Last updated: 2025-05-25
Checks: 7 0
Knit directory: GuadalShiftR/
This reproducible R Markdown analysis was created with workflowr (version 1.7.1). The Checks tab describes the reproducibility checks that were applied when the results were created. The Past versions tab lists the development history.
Great! Since the R Markdown file has been committed to the Git repository, you know the exact version of the code that produced these results.
Great job! The global environment was empty. Objects defined in the global environment can affect the analysis in your R Markdown file in unknown ways. For reproduciblity it’s best to always run the code in an empty environment.
The command set.seed(20240207) was run prior to running
the code in the R Markdown file. Setting a seed ensures that any results
that rely on randomness, e.g. subsampling or permutations, are
reproducible.
Great job! Recording the operating system, R version, and package versions is critical for reproducibility.
Nice! There were no cached chunks for this analysis, so you can be confident that you successfully produced the results during this run.
Great job! Using relative paths to the files within your workflowr project makes it easier to run your code on other machines.
Great! You are using Git for version control. Tracking code development and connecting the code version to the results is critical for reproducibility.
The results in this page were generated with repository version df41d36. See the Past versions tab to see a history of the changes made to the R Markdown and HTML files.
Note that you need to be careful to ensure that all relevant files for
the analysis have been committed to Git prior to generating the results
(you can use wflow_publish or
wflow_git_commit). workflowr only checks the R Markdown
file, but you know if there are other scripts or data files that it
depends on. Below is the status of the Git repository when the results
were generated:
Ignored files:
Ignored: .Rhistory
Ignored: .Rproj.user/
Ignored: MATERIAL/
Ignored: README.html
Ignored: pipeline.R
Note that any generated files, e.g. HTML, png, CSS, etc., are not included in this status report because it is ok for generated content to have uncommitted changes.
These are the previous versions of the repository in which changes were
made to the R Markdown (analysis/Reproduce.Rmd) and HTML
(docs/Reproduce.html) files. If you’ve configured a remote
Git repository (see ?wflow_git_remote), click on the
hyperlinks in the table below to view the files as they were in that
past version.
| File | Version | Author | Date | Message |
|---|---|---|---|---|
| html | 74dccb4 | Pablo Almaraz | 2025-05-25 | Build site. |
| html | e8c0882 | Pablo Almaraz | 2025-05-25 | Build site. |
| Rmd | 66150bc | Pablo Almaraz | 2025-05-25 | Meso-commit |
| html | 87d923e | Pablo Almaraz | 2024-11-05 | Build site. |
| Rmd | 5342c0b | Pablo Almaraz | 2024-11-05 | Start my new project |
| html | 640ccfd | Pablo Almaraz | 2024-02-08 | Build site. |
| html | 4708b57 | Pablo Almaraz | 2024-02-08 | Build site. |
| html | 558f454 | Pablo Almaraz | 2024-02-08 | Build site. |
| html | 3d3d0d9 | Pablo Almaraz | 2024-02-08 | Build site. |
| html | 106257e | Pablo Almaraz | 2024-02-08 | Build site. |
| html | 023f958 | Pablo Almaraz | 2024-02-08 | Build site. |
| Rmd | ea46d83 | Pablo Almaraz | 2024-02-08 | Start my new project |
| Rmd | 8a7d061 | Pablo Almaraz | 2024-02-08 | Big commit |
| html | 8a7d061 | Pablo Almaraz | 2024-02-08 | Big commit |
| Rmd | df580e2 | Pablo Almaraz | 2024-02-08 | Commit |
| html | e88c786 | Pablo Almaraz | 2024-02-08 | Build site. |
| Rmd | 83cf2e4 | Pablo Almaraz | 2024-02-08 | Start my new project |
| html | c29b0cf | Pablo Almaraz | 2024-02-07 | Build site. |
| Rmd | 5cda8e9 | Pablo Almaraz | 2024-02-07 | Start my new project |
| html | febb514 | Pablo Almaraz | 2024-02-07 | Build site. |
| html | bd8c1e8 | Pablo Almaraz | 2024-02-07 | Build site. |
| html | 5f474b6 | Pablo Almaraz | 2024-02-07 | Build site. |
| html | 602ce6d | Pablo Almaraz | 2024-02-07 | Build site. |
| html | 8ed1ed6 | Pablo Almaraz | 2024-02-07 | Build site. |
| html | b6dcf46 | Pablo Almaraz | 2024-02-07 | Build site. |
| Rmd | c4037f1 | Pablo Almaraz | 2024-02-07 | Start my new project |
The simplest way to reproduce the analyses is to run the
MAKEFILE.R file. Executing this chunck will perform all the
statistical analyses and produce all the figures for the main and
supplementary files.
source('MAKEFILE.R')As with all Bayesian numerical integration techniques, either using Gibbs sampling, (JAGS), HMC-NUTS, (Stan) or other sampling methods, the posterior distribution will always be an approximation to the target distribution, up to a constant and accounting for MC error. In this case, both the Bayesian DFA (using HMC-NUTS), and the SSRDLVR model (using Gibbs MCMC sampling), are very high-dimensional approaches, yielding very high-dimensional posterior distributions. As a consequence, the values of summaries from the posterior distributions will be somewhat sensitive not only to prior specification, but also to initial values of the MC chains. In particular, providing informative initial values for the MC chains (that is, abusing of language, values that are within the basin of attraction of the target distribution), can greatly speed up the convergence and provide meaningful posterior values. Keep this in mind when modifying the prior or the initial values of these or other models!
The following code is structured with these issues in mind:
Depending on your computing architecture, the fitting of the
Bayesian DFA and the SSRDLVR model can take up to several hours (most
likely), or even days (unlikely). Therefore, the essential output from
the fitted posterior distributions is already saved in the
output folder, and the analyses from this file and the
MAKEFILE.R and Makefile files use this default
procedure for quickly producing the figures and compiling the
manuscript.
If you decide to re-fit all the models, set
fit_DFA=TRUE and fit_SSRDLVR=TRUE here and
elsewhere. Keep in mind the issue with initial and prior values
described above! In particular, do not modify the random seeds used in
this project. If you do, expect slightly diverging numerical results. In
particular, the estimation of the empirical probability of feasibility
from the SSRDLVR is particularly sensitive to these issues. Even
depending on the particular version of R, Stan, JAGS, gcc, etc.,
slightly different results are expected. This is
annoying.
Finally, keep in mind that the chunks of this Rmarkdown are set
to eval=FALSE, if you want to run the chuncks, set to
eval=TRUE.
Please look the README file for further details on
reproducing the results of this project. Note that the following chunks
are supposed to run sequentially.
Check that the librarian package
is installed. Alternatively, note that you can use the checkpoint
package to build an environment identical to the date of completion of
the present project. Look at the Session information tab
below. After librarian is
installed, the shelf function will check that the selected
packages are installed. If they are, it will load them; if not, it will
first install them and then load them.
# Load libraries ####
# You can choose to make a checkpoint to reproduce all the analyses with a
# CRAN snapshot of the day of acceptance of the paper. If so, uncomment the
# two following lines:
# if (!require(checkpoint)) install.packages('checkpoint')
# checkpoint::checkpoint("2024-01-20")
if (!require(librarian)) install.packages('librarian') # Install, if necessary, the 'librarian' packageCargando paquete requerido: librarian# Automatically install, if necessary, from whichever source (CRAN, Bioconductor...), the necessary
# packages to conduct the analyses, and load them
librarian::shelf(tidyverse,readr,runjags,rstan,bayesdfa,coda,ggmcmc,bayesplot,qgraph,
imputeTS,ggridges,viridis,bayestestR,cusp,mvtnorm,ggbreak,psych,data.table,
reshape2,patchwork,ggrepel)source('code/functions.R')This will run the analyses in a temporal sequence.
