-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathsimulation2_frequencies.R
241 lines (211 loc) · 10.1 KB
/
simulation2_frequencies.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
#rm(list=ls())
# setwd("~/Dropbox/square_ecology_simulation")
library(RDAforest)
library(ggplot2)
#setting up environment
a=c(1:50)
b=c(1:50)
env=expand.grid(a,b)
names(env)=c("a","b")
# adding correlated variables
env$c=env$a+rnorm(nrow(env),0,10)
env$d=env$b+rnorm(nrow(env),0,10)
env$e=env$a+rnorm(nrow(env),0,25)
env$f=env$a+rnorm(nrow(env),0,15)
env$g=env$a+rnorm(nrow(env),0,15)
env$h=env$b+rnorm(nrow(env),0,30)
env$i=rnorm(nrow(env),0,40)
env$i[env$i<0]=0
env$j=rnorm(nrow(env),0,20)
env$k=rnorm(nrow(env),0,10)
pdf(file="frequencies_envCorrsTree.pdf",height=3.5,width=3.5)
hc=hclust(as.dist(1-cor(env)),method="ave")
plot(hc)
dev.off()
# rotated x, y coordinates
th=0.8
rot=diag(ncol(env))
rot[1:2,1:2]=t(matrix(c(cos(th),-sin(th),sin(th),cos(th)),ncol=2))
xy=t(rot %*% t(env))[,1:2]
env.col=decostand(env,method="range")
# plotting environments
pdf(file="frequencies_envs.pdf",height=4.5,width=4.5)
plot(xy,pch=18,cex=0.75,col=rgb(env.col$a,0.1,env.col$b,1),asp=1,main="a + b",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$a,0.1,0.1,1),asp=1,main="a",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$b,0.1,0.1,1),asp=1,main="b",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$c,0.1,0.1,1),asp=1,main="c",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$d,0.1,0.1,1),asp=1,main="d",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$e,0.1,0.1,1),asp=1,main="e",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$f,0.1,0.1,1),asp=1,main="f",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$g,0.1,0.1,1),asp=1,main="g",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$h,0.1,0.1,1),asp=1,main="h",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$i,0.1,0.1,1),asp=1,main="i",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$j,0.1,0.1,1),asp=1,main="j",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(env.col$k,0.1,0.1,1),asp=1,main="k",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
dev.off()
# trends for species distributions
a.trend=(a/max(a))^2
a.trend[a.trend<0.01]=0.01
a.trend[a.trend>0.99]=0.99
plot(a.trend)
b.trend=(b/max(b))^2
b.trend[b.trend<0.01]=0.01
b.trend[b.trend>0.99]=0.99
plot(b.trend)
a.peak=(1-decostand(abs(scale(a)[,1]),method="range")[,1])
b.peak=(1-decostand(abs(scale(b)[,1]),method="range")[,1])
a.peak[a.peak<0.1]=0.1
a.peak[a.peak>0.9]=0.9
b.peak[b.peak<0.1]=0.1
b.peak[b.peak>0.9]=0.9
plot(a.peak)
plot((a.trend*b.trend))
# setting up species
sp1=c();sp2=c();sp3=c();sp2a=c();sp4=c();sp5=c();sp0=sp01=sp02=sp03=sp04=sp5a=sp5b=c()
for(r in 1:nrow(env)){
i=env[r,"a"];j=env[r,"b"]
sp0=c(sp0,1) # everywhere (so jaccard distance would work)
sp01=c(sp01,sample(c(0,1),1)) # random
sp02=c(sp02,sample(c(0,1),1)) # random
sp03=c(sp03,sample(c(0,1),1)) # random
sp04=c(sp04,sample(c(0,1),1)) # random
sp1=c(sp1,sample(c(0,1),1,prob=c(1-a.trend[i],a.trend[i]))) # a-trend
sp2=c(sp2,sample(c(0,1),1,prob=c(1-b.trend[j],b.trend[j]))) # b-trend
sp2a=c(sp2a,sample(c(0,1),1,prob=c(1-b.trend[j]^2,b.trend[j]^2))) # b-trend
apeakprob=a.peak[i]^2
sp3=c(sp3,sample(c(0,1),1,prob=c(1-apeakprob,apeakprob))) # a-peak-driven
