-
Notifications
You must be signed in to change notification settings - Fork 0
/
CoastlightSDMRF.R
283 lines (263 loc) · 13.2 KB
/
CoastlightSDMRF.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
rm(list=ls())
require(sp)
require(raster)
require(maptools)
require(rgdal)
require(dismo)
require(rJava)
require(arm)
require(plyr)
require(dplyr, warn.conflicts = FALSE)
require(sf)
require(ENMeval)
require(randomForest)
require(caret)
require(pdp)
require(ggplot2)
require(cowplot)
require(splines)
require(rfUtilities)
#To deal with functions with the same name in tidyr
#.rs.unloadPackage("tidyr")
#To deal with java issues
.jinit()
wd <- "/home/cmb-07/sn1/alsimons/Coastlight"
#wd <- "~/Desktop/Coastlight/SDM"
setwd(wd)
##This part is to be run first to determine variables to keep for parsimonious random forest models
#Set species types for observation data.
#Set number of model iterations
modelNum <- 100
speciesList <- c("Grunion","Plover")
for(species in speciesList){
# Read in grunion or plover presence and pseudo-absence points.
if(species=="Grunion"){
obs.data <- read.csv(file="RandomGrunionPointsWGS84.csv")
abs.data <- read.csv(file="RandomGrunionAbsencesWGS84.csv")
# Read in environmental map layers.
env.files <- list.files(pattern="10mAligned.tif$",full.names=TRUE)
env.files <- env.files[env.files != "./DEM10mAligned.tif" & env.files != "./Saltwater10mAligned.tif"]
}
if(species=="Plover"){
obs.data <- read.csv(file="RandomPloverPointsWGS84.csv")
abs.data <- read.csv(file="RandomPloverAbsencesWGS84.csv")
# Read in environmental map layers.
env.files <- list.files(pattern="10mAligned.tif$",full.names=TRUE)
}
# Drop unused columns
obs.data <- obs.data[, c("xcoord", "ycoord")]
abs.data <- abs.data[, c("xcoord", "ycoord")]
# Read in environmental map layers.
env.data <- stack(c(env.files))
# Initialize data containing environmental layer values at presence and pseudo-absence locations.
presvals <- obs.data
absvals <- abs.data
for(env.file in env.files){
# Get layer names for data frame.
env.filename <- gsub("^./","",gsub(".tif","",env.file))
# Extract environmental map layer values at presence points.
tmp1 <- as.data.frame(extract(raster(env.file),obs.data))
colnames(tmp1) <- env.filename
presvals <- cbind(presvals,tmp1)
# Extract environmental map layer values at presence points.
tmp2 <- as.data.frame(extract(raster(env.file),abs.data))
colnames(tmp2) <- env.filename
absvals <- cbind(absvals,tmp2)
}
# Standardize missing data
presvals[is.na(presvals)] <- NA
presvals <- na.omit(presvals)
absvals[is.na(absvals)] <- NA
absvals <- na.omit(absvals)
#Intialize summary statistics variables
RFImportanceTotal <- data.frame()
for(i in 1:modelNum){
#Subsample presence and pseudo-absence points for training and testing sets for SDMs.
sample_Num <- 100
presSubset <- presvals
presSubset$pa <- 1
presSubset <- presSubset[,c(ncol(presSubset),1:ncol(presSubset)-1)]
presSubset <- presSubset[sample(nrow(presSubset),sample_Num),]
absSubset <- absvals
absSubset$pa <- 0
absSubset <- absSubset[sample(nrow(absSubset),10*sample_Num),]
absSubset <- absSubset[,c(ncol(absSubset),1:ncol(absSubset)-1)]
#Construct a training and testing set for the presence data.
group <- kfold(presSubset,5)
pres_train <- presSubset[group!=1,]
pres_test <- presSubset[group==1,]
#Construct a training and testing set for the pseudo-absence data.
group <- kfold(absSubset,5)
backgr_train <- absSubset[group!=1,]
backgr_test <- absSubset[group==1,]
#Construct presence / pseudo-absence training sets.
