multiStateLangevin-supporting-code/sslAnalysis.R at master · bmcclintock/multiStateLangevin-supporting-code · GitHub

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
library(tidyverse)
library(terra)
library(sf)
library(sp)
library(marmap)
library(nngeo)
library(raster)
library(viridis)
library(scales)
library(ggplot2)
library(patchwork)
if(!requireNamespace("RandomFieldsUtils",quietly=TRUE)){
  install.packages("http://cran.r-project.org/src/contrib/Archive/RandomFieldsUtils/RandomFieldsUtils_1.2.5.tar.gz", repos = NULL, type = "source") # most recent archived version; required by RandomFields
}
if(!requireNamespace("RandomFields",quietly=TRUE)){
  install.packages("http://cran.r-project.org/src/contrib/Archive/RandomFields/RandomFields_3.3.14.tar.gz", repos = NULL, type = "source") # most recent archived version; required by Rhabit
}
remotes::install_github("papayoun/Rhabit@31ddf44",dependencies = TRUE) # last commit before RandomFields was removed from dependencies
library(Rhabit)
if(!requireNamespace("momentuHMM",quietly=TRUE) || as.numeric(substr(packageVersion("momentuHMM"),1,1))<2){
  remotes::install_github("bmcclintock/momentuHMM@develop",dependencies = TRUE) # requires momentuHMM version >= 2.0.0
}
library(momentuHMM)

source("supportingScripts/sslCovariates.R")
#covlist <- getCovariates() # takes a while and a bit of memory, so covlist used in analysis can be loaded with "sslCovs.RData"
load("data/sslCovs.RData") # load the covariates (queried from ETOPO1 database hosted by NOAA on 6 June 2022)
load("data/tracks.RData") # load the tracks
load("data/Par0.RData") # initial parameter values for optimization

source("supportingScripts/utility.R") # utility functions

# standardize covariate rasters based on observations
covlist0 <- list()
cellInd <- cellFromXY(covlist[[1]],tracks[,c("mu.x","mu.y")])
for(i in names(covlist)){
  covlist0[[i]] <- (covlist[[i]]-mean(covlist[[i]][cellInd]))/sd(covlist[[i]][cellInd])
}

# clip raster to smaller area for plotting and extracting
covlistCrop <- lapply(covlist0,function(x) raster::crop(x,extent(-2247.282,-1212.282,198.0775,918.0775)))

# prepare data for momentuHMM and calculate gradients
tracks <- prepData(tracks,CT=TRUE,Time.name="date_time",Time.unit="hours",coordNames=c("mu.x","mu.y"),altCoordNames = "mu",spatialCovs=covlist0,gradient=TRUE)
tracks$intercept <- 1

# define data stream distribution
dist <- list(mu="rw_mvnorm2",dry="bern")

source("supportingScripts/DM.R") # model definitions

fitLangevin <- UD <- list()

# original single-state model of Michelot et al. (2019)
fitLangevinOrig <- fitCTHMM(tracks,dist=dist,nbStates=1,
                            DM=list(mu = list(mean.x=~mu.x_tm1+langevin(depth.x)+langevin(slope.x)+langevin(d2site.x),
                                              mean.y=~mu.y_tm1+langevin(depth.y)+langevin(slope.y)+langevin(d2site.y),
                                              sd.x=~1,
                                              sd.y=~1,
                                              corr.xy=~1),
                                    dry = list(prob=~1)),
                            Par0=list(mu=c(1, -0.15059375, -0.02236392, -2.05770669,  1, -0.15059375, -0.02236392, -2.05770669, 1.00566450, 1.00566450, 0),
                                      dry=-0.4844352),
                            fixPar=list(mu=c(NA,1,2,3,NA,1,2,3,4,4,NA)),
                            workBounds=list(dry=matrix(c(-Inf,0),1,2)),
                            mvnCoords = "mu",
                            optMethod="TMB",control=list(silent=TRUE,trace=1),
                            stateNames="outbound",modelName="S=1 sigma(.) beta(depth+slope+d2site)")
UDO <- plotUD(nbUD=1,parIndex=list(2:4),covNames=list(c("depth","slope","d2site")),model=fitLangevinOrig,UDnames="outbound",UDstates=list(1),return=TRUE)

# fit models with additional states and covariates defined in `DM.R`
for(i in 1:length(DM)){
  message("\n",modelName[[i]])
  fitLangevin[[i]]<- fitCTHMM(tracks,dist=dist,nbStates=nbStates[[i]],DM=DM[[i]],
                               Par0=Par[[i]]$Par,
                               beta0=Par[[i]]$beta,
                               delta0=Par[[i]]$delta,
                               formula=formula[[i]],
                               fixPar=fixPar[[i]],
                               prior=prior[[i]],
                               workBounds=workBounds[[i]],
                               mvnCoords = "mu",
                               optMethod="TMB",stateNames=stateNames[[i]],modelName=modelName[[i]],kappa=kappa[[i]],
                               control=list(silent=TRUE,trace=1))
  UD[[i]] <- plotUD(inputUD[[i]]$nbUD,inputUD[[i]]$parIndex,inputUD[[i]]$covNames,model=fitLangevin[[i]],inputUD[[i]]$UDnames,inputUD[[i]]$UDstates,inputUD[[i]]$sign,return=TRUE)
}

# AIC excluding models with better fit but intermediary "foraging" state
penAIC(append(list(fitLangevinOrig),fitLangevin[c(1:5,7,9,12,14)]))

pdf("4stateUD.pdf",width=14,height=10)
statePlot(fitLangevin[[9]],UD[[9]])
dev.off()
pdf("5stateUD.pdf",width=14,height=14)
statePlot(fitLangevin[[14]],UD[[14]])
dev.off()

