-
Notifications
You must be signed in to change notification settings - Fork 0
/
fwd-sim.R
312 lines (261 loc) · 12.3 KB
/
fwd-sim.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
#fwd sim and some wrangling
library(tidyverse)
library(here)
library(gsl)
source(here("analysis/R/functions.R"))
set.seed(123)
# read in data ---------------------------------------------------------------------------
data <- read.csv(here("analysis/data/raw/fr_pk_spw_har.csv")) |>
mutate(harvest = round(harvest/1000000, 2),
spawn = round(spawn/1000000, 2),
ER = harvest/(harvest+spawn))
AR1.stan.fit <- readRDS(here("analysis/data/generated/AR1-SS-SR.stan.fit.rds"))
AR1.model.pars <- rstan::extract(AR1.stan.fit)
TV.stan.fit <- readRDS(here("analysis/data/generated/TV-alpha-SS-SR.stan.fit.rds"))
TV.model.pars <- rstan::extract(TV.stan.fit)
# calc forecast error ---
for.error <- read.csv(here("analysis/data/raw/PinkSalmonPrediction&ObservationDataFile(from Merran).csv")) |>
mutate(error = (abs(PreSeasonForecast-FinalRunSize)/FinalRunSize)) #by year
if(FALSE){ #check trends and distribution of error
ggplot(for.error, aes(YearX, error)) +
geom_point() +
stat_smooth(method = "lm") +
labs(y="forecast error (CV)", x = "year")
ggplot(lm(error~YearX, data = for.error), aes(x = .fitted, y = .resid)) +
geom_point() +
stat_smooth()
}
for.error <- for.error |>
pull(error) |>
mean() #mean of annual means (MAPE)
# get benchmarks & pars ------------------------------------------------------------------
bench <- matrix(NA,1000,4,
dimnames = list(seq(1:1000), c("Sgen","Smsy","Umsy", "Seq")))
for(i in 1:1000){
r <- sample(seq(1,1000),1,replace=TRUE)
ln_a <- AR1.model.pars$ln_alpha[r]
b <- AR1.model.pars$beta[r]
bench[i,2] <- get_Smsy(ln_a, b) #S_MSY
bench[i,1] <- get_Sgen(exp(ln_a),b,-1,1/b*2, bench[i,2]) #S_gen
bench[i,3] <- (1 - lambert_W0(exp(1 - ln_a))) #U_MSY
bench[i,4] <- ln_a/b #S_eq
}
bench[,2] <- bench[,2]*0.8 #make it 80% Smsy
#get some "posteriors" for plotting later
#quotes because it's 1000 random not 4000; will generate the same plot...
Smsy.8.post <- bench[,2]
Sgen.post <- bench[,1]
bench.quant <- apply(bench, 2, quantile, probs=c(0.1,0.5,0.9), na.rm=T) |>
t()
mean <- apply(bench,2,mean, na.rm=T) #get means of each
benchmarks <- cbind(bench.quant, mean) |>
as.data.frame() |>
relocate('50%', 1)
if(FALSE){ #some benchmarks to pass Sue for salmon scanner
bench.quant <- t(apply(bench, 2, quantile, probs=c(0.025,0.05,0.1,0.25,0.5,0.75,0.9,0.95,0.975), na.rm=T))
rownames(bench.quant) <- c("80% Smsy","Sgen","Umsy")
mean <- apply(bench, 2, mean, na.rm=T)
sd <- apply(bench, 2, sd, na.rm=T)
bench.sue <- cbind(bench.quant,mean,sd)
write.csv(bench.sue, "C:/Users/GLASERD/Desktop/Sues.bench.csv")
}
#pull some values to use later and in text
Sgen <- benchmarks[1,1]
Smsy.8 <- benchmarks[2,1]
Umsy <- benchmarks[3,1]
Seq <- benchmarks[4,1]
R.Smsy.8 <- Smsy.8/(1-Umsy) #recruitment @ Smsy to be used as BB USR
percentiles <- quantile(data$spawn, probs=c(0.25, 0.5))
lower.25th.Sp <- percentiles[1]
lower.50th.Sp <- percentiles[2]
alpha.CI <- exp(quantile(AR1.model.pars$ln_alpha, probs = c(.1, .5, .9)))
beta.CI <- quantile(AR1.model.pars$beta, probs = c(.1, .5, .9))
sigma.CI <- quantile(AR1.model.pars$sigma_R_corr, probs = c(.1, .5, .9))
phi.CI <- quantile(AR1.model.pars$phi, probs = c(.1, .5, .9))
par.quants <- rbind(alpha.CI, sigma.CI, beta.CI, phi.CI)
#make big table of bench and pars
par.summary <- as.data.frame(rstan::summary(AR1.stan.fit)$summary) |>
select(mean, n_eff, Rhat)
#summarise not alphas...
