-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodel_performance.R
198 lines (181 loc) · 10 KB
/
model_performance.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
# options(java.parameters = "-Xmx32g")
library(maptools)
library(raster)
library(dismo)
library(sp)
library(rJava)
library(readxl)
library(mapview)
library(ggplot2)
library(xlsx)
library(readxl)
library(ENMeval)
library(gridExtra)
library(dplyr)
library(ggpubr)
library(blockCV)
#Identify block size
latlong <- CRS("+proj=longlat +datum=WGS84")
SR.ORG8287 <- CRS('+proj=cea +lon_0=0 +lat_ts=0 +x_0=0 +y_0=0 +ellps=WGS84 +datum=WGS84 +units=m +no_defs')
#import environmental variables
# Current
list.raster.full <- list.files("E:/Working/2018/PhD_research/enviromental_variables_new/current", full.names = T, pattern = ".tif")
predictors <- stack(list.raster.full)
names(predictors) <- c('bio1','bio10','bio11','bio12','bio13','bio14','bio15','bio16','bio17','bio18','bio19','bio2','bio3','bio4','bio5','bio6','bio7','bio8','bio9',
'cropland','decidous_forest','evergreen_forest','flooded_vegetation','grassland_scrub','karst','urban')
predictors <- subset(predictors, c("bio2","bio10","bio11","bio12","bio18","bio19",'cropland','decidous_forest','evergreen_forest','flooded_vegetation','grassland_scrub','karst','urban')) #select the important variables
#import occurrence data
vnbats.asia <- read_excel('E:/Working/2018/PhD_research/SDM output R/10_7_23_Transboundary/final_occurrences.xlsx', sheet = "Sheet1") # import bat occurence data
coordinates(vnbats.asia)= ~ decimalLongitude + decimalLatitude
crs(vnbats.asia) <- latlong
vnbats.asia <- spTransform(vnbats.asia, SR.ORG8287)
vnbats.asia.env <- raster::extract(predictors, vnbats.asia, df =T, na.rm=T, cellnumber =T)
vnbats.asia.env <- vnbats.asia.env[!is.na(as.character(vnbats.asia.env$bio2)),]#remove records without values
vnbats.asia <- vnbats.asia[vnbats.asia$cell_number%in%vnbats.asia.env$cells,]#remove records without environmental information
#import species list and select species have >= 20 records
species.list <- read_excel('E:/Working/2018/PhD_research/SDM output R/10_7_23_Transboundary/list of species.xlsx')
species.list <- species.list%>%filter(Num_occurence_all > 19)
nrow(species.list)
model.perfome <- data.frame(species.list$Scientific_name)
colnames(model.perfome) <- 'species'
model.perfome <- model.perfome%>%mutate(auc.m0.5=0,auc.m1=0,auc.m1.5=0,auc.m2=0,auc.m2.5=0,auc.m3=0,auc.m3.5=0,auc.m4=0,auc.m4.5=0,auc.m5=0)
vnbats.asia.env <- raster::extract(predictors, vnbats.asia, df =T, na.rm=T, cellnumber =T)
# vnbats.asia.env_2 <- vnbats.asia.env[!is.na(as.character(vnbats.asia.env$bio2)),]
# vnbats.asia.env_2 <- vnbats.asia.env_2[!is.na(as.character(vnbats.asia.env_2$karst)),]
# vnbats.asia.env_2 <- vnbats.asia.env_2[!is.na(as.character(vnbats.asia.env_2$forest)),]
# vnbats.asia.2 <- vnbats.asia[vnbats.asia$cell_number%in%vnbats.asia.env_2$cells,]#remove records without enviromental information
for(i in 1:length(species.list$Scientific_name)){