# pseudo-residuals
pdf("pseudoRes.pdf",width=21,height=7)
plotRes(fitLangevin[[2]])
plotRes(fitLangevin[[9]])
plotRes(fitLangevin[[14]])
dev.off()

# predicted and residuals
pdf("predRes.pdf",width=14,height=7)
plotPredRes(fitLangevin[[2]],inputUD[[2]])
plotPredRes(fitLangevin[[9]],inputUD[[9]])
plotPredRes(fitLangevin[[14]],inputUD[[14]])
dev.off()

# TPM plot (at median \Delta_t)
pdf("tpm.pdf",width=16,height=16)
tpmPlot(fitLangevin[[9]])
tpmPlot(fitLangevin[[14]])
dev.off()

# stationary distribution plot
pdf("stationary.pdf",width=13,height=7)
statPlot(fitLangevin[[9]])
statPlot(fitLangevin[[14]])
dev.off()

# sojourn time
pdf("sojourn.pdf",width=14,height=7)
sojournTimePlot(fitLangevin[[9]])
sojournTimePlot(fitLangevin[[14]])
dev.off()

# compare simulated tracks from models 2, 9, and 14
sim <- expandUD <- list()
initPos <- mapply(function(x) c(tracks[which(tracks$ID==x)[1],c("mu.x")],tracks[which(tracks$ID==x)[1],c("mu.y")]),unique(tracks$ID),SIMPLIFY = FALSE)
for(i in c(2,9,14)){
  message("\n",modelName[[i]])
  set.seed(1,kind="Mersenne-Twister",normal.kind = "Inversion")
  sim[[i]] <- simCTHMM(model=fitLangevin[[i]],matchModelObs=TRUE,spatialCovs=covlist0,initialPosition=initPos,states=TRUE,retrySims=5)
  expandUD[[i]] <- plotUD(inputUD[[i]]$nbUD,inputUD[[i]]$parIndex,inputUD[[i]]$covNames,model=fitLangevin[[i]],inputUD[[i]]$UDnames,inputUD[[i]]$UDstates,inputUD[[i]]$sign,cropRast=FALSE,return=TRUE)
}
slim <- apply(rbind(tracks[,c("mu.x","mu.y")],dplyr::bind_rows(sim)[,c("mu.x","mu.y")]),2,range)+c(-100,100,-50,50)

land <- (covlist0$depth*sd(covlist$depth[cellInd])+mean(covlist$depth[cellInd])-attr(covlist$depth,"absmin"))>=0

cextheme <- theme(legend.text = element_text(size = rel(1.15)),legend.title=element_text(size=rel(1.35)),axis.title=element_text(size = rel(1.25)),axis.text=element_text(size = rel(1.15)))
p1<-plotSpatialCov(tracks,crop(land,slim),colors=gray.colors(10, start = 1, end = 0.4, gamma = 2.2, alpha = NULL),alpha=0.5,col=stateCols[[4]],return=TRUE)+ggplot2::labs(fill = NULL,x=NULL,y="northing (km)")+guides(fill = "none")+theme(panel.border = element_blank())+cextheme+coord_fixed(ratio = 1.25,xlim=c(NA,-1250),ylim=c(200,650))
p2<-plotSpatialCov(sim[[2]],crop(expandUD[[2]]$outbound,slim),colors=viridis_pal()(100),alpha=0.5,col=stateCols[[4]],return=TRUE)+ggplot2::labs(fill = expression("log"(pi)),x=NULL,y=NULL)+guides(col = "none",shape="none")+theme(panel.border = element_blank())  + ggtitle("1 State")+ theme(plot.title = element_text(size=14,hjust=0.5))+cextheme+coord_fixed(ratio = 1.25,xlim=c(NA,-1250),ylim=c(200,650))
p3<-plotSpatialCov(sim[[9]],crop(expandUD[[9]]$foraging,slim),colors=viridis_pal()(100),alpha=0.5,col=stateCols[[4]],return=TRUE)+ggplot2::labs(fill = expression("log"(pi[2])),x="easting (km)",y="northing (km)")+guides(col = "none",shape="none")+theme(panel.border = element_blank())  + ggtitle("4 States")+ theme(plot.title = element_text(size=14,hjust=0.5))+cextheme+coord_fixed(ratio = 1.25,xlim=c(NA,-1250),ylim=c(200,650))
p4<-plotSpatialCov(sim[[14]],crop(expandUD[[14]]$foraging2,slim),colors=viridis_pal()(100),alpha=0.5,col=stateCols[[4]],return=TRUE)+ggplot2::labs(fill = expression("log"(pi[3])),x="easting (km)",y=NULL)+guides(col = "none",shape="none")+theme(panel.border = element_blank())  + ggtitle("5 States")+ theme(plot.title = element_text(size=14,hjust=0.5))+cextheme+coord_fixed(ratio = 1.25,xlim=c(NA,-1250),ylim=c(200,650))
pdf("simTracks.pdf",width=14,height=14)
plot(p1 + p2 + p3 + p4 + plot_layout(ncol = 2, nrow=2) )
dev.off()


# MALA simulations
mala <- list()
mala[[2]] <- malaSim(fitLangevin[[2]],inputUD[[2]],niter=50)
apply(mala[[2]] ,2,mean)
mala[[9]] <- malaSim(fitLangevin[[9]],inputUD[[9]],niter=50)
apply(mala[[9]] ,2,mean)
mala[[14]] <- malaSim(fitLangevin[[14]],inputUD[[14]],niter=50)
apply(mala[[14]] ,2,mean)