1
|
################## MULTI SAMPLING GAM FUSION METHOD ASSESSMENT ####################################
|
2
|
############################ Merging LST and station data ##########################################
|
3
|
#This script interpolates tmax values using MODIS LST and GHCND station data
|
4
|
#interpolation area. It requires the text file of stations and a shape file of the study area.
|
5
|
#Note that the projection for both GHCND and study area is lonlat WGS84.
|
6
|
#Options to run this program are:
|
7
|
#1) Multisampling: vary the porportions of hold out and use random samples for each run
|
8
|
#2)Constant sampling: use the same sample over the runs
|
9
|
#3)over dates: run over for example 365 dates without mulitsampling
|
10
|
#4)use seed number: use seed if random samples must be repeatable
|
11
|
#5)GAM fusion: possibilty of running GAM+FUSION or GAM separately
|
12
|
#AUTHOR: Benoit Parmentier
|
13
|
#DATE: 02/06/2013
|
14
|
#PROJECT: NCEAS INPLANT: Environment and Organisms --TASK#363--
|
15
|
###################################################################################################
|
16
|
|
17
|
###Loading R library and packages
|
18
|
library(gtools) # loading some useful tools
|
19
|
library(mgcv) # GAM package by Simon Wood
|
20
|
library(sp) # Spatial pacakge with class definition by Bivand et al.
|
21
|
library(spdep) # Spatial pacakge with methods and spatial stat. by Bivand et al.
|
22
|
library(rgdal) # GDAL wrapper for R, spatial utilities
|
23
|
library(gstat) # Kriging and co-kriging by Pebesma et al.
|
24
|
library(fields) # NCAR Spatial Interpolation methods such as kriging, splines
|
25
|
library(raster) # Hijmans et al. package for raster processing
|
26
|
library(rasterVis)
|
27
|
library(parallel) # Urbanek S. and Ripley B., package for multi cores & parralel processing
|
28
|
library(reshape)
|
29
|
library(plotrix)
|
30
|
### Parameters and argument
|
31
|
|
32
|
infile1<- "ghcn_or_tmax_covariates_06262012_OR83M.shp" #GHCN shapefile containing variables for modeling 2010
|
33
|
infile2<-"list_365_dates_04212012.txt"
|
34
|
#infile3<-"LST_dates_var_names.txt" #LST dates name
|
35
|
#infile4<-"models_interpolation_05142012.txt" #Interpolation model names
|
36
|
#infile5<-"mean_day244_rescaled.rst" #Raster or grid for the locations of predictions
|
37
|
#infile6<-"lst_climatology.txt"
|
38
|
#infile6<-"LST_files_monthly_climatology.txt"
|
39
|
#inlistf<-"list_files_05032012.txt" #Stack of images containing the Covariates
|
40
|
|
41
|
infile_monthly<-"monthly_covariates_ghcn_data_TMAXy2010_2010_VE_02062013.shp"
|
42
|
infile_daily<-"daily_covariates_ghcn_data_TMAXy2010_2010_VE_02062013.shp"
|
43
|
infile_locs<-"stations_venezuela_region_y2010_2010_VE_02062013.shp"
|
44
|
infile3<-"covariates__venezuela_region__VE_01292013.tif" #this is an output from covariate script
|
45
|
|
46
|
in_path<-"/home/parmentier/Data/IPLANT_project/Venezuela_interpolation/Venezuela_01142013/input_data"
|
47
|
out_path<-"/home/parmentier/Data/IPLANT_project/Venezuela_interpolation/Venezuela_01142013/output_data"
|
48
|
setwd(in_path)
|
49
|
|
50
|
nmodels<-9 #number of models running
|
51
|
y_var_name<-"dailyTmax"
|
52
|
predval<-1
|
53
|
prop<-0.3 #Proportion of testing retained for validation
|
54
|
#prop<-0.25
|
55
|
seed_number<- 100 #if seed zero then no seed? #Seed number for random sampling
|
56
|
out_prefix<-"_10d_GAM_fus5_all_lstd_020632013" #User defined output prefix
|
57
|
#out_prefix<-"_365d_GAM_12272012" #User defined output prefix
|
58
|
|
59
|
bias_val<-0 #if value 1 then training data is used in the bias surface rather than the all monthly stations
|
60
|
bias_prediction<-1 #if value 1 then use GAM for the BIAS prediction otherwise GAM direct repdiction for y_var (daily tmax)
|
61
|
nb_sample<-1 #number of time random sampling must be repeated for every hold out proportion
|
62
|
prop_min<-0.3 #if prop_min=prop_max and step=0 then predicitons are done for the number of dates...