# Load data ####
message('\nLoad data\n')
Count_data = read_delim("data/Aerial_count_data.csv",delim = ";", escape_double = FALSE, trim_ws = TRUE)
Sp_names_long = c("Pintail","Shoveler","Common teal","Eurasian wigeon","Mallard",
"Gadwall","Greylag goose","Common pochard","Red-crested pochard","Shelduck")
Sp_names_short = c("Pintail","Shoveler","C. teal","E. wigeon","Mallard",
"Gadwall","G. goose","C. pochard","R-c. pochard","Shelduck")
Env_data = read_delim("data/Environmental_data.csv",delim = ";", escape_double = FALSE, trim_ws = TRUE)
if (!dir.exists("output/figures")) { dir.create("output/figures") }# Bayesian DFA ####
message('\nBegin the fitting of the Bayesian Dynamic Factor Analysis:\n')
set.seed(498468)
options(mc.cores = parallel::detectCores()) # Use all the available cores (if using a multi-core processor)
## Long format for the waterfowl dataset
years = 1978:2013
WaterfowlData = Count_data[-nrow(Count_data),] %>%
dplyr::mutate(time=rep(1:length(years), each=2)) %>%
dplyr::select(time,
Anas_acuta,
Anas_clypeata,
Anas_crecca,
Anas_penelope,
Anas_platyrhynchos,
Anas_strepera,
Anser_anser,
Aythya_ferina,
Netta_rufina,
Tadorna_tadorna) %>%
reshape2::melt(., id=c("time")) %>%
dplyr::mutate(obs=log(value+1),
ts=as.numeric(variable),
time=time) %>%
dplyr::select(obs,ts,time)
# Specify some items
chains = 3
iter = 6000
n_knots = 16
fit_DFA=FALSE
if(fit_DFA==TRUE){
## DFA with 1 trend ####
message('\nFit a Bayesian DFA with 1 trend\n')
DFA_1trend_PS = fit_dfa(
y = WaterfowlData, num_trends = 1, data_shape = "long", estimation = "sampling",
scale="zscore", trend_model="ps", n_knots = n_knots,
iter = iter, chains = chains, thin = 1, refresh = 100, verbose = T, expansion_prior=T,
seed = 1083941809)
save(DFA_1trend_PS, file = 'output/BDFA_model/DFA_1trend_PS.Rdata')
# LOO cross-validation
loocv_DFA_1trend_PS = loo(DFA_1trend_PS)
rotate_DFA_1trend_PS <- rotate_trends(DFA_1trend_PS, conf_level=0.9, invert = F)
plot_trends(rotate_DFA_1trend_PS, years=years) +
xlab("Year") + ylab("Dynamic factor") + theme_bw() + theme(panel.grid = element_blank())
# DFA with 2 trends ####
message('\nFit a Bayesian DFA with 2 trends\n')
DFA_2trends_PS = fit_dfa(
y = WaterfowlData, num_trends = 2, data_shape = "long", estimation = "sampling",
scale="zscore", trend_model="ps", n_knots = n_knots,
iter = iter, chains = chains, thin = 1, refresh = 100, verbose = T, expansion_prior=T,
seed=4128785653)
save(DFA_2trends_PS, file = 'output/BDFA_model/DFA_2trends_PS.Rdata')
# LOO cross-validation
loocv_DFA_2trends_PS = loo(DFA_2trends_PS)
rotate_DFA_2trends <- rotate_trends(DFA_2trends_PS, conf_level=0.9, invert = F)
# Make some changes to display the major trend first
trends <- dfa_trends(rotate_DFA_2trends)
trends$trend_number = as.factor(trends$trend_number)
levels(trends$trend_number) <- c("Trend 2","Trend 1")
trends$trend_number = as.character(trends$trend_number)
trends$trend_number = factor(trends$trend_number, labels = c("Trend 1","Trend 2"))
trends$year = rep(years,2)
plot_rotate_DFA_2trends = ggplot(trends, aes(x = year, y = estimate)) +
geom_ribbon(aes(ymin = lower, ymax = upper),
alpha = 0.4) + geom_line() + facet_wrap("trend_number") +
xlab("Year") + ylab("Dynamic factor") + theme_bw() + theme(panel.grid = element_blank())
# Plot the major common trends of the community
pdf("output/figures/Common_Trends_raw.pdf",height=4,width=8)
print(plot_rotate_DFA_2trends)
dev.off()
Common_Trends_by_Species_plot = plot_fitted(DFA_2trends_PS, conf_level = 0.9, names=Sp_names_long, time_labels=years, spaghetti=F) +
geom_point(aes(x = time,
y = y), col = "royalblue", size = 1, alpha = 1) +
theme_bw() +
ylab('Abdundance (standardised)') + xlab('Year') + theme(panel.grid = element_blank())
pdf("output/figures/Common_Trends_by_Species.pdf",height=6,width=7)
print(Common_Trends_by_Species_plot)
dev.off()
# Plot the factor loadings
FacLoad = dfa_loadings(rotate_DFA_2trends, summary = FALSE, names = Sp_names_long, conf_level = 0.95)
levels(FacLoad$trend) <- c("Trend 2","Trend 1")
FacLoad$loading = 1*FacLoad$loading
FacLoad$trend = relevel(FacLoad$trend, "Trend 1")
FacLoad_plot = ggplot(FacLoad, aes(x = name, y = loading, fill = trend, alpha = prob_diff0)) +
geom_violin(color = NA) +
scale_fill_manual(values = c("royalblue", "orange2")) +
scale_y_continuous(limits = c(-2.5, 5)) +
geom_hline(yintercept = 0, lty = 2) +
coord_flip() +
xlab("Species") + ylab("Loading")
FacLoad_plot = FacLoad_plot + facet_wrap(~trend, scales = "free_x")
pdf("output/figures/Factor_loadings.pdf",height=5,width=7)
print(FacLoad_plot)
dev.off()
# Fit a Hidden Markov Model to identify regime shifts in the dynamic common trends
message('\nFit a Hidden Markov Model to identify regime shifts in the dynamic common trends:\n')
# HMM with 1 regime
HMM_model_DFA_1trend = fit_regimes(
y = rotate_DFA_2trends$trends_mean[2, ],
sds = (rotate_DFA_2trends$trends_upper - rotate_DFA_2trends$trends_mean)[2, ] / 1.96,
n_regimes = 1,
iter = iter, chains = chains, refresh=iter, verbose = F)
# HMM with 2 regimes
HMM_model_DFA_2trends <- fit_regimes(
y = rotate_DFA_2trends$trends_mean[2, ],
sds = (rotate_DFA_2trends$trends_upper - rotate_DFA_2trends$trends_mean)[2, ] / 1.96,
n_regimes = 2,