envtrain <- rbind(pres_train,backgr_train)
envtrain$SoCalBeachType10mAligned <- factor(envtrain$SoCalBeachType10mAligned,levels=1:6)
testpres <- pres_test
testbackgr <- backgr_test
testpres$SoCalBeachType10mAligned <- factor(testpres$SoCalBeachType10mAligned,levels=1:6)
testbackgr$SoCalBeachType10mAligned <- factor(testbackgr$SoCalBeachType10mAligned,levels=1:6)
#Parsimonious random forest model
rf.regress <- rf.modelSel(envtrain[,4:ncol(envtrain)],envtrain[,c(1)], imp.scale="mir", parsimony=0.03,final.model=TRUE,seed=1)
RFImportance <- importance(rf.regress$rf.final)
RFImportance <- data.frame(names=row.names(RFImportance),RFImportance)
RFImportanceTotal <- rbind(RFImportanceTotal,RFImportance)
}
#Summary statistics on the frequency and importance of environmental parameters in random forest model.
tmp <- as.data.frame(table(RFImportanceTotal$names))
colnames(tmp) <- c("Variable","Freq")
RFImportanceTotal <- ddply(RFImportanceTotal, .(names), summarize, MeanIncNodePurity=mean(IncNodePurity), SDIncNodePurity=sd(IncNodePurity))
colnames(RFImportanceTotal) <- c("Variable","MeanIncNodePurity","SDIncNodePurity")
RFImportanceTotal <- left_join(tmp,RFImportanceTotal)
#To save aggregated data frame.
write.table(RFImportanceTotal,paste(species,"ParsimonyRFImportance.txt",sep=""),quote=FALSE,sep="\t",row.names = FALSE)
}
#Using the environmental variables found to be included in parsimonious
#random forest models rerun the model iterations.
speciesList <- c("Grunion","Plover")
for(species in speciesList){
#Select the variables to include in the parsimonious random forest model.
# Read in environmental map layers.
env.files <- list.files(pattern="10mAligned.tif$",full.names=TRUE)
#
# Read in grunion or plover presence and pseudo-absence points.
if(species=="Grunion"){
obs.data <- read.csv(file="RandomGrunionPointsWGS84.csv")
abs.data <- read.csv(file="RandomGrunionAbsencesWGS84.csv")
RFVars <- c("LogSI10mAligned","SoCalBeachWidth10mAligned")
env.files <- env.files[grep(paste(RFVars,collapse="|"),env.files)]
}
if(species=="Plover"){
obs.data <- read.csv(file="RandomPloverPointsWGS84.csv")
abs.data <- read.csv(file="RandomPloverAbsencesWGS84.csv")
RFVars <- c("Freshwater10mAligned","LogSI10mAligned","SoCalBeachWidth10mAligned")
env.files <- env.files[grep(paste(RFVars,collapse="|"),env.files)]
}
# Drop unused columns
obs.data <- obs.data[, c("xcoord", "ycoord")]
abs.data <- abs.data[, c("xcoord", "ycoord")]
# Read in environmental map layers.
env.data <- stack(c(env.files))
# Initialize data containing environmental layer values at presence and pseudo-absence locations.
presvals <- obs.data
absvals <- abs.data
for(env.file in env.files){
# Get layer names for data frame.
env.filename <- gsub("^./","",gsub(".tif","",env.file))
# Extract environmental map layer values at presence points.
tmp1 <- as.data.frame(extract(raster(env.file),obs.data))
colnames(tmp1) <- env.filename
presvals <- cbind(presvals,tmp1)
# Extract environmental map layer values at presence points.
tmp2 <- as.data.frame(extract(raster(env.file),abs.data))
colnames(tmp2) <- env.filename
absvals <- cbind(absvals,tmp2)
}
# Standardize missing data
presvals[is.na(presvals)] <- NA
presvals <- na.omit(presvals)
absvals[is.na(absvals)] <- NA
absvals <- na.omit(absvals)
#Intialize summary statistics variables
RFImportanceTotal <- data.frame()
RFEvaluationTotal <- data.frame()
RFp1Total <- data.frame()
RFp2Total <- data.frame()
RFp3Total <- data.frame()
for(i in 1:modelNum){
#Subsample presence and pseudo-absence points for training and testing sets for SDMs.