|
63
|
prop_max<-0.3
|
64
|
step<-0
|
65
|
constant<-0 #if value 1 then use the same samples as date one for the all set of dates
|
66
|
#projection used in the interpolation of the study area: should be read directly from the outline of the study area
|
67
|
CRS_interp<-"+proj=lcc +lat_1=43 +lat_2=45.5 +lat_0=41.75 +lon_0=-120.5 +x_0=400000 +y_0=0 +ellps=GRS80 +units=m +no_defs";
|
68
|
CRS_locs_WGS84<-CRS("+proj=longlat +ellps=WGS84 +datum=WGS84 +towgs84=0,0,0") #Station coords WGS84
|
69
|
|
70
|
source("GAM_fusion_function_multisampling_02062013.R")
|
71
|
|
72
|
###################### START OF THE SCRIPT ########################
|
73
|
|
74
|
###Reading the daily station data and setting up for models' comparison
|
75
|
#filename<-sub(".shp","",infile1) #Removing the extension from file.
|
76
|
#ghcn<-readOGR(".", filename) #reading shapefile
|
77
|
ghcn<-readOGR(dsn=in_path,layer=sub(".shp","",infile_daily))
|
78
|
CRS_interp<-proj4string(ghcn) #Storing projection information (ellipsoid, datum,etc.)
|
79
|
|
80
|
#mean_LST<- readGDAL(infile5) #Reading the whole raster in memory. This provides a grid for kriging
|
81
|
#proj4string(mean_LST)<-CRS_interp #Assigning coordinate information to prediction grid.
|
82
|
|
83
|
#Station location of the study area
|
84
|
#stat_loc<-read.table(paste(path,"/","location_study_area_OR_0602012.txt",sep=""),sep=",", header=TRUE)
|
85
|
stat_loc<-readOGR(dsn=in_path,layer=sub(".shp","",infile_locs))
|
86
|
|
87
|
#GHCN Database for 1980-2010 for study area (OR)
|
88
|
#data3<-read.table(paste(path,"/","ghcn_data_TMAXy1980_2010_OR_0602012.txt",sep=""),sep=",", header=TRUE)
|
89
|
#data3<-file.path(in_path,infile_monthly)
|
90
|
data3<-readOGR(dsn=in_path,layer=sub(".shp","",infile_monthly))
|
91
|
|
92
|
#Remove NA for LC and CANHEIGHT: Need to check this part
|
93
|
ghcn$LC1[is.na(ghcn$LC1)]<-0
|
94
|
ghcn$LC3[is.na(ghcn$LC3)]<-0
|
95
|
ghcn$CANHEIGHT[is.na(ghcn$CANHEIGHT)]<-0
|
96
|
ghcn$LC4[is.na(ghcn$LC4)]<-0
|
97
|
ghcn$LC6[is.na(ghcn$LC6)]<-0
|
98
|
|
99
|
dates <-readLines(paste(file.path(in_path,infile2))
|
100
|
#LST_dates <-readLines(file.path(in_path,infile3))
|
101
|
#models <-readLines(paste(path,"/",infile4, sep=""))
|
102
|
|
103
|
##Extracting the variables values from the raster files
|
104
|
|
105
|
#The names of covariates can be changed...