abpeakprob=a.peak[i]*b.peak[j]
if(abpeakprob>0.5) { abprob=0.9}
if(abpeakprob<0.4) { abprob=0.1}
sp4=c(sp4,sample(c(0,1),1,prob=c(1-(abpeakprob),(abpeakprob)))) # interaction center
abprob=(a.trend[i]*b.trend[j])
# if(i<45 | j<45) { abrpob=0.001 }
if(abprob>0.6) { abprob=0.9}
if(abprob<0.4) { abprob=0.1}
sp5=c(sp5,sample(c(0,1),1,prob=c(1-abprob,abprob))) # interaction top corner
sp5a=c(sp5a,sample(c(0,1),1,prob=c(1-abprob,abprob))) # interaction top corner
sp5b=c(sp5b,sample(c(0,1),1,prob=c(1-abprob,abprob))) # interaction top corner
}
samp=data.frame(cbind(sp0,sp2,sp3,sp4,sp5,sp5a,sp5b))
names(samp)
dis0=data.frame(as.matrix((vegdist(samp,method="jaccard"))))
caps0=capscale(dis0~1)
nn=length(caps0$CA$eig)
cpcs=scores(caps0,scaling=1,display="sites",choices=c(1:nn))
cpcs0=decostand(cpcs,method="range")
pdf(file="frequencies_species.pdf",height=4.5,width=4.5)
plot(xy,pch=18,cex=1,col=rgb(cpcs0[,1],cpcs0[,3],cpcs0[,2],1),asp=1,main="original pcs of community",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(1-(0.2+samp[,1]/1.5),1-(0.2+samp[,1]/1.5),1-(0.2+samp[,1]/1.5),1),asp=1,main="sp1",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(1-(0.2+samp[,2]/1.5),1-(0.2+samp[,2]/1.5),1-(0.2+samp[,2]/1.5),1),asp=1,main="sp2",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(1-(0.2+samp[,3]/1.5),1-(0.2+samp[,3]/1.5),1-(0.2+samp[,3]/1.5),1),asp=1,main="sp3",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(1-(0.2+samp[,4]/1.5),1-(0.2+samp[,4]/1.5),1-(0.2+samp[,4]/1.5),1),asp=1,main="sp4",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
plot(xy,pch=18,cex=0.75,col=rgb(1-(0.2+samp[,5]/1.5),1-(0.2+samp[,5]/1.5),1-(0.2+samp[,5]/1.5),1),asp=1,main="sp5",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
dev.off()
# choosing samples to fit the model (1/10 of total), plotting them
picks=sample(1:nrow(samp),round(nrow(samp)/10))
samp.fit=samp[picks,]
env.fit=env[picks,]
XY=data.frame(xy)
names(XY)=c("x","y")
XY.fit=XY[picks,]
cpcs0.fit=cpcs0[picks,]
pdf(file="frequencies_samples4model.pdf",height=4.5,width=4.5)
plot(XY.fit,pch=18,cex=1,col=rgb(cpcs0.fit[,1],cpcs0.fit[,3],cpcs0.fit[,2],1),asp=1,main="original pcs of community",xaxt="na",yaxt="na",ylab="",xlab="",bty="none")
dev.off()
# generating distance matrix for fitting, also scores (for direct RF prediction)
dis=data.frame(as.matrix((vegdist(samp.fit,method="jaccard"))))
caps=capscale(dis~1)
nc=length(caps$CA$eig)
sc.fit=data.frame(scores(caps,scaling=1,display="sites",choices=c(1:nc)))
# ------ Gradient Forest: exploring turnover curves
# fitting gradient forest model to ordination
gf=makeGF(caps,env.fit,ntree=1500)
gf$result
nc=length(gf$result)
# plot importances (R2)
imps=data.frame(importance_RDAforest(gf,caps))
# total R2
sum(imps)
names(imps)="R2"
imps$var=row.names(imps)
imps$var=factor(imps$var,levels=imps$var[order(imps$R2)])
pdf(file="frequencies_GFimportances.pdf",width=2,height=2.5)
ggplot(imps,aes(var,R2))+geom_bar(stat="identity")+coord_flip()+theme_bw()
dev.off()
# plot turnover curves (first 4 most important environmental variables)
vars=names(importance(gf))[1:2]
pdf(file="frequencies_GF_turnovers.pdf",width=4.4,height=4.4)
plot_gf_turnovers(gf,vars)
dev.off()
#--------- variable selection
mm=mtrySelJack(dis,env.fit,nreps=20,top.pcs = length(gf$result),importance.type = "GF")