par.summary <- filter(par.summary, row.names(par.summary) %in% c('ln_alpha', 'beta',
'phi', 'sigma_R')) |>
slice(1,4,3,2)
pars <- cbind(par.quants, par.summary)
bench.par.table <- bind_rows(benchmarks, pars) |>
mutate(n_eff = round(n_eff, 0),
Rhat = round(Rhat, 4))
bench.par.table[,1:4] <- round(bench.par.table[,1:4], 2)
bench.par.table[is.na(bench.par.table)] <- ""
rownames(bench.par.table) <- c("$S_{gen}$", "80% $S_{MSY}$", "$U_{MSY}$", "$S_{eq}$",
"$\\alpha$", "$\\beta$", "$\\phi$", "$\\sigma_{R}$")
colnames(bench.par.table) <- c("Median", "10th percentile", "90th percentile", "Mean",
"$N_{eff}$", "$\\hat{R}$")
# initialize the sim ---------------------------------------------------------------------
last.yr <- max(data$year) #final yr of model fit
sim.gens <- 1+5 #final state in model + nyrs (i.e. gens for pinks) to fwd sim
n.sims <- 1000
states <- c("R", "S", "C", "U", "under_Sgen", "over_Smsy") #don't THINK I need to track residual?
last.yr.ind <- ncol(TV.model.pars$lnR) #index the last year of data
scenarios <- c("base", "low_prod")
HCRs <- c("current", "PA_alt", "no_fishing")
OU.CV <- 0.1 #assumed outcome uncertainty, did forecast error earlier
#subset posterior for low productivity
recent.a <- apply(TV.model.pars$ln_alpha[,(last.yr.ind-2):last.yr.ind], 1, median) #median of last 3 gens
recent.a.CIs <- exp(quantile(recent.a, probs = c(.1, .5, .9))) #CIs for reporting
recent.a.50s <- exp(quantile(recent.a, probs = c(.25, .75))) #range to trigger reassessment
recent.a.10th <- quantile(recent.a, .1)
low.a.rows <- which(recent.a<recent.a.10th) #rows to subset later
#overwrite low.a pop dynamics to subset later
low.a.ln.alpha <- recent.a[low.a.rows]
low.a.beta <- TV.model.pars$beta[low.a.rows]
low.a.sigma_R_corr <- TV.model.pars$sigma_R_corr[low.a.rows]
fwd.states <- array(NA, dim = c(length(scenarios), length(HCRs), n.sims, sim.gens, length(states)))
for(i in 1:length(scenarios)){
for(j in 1:length(HCRs)){
scenario <- scenarios[i]
HCR <- HCRs[j]
for(k in 1:n.sims){
#take a random correlated slice
r <- sample(length(TV.model.pars$beta), 1, replace = TRUE) #random draw
#draw FULLY CORRELATED pars, states, and benchmarks for the sim
ln_alpha <- recent.a[r]
beta <- TV.model.pars$beta[r]
sigma_R_corr <- TV.model.pars$sigma_R_corr[r]