|
106
|
rnames<-c("x","y","lon","lat","N","E","N_w","E_w","elev","slope","aspect","CANHEIGHT","DISTOC")
|
107
|
lc_names<-c("LC1","LC2","LC3","LC4","LC5","LC6","LC7","LC8","LC9","LC10","LC11","LC12")
|
108
|
lst_names<-c("mm_01","mm_02","mm_03","mm_04","mm_05","mm_06","mm_07","mm_08","mm_09","mm_10","mm_11","mm_12",
|
109
|
"nobs_01","nobs_02","nobs_03","nobs_04","nobs_05","nobs_06","nobs_07","nobs_08",
|
110
|
"nobs_09","nobs_10","nobs_11","nobs_12")
|
111
|
|
112
|
covar_names<-c(rnames,lc_names,lst_names)
|
113
|
|
114
|
s_raster<-stack(infile3) #read in the data stack
|
115
|
names(s_raster)<-covar_names #Assigning names to the raster layers: making sure it is included in the extraction
|
116
|
|
117
|
#Deal with no data value and zero
|
118
|
pos<-match("LC1",layerNames(s_raster)) #Find column with name "value"
|
119
|
LC1<-raster(s_raster,layer=pos) #Select layer from stack
|
120
|
s_raster<-dropLayer(s_raster,pos)
|
121
|
LC1[is.na(LC1)]<-0
|
122
|
|
123
|
pos<-match("LC3",layerNames(s_raster)) #Find column with name "value"
|
124
|
LC3<-raster(s_raster,layer=pos) #Select layer from stack
|
125
|
s_raster<-dropLayer(s_raster,pos)
|
126
|
LC3[is.na(LC3)]<-0
|
127
|
|
128
|
pos<-match("CANHEIGHT",layerNames(s_raster)) #Find column with name "value"
|
129
|
CANHEIGHT<-raster(s_raster,layer=pos) #Select layer from stack
|
130
|
s_raster<-dropLayer(s_raster,pos)
|
131
|
CANHEIGHT[is.na(CANHEIGHT)]<-0
|
132
|
pos<-match("ELEV_SRTM",layerNames(s_raster)) #Find column with name "ELEV_SRTM"
|
133
|
ELEV_SRTM<-raster(s_raster,layer=pos) #Select layer from stack on 10/30
|
134
|
s_raster<-dropLayer(s_raster,pos)
|
135
|
ELEV_SRTM[ELEV_SRTM <0]<-NA
|
136
|
|
137
|
#s_sgdf<-as(s_raster,"SpatialGridDataFrame") #Conversion to spatial grid data frame
|
138
|
|
139
|
####### Preparing LST stack of climatology...
|
140
|
|
141
|
#l=list.files(pattern="mean_month.*rescaled.rst")
|
142
|
l <-readLines(paste(path,"/",infile6, sep=""))
|
143
|
molst<-stack(l) #Creating a raster stack...
|
144
|
#setwd(old)
|
145
|
molst<-molst-273.16 #K->C #LST stack of monthly average...
|
146
|
idx <- seq(as.Date('2010-01-15'), as.Date('2010-12-15'), 'month')
|
147
|
molst <- setZ(molst, idx)
|
148
|
layerNames(molst) <- month.abb
|
149
|
|
150
|
###### Preparing tables for model assessment: specific diagnostic/metrics
|
151
|
|
152
|
#Model assessment: specific diagnostics/metrics
|
153
|
results_AIC<- matrix(1,1,nmodels+3)
|
154
|
results_GCV<- matrix(1,1,nmodels+3)
|
155
|
results_DEV<- matrix(1,1,nmodels+3)
|
156
|
#results_RMSE_f<- matrix(1,length(models)+3)
|
157
|
|
158
|
#Model assessment: general diagnostic/metrics
|
159
|
results_RMSE <- matrix(1,1,nmodels+4)
|
160
|
results_MAE <- matrix(1,1,nmodels+4)
|
161
|
results_ME <- matrix(1,1,nmodels+4) #There are 8+1 models
|
162
|
results_R2 <- matrix(1,1,nmodels+4) #Coef. of determination for the validation dataset
|
163
|
|
164
|
results_RMSE_f<- matrix(1,1,nmodels+4) #RMSE fit, RMSE for the training dataset
|
165
|
results_MAE_f <- matrix(1,1,nmodels+4)
|
166
|
|
167
|
######## Preparing monthly averages from the ProstGres database and extracting covarvariates from stack