iter = iter, chains = chains, refresh=iter, verbose = F)
message(paste0('\nThe looic for the HMM with 1 regime is ',
round(HMM_model_DFA_1trend$looic,3), '\n',
'The looic for the HMM with 2 regimes is ',
round(HMM_model_DFA_2trends$looic,3), '\n',
'The difference is of ', round(HMM_model_DFA_1trend$looic,3) -
round(HMM_model_DFA_2trends$looic,3),'\n',
sep=''))
# Plot the regimes
pdf("output/figures/Regime_shift.pdf",height=5,width=5)
plot_hmm_regimes(HMM_model_DFA_2trends,years=years)
dev.off()
# save(DFA_1trend_PS,DFA_2trends_PS,HMM_model_DFA_1trend,HMM_model_DFA_2trends,
# file='output/BDFA_model/BayesDFA_results.Rdata')
}
if(fit_DFA==FALSE) {
load('output/BDFA_model/BDFA_results.Rdata')
# Plot the major common trends of the community
plot_rotate_DFA_2trends = ggplot(trends, aes(x = year, y = estimate)) +
geom_ribbon(aes(ymin = lower, ymax = upper),
alpha = 0.4) + geom_line() + facet_wrap("trend_number") +
xlab("Year") + ylab("Dynamic factor") + theme_bw() + theme(panel.grid = element_blank())
pdf("output/figures/Common_Trends_raw.pdf",height=4,width=8)
print(plot_rotate_DFA_2trends)
dev.off()
# Difference in looic between models
message(paste0('\nThe looic for the DFA with 1 trend is ',
round(loocv_DFA_1trend_PS$estimates['looic','Estimate'],3), '\n',
'The looic for the DFA with 2 trends is ',
round(loocv_DFA_2trends_PS$estimates['looic','Estimate'],3), '\n',
'The difference is of ', round(loocv_DFA_1trend_PS$estimates['looic','Estimate'],3) -
round(loocv_DFA_2trends_PS$estimates['looic','Estimate'],3),'\n',
sep=''))
Common_Trends_by_Species_plot = ggplot(trends_by_sp) +
geom_ribbon(aes(x = time, ymin = lower, ymax = upper), alpha = 0.4) +
geom_line(aes(x = time, y = estimate)) +
geom_point(col = "royalblue", size = 1, alpha = 1, aes(x = time, y = y)) +
facet_wrap("ID", scales = "free_y") + xlab("Time") +
ylab("") +
theme_bw() +
ylab('Abdundance (standardised)') + xlab('Year') +
theme(panel.grid = element_blank())
pdf("output/figures/Common_Trends_by_Species.pdf",height=6,width=7)
print(Common_Trends_by_Species_plot)
dev.off()
# Plot the factor loadings
FacLoad_plot = ggplot(FacLoad, aes(x = name, y = loading, fill = trend, alpha = prob_diff0)) +
geom_violin(color = NA) +
scale_fill_manual(values = c("royalblue", "orange2")) +
scale_y_continuous(limits = c(-2.5, 5)) +
geom_hline(yintercept = 0, lty = 2) +
coord_flip() +
xlab("Species") + ylab("Loading")
FacLoad_plot = FacLoad_plot + facet_wrap(~trend, scales = "free_x")
pdf("output/figures/Factor_loadings.pdf",height=5,width=7)
print(FacLoad_plot)
dev.off()
# HMM with 2 regimes
HMM_model_DFA_2trends <- fit_regimes(
y = rotate_DFA_2trends_trends_mean,
sds = sds,
n_regimes = 2,
iter = iter, chains = chains, refresh=iter, verbose = F)
message(paste0('\nThe looic for the HMM with 1 regime is ',
round(HMM_model_DFA_1trend_looic,3), '\n',
'The looic for the HMM with 2 regimes is ',
round(HMM_model_DFA_2trends_looic,3), '\n',
'The difference is of ', round(HMM_model_DFA_1trend_looic,3) -
round(HMM_model_DFA_2trends_looic,3),'\n',
sep=''))
# Plot the regimes
pdf("output/figures/Regime_shift.pdf",height=5,width=5)
plot_hmm_regimes(HMM_model_DFA_2trends,years=years)
dev.off()
}
# END Bayesian DFA ####
# State-space regime-dependent LVR model ####
seed=827545
set.seed(seed)
runjags.options(inits.warning=F,
rng.warning=F,
blockignore.warning=F,
blockcombine.warning=F,
nodata.warning=F,
silent.jags=F,
silent.runjags=F)
testjags()
scal_Fact=1000 # Scaling factor for abundance data
NSpecies = 10
WaterfowlData = Count_data[-nrow(Count_data),] %>%
dplyr::mutate(time=rep(1:36, each=2)) %>%
dplyr::select(time,
Anas_acuta,
Anas_clypeata,
Anas_crecca,
Anas_penelope,
Anas_platyrhynchos,
Anas_strepera,
Anser_anser,
Aythya_ferina,
Netta_rufina,
Tadorna_tadorna) %>%
reshape2::melt(., id=c("time")) %>%
dplyr::mutate(obs=value,
ts=as.numeric(variable),
time=time) %>%
dplyr::select(obs,ts,time)
Period = c('PrePinatubo', 'PostPinatubo')
## Loop ####
for(period in Period) {
message(paste0('\nFit the regime-dependent SSLVR model to the ', period, ' period\n'))
## Data ####
if(period=='PrePinatubo') NYears=length(1978:1991)
if(period=='PostPinatubo') {
FirstYear = 21
FinalYear = max(WaterfowlData$time)
NYears = FinalYear-FirstYear
}
if(period=='PrePinatubo') {
est.k = WaterfowlData %>% dplyr::filter(time < NYears) %>% group_by(ts) %>% summarise_all(mean, na.rm=T) %>% select(-ts, -time)
est.k.var = WaterfowlData %>% dplyr::filter(time < NYears) %>% group_by(ts) %>% mutate(obs.var = obs/scal_Fact) %>% select(-obs) %>% summarise_all(var, na.rm=T) %>% select(-ts, -time)
data_list_SSLVR_regime = list(
NSpecies = NSpecies,
NYears = NYears,
n1 = as.matrix((log((Count_data[-1,] %>% dplyr::filter(Month == 12) %>% dplyr::select(-Year, -Month))/scal_Fact + 1)))[1:NYears,],
n2 = as.matrix((log((Count_data[-1,] %>% dplyr::filter(Month == 1) %>% dplyr::select(-Year, -Month))/scal_Fact + 1)))[1:NYears,],
flood = scale(Env_data[1:NYears,'Flood'])[,1],
est.k = est.k$obs/scal_Fact,
est.k.var = est.k.var$obs.var)
}
if(period=='PostPinatubo') {
est.k = WaterfowlData %>% dplyr::filter(time > FirstYear) %>% group_by(ts) %>% summarise_all(mean, na.rm=T) %>% select(-ts, -time)
