1
|
|
2
|
runKriging <- function(i) { # loop over dates
|
3
|
|
4
|
#This allows to change only one name of the
|
5
|
|
6
|
date<-strptime(dates[i], "%Y%m%d")
|
7
|
month<-strftime(date, "%m")
|
8
|
LST_month<-paste("mm_",month,sep="")
|
9
|
|
10
|
mod <-ghcn.subsets[[i]][,match(LST_month, names(ghcn.subsets[[i]]))]
|
11
|
ghcn.subsets[[i]]$LST <-mod[[1]]
|
12
|
#
|
13
|
# n<-nrow(ghcn.subsets[[i]])
|
14
|
# ns<-n-round(n*prop) #Create a sample from the data frame with 70% of the rows
|
15
|
# nv<-n-ns #create a sample for validation with prop of the rows
|
16
|
# ind.training <- sample(nrow(ghcn.subsets[[i]]), size=ns, replace=FALSE) #This selects the index position for 70% of the rows taken randomly
|
17
|
ind.training<-sampling[[i]]
|
18
|
ind.testing <- setdiff(1:nrow(ghcn.subsets[[i]]), ind.training) #This selects the index position for testing subset stations.
|
19
|
data_s <- ghcn.subsets[[i]][ind.training, ]
|
20
|
data_v <- ghcn.subsets[[i]][ind.testing, ]
|
21
|
|
22
|
#adding to SpatialGridDataFrame
|
23
|
|
24
|
pos<-match(LST_month,layerNames(s_raster)) #Find column with the current month for instance mm12
|
25
|
r1<-raster(s_raster,layer=pos) #Select layer from stack
|
26
|
layerNames(r1)<-"LST"
|
27
|
s_raster<-addLayer(s_raster,r1)
|
28
|
s_sgdf<-as(s_raster,"SpatialGridDataFrame") #Conversion to spatial grid data frame
|
29
|
|
30
|
###BEFORE Kringing the data object must be transformed to SDF
|
31
|
|
32
|
coords<- data_v[,c('x_OR83M','y_OR83M')]
|
33
|
coordinates(data_v)<-coords
|
34
|
proj4string(data_v)<-CRS #Need to assign coordinates...
|
35
|
coords<- data_s[,c('x_OR83M','y_OR83M')]
|
36
|
coordinates(data_s)<-coords
|
37
|
proj4string(data_s)<-CRS #Need to assign coordinates..
|
38
|
|
39
|
#This allows to change only one name of the data.frame
|
40
|
pos<-match("value",names(data_s)) #Find column with name "value"
|
41
|
names(data_s)[pos]<-c("tmax")
|
42
|
data_s$tmax<-data_s$tmax/10 #TMax is the average max temp for months
|
43
|
pos<-match("value",names(data_v)) #Find column with name "value"
|
44
|
names(data_v)[pos]<-c("tmax")
|
45
|
data_v$tmax<-data_v$tmax/10
|
46
|
#dstjan=dst[dst$month==9,] #dst contains the monthly averages for tmax for every station over 2000-2010
|
47
|
##############
|
48
|
###STEP 2 KRIGING###
|
49
|
|
50
|
#Kriging tmax
|
51
|
|
52
|
# hscat(tmax~1,data_s,(0:9)*20000) # 9 lag classes with 20,000m width
|
53
|
# v<-variogram(tmax~1, data_s) # This plots a sample varigram for date 10 fir the testing dataset
|
54
|
# plot(v)
|
55
|
# v.fit<-fit.variogram(v,vgm(2000,"Sph", 150000,1000)) #Model variogram: sill is 2000, spherical, range 15000 and nugget 1000
|
56
|
# plot(v, v.fit) #Compare model and sample variogram via a graphical plot
|
57
|
# tmax_krige<-krige(tmax~1, data_s,mean_LST, v.fit) #mean_LST provides the data grid/raster image for the kriging locations to be predicted.