|
168
|
|
169
|
# do this work outside of (before) this function
|
170
|
# to avoid making a copy of the data frame inside the function call
|
171
|
# date1<-ISOdate(data3$year,data3$month,data3$day) #Creating a date object from 3 separate column
|
172
|
# date2<-as.POSIXlt(as.Date(date1))
|
173
|
# data3$date<-date2
|
174
|
# d<-subset(data3,year>=2000 & mflag=="0" ) #Selecting dataset 2000-2010 with good quality: 193 stations
|
175
|
# #May need some screeing??? i.e. range of temp and elevation...
|
176
|
# d1<-aggregate(value~station+month, data=d, mean) #Calculate monthly mean for every station in OR
|
177
|
# id<-as.data.frame(unique(d1$station)) #Unique station in OR for year 2000-2010: 193 but 7 loss of monthly avg
|
178
|
#
|
179
|
# dst<-merge(d1, stat_loc, by.x="station", by.y="STAT_ID") #Inner join all columns are retained
|
180
|
#
|
181
|
# #This allows to change only one name of the data.frame
|
182
|
# pos<-match("value",names(dst)) #Find column with name "value"
|
183
|
# names(dst)[pos]<-c("TMax")
|
184
|
# dst$TMax<-dst$TMax/10 #TMax is the average max temp for monthy data
|
185
|
# #dstjan=dst[dst$month==9,] #dst contains the monthly averages for tmax for every station over 2000-2010
|
186
|
#
|
187
|
# #Extracting covariates from stack
|
188
|
# coords<- dst[c('lon','lat')] #Define coordinates in a data frame
|
189
|
# coordinates(dst)<-coords #Assign coordinates to the data frame
|
190
|
# proj4string(dst)<-CRS_locs_WGS84 #Assign coordinates reference system in PROJ4 format
|
191
|
# dst_month<-spTransform(dst,CRS(CRS_interp)) #Project from WGS84 to new coord. system
|
192
|
#
|
193
|
# stations_val<-extract(s_raster,dst_month) #extraction of the infomration at station location
|
194
|
# stations_val<-as.data.frame(stations_val)
|
195
|
# dst_extract<-cbind(dst_month,stations_val)
|
196
|
# dst<-dst_extract
|
197
|
|
198
|
#Now clean and screen monthly values
|
199
|
#dst_all<-dst
|
200
|
dst_all<-data3
|
201
|
dst<-data3
|
202
|
#dst<-subset(dst,dst$TMax>-15 & dst$TMax<45) #may choose different threshold??
|
203
|
#dst<-subset(dst,dst$ELEV_SRTM>0) #This will drop two stations...or 24 rows
|
204
|
|
205
|
######### Preparing daily values for training and testing
|
206
|
|
207
|
#Screening for bad values: value is tmax in this case
|
208
|
#ghcn$value<-as.numeric(ghcn$value)
|
209
|
#ghcn_all<-ghcn
|
210
|
#ghcn_test<-subset(ghcn,ghcn$value>-150 & ghcn$value<400)
|
211
|
#ghcn_test<-ghcn
|
212
|
#ghcn_test2<-subset(ghcn_test,ghcn_test$elev_1>0)
|
213
|
#ghcn<-ghcn_test2
|
214
|
#coords<- ghcn[,c('x_OR83M','y_OR83M')]
|
215
|
|
216
|
##Sampling: training and testing sites.
|
217
|
|
218
|
#Make this a a function
|
219
|
|
220
|
if (seed_number>0) {
|
221
|
set.seed(seed_number) #Using a seed number allow results based on random number to be compared...