est.k.var = WaterfowlData %>% dplyr::filter(time > FirstYear) %>% group_by(ts) %>% mutate(obs.var = obs/scal_Fact) %>% select(-obs) %>% summarise_all(var, na.rm=T) %>% select(-ts, -time)
data_list_SSLVR_regime = list(
NSpecies = NSpecies,
NYears = NYears,
n1 = as.matrix((log((Count_data[-1,] %>% dplyr::filter(Month == 12) %>% dplyr::select(-Year, -Month))/scal_Fact + 1)))[(FirstYear+1):FinalYear,],
n2 = as.matrix((log((Count_data[-1,] %>% dplyr::filter(Month == 1) %>% dplyr::select(-Year, -Month))/scal_Fact + 1)))[(FirstYear+1):FinalYear,],
flood = scale(Env_data[(FirstYear+1):FinalYear,'Flood'])[,1],
est.k = est.k$obs/scal_Fact,
est.k.var = est.k.var$obs.var)
}
parameter_list_SSLVR_regime = c("prob_inter","g.alpha","alpha","k","r","b","gamma",
"Corr.sigma","Corr.tau","state","n1_ppc","n2_ppc")
## Inits for the SSLVR model ####
if(period=='PrePinatubo'){
# Informative initial conditions, to speed up convergence
b_mean=c(-0.8544,-1.05,0.6868,0.03144,0.3442,-0.06122,-0.3627,-1.763,-3.054,-1.235)
n1 = matrix(NA, NYears, NSpecies)
n1[12,] = unname(data_list_SSLVR_regime$n1[11,]+data_list_SSLVR_regime$n1[13,])/2
n2 = matrix(NA, NYears, NSpecies)
n2[7,] = unname(data_list_SSLVR_regime$n1[6,]+data_list_SSLVR_regime$n1[8,])/2
inits_list_SSLVR_regime=function(){
list(
prob_inter=rbeta(1,2,8),
active_alpha=replicate(NSpecies,rnorm(NSpecies,0,0.1)) + diag(NA, NSpecies),
g.alpha=replicate(NSpecies,rbinom(NSpecies,1,0.5)) + diag(NA, NSpecies),
Obs.prec.mat=diag(0.1,NSpecies),
Sys.prec.mat=diag(0.1,NSpecies),
k=pmax(rnorm(NSpecies, data_list_SSLVR_regime$est.k, 1), 0.1),
b=rnorm(NSpecies, b_mean,0.1),
gamma=rnorm(NSpecies, b_mean, 0.1),
n1=n1,
n2=n2,
state=matrix(c(NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,
1.54,3.06,2.76,3.54,1.99,1.36,3.77,0.87,2,1.15,
1.99,3.82,2.88,3.78,1.8,1.52,4.1,1.49,2.74,1.09,
1.66,2.77,2.44,3.31,1.29,0.76,4.02,0.76,0.9,1.1,
2.49,3.61,2.6,3.89,1.03,0.94,4,1.35,2.03,0.92,
2.37,3.65,3.34,3.66,0.9,0.99,4.08,1.45,2.13,0.96,
2.78,3.41,3.48,3.88,1.41,1.19,4.04,1.78,2.08,1.01,
2.58,3.53,3.47,3.89,1.75,1.5,4.1,1.55,1.95,1.1,
2.5,3.65,3.63,3.61,1.83,1.39,4.07,1.7,2.46,1.07,
2.59,3.41,3.88,3.94,1.6,1.07,3.97,1.65,2.32,1.12,
2.84,3.48,4.23,3.72,2.16,1.53,4,1.67,2.11,1.32,
2.26,3.74,4.06,3.97,2.29,1.85,4.14,1.04,2.42,1.39,
2.73,3.4,4.22,3.72,2.34,1.36,3.79,0.93,2.01,1.25,
2.41,3.92,4.15,3.81,2.24,1.44,3.98,1.07,2.68,1.13),
NYears,NSpecies,byrow = T),
.RNG.seed=seed)
}
}
if(period=='PostPinatubo'){
# Informative initial conditions, to speed up convergence
b_mean=c(-0.8544,-1.05,0.6868,0.03144,0.3442,-0.06122,-0.3627,-1.763,-3.054,-1.235)
inits_list_SSLVR_regime=function(){
list(
prob_inter=rbeta(1,2,8),
active_alpha=replicate(NSpecies,rnorm(NSpecies,0,0.1)) + diag(NA, NSpecies),
g.alpha=replicate(NSpecies,rbinom(NSpecies,1,0.5)) + diag(NA, NSpecies),
Obs.prec.mat=diag(0.1,NSpecies),
Sys.prec.mat=diag(0.1,NSpecies),
k=pmax(rnorm(NSpecies, data_list_SSLVR_regime$est.k, 1), 0.1),
b=rnorm(NSpecies, b_mean,0.1),
gamma=rnorm(NSpecies, b_mean, 0.1),
state=matrix(c(NA,NA,NA,NA,NA,NA,NA,NA,NA,NA,
0.5,2.25,1.97,0.91,1.47,0.9,3.08,0.98,3.24,0.77,
1.18,2.87,2.83,1.17,1.69,-2.06,3.62,0.77,0.99,0.47,
2.1,3.09,2.73,1.66,1.91,-2.04,3.85,0.8,0.54,0.37,
2.6,3,2.71,1.92,1.7,-2.24,4.09,0.75,0.8,0.5,
3.12,3.22,2.8,1.93,1.96,-2.19,3.94,0.63,0.86,0.4,
3.47,3.15,2.71,2.03,1.88,-2.21,3.88,0.71,0.59,0.19,
2.63,3.58,2.07,1.96,1.47,-1.86,4.09,0.5,2.3,0.74,
2.42,3.98,2.64,2.09,1.2,-2.24,3.95,0.47,3.11,0.74,
3.09,3.87,2.91,2.12,1.98,-5.99,3.92,0.64,0.6,0.58,
3.23,3.79,2.71,2.19,1.78,-5.95,4.15,0.7,0.68,0.49,
3.33,3.77,2.79,2.15,1.94,-5.94,3.95,0.68,0.89,0.57,
3.39,3.57,2.59,2.2,1.8,-5.98,4.08,0.83,0.64,0.72,
3.64,3.58,2.92,2.24,1.73,-6,3.81,0.88,0.58,0.85,
3.31,3.82,2.57,2.1,1.75,-5.87,4,0.81,0.98,0.8),
NYears,NSpecies,byrow = T),
.RNG.seed=seed)
}
}
## Model fit ####
fit_SSRDLVR=FALSE
if(fit_SSRDLVR==TRUE) {
SSRDLVR_mcmc_analysis = runjags::run.jags(
data=data_list_SSLVR_regime,
inits=inits_list_SSLVR_regime,
monitor=parameter_list_SSLVR_regime,
model=read.jagsfile("code/SSRDLVR_model.JAGS"),
n.chains = 3,
adapt = 5000,
burnin = 100000,
sample = 1000,
thin = 50,
method='parallel',
modules = 'glm',
keep.jags.files=file.path(paste0("output/SSRDLVR_model/","mcmc_run_",period,sep = "")),
summarise=TRUE)
save(SSRDLVR_mcmc_analysis, file = paste0('output/SSRDLVR_model/SSRDLVR_model_results_',period,'.Rdata'))
# Convert the posterior object
post_vars = as.data.frame(combine.mcmc(as.mcmc.list(SSRDLVR_mcmc_analysis)))
nloops = nrow(post_vars)
}
if(fit_SSRDLVR==FALSE) {
load(paste0('output/SSRDLVR_model/SSRDLVR_model_results_',period,'.Rdata'))
if(period=='PrePinatubo') post_vars = as.data.frame(combine.mcmc(as.mcmc.list(SSRDLVR_model_results_PrePinatubo)))
if(period=='PostPinatubo') post_vars = as.data.frame(combine.mcmc(as.mcmc.list(SSRDLVR_model_results_PostPinatubo)))
nloops = nrow(post_vars)
}
## MCMC diagnostics ####
plot=FALSE
if(plot){
if (!dir.exists("output/SSRDLVR_model/MCMC_checks")) { dir.create("output/SSRDLVR_model/MCMC_checks") }
if(period=='PrePinatubo') ggs_object = ggs(as.mcmc.list(SSRDLVR_model_results_PrePinatubo))
if(period=='PostPinatubo') ggs_object = ggs(as.mcmc.list(SSRDLVR_model_results_PostPinatubo))
ggmcmc(ggs_object,family="prob_inter",
file = paste0("output/SSRDLVR_model/MCMC_checks/MCMC_diagnostics_prob_inter_",period,".pdf"),