|
58
|
|
59
|
# krmod1<-try(autoKrige(tmax~1, data_s,s_sgdf,data_s)) #Use autoKrige instead of krige: with data_s for fitting on a grid
|
60
|
# krmod2<-try(autoKrige(tmax~x_OR83M+y_OR83M,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
61
|
# krmod3<-try(autoKrige(tmax~x_OR83M+y_OR83M+ELEV_SRTM,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
62
|
# krmod4<-try(autoKrige(tmax~x_OR83M+y_OR83M+DISTOC,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
63
|
# krmod5<-try(autoKrige(tmax~x_OR83M+y_OR83M+ELEV_SRTM+DISTOC,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
64
|
# krmod6<-try(autoKrige(tmax~x_OR83M+y_OR83M+Northness+Eastness,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
65
|
# krmod7<-try(autoKrige(tmax~x_OR83M+y_OR83M+Northness+Eastness,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
66
|
#
|
67
|
krmod1<-try(autoKrige(tmax~1, data_s,s_sgdf,data_s)) #Use autoKrige instead of krige: with data_s for fitting on a grid
|
68
|
krmod2<-try(autoKrige(tmax~lat+lon,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
69
|
krmod3<-try(autoKrige(tmax~lat+lon+ELEV_SRTM,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
70
|
krmod4<-try(autoKrige(tmax~lat+lon+DISTOC,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
71
|
krmod5<-try(autoKrige(tmax~lat+lon+ELEV_SRTM+DISTOC,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
72
|
krmod6<-try(autoKrige(tmax~lat+lon+Northness+Eastness,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
73
|
krmod7<-try(autoKrige(tmax~lat+lon+Northness+Eastness,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
74
|
|
75
|
krmod8<-try(autoKrige(tmax~LST,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
76
|
krmod9<-try(autoKrige(tmax~lat+lon+LST,input_data=data_s,new_data=s_sgdf,data_variogram=data_s))
|
77
|
|
78
|
# krig1<-krmod1$krige_output #Extracting Spatial Grid Data frame
|
79
|
# krig2<-krmod2$krige_output
|
80
|
# krig3<-krmod3$krige_outpu
|
81
|
# krig4<-krmod4$krige_output
|
82
|
# krig5<-krmod5$krige_output
|
83
|
# krig6<-krmod6$krige_output #Extracting Spatial Grid Data frame
|
84
|
# krig7<-krmod7$krige_output
|
85
|
#krig8<-krmod8$krige_outpu
|
86
|
#krig9<-krmod9$krige_output
|
87
|
|
88
|
#tmax_krig1_s <- overlay(krige,data_s) #This overlays the kriged surface tmax and the location of weather stations
|
89
|
#tmax_krig1_v <- overlay(krige,data_v)
|
90
|
#
|
91
|
# #Cokriging tmax
|
92
|
# g<-gstat(NULL,"tmax", tmax~1, data_s) #This creates a gstat object "g" that acts as container for kriging specifications.
|
93
|
# g<-gstat(g, "SRTM_elev",ELEV_SRTM~1,data_s) #Adding variables to gstat object g
|
94
|
# g<-gstat(g, "LST", LST~1,data_s)
|
95
|
|
96
|
# vm_g<-variogram(g) #Visualizing multivariate sample variogram.
|
97
|
# vm_g.fit<-fit.lmc(vm_g,g,vgm(2000,"Sph", 100000,1000)) #Fitting variogram for all variables at once.
|
98
|
# plot(vm_g,vm_g.fit) #Visualizing variogram fit and sample
|
99
|
# vm_g.fit$set <-list(nocheck=1) #Avoid checking and allow for different range in variogram
|
100
|
# co_kriged_surf<-predict(vm_g.fit,mean_LST) #Prediction using co-kriging with grid location defined from input raster image.