|
222
|
}
|
223
|
nel<-length(dates)
|
224
|
dates_list<-vector("list",nel) #list of one row data.frame
|
225
|
|
226
|
prop_range<-(seq(from=prop_min,to=prop_max,by=step))*100 #range of proportion to run
|
227
|
sn<-length(dates)*nb_sample*length(prop_range) #Number of samples to run
|
228
|
|
229
|
for(i in 1:length(dates)){
|
230
|
d_tmp<-rep(dates[i],nb_sample*length(prop_range)) #repeating same date
|
231
|
s_nb<-rep(1:nb_sample,length(prop_range)) #number of random sample per proportion
|
232
|
prop_tmp<-sort(rep(prop_range, nb_sample))
|
233
|
tab_run_tmp<-cbind(d_tmp,s_nb,prop_tmp)
|
234
|
dates_list[[i]]<-tab_run_tmp
|
235
|
}
|
236
|
|
237
|
sampling_dat<-as.data.frame(do.call(rbind,dates_list))
|
238
|
names(sampling_dat)<-c("date","run_samp","prop")
|
239
|
|
240
|
for(i in 2:3){ # start of the for loop #1
|
241
|
sampling_dat[,i]<-as.numeric(as.character(sampling_dat[,i]))
|
242
|
}
|
243
|
|
244
|
sampling_dat$date<- as.character(sampling_dat[,1])
|
245
|
#ghcn.subsets <-lapply(dates, function(d) subset(ghcn, date==d)) #this creates a list of 10 or 365 subsets dataset based on dates
|
246
|
ghcn.subsets <-lapply(as.character(sampling_dat$date), function(d) subset(ghcn, date==d)) #this creates a list of 10 or 365 subsets dataset based on dates
|
247
|
|
248
|
## adding choice of constant sample
|
249
|
if (seed_number>0) {
|
250
|
set.seed(seed_number) #Using a seed number allow results based on random number to be compared...
|
251
|
}
|
252
|
|
253
|
sampling<-vector("list",length(ghcn.subsets))
|
254
|
sampling_station_id<-vector("list",length(ghcn.subsets))
|
255
|
for(i in 1:length(ghcn.subsets)){
|
256
|
n<-nrow(ghcn.subsets[[i]])
|
257
|
prop<-(sampling_dat$prop[i])/100
|
258
|
ns<-n-round(n*prop) #Create a sample from the data frame with 70% of the rows
|
259
|
nv<-n-ns #create a sample for validation with prop of the rows
|
260
|
ind.training <- sample(nrow(ghcn.subsets[[i]]), size=ns, replace=FALSE) #This selects the index position for 70% of the rows taken randomly
|
261
|
ind.testing <- setdiff(1:nrow(ghcn.subsets[[i]]), ind.training)
|
262
|
#Find the corresponding
|
263
|
data_sampled<-ghcn.subsets[[i]][ind.training,] #selected the randomly sampled stations
|
264
|
station_id.training<-data_sampled$station #selected id for the randomly sampled stations (115)
|
265
|
#Save the information
|
266
|
sampling[[i]]<-ind.training
|
267
|
sampling_station_id[[i]]<- station_id.training
|
268
|
}
|
269
|
## Use same samples across the year...
|
270
|
if (constant==1){
|
271
|
sampled<-sampling[[1]]
|
272
|
data_sampled<-ghcn.subsets[[1]][sampled,] #selected the randomly sampled stations
|
273
|
station_sampled<-data_sampled$station #selected id for the randomly sampled stations (115)
|
274
|
list_const_sampling<-vector("list",sn)
|
275
|
list_const_sampling_station_id<-vector("list",sn)
|
276
|
for(i in 1:sn){
|
277
|
station_id.training<-intersect(station_sampled,ghcn.subsets[[i]]$station)
|
278
|
ind.training<-match(station_id.training,ghcn.subsets[[i]]$station)
|
279
|
list_const_sampling[[i]]<-ind.training
|
280
|
list_const_sampling_station_id[[i]]<-station_id.training
|
281
|
}
|
282
|
sampling<-list_const_sampling
|
283
|
sampling_station_id<-list_const_sampling_station_id
|
284
|
}
|
285
|
|
286
|
######## Prediction for the range of dates and sampling data
|
287
|
|
288
|
#gam_fus_mod<-mclapply(1:length(dates), runGAMFusion,mc.preschedule=FALSE,mc.cores = 8) #This is the end bracket from mclapply(...) statement
|
289
|
#gam_fus_mod_s<-mclapply(1:1, runGAMFusion,mc.preschedule=FALSE,mc.cores = 1) #This is the end bracket from mclapply(...) statement
|
290
|
gam_fus_mod_s<-mclapply(1:length(ghcn.subsets), runGAMFusion,mc.preschedule=FALSE,mc.cores = 9) #This is the end bracket from mclapply(...) statement
|
291
|
#gam_fus_mod2<-mclapply(4:4, runGAMFusion,mc.preschedule=FALSE,mc.cores = 1) #This is the end bracket from mclapply(...) statement
|
292
|
|
293
|
save(gam_fus_mod_s,file= paste(path,"/","results2_fusion_Assessment_measure_all",out_prefix,".RData",sep=""))
|
294
|
|
295
|
## Plotting and saving diagnostic measures
|
296
|
|
297
|
tb<-gam_fus_mod_s[[1]][[3]][0,] #empty data frame with metric table structure that can be used in rbinding...