param_page = 5, width=6, height=6)
ggmcmc(ggs_object,family="b",
file = paste0("output/SSRDLVR_model/MCMC_checks/MCMC_diagnostics_b_",period,".pdf"),
param_page = 11, width=5, height=12)
ggmcmc(ggs_object,family="k",
file = paste0("output/SSRDLVR_model/MCMC_checks/MCMC_diagnostics_k_",period,".pdf"),
param_page = 10, width=5, height=12)
ggmcmc(ggs_object,family="r",
file = paste0("output/SSRDLVR_model/MCMC_checks/MCMC_diagnostics_r_",period,".pdf"),
param_page = 10, width=5, height=12)
ggmcmc(ggs_object,family="alpha",
file = paste0("output/SSRDLVR_model/MCMC_checks/MCMC_diagnostics_alpha_",period,".pdf"),
param_page = 10, width=5, height=12)
}
## Plot prior vs. posterior ####
nu=dplyr::select(post_vars,starts_with("prob_inter"))
pdf(file=paste0('output/figures/Prior_post_',period,'.pdf'), height = 6, width = 6)
prior_vs_posterior(nu$prob_inter, dist = "beta", xlim = c(0, 1), alpha=2, beta=8, pos.legend="right",
title = "Prior vs. posterior distribution")
dev.off()
## PPC plot ####
Obs_mean_Dec = data_list_SSLVR_regime$n1
Obs_mean_Jan = data_list_SSLVR_regime$n2
colnames(Obs_mean_Dec) = colnames(Obs_mean_Jan) = Sp_names_long
yobs_n1_ppc_mcmc = dplyr::select(post_vars, starts_with("n1_ppc"))
yobs_n2_ppc_mcmc = dplyr::select(post_vars, starts_with("n2_ppc"))
ppc_n1_mcmc = array(NA, dim = c(nloops, NYears-1, NSpecies))
for(j in 1:NSpecies){
ppc_n1_mcmc[,,j] = yobs_n1_ppc_mcmc %>%
dplyr::select(bayesplot::param_glue("n1_ppc[{time},{type}]", type = j, time = 2:NYears)) %>%
as.matrix()
}
ppc_n1_yobs = array(NA, dim = c(3, NYears-1, NSpecies))
for(l in 1:NSpecies){
ppc_n1_yobs[,,l] = apply(ppc_n1_mcmc[,,l], 2 , quantile , probs = c(0.10, 0.5, 0.9) , na.rm = TRUE )
}
ppc_n2_mcmc = array(NA, dim = c(nloops, NYears-1, NSpecies))
for(j in 1:NSpecies){
ppc_n2_mcmc[,,j] = yobs_n2_ppc_mcmc %>%
dplyr::select(bayesplot::param_glue("n2_ppc[{time},{type}]", type = j, time = 2:NYears)) %>%
as.matrix()
}
ppc_n2_yobs = array(NA, dim = c(3, NYears-1, NSpecies))
for(l in 1:NSpecies){
ppc_n2_yobs[,,l] = apply(ppc_n2_mcmc[,,l], 2 , quantile , probs = c(0.025, 0.5, 0.975) , na.rm = TRUE )
}
# for December counts
Yobs_n1 = array(NA, dim=c((NYears-1)*NSpecies, 4))
temp = array(NA, dim=c((NYears-1)*NSpecies, 4))
for(l in 1:NSpecies){
if(l==1){
temp = as.data.frame(t(ppc_n1_yobs[,,l])) %>%
dplyr::mutate(Type = rep(Sp_names_long[l], NYears-1),
ppc_Lower95 = V1,
ppc_mean = V2,
ppc_Upper95 = V3) %>%
dplyr::select(Type, ppc_Lower95, ppc_mean, ppc_Upper95)
Yobs_n1 = temp
}
else{
temp = as.data.frame(t(ppc_n1_yobs[,,l])) %>%
dplyr::mutate(Type = rep(Sp_names_long[l], NYears-1),
ppc_Lower95 = V1,
ppc_mean = V2,
ppc_Upper95 = V3) %>%
dplyr::select(Type, ppc_Lower95, ppc_mean, ppc_Upper95)
Yobs_n1 = rbind.data.frame(Yobs_n1, temp)
}
}
# for January counts
Yobs_n2 = array(NA, dim=c((NYears-1)*NSpecies, 4))
temp = array(NA, dim=c((NYears-1)*NSpecies, 4))
for(l in 1:NSpecies){
if(l==1){
temp = as.data.frame(t(ppc_n2_yobs[,,l])) %>%
dplyr::mutate(Type = rep(Sp_names_long[l], NYears-1),
ppc_Lower95 = V1,
ppc_mean = V2,
ppc_Upper95 = V3) %>%
dplyr::select(Type, ppc_Lower95, ppc_mean, ppc_Upper95)
Yobs_n2 = temp
}
else{
temp = as.data.frame(t(ppc_n2_yobs[,,l])) %>%
dplyr::mutate(Type = rep(Sp_names_long[l], NYears-1),
ppc_Lower95 = V1,
ppc_mean = V2,
ppc_Upper95 = V3) %>%
dplyr::select(Type, ppc_Lower95, ppc_mean, ppc_Upper95)
Yobs_n2 = rbind.data.frame(Yobs_n2, temp)
}
}
# Latent states
latent_state_mcmc = dplyr::select(post_vars, starts_with("state"))
state_mcmc = array(NA, dim = c(nloops, NYears-1, NSpecies))
for(j in 1:NSpecies){
state_mcmc[,,j] = latent_state_mcmc %>%
dplyr::select(bayesplot::param_glue("state[{time},{type}]", type = j, time = 2:NYears)) %>%
as.matrix()
}
latent_state = array(NA, dim = c(3, NYears-1, NSpecies))
for(l in 1:NSpecies){
latent_state[,,l] = apply(state_mcmc[,,l], 2 , quantile , probs = c(0.025, 0.5, 0.975) , na.rm = TRUE )
}
state = array(NA, dim=c((NYears-1)*NSpecies, 4))
temp = array(NA, dim=c((NYears-1)*NSpecies, 4))
for(l in 1:NSpecies){
if(l==1){
temp = as.data.frame(t(latent_state[,,l])) %>%
dplyr::mutate(Type = rep(Sp_names_long[l], NYears-1),
state_Lower95 = V1,
state_mean = V2,
state_Upper95 = V3) %>%
dplyr::select(Type, state_Lower95, state_mean, state_Upper95)
state = temp
}
else{
temp = as.data.frame(t(latent_state[,,l])) %>%
dplyr::mutate(Type = rep(Sp_names_long[l], NYears-1),
state_Lower95 = V1,
state_mean = V2,
state_Upper95 = V3) %>%
dplyr::select(Type, state_Lower95, state_mean, state_Upper95)
state = rbind.data.frame(state, temp)
}
}
plot_ppc_ts = as.data.frame(cbind(melt(Obs_mean_Dec[2:NYears,]),
melt(Obs_mean_Jan[2:NYears,])["value"],
Yobs_n1[,c("ppc_Lower95","ppc_mean","ppc_Upper95")],
Yobs_n2[,c("ppc_Lower95","ppc_mean","ppc_Upper95")],
state[,c("state_Lower95", "state_mean", "state_Upper95")]))
colnames(plot_ppc_ts) = c("Year","Species","Dec","Jan",
"ppc_Lower95_Dec","ppc_mean_Dec","ppc_Upper95_Dec",
"ppc_Lower95_Jan","ppc_mean_Jan","ppc_Upper95_Jan",
"state_Lower95", "state_mean", "state_Upper95")
# Plot
if(period=='PrePinatubo') years = 1979:1991
if(period=='PostPinatubo') years = 2000:2013
ppc_ts_plots = plot_ppc_ts %>% group_by(Species = factor(Species, levels = unique(Species))) %>%