sample_Num <- 100
presSubset <- presvals
presSubset$pa <- 1
presSubset <- presSubset[,c(ncol(presSubset),1:ncol(presSubset)-1)]
presSubset <- presSubset[sample(nrow(presSubset),sample_Num),]
absSubset <- absvals
absSubset$pa <- 0
absSubset <- absSubset[sample(nrow(absSubset),10*sample_Num),]
absSubset <- absSubset[,c(ncol(absSubset),1:ncol(absSubset)-1)]
#Construct a training and testing set for the presence data.
group <- kfold(presSubset,5)
pres_train <- presSubset[group!=1,]
pres_test <- presSubset[group==1,]
#Construct a training and testing set for the pseudo-absence data.
group <- kfold(absSubset,5)
backgr_train <- absSubset[group!=1,]
backgr_test <- absSubset[group==1,]
#Construct presence / pseudo-absence training sets.
envtrain <- rbind(pres_train,backgr_train)
#envtrain$SoCalBeachType10mAligned <- factor(envtrain$SoCalBeachType10mAligned,levels=1:6)
testpres <- pres_test
testbackgr <- backgr_test
#testpres$SoCalBeachType10mAligned <- factor(testpres$SoCalBeachType10mAligned,levels=1:6)
#testbackgr$SoCalBeachType10mAligned <- factor(testbackgr$SoCalBeachType10mAligned,levels=1:6)
#Parsimonious random forest model
rf.regress <- suppressWarnings(randomForest(envtrain[,4:ncol(envtrain)],envtrain[,c(1)], imp.scale="mir", parsimony=0.03,final.model=TRUE,seed=1))
RFImportance <- importance(rf.regress)
RFImportance <- data.frame(names=row.names(RFImportance),RFImportance)
RFImportanceTotal <- rbind(RFImportanceTotal,RFImportance)
erf <- suppressWarnings(evaluate(testpres,testbackgr,rf.regress))
RFEvaluation <- data.frame(matrix(nrow=1,ncol=5))
colnames(RFEvaluation) <- c("AUC","cor","kappa","Q","TSS")
RFEvaluation$AUC <- erf@auc
RFEvaluation$cor <- erf@cor
RFEvaluation$kappa <- max(erf@kappa)
# Calculate Yule's Q.
tmp <- erf@OR
tmp[!is.finite(tmp)] <- NA
RFEvaluation$Q <- (mean(tmp,na.rm=T)-1)/(mean(tmp,na.rm=T)+1)
RFEvaluation$TSS <- mean(erf@TPR,na.rm=T)+mean(erf@TNR,na.rm=T)-1
RFEvaluationTotal <- rbind(RFEvaluationTotal,RFEvaluation)
#Store partial response data for each environmental factor in the random forest model.
if(species=="Plover"){
RFp1 <- partial(rf.regress,pred.var = "Freshwater10mAligned",train=envtrain[,c(4:ncol(envtrain))])
RFp1Total <- rbind(RFp1Total,RFp1)
}
RFp2 <- partial(rf.regress,pred.var = "LogSI10mAligned",train=envtrain[,c(4:ncol(envtrain))])
RFp2Total <- rbind(RFp2Total,RFp2)
RFp3 <- partial(rf.regress,pred.var = "SoCalBeachWidth10mAligned",train=envtrain[,c(4:ncol(envtrain))])
RFp3Total <- rbind(RFp3Total,RFp3)
}
#Summary statistics on the frequency and importance of environmental parameters in random forest model.
tmp <- as.data.frame(table(RFImportanceTotal$names))
colnames(tmp) <- c("Variable","Freq")
RFImportanceTotal <- ddply(RFImportanceTotal, .(names), summarize, MeanIncNodePurity=mean(IncNodePurity), SDIncNodePurity=sd(IncNodePurity))
colnames(RFImportanceTotal) <- c("Variable","MeanIncNodePurity","SDIncNodePurity")
RFImportanceTotal <- left_join(tmp,RFImportanceTotal)
#To save aggregated data frame.