|
101
|
# #co_kriged_surf$tmax.pred #Results stored in SpatialGridDataFrame with tmax prediction accessible in dataframe.
|
102
|
|
103
|
#spplot.vcov(co_kriged_surf) #Visualizing the covariance structure
|
104
|
|
105
|
# tmax_cokrig1_s<- overlay(co_kriged_surf,data_s) #This overalys the cokriged surface tmax and the location of weather stations
|
106
|
# tmax_cokrig1_v<- overlay(co_kriged_surf,data_v)
|
107
|
|
108
|
for (j in 1:models){
|
109
|
|
110
|
#mod<-paste("krig",j,sep="")
|
111
|
mod<-paste("krmod",j,sep="")
|
112
|
|
113
|
krmod_auto<-get(mod)
|
114
|
|
115
|
#If mod "j" is not a model object
|
116
|
if (inherits(krmod_auto,"try-error")) {
|
117
|
|
118
|
#Model assessment:results are NA
|
119
|
|
120
|
results_RMSE[1,1]<- dates[i] #storing the interpolation dates in the first column
|
121
|
results_RMSE[1,2]<- ns #number of stations used in the training stage
|
122
|
results_RMSE[1,3]<- "RMSE"
|
123
|
results_RMSE[1,j+3]<- NA
|
124
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
125
|
|
126
|
results_MAE[1,1]<- dates[i] #storing the interpolation dates in the first column
|
127
|
results_MAE[1,2]<- ns #number of stations used in the training stage
|
128
|
results_MAE[1,3]<- "MAE"
|
129
|
results_MAE[1,j+3]<- NA
|
130
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
131
|
|
132
|
results_ME[1,1]<- dates[i] #storing the interpolation dates in the first column
|
133
|
results_ME[1,2]<- ns #number of stations used in the training stage
|
134
|
results_ME[1,3]<- "ME"
|
135
|
results_ME[1,j+3]<- NA
|
136
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
137
|
|
138
|
results_R2[1,1]<- dates[i] #storing the interpolation dates in the first column
|
139
|
results_R2[1,2]<- ns #number of stations used in the training stage
|
140
|
results_R2[1,3]<- "R2"
|
141
|
results_R2[1,j+3]<- NA
|
142
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
143
|
|
144
|
results_RMSE_f[1,1]<- dates[i] #storing the interpolation dates in the first column
|
145
|
results_RMSE_f[1,2]<- ns #number of stations used in the training stage
|
146
|
results_RMSE_f[1,3]<- "RMSE_f"
|
147
|
results_RMSE_f[1,j+3]<- NA
|
148
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
149
|
|
150
|
results_MAE_f[1,1]<- dates[i] #storing the interpolation dates in the first column
|
151
|
results_MAE_f[1,2]<- ns #number of stations used in the training stage
|
152
|
results_MAE_f[1,3]<- "MAE_f"
|
153
|
results_MAE_f[1,j+3]<- NA
|
154
|
name3<-paste("res_kr_mod",j,sep="")
|
155
|
|
156
|
}
|
157
|
|
158
|
#If mod "j" is not a model object
|
159
|
if (inherits(krmod_auto,"autoKrige")) {
|
160
|
krmod<-krmod_auto$krige_output #Extracting Spatial Grid Data frame
|
161
|
|
162
|
krig_val_s <- overlay(krmod,data_s) #This overlays the kriged surface tmax and the location of weather stations
|
163
|
krig_val_v <- overlay(krmod,data_v) #This overlays the kriged surface tmax and the location of weather stations
|
164
|
|
165
|
pred_krmod<-paste("pred_krmod",j,sep="")
|
166
|
#Adding the results back into the original dataframes.
|
167
|
data_s[[pred_krmod]]<-krig_val_s$var1.pred
|
168
|
data_v[[pred_krmod]]<-krig_val_v$var1.pred
|
169
|
|
170
|
#Model assessment: RMSE and then krig the residuals....!
|
171
|
|
172
|
res_mod_kr_s<- data_s$tmax - data_s[[pred_krmod]] #Residuals from kriging training
|
173
|
res_mod_kr_v<- data_v$tmax - data_v[[pred_krmod]] #Residuals from kriging validation
|
174
|
|
175
|
RMSE_mod_kr_s <- sqrt(sum(res_mod_kr_s^2,na.rm=TRUE)/(nv-sum(is.na(res_mod_kr_s)))) #RMSE from kriged surface training
|
176
|
RMSE_mod_kr_v <- sqrt(sum(res_mod_kr_v^2,na.rm=TRUE)/(nv-sum(is.na(res_mod_kr_v)))) #RMSE from kriged surface validation
|
177
|
MAE_mod_kr_s<- sum(abs(res_mod_kr_s),na.rm=TRUE)/(nv-sum(is.na(res_mod_kr_s))) #MAE from kriged surface training #MAE, Mean abs. Error FOR REGRESSION STEP 1: GAM
|
178
|
MAE_mod_kr_v<- sum(abs(res_mod_kr_v),na.rm=TRUE)/(nv-sum(is.na(res_mod_kr_v))) #MAE from kriged surface validation