#estimate draw-specific benchmarks for relative performance measures later
sub.Smsy <- get_Smsy(ln_alpha, beta)
sub.Smsy.8 <- sub.Smsy*0.8
sub.Sgen <- get_Sgen(exp(ln_alpha), beta, -1, 1/beta*2, sub.Smsy)
#draw final states from model to start fwd sim from
R <- TV.model.pars$R[r, last.yr.ind]
S <- TV.model.pars$S[r, last.yr.ind]
C <- TV.model.pars$C[r, last.yr.ind]
U <- TV.model.pars$U[r, last.yr.ind]
fwd.states[i,j,k,1, ] <- c(R,S,C,U,NA,NA) #states and holders for above/below bench
if(scenario == "low_prod"){#subset the low.a dynamics only
r.2 <- sample(length(low.a.rows), 1, replace = TRUE)
ln_alpha <- low.a.ln.alpha[r.2]
beta <- low.a.beta[r.2]
sigma_R_corr <- low.a.sigma_R_corr[r.2]
}
for(l in 2:sim.gens){
#forward estimate recruits
last.S <- fwd.states[i,j,k,l-1,2]
R <- exp(ln_alpha)*last.S*exp(-beta*last.S+sigma_R_corr) #predict R
R <- R*rlnorm(1, 0, for.error) #add forecast error
#then apply the HCR
if(HCR == "current"){post_HCR <- current_HCR(R, OU=1+rnorm(1, 0, OU.CV))}
if(HCR == "PA_alt"){post_HCR <- PA_HCR(R, OU=1+rnorm(1, 0, OU.CV),
Sgen=Sgen, R.Smsy=R.Smsy.8, Umsy=Umsy)}
if(HCR == "no_fishing"){post_HCR <- c(R, 0, 0)}
under.Sgen <- post_HCR[1] < sub.Sgen
over.Smsy.8 <- post_HCR[1] > sub.Smsy.8
fwd.states[i,j,k,l, ] <- c(R, post_HCR, under.Sgen, over.Smsy.8) #write state to array
}
}
}
}
# pull some random individual sims for plotting...
ind.sims <- NULL
for(i in 1:length(scenarios)){
for(j in 1:length(HCRs)){
for(k in 1:2){ #can toggle number of sims here
sub <- fwd.states[i,j,,,]
r <- sample(n.sims, 1, replace = TRUE)
single.sim <- sub[r,,] |>
as.data.frame() |>
select(2,3) |>
mutate(year = seq(max(data$year), max(data$year)+((sim.gens-1)*2), by = 2),
scenario = scenarios[i],
HCR = HCRs[j],
sim = as.factor(k))
ind.sims <- rbind(ind.sims, single.sim)
}
}
}
colnames(ind.sims)[1:2] <- c("spawners", "catch")
#wrangle fwd.sim into summary array of [yr, states(+CIs), HCR] for plotting...
fwd.sim <- NULL
yrs <- as.character(seq(from = last.yr,
to = last.yr+((sim.gens-1)*2), by=2)) #final yr of sim + fwd sims
for(i in 1:length(scenarios)){
for(j in 1:length(HCRs)){
scenario <- scenarios[i]
HCR <- HCRs[j]
sub <- fwd.states[i,j,,,]
for(k in 1:sim.gens){
sub_sub <- as.data.frame(sub[,k,]) |>
reframe(across(1:4, quantile_df, .unpack = TRUE)) |> #could use a better fun.?