write.table(RFImportanceTotal,paste(species,"ParsimonyFinalRFImportance.txt",sep=""),quote=FALSE,sep="\t",row.names = FALSE)
tmpMean <- colMeans(RFEvaluationTotal)
tmpSD <- apply(RFEvaluationTotal,2,sd)
RFEvaluationTotal <- as.data.frame(rbind(tmpMean,tmpSD))
#To save aggregated data frame.
write.table(RFEvaluationTotal,paste(species,"ParsimonyFinalRFEvaluation.txt",sep=""),quote=FALSE,sep="\t",row.names = TRUE)
#To save all of the raw points used in the partial response plots for the random forest model.
if(species=="Plover"){
colnames(RFp1Total)[which(names(RFp1Total) == "yhat")] <- paste(colnames(RFp1Total)[2],colnames(RFp1Total)[1],sep="")
}
colnames(RFp2Total)[which(names(RFp2Total) == "yhat")] <- paste(colnames(RFp2Total)[2],colnames(RFp2Total)[1],sep="")
colnames(RFp3Total)[which(names(RFp3Total) == "yhat")] <- paste(colnames(RFp3Total)[2],colnames(RFp3Total)[1],sep="")
if(species=="Plover"){
RFTotal <- bind_cols(RFp1Total,RFp2Total,RFp3Total)
}
if(species=="Grunion"){
RFTotal <- bind_cols(RFp2Total,RFp3Total)
}
write.table(RFTotal,paste(species,"RFParsimonyFinalPartialResponse.txt",sep=""),quote=FALSE,sep="\t",row.names = FALSE)
#Create 2d histograms with best-fit splines for the partial response curves.
RFTotal <- read.table(paste(species,"RFParsimonyFinalPartialResponse.txt",sep=""), header=TRUE, sep="\t",as.is=T,skip=0,fill=TRUE,check.names=FALSE, encoding = "UTF-8")
if(species=="Plover"){
RFp1Plot <- ggplot(RFTotal, aes(x=Freshwater10mAligned, y=yhatFreshwater10mAligned) )+xlab("Distance to Freshwater (m)")+ylab("Detection\nProbability")+geom_bin2d(bins = 50)+scale_fill_continuous(type = "viridis")+stat_smooth(aes(y = yhatFreshwater10mAligned, fill=yhatFreshwater10mAligned),method="auto",formula=y~x,color="violet",fill="red",n=0.1*nrow(RFTotal))+theme_bw(base_size=25)
}
RFp2Plot <- ggplot(RFTotal, aes(x=LogSI10mAligned, y=yhatLogSI10mAligned) )+xlab("Log(SI) log(mlx)")+ylab("Detection\nProbability")+geom_bin2d(bins = 50)+scale_fill_continuous(type = "viridis")+stat_smooth(aes(y = yhatLogSI10mAligned, fill=yhatLogSI10mAligned),method="auto",formula=y~x,color="violet",fill="red",n=0.1*nrow(RFTotal))+theme_bw(base_size=25)
RFp3Plot <- ggplot(RFTotal, aes(x=SoCalBeachWidth10mAligned, y=yhatSoCalBeachWidth10mAligned) )+xlab("Beach width (m)")+ylab("Detection\nProbability")+geom_bin2d(bins = 50)+scale_fill_continuous(type = "viridis")+stat_smooth(aes(y = yhatSoCalBeachWidth10mAligned, fill=yhatSoCalBeachWidth10mAligned),method="auto",formula=y~x,color="violet",fill="red",n=0.1*nrow(RFTotal))+theme_bw(base_size=25)
if(species=="Plover"){
RFPlots <- plot_grid(RFp1Plot,RFp2Plot,RFp3Plot,ncol=2,labels="AUTO")
}
if(species=="Grunion"){
RFPlots <- plot_grid(RFp2Plot,RFp3Plot,ncol=2,labels="AUTO")
}
png(paste(species,"RFParsimonyFinalPlots.png",sep=""),width=2*800,height=400*length(env.files))
RFPlots
print(RFPlots)
dev.off()
}