group_map(~tibble(plots=list(
ggplot(.) + aes(x= years, y=state_mean) +
geom_line(col="red", linewidth=1.5) +
geom_ribbon(aes(ymax = state_Lower95, ymin = state_Upper95), fill = "red", alpha = 0.2) +
geom_point(aes(x=years,y=Dec),col="blue3",size=1.5) +
geom_point(aes(x=years,y=Jan),col="green4",size=1.5) +
geom_abline(slope=0, intercept=0) +
theme_classic() +
# scale_y_continuous(limits = c(-0.5, 9)) +
scale_y_continuous(limits = c(NA, 10)) +
labs(x="Year",y=expression('Abundance, (' ~ italic(ln) ~ '[' ~ N ~ '/' ~ 1000 ~ '])')) +
annotate("point", x = years[5], y = 8.3, color="blue3", fill="blue3", size=2) +
annotate("text", x = years[7], y = 8.3, label = "December count", color="black", size=3) +
annotate("point", x = years[5], y = 7.3, color="green4", fill="green4", size=2) +
annotate("text", x = years[7], y = 7.3, label = "January count", color="black", size=3) +
annotate("segment", x = years[5]-0.5, xend = years[5], y = 6.3, yend = 6.3, colour = "red", linewidth=1) +
annotate("text", x = years[7], y = 6.3, label = "Latent abundance", color="black", size=3) +
ggtitle(.y[[1]]))))
PPC_TS_fullplot =
(ppc_ts_plots[[1]]$plots[[1]]+ppc_ts_plots[[2]]$plots[[1]])/
(ppc_ts_plots[[3]]$plots[[1]]+ppc_ts_plots[[4]]$plots[[1]])/
(ppc_ts_plots[[5]]$plots[[1]]+ppc_ts_plots[[6]]$plots[[1]])/
(ppc_ts_plots[[7]]$plots[[1]]+ppc_ts_plots[[8]]$plots[[1]])/
(ppc_ts_plots[[9]]$plots[[1]]+ppc_ts_plots[[10]]$plots[[1]]) +
plot_annotation(
title = 'Posterior predicted time series',
# subtitle = 'First stable state, before Mt. Pinatubo eruption',
subtitle = 'Second stable state, after Mt. Pinatubo eruption',
theme = theme(plot.title = element_text(size = 25),
plot.subtitle = element_text(size = 15)))
# print(PPC_TS_fullplot)
ggsave(file=paste0('output/figures/PPC_TS_fullplot_',period,'.pdf'), plot=PPC_TS_fullplot, width = 10, height = 13)
## Plot results ####
### Environmental effects ####
gamma = dplyr::select(post_vars, starts_with("gamma"))
colnames(gamma) = Sp_names_long
gamma_PrePinatubo = ggplot(reshape2::melt(gamma), aes(x=value, y=reorder(variable, desc(variable)), fill= after_stat(x))) +
geom_density_ridges_gradient(stat = "binline", bins = 100, scale = 1.5, draw_baseline = FALSE) +
geom_vline(xintercept=0) +
scale_x_continuous(limits = c(-1.5,1.5)) +
scale_y_discrete(expression(N)) +
scale_fill_viridis(name = "Posterior\nvalue", option = "magma") +
labs(x="Posterior value",
title = "Effect of flooding extension") +
theme_ridges(font_size = 13, grid = TRUE) +
theme(axis.title.y = element_blank(),
plot.title = element_text(size=18))
# gamma_PrePinatubo
ggsave(file=paste0('output/figures/Flooding_impact_',period,'.pdf'), plot=gamma_PrePinatubo,
width=7.5, height=5.5, limitsize = FALSE)
### Stability measures ####
alpha = dplyr::select(post_vars, starts_with("alpha"))
alpha_mean = matrix(colMeans(alpha),NSpecies,NSpecies)
k_vec = dplyr::select(post_vars, starts_with("k"))
k_mean = as.numeric(colMeans(k_vec))
r_vec = dplyr::select(post_vars, starts_with("r"))[,1:NSpecies]
r_mean = as.numeric(colMeans(r_vec))
# Create the objects to be populated:
Resilience=c()
jacobi_mat = matrix(rep(NA,(NSpecies^2)*nloops),ncol=NSpecies^2,nrow=nloops)
Eigenvals = matrix(rep(NA,NSpecies*nloops),ncol=NSpecies,nrow=nloops)
EqAbund = matrix(rep(NA,NSpecies*nloops),ncol=NSpecies,nrow=nloops)
initial_time = proc.time()
pb = txtProgressBar(min = 0, max = nloops, initial = 0, char = "*", width = 60)
for (j in 1:nloops){
# Interaction matrix
alpha_mat = matrix(as.matrix(as.numeric(alpha[j,])), nrow = NSpecies, ncol = NSpecies)
k_vector = as.vector(as.numeric(k_vec[j,]))
r_vector = as.vector(as.numeric(r_vec[j,]))
EqAbund[j, ] = as.vector(solve(alpha_mat) %*% k_vector)
if(all(EqAbund[j, ] >= 0)) {
jacobian = numDeriv::jacobian(function(N) N*exp(r_vector*(1 - (alpha_mat %*% N)/k_vector)), EqAbund[j, ], "complex")
jacobi_mat[j,] = as.vector(jacobian)
Eigenvals[j,] = matrix(eigen(jacobian, only.values = TRUE)$values, nrow=1)
Resilience[j] = 1/max(abs(Re(Eigenvals[j,])))
}
setTxtProgressBar(pb,j)
final_time = proc.time() - initial_time
if (j == nloops*0.25) cat(" 25% completed in", final_time[3], "sec!\n")
final_time = proc.time() - initial_time
if (j == nloops*0.5) cat(" 50% completed in", final_time[3], "sec!\n")
final_time = proc.time() - initial_time
if (j == nloops*0.75) cat(" 75% completed in", final_time[3], "sec!\n")
final_time = proc.time() - initial_time
if (j == nloops) cat(" 100% completed in", final_time[3], "sec!\n")
}
(HDInterval_stability = bayestestR::hdi(as.mcmc(Resilience), credMass = 0.9, allowSplit = TRUE))
### Plot unit circle and prob. of dynamic stability ####
Eigenvals = Eigenvals[complete.cases(Eigenvals),]
ReEigenvals = reshape2::melt(Re(Eigenvals), value.name="Real_part")
ImEigenvals = reshape2::melt(Im(Eigenvals), value.name="Imaginary_part")
Eigenvals_toplot = as.data.frame(cbind(ReEigenvals[,c("Real_part")],ImEigenvals[,c("Imaginary_part")]))
colnames(Eigenvals_toplot) = c("Real_part","Imaginary_part")
Eigenvals_toplot$Modulus = sqrt((Eigenvals_toplot$Real_part)^2 + (Eigenvals_toplot$Imaginary_part)^2)
(EmpProbDynStab = sum(Eigenvals_toplot$Modulus < 1)/length(Eigenvals_toplot$Modulus))
th = seq(-pi, pi, len = 100)
z = exp((0+1i) * th)