mm$goodvars
# plot mtrySelJack diagnostic plots
pdf(file="frequencies_mtrySelJack.pdf",,width=2,height=2.5)
# bar chart of proportion of positive change in response to higher mtry
# good variables would be the ones above the red line
ggplot(mm$prop.positive,aes(var,prop.positive))+
geom_bar(stat="identity")+
coord_flip()+
theme_bw()+
geom_hline(yintercept=0.5,col="red")
dev.off()
#------- measuring importance of passing variables, forming predictions
sb=ordinationJackknife(Y=dis,X=env.fit[,mm$goodvars],newX=env,nreps=10,top.pcs=nc)
pdf(file="frequencies_sb.pdf",height=1.5,width=3)
important=names(sb$median.importance)
ggplot(sb$all.importances[sb$all.importances$variable %in% important,],aes(variable,importance))+
geom_boxplot()+
coord_flip()+
# ylim(0,0.06)+
theme_bw()
dev.off()
#-------------- plotting community maps / adaptive neighborhoods
dirs=sb$predictions.direct
turnovers=sb$predictions.turnover
nbests=2
bests=names(sb$median.importance)[c(1:nbests)]
# subsetting for spots within modeled range
table(sb$goodrows)
xy2=xy[sb$goodrows,]
ras2=env[which(sb$goodrows),]
?plot_adaptation
pdf("frequencies_adaptation.pdf",width=5,heigh=4)
# unclustered
pa0=plot_adaptation(dirs,ras2[,bests],xy2,lighten=0,color.scheme="000",ordinate=T,scal=8,cex=0.8,jitscale=0.05,rangeExp=1.5,col.envs="red",main="unclustered RF predictions")
# RF clusters
pa1=plot_adaptation(dirs,ras2[,bests],xy2,lighten=0,nclust=12,color.scheme="000",ordinate=T,cluster.merge.level=0.25,scal=8,cex=0.8,jitscale=0.05,cluster.labels=F,rangeExp=1.5,col.envs="red",main="clustered by RF, nclust=12")
# turnover clusters
pa2=plot_adaptation(turnovers,ras2[,bests],xy2,nclust=12,cluster.merge.level=0.1,lighten=0,color.scheme="000",scal=8,cex=0.8,jitscale=0.05,cluster.labels=F,rangeExp=1.5,col.envs="red",main="clustered by GF, nclust=12")
# turnover clusters, merged by RF
pa3=plot_adaptation(dirs,ras2[,bests],xy2,nclust=12,lighten=0,ordinate=T,color.scheme="000",cluster.merge.level=0.25,cluster.guide = turnovers,cluster.labels=F,col.envs="red",scal=8,cex=0.8,jitscale=0.05,rangeExp=1.5,main="turnover clusters\nmerged by RF, nclust=12")
dev.off()
#-----predictions based on just spatial coordinates
# choosing 1/5 of total samples to fit
picks=sample(1:nrow(samp),round(nrow(samp)/5))
samp.fit=samp[picks,]
env.fit=env[picks,]
XY=data.frame(xy)
names(XY)=c("x","y")
XY.fit=XY[picks,]
cpcs0.fit=cpcs0[picks,]
# generating distance matrix for fitting, also scores (for direct RF prediction)
dis=data.frame(as.matrix(vegdist(samp.fit,method="jaccard")))
#caps=capscale(dis~1+Condition(as.matrix(XY.fit)))
caps=capscale(dis~1)
nc=length(caps$CA$eig)
sc.fit=data.frame(scores(caps,scaling=1,display="sites",choices=c(1:nc)))
# predictions based only on spatial coordinates
sbxy=ordinationJackknife(Y=dis,X=XY.fit,newX=XY,nreps=7,top.pcs=nc)
dirs=sbxy$predictions.direct
turnovers=sbxy$predictions.turnover
nbests=2
bests=names(sbxy$median.importance)[c(1:nbests)]
# subsetting for spots within modeled range
table(sbxy$goodrows)
xy2=xy[sbxy$goodrows,]
ras2=XY[which(sbxy$goodrows),]
pdf("frequencies_adaptation_XY_2xSamples.pdf",width=5,heigh=4)
# unclustered
pa0=plot_adaptation(dirs,ras2[,bests],xy2,color.scheme="000",scal=8,cex=0.8,jitscale=0.05,rangeExp=1.5,col.envs="red",main="unclustered RF predictions")
dev.off()