|
179
|
ME_mod_kr_s<- sum(res_mod_kr_s,na.rm=TRUE)/(nv-sum(is.na(res_mod_kr_s))) #ME, Mean Error or bias FOR REGRESSION STEP 1: GAM
|
180
|
ME_mod_kr_v<- sum(res_mod_kr_v,na.rm=TRUE)/(nv-sum(is.na(res_mod_kr_v))) #ME, Mean Error or bias FOR REGRESSION STEP 1: GAM
|
181
|
R2_mod_kr_s<- cor(data_s$tmax,data_s[[pred_krmod]],use="complete.obs")^2 #R2, coef. of determination FOR REGRESSION STEP 1: GAM
|
182
|
R2_mod_kr_v<- cor(data_v$tmax,data_v[[pred_krmod]],use="complete.obs")^2 #R2, coef. of determinationFOR REGRESSION STEP 1: GAM
|
183
|
#(nv-sum(is.na(res_mod2)))
|
184
|
#Writing out results
|
185
|
|
186
|
results_RMSE[1,1]<- dates[i] #storing the interpolation dates in the first column
|
187
|
results_RMSE[1,2]<- ns #number of stations used in the training stage
|
188
|
results_RMSE[1,3]<- "RMSE"
|
189
|
results_RMSE[1,j+3]<- RMSE_mod_kr_v
|
190
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
191
|
|
192
|
results_MAE[1,1]<- dates[i] #storing the interpolation dates in the first column
|
193
|
results_MAE[1,2]<- ns #number of stations used in the training stage
|
194
|
results_MAE[1,3]<- "MAE"
|
195
|
results_MAE[1,j+3]<- MAE_mod_kr_v
|
196
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
197
|
|
198
|
results_ME[1,1]<- dates[i] #storing the interpolation dates in the first column
|
199
|
results_ME[1,2]<- ns #number of stations used in the training stage
|
200
|
results_ME[1,3]<- "ME"
|
201
|
results_ME[1,j+3]<- ME_mod_kr_v
|
202
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
203
|
|
204
|
results_R2[1,1]<- dates[i] #storing the interpolation dates in the first column
|
205
|
results_R2[1,2]<- ns #number of stations used in the training stage
|
206
|
results_R2[1,3]<- "R2"
|
207
|
results_R2[1,j+3]<- R2_mod_kr_v
|
208
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
209
|
|
210
|
results_RMSE_f[1,1]<- dates[i] #storing the interpolation dates in the first column
|
211
|
results_RMSE_f[1,2]<- ns #number of stations used in the training stage
|
212
|
results_RMSE_f[1,3]<- "RMSE_f"
|
213
|
results_RMSE_f[1,j+3]<- RMSE_mod_kr_s
|
214
|
#results_RMSE_kr[i,3]<- res_mod_kr_v
|
215
|
|
216
|
results_MAE_f[1,1]<- dates[i] #storing the interpolation dates in the first column
|
217
|
results_MAE_f[1,2]<- ns #number of stations used in the training stage
|
218
|
results_MAE_f[1,3]<- "MAE_f"
|
219
|
results_MAE_f[1,j+3]<- MAE_mod_kr_s
|
220
|
name3<-paste("res_kr_mod",j,sep="")
|
221
|
|
222
|
#as.numeric(res_mod)
|
223
|
#data_s[[name3]]<-res_mod_kr_s
|
224
|
data_s[[name3]]<-as.numeric(res_mod_kr_s)
|
225
|
#data_v[[name3]]<-res_mod_kr_v
|
226
|
data_v[[name3]]<-as.numeric(res_mod_kr_v)
|
227
|
#Writing residuals from kriging
|
228
|
|
229
|
#Saving kriged surface in raster images
|
230
|
data_name<-paste("mod",j,"_",dates[[i]],sep="")
|
231
|
#krig_raster_name<-paste("krmod_",data_name,out_prefix,".tif", sep="")
|
232
|
#writeGDAL(krmod,fname=krig_raster_name, driver="GTiff", type="Float32",options ="INTERLEAVE=PIXEL", overwrite=TRUE)
|
233
|
krig_raster_name<-paste("krmod_",data_name,out_prefix,".rst", sep="")
|
234
|
writeRaster(raster(krmod), filename=krig_raster_name, overwrite=TRUE) #Writing the data in a raster file format...(IDRISI)
|
235
|
|
236
|
#krig_raster_name<-paste("Kriged_tmax_",data_name,out_prefix,".tif", sep="")
|
237
|
#writeGDAL(tmax_krige,fname=krig_raster_name, driver="GTiff", type="Float32",options ="INTERLEAVE=PIXEL")
|
238
|
#X11()
|
239
|
#plot(raster(co_kriged_surf))
|
240
|
#title(paste("Tmax cokriging for date ",dates[[i]],sep=""))
|
241
|
#savePlot(paste("Cokriged_tmax",data_name,out_prefix,".png", sep=""), type="png")
|
242
|
#dev.off()
|
243
|
#X11()
|
244
|
#plot(raster(tmax_krige))
|
245
|
#title(paste("Tmax Kriging for date ",dates[[i]],sep=""))
|
246
|
#savePlot(paste("Kriged_res_",data_name,out_prefix,".png", sep=""), type="png")
|
247
|
#dev.off()