UnitCircle=as.data.frame(cbind(Re(z),Im(z)))
Origin=data.frame(x=0,y=0)
unitcircle_plot = ggplot(data=UnitCircle,aes(Re(z),Im(z))) +
geom_path() +
geom_vline(xintercept = 0,lty="dotted") +
geom_hline(yintercept = 0,lty="dotted") +
geom_point(data=Eigenvals_toplot,aes(x=Real_part,y=Imaginary_part),size=0.5,col="red") +
scale_x_continuous(limits = c(-1.5,1.5)) +
scale_y_continuous(limits = c(-1.5,1.5)) +
labs(x="Real", y = "Imaginary") +
theme(
axis.line.x=element_line(linewidth=0.5,colour="Black"),
axis.line.y=element_line(linewidth=0.5,colour="Black"),
axis.text=element_text(size=18,colour="Black"),
axis.title=element_text(size=18,colour="Black"),
plot.title = element_text(size=15),
legend.position = "right",
legend.text = element_text(size = 12),
legend.key = element_blank(),
panel.grid = element_blank(),
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
panel.background = element_blank(),
plot.background = element_rect(fill = "transparent", colour = NA)) +
ggtitle(paste0("Resilience: ", round(map_estimate(Resilience), 3),
" (", round(bayestestR::hdi(Resilience, credMass = 0.9, allowSplit = TRUE)[["CI_low"]], 3),",",
round(bayestestR::hdi(Resilience, credMass = 0.9, allowSplit = TRUE)[["CI_high"]],3), ") MAP, 90% HDI", sep=""),
paste0("Probability of dynamic stability: ", round(EmpProbDynStab, 4),"\n", sep=""))
# unitcircle_plot
ggsave(paste0('output/figures/Probability_of_stability_',period,'.pdf'), unitcircle_plot, height = 6.5, width = 6)
if(period=='PrePinatubo') unitcircle_plot_PrePinatubo = unitcircle_plot
if(period=='PostPinatubo') unitcircle_plot_PostPinatubo = unitcircle_plot
### Plot feasibility and extinction probability ####
EquilAbund = as.data.frame(EqAbund)
colnames(EquilAbund) = Sp_names_long
# Probability of species extinction
ProbSpExtinct=matrix(NA,1,NSpecies)
for(g in 1:NSpecies){
ProbSpExtinct[,g] = length(which(apply(as.data.frame(EquilAbund[,g]), 1, function(row) any(row <= 0))))/nrow(EquilAbund)
}
colnames(ProbSpExtinct) = Sp_names_long
if(period=='PrePinatubo') ProbSpExtinct_PrePinatubo = ProbSpExtinct
if(period=='PostPinatubo') ProbSpExtinct_PostPinatubo = ProbSpExtinct
(EmpProbFeas = (dim(EquilAbund)[1] - length(which(apply(EquilAbund, 1, function(row) any(row < 0)))))/dim(EquilAbund)[1])
k_posterior = as.data.frame(map_estimate(k_vec))
rownames(k_posterior) = Sp_names_long
k_posterior = as.data.frame(cbind(rownames(k_posterior),as.data.frame(k_posterior$MAP_Estimate)))
colnames(k_posterior) = c("variable","value")
plot_feasibility = ggplot(reshape2::melt(EquilAbund), aes(x=value, y=reorder(variable, desc(variable)), fill = after_stat(x))) +
geom_density_ridges_gradient(stat = "binline", bins = 100, scale = 1.5, draw_baseline = FALSE) +
geom_vline(xintercept=0) +
scale_x_continuous(limits = c(-20,100)) +
scale_y_discrete(expression(N)) +
scale_fill_viridis(name = expression("Equilibrium \nabundance,"~italic(N)^"*"), option = "viridis") +
geom_point(data = k_posterior, aes(col="k_posterior"), size=2) +
scale_color_manual(name = element_blank(),
values = c(k_posterior = "brown"),
labels = "Carrying capacity") +
labs(x="Abundance (n. ind x1000)",
title = paste("Probability of feasibility: ",round(EmpProbFeas,3))) +
theme_ridges(font_size = 13, grid = TRUE) +
theme(axis.title.y = element_blank(),
plot.title = element_text(size=18))
# plot_feasibility
ggsave(file=paste0('output/figures/Prob_of_feasibility_',period,'.pdf'), plot=plot_feasibility,
width=7.5, height=5.5, limitsize = FALSE)
if(period=='PrePinatubo') plot_feasibility_PrePinatubo = plot_feasibility
if(period=='PostPinatubo') plot_feasibility_PostPinatubo = plot_feasibility
}
## Plot wrapped figures ####
### Stability and feasibility ####
Wrapped_stability = wrap_plots(unitcircle_plot_PrePinatubo,
unitcircle_plot_PostPinatubo,
plot_feasibility_PrePinatubo,
plot_feasibility_PostPinatubo) + plot_annotation(tag_levels = 'A', tag_suffix = ')')
# Wrapped_stability
ggsave(file='output/figures/Wrapped_stability.pdf', plot=Wrapped_stability,
width=14.5, height=9, limitsize = FALSE)
## Probability of species extinction ####
probspExt = as.data.frame(t(rbind(ProbSpExtinct_PrePinatubo,ProbSpExtinct_PostPinatubo)))
colnames(probspExt) = c('ProbSpExtinct_pre','ProbSpExtinct_post')
probspExt <- probspExt %>%
mutate(Names = Sp_names_long,
'Before Pinatubo' = ProbSpExtinct_pre,
'After Pinatubo' = ProbSpExtinct_post) %>%
dplyr::select(Names, `Before Pinatubo`, `After Pinatubo`)
SpExtPr = ggplot(reshape2::melt(probspExt, id="Names")) +
geom_vline(xintercept=0) +
geom_point(aes(x=value, y=reorder(Names, desc(Names)), col = variable), size=5) +
scale_color_manual(values = c("purple", "green")) +
scale_x_continuous(limits = c(0,0.4), trans = "sqrt") +
theme_bw() +
theme(axis.title.y = element_blank(),
axis.text.x = element_text(size = 12),
axis.text.y = element_text(size = 12),
plot.title = element_text(size=18)) +
labs(col = "Regime",
x="Species extinction probability",
title = "Probability of extinction at equilibrium")
# SpExtPr
for(i in 1:nrow(probspExt)){
SpExtPr = SpExtPr + geom_segment(data=probspExt[i,],
aes(x=`Before Pinatubo`, xend = `After Pinatubo`, y=Names, yend = Names),
arrow = arrow(type = "closed", length = unit(0.2, "cm"), ends = "last"))
}