|
298
|
tb_tmp<-gam_fus_mod_s #copy
|
299
|
|
300
|
for (i in 1:length(tb_tmp)){
|
301
|
tmp<-tb_tmp[[i]][[3]]
|
302
|
tb<-rbind(tb,tmp)
|
303
|
}
|
304
|
rm(tb_tmp)
|
305
|
|
306
|
for(i in 4:ncol(tb)){ # start of the for loop #1
|
307
|
tb[,i]<-as.numeric(as.character(tb[,i]))
|
308
|
}
|
309
|
|
310
|
metrics<-as.character(unique(tb$metric)) #Name of accuracy metrics (RMSE,MAE etc.)
|
311
|
tb_metric_list<-vector("list",length(metrics))
|
312
|
|
313
|
for(i in 1:length(metrics)){ # Reorganizing information in terms of metrics
|
314
|
metric_name<-paste("tb_",metrics[i],sep="")
|
315
|
tb_metric<-subset(tb, metric==metrics[i])
|
316
|
tb_metric<-cbind(tb_metric,sampling_dat[,2:3])
|
317
|
assign(metric_name,tb_metric)
|
318
|
tb_metric_list[[i]]<-tb_metric
|
319
|
}
|
320
|
|
321
|
tb_diagnostic<-do.call(rbind,tb_metric_list)
|
322
|
tb_diagnostic[["prop"]]<-as.factor(tb_diagnostic[["prop"]])
|
323
|
|
324
|
mod_pat<-glob2rx("mod*")
|
325
|
mod_var<-grep(mod_pat,names(tb_diagnostic),value=TRUE) # using grep with "value" extracts the matching names
|
326
|
|
327
|
t<-melt(tb_diagnostic,
|
328
|
measure=mod_var,
|
329
|
id=c("dates","metric","prop"),
|
330
|
na.rm=F)
|
331
|
avg_tb<-cast(t,metric+prop~variable,mean)
|
332
|
median_tb<-cast(t,metric+prop~variable,median)
|
333
|
avg_tb[["prop"]]<-as.numeric(as.character(avg_tb[["prop"]]))
|
334
|
avg_RMSE<-subset(avg_tb,metric=="RMSE")
|
335
|
|
336
|
sampling_obj<-list(sampling_dat=sampling_dat,training=sampling, training_id=sampling_station_id, tb=tb_diagnostic)
|
337
|
|
338
|
write.table(avg_tb, file= paste(path,"/","results2_fusion_Assessment_measure_avg_",out_prefix,".txt",sep=""), sep=",")
|
339
|
write.table(median_tb, file= paste(path,"/","results2_fusion_Assessment_measure_median_",out_prefix,".txt",sep=""), sep=",")
|
340
|
write.table(tb_diagnostic, file= paste(path,"/","results2_fusion_Assessment_measure",out_prefix,".txt",sep=""), sep=",")
|
341
|
write.table(tb, file= paste(path,"/","results2_fusion_Assessment_measure_all",out_prefix,".txt",sep=""), sep=",")
|
342
|
|
343
|
save(sampling_obj, file= paste(path,"/","results2_fusion_sampling_obj",out_prefix,".RData",sep=""))
|
344
|
#save(gam_fus_mod_s,file= paste(path,"/","results2_fusion_Assessment_measure_all",out_prefix,".RData",sep=""))
|
345
|
gam_fus_mod_obj<-list(gam_fus_mod=gam_fus_mod_s,sampling_obj=sampling_obj)
|
346
|
save(gam_fus_mod_obj,file= paste(path,"/","results_mod_obj_",out_prefix,".RData",sep=""))
|
347
|
|
348
|
#### END OF SCRIPT
|