(i in 1:ni){
    for(y in 1:ny){
      if(a == 1) ages.error[1, , y, i] <- rdirichlet(1, c(0.85, 0.1, 0.05, rep(0, 7))) 
      if(a == 2) ages.error[2, , y, i] <- rdirichlet(1, c(0.1, 0.75, 0.1, 0.05, rep(0, 6)))
      if(a == 3) ages.error[3, , y, i] <- rdirichlet(1, c(0.05, 0.1, 0.7, 0.1, 0.05, rep(0, 5)))
      if(a == 4) ages.error[4, , y, i] <- rdirichlet(1, c(rep(0, 1), 0.05, 0.1, 0.7, 0.1, 0.05, rep(0, 4)))
      if(a == 5) ages.error[5, , y, i] <- rdirichlet(1, c(rep(0, 2), 0.05, 0.1, 0.7, 0.1, 0.05, rep(0, 3)))
      if(a == 6) ages.error[6, , y, i] <- rdirichlet(1, c(rep(0, 3), 0.05, 0.1, 0.7, 0.1, 0.05, rep(0, 2)))
      if(a == 7) ages.error[7, , y, i] <- rdirichlet(1, c(rep(0, 4), 0.05, 0.1, 0.7, 0.1, 0.05, rep(0, 1)))
      if(a == 8) ages.error[8, , y, i] <- rdirichlet(1, c(rep(0, 5), 0.05, 0.1, 0.7, 0.1 ,0.05))
      if(a == 9) ages.error[9, , y, i] <- rdirichlet(1, c(rep(0, 6), 0.05, 0.1, 0.7, 0.1))
      if(a == 10) ages.error[10, , y, i] <- rdirichlet(1, c(rep(0, 7), 0.05, 0.1, 0.85))
    }
  }
} 

## set dataframes for uncertainty parameters
nmort.error <- fec.error <- land.wgt.error <- stk.nage.error <- stk.wgt.error <- disc.wgt.error <- 
  land.nage.error <- disc.nage.error  <- flq
TAC.ovrsht  <- flq[1, , , ]
dimnames(TAC.ovrsht)[[1]] <- 'all' 
obs.ctrl.age$stk1$stkObs$TAC.ovrsht <- TAC.ovrsht
obs.ctrl.age$stk1$stkObs$TAC.ovrsht[] <- 10
obs.ctrl.age$stk1$stkObs$land.bio.error <- TAC.ovrsht
obs.ctrl.age$stk1$stkObs$land.bio.error[] <- 50
obs.ctrl.age$stk1$stkObs$disc.bio.error <- TAC.ovrsht
obs.ctrl.age$stk1$stkObs$disc.bio.error[] <- 10
obs.ctrl.age[['stk1']][['stkObs']][['ages.error']] <- ages.error
slts <- c('nmort.error', 'fec.error', 'land.wgt.error', 'stk.nage.error', 'stk.wgt.error', 'disc.wgt.error', 
          'land.nage.error', 'disc.nage.error')
for(sl in slts){
  obs.ctrl.age[['stk1']][['stkObs']][[sl]] <- get(sl)
  obs.ctrl.age[['stk1']][['stkObs']][[sl]][] <- rnorm(prod(dim(flq)), 1, .1)
}

## advice: 
## TAC is set by historical catch for this example
data_tac <- fit$stock.prey %>% select(year, area, biomass_consumed) %>% rename(data = biomass_consumed)
data_tac$data[(proj.yr-first.yr+1):(last.yr-first.yr+1)] = NA
stk1_advice.TAC.flq <- iter(as.FLQuant(as.data.frame(data_tac)), it)
stk1_advice.TAC.flq <- window(stk1_advice.TAC.flq, first.yr, last.yr)
stk1_advice.quota.share.flq <- FLQuant(1, 
                                       dim = c(1, last.yr-first.yr+1, 1), 
                                       dimnames = list(year = first.yr:last.yr))
stk1_advice.avg.yrs <- c((proj.yr-3):(proj.yr-1)) 

## create advice object
stks.data <- list(stk1 = ls(pattern = "^stk1")) 
advice   <- create.advice.data(yrs, ns, ni, stks.data, fleets)