# print(SpExtPr)
ggsave(file="output/figures/Prob_of_extinction.pdf", plot=SpExtPr,
width=7, height=6, limitsize = FALSE)
save(ProbSpExtinct_PrePinatubo, ProbSpExtinct_PostPinatubo,
unitcircle_plot_PrePinatubo,
unitcircle_plot_PostPinatubo,
plot_feasibility_PrePinatubo,
plot_feasibility_PostPinatubo, file='output/SSRDLVR_model/plots.Rdata')
# Cusp ####
Cusp_db <- read_delim("data/Environmental_data.csv", delim = ";", escape_double = FALSE, trim_ws = TRUE)
trend_main = trends %>% dplyr::filter(trend_number == 'Trend 1')
Cusp_db$DFA_trend = trend_main$estimate
pdf("output/figures/Transition.pdf",height=4,width=5)
Cusp_db %>%
mutate(Period = factor(Period,
levels = c("Pre-Volcano","Transient","Post-Volcano"),
labels = c("Before Pinatubo","Transient period","After Pinatubo"))) %>%
ggplot(.,aes(x=Flood, y=DFA_trend, label=Year,color=Period)) +
scale_color_manual(values = c("purple","grey","green")) +
geom_segment(aes(
xend=c(tail(Flood, n=-1), NA),
yend=c(tail(DFA_trend, n=-1), NA),
color=Period
),
arrow=arrow(length=unit(0.3,"cm"),type="closed")) +
xlab(expression(paste("Flooding extension (", italic("Ln"),"Has)"))) +
ylab("Community trend \n(abundance)") +
geom_text_repel(max.overlaps=10) +
theme_bw() +
theme(panel.grid = element_blank())
dev.off()
# Standardise data
Cusp_db = Cusp_db %>%
mutate(Flood = scale(Flood)[,1],
SAOD = scale(SAOD)[,1])
# Fit the stochastic cusp catastrophe model
fit <- cusp(
y ~ DFA_trend,
alpha ~ -1 + Flood ,
beta ~ SAOD,
data = Cusp_db)
# save(fit, file='output/Cusp_fit.Rdata')
print(summary(fit, logist = TRUE))
print(confint(fit,level = 0.95))
pdf("output/figures/Cusp_Model_Fit.pdf",height=8,width=7)
plot(fit)
dev.off()
pdf("output/figures/Cusp_surface.pdf",height=8,width=7)
cusp3d(y=fit[["fitted.values"]],
alpha = fit[["linear.predictors"]][,1],
beta = fit[["linear.predictors"]][,2],
w = 0.025,
theta = 150, phi = 25,
B = 6.5, Y = 4, Yfloor = -10,
np = 180, n.surface = 45, surface.plot = TRUE,
surf.alpha = 0.95, surf.gamma = 1.9, surf.chroma = 45, surf.hue = 150,
surf.ltheta = 0, surf.lphi = 45,
main="Cusp Equilibrium Surface")
dev.off()
# END ####Once the analyses are completed and the figures produced, this shell
call to the Makefile will compile and open the manuscript
and related supplementary material:
make compile
sessionInfo()R version 4.5.0 (2025-04-11)
Platform: x86_64-pc-linux-gnu
Running under: Ubuntu 24.04.2 LTS
Matrix products: default
BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so; LAPACK version 3.12.0
locale:
[1] LC_CTYPE=es_ES.UTF-8 LC_NUMERIC=C
[3] LC_TIME=es_ES.UTF-8 LC_COLLATE=es_ES.UTF-8
[5] LC_MONETARY=es_ES.UTF-8 LC_MESSAGES=es_ES.UTF-8
[7] LC_PAPER=es_ES.UTF-8 LC_NAME=C
[9] LC_ADDRESS=C LC_TELEPHONE=C
[11] LC_MEASUREMENT=es_ES.UTF-8 LC_IDENTIFICATION=C
time zone: Europe/Madrid
tzcode source: system (glibc)
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] ggrepel_0.9.6 patchwork_1.3.0 reshape2_1.4.4
[4] data.table_1.17.2 psych_2.5.3 ggbreak_0.1.4
[7] mvtnorm_1.3-3 cusp_2.3.8 bayestestR_0.16.0
[10] viridis_0.6.5 viridisLite_0.4.2 ggridges_0.5.6
[13] imputeTS_3.3 qgraph_1.9.8 bayesplot_1.11.1
[16] ggmcmc_1.5.1.1 coda_0.19-4.1 bayesdfa_1.3.4
[19] rstan_2.32.7 StanHeaders_2.32.10 runjags_2.2.2-5
[22] lubridate_1.9.4 forcats_1.0.0 stringr_1.5.1
[25] dplyr_1.1.4 purrr_1.0.4 readr_2.1.5
[28] tidyr_1.3.1 tibble_3.2.1 ggplot2_3.5.2
[31] tidyverse_2.0.0 librarian_1.8.1 workflowr_1.7.1
loaded via a namespace (and not attached):
[1] RColorBrewer_1.1-3 rstudioapi_0.17.1 jsonlite_2.0.0
[4] magrittr_2.0.3 farver_2.1.2 rmarkdown_2.29.1
[7] fs_1.6.6 vctrs_0.6.5 base64enc_0.1-3
[10] htmltools_0.5.8.1 curl_6.2.3 gridGraphics_0.5-1
[13] Formula_1.2-5 TTR_0.24.4 sass_0.4.10
[16] bslib_0.9.0 htmlwidgets_1.6.4 plyr_1.8.9
[19] zoo_1.8-13 cachem_1.1.0 whisker_0.4.1
[22] igraph_2.1.4 lifecycle_1.0.4 pkgconfig_2.0.3
[25] Matrix_1.7-3 R6_2.6.1 fastmap_1.2.0
[28] aplot_0.2.5 digest_0.6.37 fdrtool_1.2.18
[31] colorspace_2.1-1 GGally_2.2.1 ps_1.9.1
[34] rprojroot_2.0.4 Hmisc_5.2-3 timechange_0.3.0
[37] httr_1.4.7 abind_1.4-8 mgcv_1.9-3
[40] compiler_4.5.0 withr_3.0.2 glasso_1.11
[43] htmlTable_2.4.3 backports_1.5.0 tseries_0.10-58
[46] inline_0.3.21 ggstats_0.9.0 QuickJSR_1.7.0
[49] pkgbuild_1.4.7 stinepack_1.5 corpcor_1.6.10
[52] gtools_3.9.5 loo_2.8.0 tools_4.5.0
[55] pbivnorm_0.6.0 foreign_0.8-90 lmtest_0.9-40
[58] quantmod_0.4.26 httpuv_1.6.16 nnet_7.3-20
[61] glue_1.8.0 quadprog_1.5-8 callr_3.7.6
[64] nlme_3.1-168 promises_1.3.2 gridtext_0.1.5
[67] grid_4.5.0 checkmate_2.3.2 getPass_0.2-4
[70] cluster_2.1.8.1 generics_0.1.4 gtable_0.3.6
[73] tzdb_0.5.0 hms_1.1.3 xml2_1.3.8
[76] pillar_1.10.2 yulab.utils_0.2.0 later_1.4.2
[79] splines_4.5.0 ggtext_0.1.2 lattice_0.22-7
[82] tidyselect_1.2.1 pbapply_1.7-2 knitr_1.50
[85] git2r_0.36.2 gridExtra_2.3 V8_6.0.3
[88] urca_1.3-4 forecast_8.24.0 stats4_4.5.0
[91] xfun_0.52 timeDate_4041.110 matrixStats_1.5.0
[94] stringi_1.8.4 ggfun_0.1.8 yaml_2.3.10
[97] evaluate_1.0.3 codetools_0.2-20 ggplotify_0.1.2
[100] cli_3.6.5 RcppParallel_5.1.10 rpart_4.1.24
[103] processx_3.8.6 jquerylib_0.1.4 lavaan_0.6-19
[106] dichromat_2.0-0.1 Rcpp_1.0.14 png_0.1-8
[109] parallel_4.5.0 fracdiff_1.5-3 jpeg_0.1-11
[112] scales_1.4.0 xts_0.14.1 insight_1.3.0
[115] rlang_1.1.6 mnormt_2.1.1