sp <- species.list$Scientific_name[i]
vnbats.asia$sp_validate <- ifelse(vnbats.asia$species==sp,1,0)
sp.block <- cv_spatial(x = vnbats.asia,
column = 'sp_validate',
r = predictors,
size = 50000,
k = 5,
selection = "random",
iteration = 100)
vnbats.asia$foldID <- sp.block$folds_ids
sp.presence <- vnbats.asia[vnbats.asia$species==sp,]
sp.absence <- vnbats.asia[vnbats.asia$species!=sp,]
user.grp <- list(occs.grp = sp.presence$foldID,
bg.grp = sp.absence$foldID)
enviromental.presence <- raster::extract(predictors,sp.presence)
enviromental.absence <- raster::extract(predictors,sp.absence)
res <- ENMevaluate(occ = enviromental.presence, RMvalues = seq(0.5,5,0.5),tune.args = list(fc = c('LQH')),
partitions = 'user',user.grp = user.grp, bg.coords = enviromental.absence,
parallel = T, numCores = 2, algorithm = "maxent.jar", clamp = F)
model.perfome$auc.m0.5[i] <- res@results[res@results$rm==0.5,]$auc.val.avg
model.perfome$auc.m1[i] <- res@results[res@results$rm==1,]$auc.val.avg
model.perfome$auc.m1.5[i] <- res@results[res@results$rm==1.5,]$auc.val.avg
model.perfome$auc.m2[i] <- res@results[res@results$rm==2,]$auc.val.avg
model.perfome$auc.m2.5[i] <- res@results[res@results$rm==2.5,]$auc.val.avg
model.perfome$auc.m3[i] <- res@results[res@results$rm==3,]$auc.val.avg
model.perfome$auc.m3.5[i] <- res@results[res@results$rm==3.5,]$auc.val.avg
model.perfome$auc.m4[i] <- res@results[res@results$rm==4,]$auc.val.avg
model.perfome$auc.m4.5[i] <- res@results[res@results$rm==4.5,]$auc.val.avg
model.perfome$auc.m5[i] <- res@results[res@results$rm==5,]$auc.val.avg
print(sp)
}
write.xlsx(model.perfome, file = "E:/Working/2018/PhD_research/SDM output R/10_7_23_Transboundary/model performance_block_LQH.xlsx")
#model performance with LQ
model.perfome <- data.frame(species.list$Scientific_name)
colnames(model.perfome) <- 'species'
model.perfome <- model.perfome%>%mutate(auc.m0.5=0,auc.m1=0,auc.m1.5=0,auc.m2=0,auc.m2.5=0,auc.m3=0,auc.m3.5=0,auc.m4=0,auc.m4.5=0,auc.m5=0)
for(i in 1:length(species.list$Scientific_name)){
sp <- species.list$Scientific_name[i]
vnbats.asia$sp_validate <- ifelse(vnbats.asia$species==sp,1,0)
sp.block <- cv_spatial(x = vnbats.asia,
column = 'sp_validate',
r = predictors,
size = 50000,
k = 5,
selection = "random",
iteration = 100)
vnbats.asia$foldID <- sp.block$folds_ids
sp.presence <- vnbats.asia[vnbats.asia$species==sp,]
sp.absence <- vnbats.asia[vnbats.asia$species!=sp,]
user.grp <- list(occs.grp = sp.presence$foldID,
bg.grp = sp.absence$foldID)
enviromental.presence <- raster::extract(predictors,sp.presence)
enviromental.absence <- raster::extract(predictors,sp.absence)
res <- ENMevaluate(occ = enviromental.presence, RMvalues = seq(0.5,5,0.5),tune.args = list(fc = c('LQ')),
partitions = 'user',user.grp = user.grp, bg.coords = enviromental.absence,
parallel = T, numCores = 8, algorithm = "maxent.jar", clamp = F)
model.perfome$auc.m0.5[i] <- res@results[res@results$rm==0.5,]$auc.val.avg
model.perfome$auc.m1[i] <- res@results[res@results$rm==1,]$auc.val.avg
model.perfome$auc.m1.5[i] <- res@results[res@results$rm==1.5,]$auc.val.avg