mutate(quant = c("lwr", "med", "upr")) |>
pivot_wider(values_from = 1:4, names_from = quant) |>
mutate(scenario = scenario,
HCR = HCR,
year = as.numeric(yrs[k])) |>
relocate(c(year, HCR), 1)
fwd.sim <- rbind(fwd.sim, sub_sub)
}
}
}
fwd.sim <- fwd.sim |>
relocate(scenario, .after=2) |>
as.data.frame()
colnames(fwd.sim) <- c("year", "HCR", "scenario", "R_lwr", "R", "R_upr", "S_lwr", "S",
"S_upr", "C_lwr", "C", "C_upr", "U_lwr", "U", "U_upr")
# summarise performance metrics------------------------------------------------------
perf.metrics <- NULL
#get relative catch index
rel.catch.index <- filter(data, year >= 2000) |>
arrange(desc(harvest)) |>
slice(1:3) |>
summarise(mean(harvest)) |>
pull()
for(i in 1:length(scenarios)){
for(j in 1:length(HCRs)){
sub.data <- fwd.states[i,j,,2:(sim.gens),] #slice HCR to NOT include final year of observed data.
Cs <- sub.data[,,3] #catch sims
#getting total count of points that dipped above/below lines
below.Sgen <- (length(which(sub.data[,,5]==1))/length(sub.data[,,5]))*100
above.Smsy.8 <- (length(which(sub.data[,,6]==1))/length(sub.data[,,6]))*100
#then treating catch more like a distribution and describing the distribution of medians
#by draw, kind of like describing the distribution of the intercepts...
catch <- 1:nrow(Cs) |>
map_dbl(\(x) median(Cs[x,])) |>
quantile(probs = c(.1, .5, .9)) |>
round(2)
catch <- as.character(paste0(catch[2], " (", catch[1], "-", catch[3], ")"))
catch.stability <- 1:nrow(Cs) |>
map_dbl(\(x) median(Cs[x,])/sd(Cs[x,])) |>
quantile(probs = c(.1, .5, .9), na.rm=T) |> #no NAs, but breaks without na.rm=T, maybe because many 0's?
round(2)
catch.stability <- as.character(paste0(catch.stability[2], " (", catch.stability[1], "-",
catch.stability[3], ")"))
catch.index <- length(which(Cs > rel.catch.index))/length(Cs)*100 |> #total points that go above index
round(2)
perf.metrics <- rbind(perf.metrics, data.frame(scenario = rep(scenarios[i],5),
HCR = rep(HCRs[j],5),
value = c(below.Sgen, above.Smsy.8,
catch, catch.stability, catch.index),
metric = c("below.Sgen", "above.Smsy.8", "median annual catch",
"catch.stability", "catch index")))
}
}
perf.metrics <- perf.metrics |>
pivot_wider(names_from = metric, values_from = value) |>
as.data.frame() |>
mutate(catch.stability = ifelse(HCR == "no_fishing", "NA", catch.stability),
scenario = gsub("_", " ", scenario),
HCR = gsub("_", " ", HCR))
write.csv(perf.metrics, here("analysis/data/generated/perf-metrics.csv"), row.names = FALSE)
perf.metrics.fr <- perf.metrics |>
mutate(scenario = case_when(scenario == "base" ~ "de base",
scenario == "low prod" ~ "faible prod"),
HCR = case_when(HCR == "current" ~ "actuelle",
HCR == "PA alt" ~ "autre, AP",
HCR == "no fishing" ~ "Pas de peche"),
catch.stability = ifelse(catch.stability== "NA", "S. O.", catch.stability))
#take the trash out --
rm(beta,ln_a, ln_alpha, C, Cs, catch, catch.stability, fwd.states, bench, bench.quant,
HCR, HCRs, i,j,k,last.S, last.yr, sub.data, n.sims, post_HCR, r, R, S, sigma_R_corr,
sim.gens, states, below.Sgen, U, last.yr.ind, above.Smsy.8, over.Smsy.8, catch.index,
sub.Sgen, under.Sgen, par.quants, par.summary, pars, sub, scenario, scenarios,
single.sim, sub_sub, b, l, low.a.beta, low.a.ln.alpha, low.a.rows, low.a.sigma_R_corr,
mean, percentiles, r.2, sub.Smsy.8, TV.stan.fit, AR1.stan.fit, recent.a, sub.Smsy, yrs)