|
248
|
# end of if krige object
|
249
|
}
|
250
|
|
251
|
}
|
252
|
|
253
|
# #Co-kriging only on the validation sites for faster computing
|
254
|
#
|
255
|
# cokrig1_dv<-predict(vm_g.fit,data_v)
|
256
|
# cokrig1_ds<-predict(vm_g.fit,data_s)
|
257
|
# # data_s$tmax_cokr<-cokrig1_ds$tmax.pred
|
258
|
# # data_v$tmax_cokr<-cokrig1_dv$tmax.pred
|
259
|
#
|
260
|
# #Calculate RMSE and then krig the residuals....!
|
261
|
#
|
262
|
# res_mod1<- data_v$tmax - data_v$tmax_kr #Residuals from kriging.
|
263
|
# res_mod2<- data_v$tmax - data_v$tmax_cokr #Residuals from cokriging.
|
264
|
#
|
265
|
# RMSE_mod1 <- sqrt(sum(res_mod1^2,na.rm=TRUE)/(nv-sum(is.na(res_mod1)))) #RMSE from kriged surface.
|
266
|
# RMSE_mod2 <- sqrt(sum(res_mod2^2,na.rm=TRUE)/(nv-sum(is.na(res_mod2)))) #RMSE from co-kriged surface.
|
267
|
# #(nv-sum(is.na(res_mod2)))
|
268
|
|
269
|
#Saving the subset in a dataframe
|
270
|
data_name<-paste("ghcn_v_",dates[[i]],sep="")
|
271
|
assign(data_name,data_v)
|
272
|
data_name<-paste("ghcn_s_",dates[[i]],sep="")
|
273
|
assign(data_name,data_s)
|
274
|
|
275
|
# results[i,1]<- dates[i] #storing the interpolation dates in the first column
|
276
|
# results[i,2]<- ns #number of stations in training
|
277
|
# results[i,3]<- RMSE_mod1
|
278
|
# results[i,4]<- RMSE_mod2
|
279
|
#
|
280
|
# results_mod_n[i,1]<-dates[i]
|
281
|
# results_mod_n[i,2]<-(nv-sum(is.na(res_mod1)))
|
282
|
# results_mod_n[i,3]<-(nv-sum(is.na(res_mod2)))
|
283
|
|
284
|
#Specific diagnostic measures related to the testing datasets
|
285
|
#browser()
|
286
|
results_table_RMSE<-as.data.frame(results_RMSE)
|
287
|
results_table_MAE<-as.data.frame(results_MAE)
|
288
|
results_table_ME<-as.data.frame(results_ME)
|
289
|
results_table_R2<-as.data.frame(results_R2)
|
290
|
results_table_RMSE_f<-as.data.frame(results_RMSE_f)
|
291
|
results_table_MAE_f<-as.data.frame(results_MAE_f)
|
292
|
|
293
|
results_table_AIC<-as.data.frame(results_AIC) #Other tables for kriging
|
294
|
results_table_GCV<-as.data.frame(results_GCV)
|
295
|
results_table_DEV<-as.data.frame(results_DEV)
|
296
|
|
297
|
tb_metrics1<-rbind(results_table_RMSE,results_table_MAE, results_table_ME, results_table_R2,results_table_RMSE_f,results_table_MAE_f) #
|
298
|
tb_metrics2<-rbind(results_table_AIC,results_table_GCV, results_table_DEV)
|
299
|
cname<-c("dates","ns","metric","mod1", "mod2","mod3", "mod4", "mod5", "mod6", "mod7")
|
300
|
colnames(tb_metrics1)<-cname
|
301
|
cname<-c("dates","ns","metric","mod1", "mod2","mod3", "mod4", "mod5", "mod6", "mod7")
|
302
|
colnames(tb_metrics2)<-cname
|
303
|
#colnames(results_table_RMSE)<-cname
|
304
|
#colnames(results_table_RMSE_f)<-cname
|
305
|
#tb_diagnostic1<-results_table_RMSE #measures of validation
|
306
|
#tb_diagnostic2<-results_table_RMSE_f #measures of fit
|
307
|
|
308
|
#write.table(tb_diagnostic1, file= paste(path,"/","results_fusion_Assessment_measure1",out_prefix,".txt",sep=""), sep=",")
|
309
|
|
310
|
#}
|
311
|
print(paste(dates[i],"processed"))
|
312
|
mod_obj<-list(krmod1,krmod2,krmod3,krmod4,krmod5,krmod6,krmod7)
|
313
|
# end of the for loop1
|
314
|
#results_list<-list(data_s,data_v,tb_metrics1,tb_metrics2)
|
315
|
results_list<-list(data_s,data_v,tb_metrics1,tb_metrics2,mod_obj)
|
316
|
return(results_list)
|
317
|
#return(tb_diagnostic1)
|
318
|
}
|