model.perfome$auc.m2[i] <- res@results[res@results$rm==2,]$auc.val.avg
model.perfome$auc.m2.5[i] <- res@results[res@results$rm==2.5,]$auc.val.avg
model.perfome$auc.m3[i] <- res@results[res@results$rm==3,]$auc.val.avg
model.perfome$auc.m3.5[i] <- res@results[res@results$rm==3.5,]$auc.val.avg
model.perfome$auc.m4[i] <- res@results[res@results$rm==4,]$auc.val.avg
model.perfome$auc.m4.5[i] <- res@results[res@results$rm==4.5,]$auc.val.avg
model.perfome$auc.m5[i] <- res@results[res@results$rm==5,]$auc.val.avg
print(sp)
}
write.xlsx(model.perfome, file = "E:/Working/2018/PhD_research/SDM output R/10_7_23_Transboundary/model performance_block_LQ.xlsx")
#plot model performance with different features types and regularization values
sum.LQ <- read_xlsx("E:/Working/2018/PhD_research/SDM output R/10_7_23_Transboundary/model performance_block_LQ.xlsx")
sum.LQH <- read_xlsx("E:/Working/2018/PhD_research/SDM output R/10_7_23_Transboundary/model performance_block_LQH.xlsx")
k <- length(sum.LQH$species)
auc <- c(sum.LQ$auc.m0.5,
sum.LQ$auc.m1,
sum.LQ$auc.m1.5,
sum.LQ$auc.m2,
sum.LQ$auc.m2.5,
sum.LQ$auc.m3,
sum.LQ$auc.m3.5,
sum.LQ$auc.m4,
sum.LQ$auc.m4.5,
sum.LQ$auc.m5,
sum.LQH$auc.m0.5,
sum.LQH$auc.m1,
sum.LQH$auc.m1.5,
sum.LQH$auc.m2,
sum.LQH$auc.m2.5,
sum.LQH$auc.m3,
sum.LQH$auc.m3.5,
sum.LQH$auc.m4,
sum.LQH$auc.m4.5,
sum.LQH$auc.m5)
parameters <- c(rep('LQ_0.5',k),
rep('LQ_1',k),
rep('LQ_1.5',k),
rep('LQ_2',k),
rep('LQ_2.5',k),
rep('LQ_3',k),
rep('LQ_3.5',k),
rep('LQ_4',k),
rep('LQ_4.5',k),
rep('LQ_5',k),
rep('LQH_0.5',k),
rep('LQH_1',k),
rep('LQH_1.5',k),
rep('LQH_2',k),
rep('LQH_2.5',k),
rep('LQH_3',k),
rep('LQH_3.5',k),
rep('LQH_4',k),
rep('LQH_4.5',k),
rep('LQH_5',k))
sum.perfome <- data.frame(parameters,auc)
auc.plot <- ggplot(sum.perfome, aes(x= reorder(parameters, auc, median), y=auc)) +
stat_summary(geom = "boxplot",
fun.data = function(x) setNames(quantile(x, c(0.0, 0.25, 0.5, 0.75, 1)), c("ymin", "lower", "middle", "upper", "ymax")),
position = "dodge")+
theme_classic()+
coord_flip()+
xlab('Feature types and Regularization values')+ylab("AUC")+
theme(text = element_text(size = 14), axis.text = element_text(size = 14), plot.title = element_text(hjust = 0))+
geom_hline(yintercept=median(sum.LQH$auc.m2.5), linetype="dashed", color = "red")
ggsave(auc.plot, filename = 'E:/Working/2018/PhD_research/SDM output R/10_7_23_Transboundary/figures/AUC_regularization values.tiff',
height = 20, width = 12, dpi = 300, units = 'cm')
#histogram AUC
figure.auc <- ggplot(sum.LQH, aes(x=auc.m2.5)) +
geom_histogram(col='black', binwidth = 0.04)+
labs(title="",x="AUC", y = "Number of species")+ theme_classic()+
theme(axis.text=element_text(size=14),axis.title.x = element_text(size = 14),
axis.title.y = element_text(size = 14))+
geom_vline(xintercept = 0.7, color = 'red')
ggsave(figure.auc, filename = 'E:/Working/2018/PhD_research/SDM output R/10_7_23_Transboundary/figures/AUC_histogram.tiff', width = 10, height = 8, units = 'cm')
nrow(sum.LQH%>%filter(auc.m2.5 >= 0.7))