## advice.ctrl: 
# hypothetical HCR and reference points for this example
HCR.models  <- c('IcesHCR') 
blim <- mean(stk1_ssb.flq)*0.15 
btrigger <- mean(stk1_ssb.flq)*0.4
fmsy <- 0.10
ref.pts.stk1      <- matrix(rep(c(blim, btrigger, fmsy), 3), 3, ni, dimnames = list(c('Blim', 'Btrigger','Fmsy'), 1:ni))
advice.ctrl      <- create.advice.ctrl(stksnames = stks, 
                                       HCR.models = HCR.models, 
                                       ref.pts.stk1 = ref.pts.stk1, 
                                       first.yr = first.yr, 
                                       last.yr = last.yr)
advice.ctrl[['stk1']][['sr']] <- list()
advice.ctrl[['stk1']][['sr']][['model']] <- 'geomean'
advice.ctrl$stk1$AdvCatch <- rep(TRUE, length(first.yr:last.yr))   #TRUE advice in catches, FALSE advice in landings
names(advice.ctrl$stk1$AdvCatch) <- as.character((first.yr:last.yr))

## assess.ctrl: 
assess.models <- 'NoAssessment'
assess.ctrl <- create.assess.ctrl(stksnames = stks, assess.models = assess.models)
assess.ctrl[['stk1']]$work_w_Iter <- TRUE
assess.ctrl.sam <- assess.ctrl
assess.ctrl.sam[["stk1"]]$assess.model <- "sam2flbeia"
assess.ctrl.sam$stk1$assess.model <- "sam2flbeia"
assess.ctrl.sam[["stk1"]]$harvest.units <- "f"
assess.ctrl.sam[["stk1"]]$control$indices.type <- "number" 

## check if the conditioning is ok
checkFLBEIAData( biols = biols, 
                 SRs = SRs, 
                 BDs = BDs, 
                 fleets = fleets, 
                 covars = covars, 
                 indices = oneIndAge, 
                 advice = advice, 
                 main.ctrl = main.ctrl, 
                 biols.ctrl = biols.ctrl, 
                 fleets.ctrl = fleets.ctrl,
                 covars.ctrl = covars.ctrl, 
                 obs.ctrl = obs.ctrl.age, 
                 assess.ctrl = assess.ctrl.sam, 
                 advice.ctrl = advice.ctrl)

## Save FLBEIA input objects
 save(biols, SRs, BDs, oneGDGT, had.params, fleets, covars, oneIndAge, advice, main.ctrl, biols.ctrl, fleets.ctrl, 
      covars.ctrl, advice.ctrl, obs.ctrl.age, assess.ctrl.sam, file="input_flbeia-gadget_age_4s_sam.RData")

 
## Run FLBEIA
## age-structured assessment models
s0 <- FLBEIA(biols = biols,
            SRs = SRs,
            BDs = BDs,
            GDGT = oneGDGT,      # GADGET as OM
            fleets = fleets,
            covars = covars,
            indices = oneIndAge,
            advice = advice,
            main.ctrl = main.ctrl,
            biols.ctrl = biols.ctrl,
            fleets.ctrl = fleets.ctrl,
            covars.ctrl = covars.ctrl,
            obs.ctrl = obs.ctrl.age,
            assess.ctrl = assess.ctrl.sam,
            advice.ctrl = advice.ctrl)
## reset the directory
setwd(dir)

## Results FLBEIA
names(s0)
plot(s0$biols[[1]]) 
plot(s0$stocks[[1]]) 

## plot for age-structured models
stk1.mp1 <- s0$stocks[['stk1']]
stk1.om1 <- FLBEIA:::perfectObs(s0$biols[['stk1']], s0$fleets, year = dim(s0$biols[['stk1']]@n)[2])
adf <- as.data.frame
s0_pop <- rbind( data.frame(population='obs', indicator='SSB', as.data.frame(ssb(stk1.mp1))),
                data.frame(population='obs', indicator='Harvest', as.data.frame(fbar(stk1.mp1))),
                data.frame(population='obs', indicator='Catch', as.data.frame(catch(stk1.mp1))),
                data.frame(population='obs', indicator='Recruitment', as.data.frame(rec(stk1.mp1))),
                data.frame(population='real', indicator='SSB', as.data.frame(ssb(stk1.om1))),
                data.frame(population='real', indicator='Harvest', as.data.frame(fbar(stk1.om1))),
                data.frame(population='real', indicator='Catch', as.data.frame(catch(stk1.om1))),
                data.frame(population='real', indicator='Recruitment', as.data.frame(rec(stk1.om1))))
plot1 <- ggplot(data=s0_pop, aes(x=year, y=data, color=population)) +
 geom_line() +
 facet_grid(indicator ~ ., scales="free") +
 geom_vline(xintercept = main.ctrl$sim.years[['initial']]-1, linetype = "longdash")+
 theme_bw()+
 theme(text=element_text(size=15),
       title=element_text(size=15,face="bold"),
       strip.text=element_text(size=15),
       legend.position="top")+
 ylab("")
print(plot1)