-
Notifications
You must be signed in to change notification settings - Fork 14
/
Copy pathprofile_soilm_neuralnet.R
822 lines (659 loc) · 38.9 KB
/
profile_soilm_neuralnet.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
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
profile_soilm_neuralnet <- function( sitename, nam_target="lue_obs_evi", use_weights=ifelse( nam_target=="lue_obs_evi", TRUE, FALSE ), use_fapar=FALSE, outdir="./data/", varnams_swc=NA, soilm_threshold=NA, packages="nnet", overwrite_profile=TRUE, nrep=1, makepdf=TRUE ){
# ## XXX debug----------------
# sitename = "FR-Pue"
# nam_target = "lue_obs_evi"
# outdir = "./data/profile/"
# soilm_threshold = seq( 0.45, 0.55, 0.05 )
# varnams_swc = c("soilm_swbm")
# packages = "nnet"
# nrep = 2
# overwrite_profile = TRUE
# use_fapar = FALSE
# makepdf = FALSE
# use_weights = FALSE
# ##--------------------------
## default is sampling full range
if (is.na(soilm_threshold)){
soilm_threshold <- seq( 0.1, 0.70, 0.05 )
}
require( dplyr )
require( abind )
require( Metrics )
# source( "repeat_neuralnet.R" )
source( paste( workingdir, "/analyse_modobs.R", sep="" ) )
source( paste( workingdir, "/add_alpha.R", sep="" ) )
source( paste( workingdir, "/predict_nn.R", sep="" ) )
source( paste( workingdir, "/clean_fluxnet.R", sep="" ) )
source( paste( workingdir, "/remove_outliers.R", sep="" ) )
source( paste( workingdir, "/cleandata_nn.R", sep="" ) )
load( paste( workingdir, "/data/modobs_fluxnet2015_s11_s12_s13_with_SWC_v3.Rdata", sep="" ) )
packages <- "nnet"
## check and override if necessary
if ( nam_target=="lue_obs" || nam_target=="lue_obs_evi" || nam_target=="lue_obs_fpar" ){
if (nam_target=="lue_obs_evi"){
fapar_data <- "evi"
} else if (nam_target=="lue_obs_fpar"){
fapar_data <- "fpar"
}
if (use_fapar){
print("WARNING: setting use_fapar to FALSE")
use_fapar <- FALSE
}
}
if (use_weights){
char_wgt <- "_wgt"
} else {
char_wgt <- ""
}
## identifier for output files
if (use_fapar){
if (nam_target=="lue_obs_evi"){
char_fapar <- "_withEVI"
} else if (nam_target=="lue_obs_fpar"){
char_fapar <- "_withFPAR"
} else {
print("ERROR: PROVIDE VALID FAPAR DATA!")
}
} else {
char_fapar <- ""
}
if (!file.exists(outdir)) system( "mkdir -p ", outdir )
outfilnam <- paste( outdir, "profile_", nam_target, char_wgt, char_fapar, "_nn_", sitename, ".Rdata", sep="" )
outfilnam_light <- paste( outdir, "profile_light_", nam_target, char_wgt, char_fapar, "_nn_", sitename, ".Rdata", sep="" )
if ( !file.exists(outfilnam) || overwrite_profile ){
# if (is.na(varnams_swc)){
# varnams_swc <- c( "soilm_splash150", "soilm_splash220", "soilm_swbm", "soilm_etobs", "soilm_etobs_ob" )
# } else {
# print("USING THE FOLLOWING SOIL MOISTURE DATA:")
# print(varnams_swc)
# }
## Get neural network for this site, each target individually and looping over repetitions to get statistics
print("=============================================")
print( paste( "get profile for", sitename, "..." ) )
profile_nn <- list()
profile_nn_light <- list()
rsq_all_by_smdata <- data.frame()
rmse_good_by_smdata <- data.frame()
## Get model output from simulation without temperature or soil moisture limitation (s13)
data <- fluxnet[[ sitename ]]$ddf$s13 ## wcont is SPLASH WITH 150 mm
data <- subset( data, select=c(year_dec, wcont, gpp, aet, pet) )
data <- data %>% dplyr::rename( soilm_splash150=wcont, gpp_pmodel=gpp, aet_pmodel=aet, pet_pmodel=pet )
# data <- rename( data, c("wcont"="soilm", "gpp"="gpp_pmodel", "aet"="aet_pmodel", "pet"="pet_pmodel" ) )
## Get alternative soil moisture data
data$soilm_splash220 <- fluxnet[[ sitename ]]$ddf$s11$wcont
data$soilm_swbm <- fluxnet[[ sitename ]]$ddf$s12$wcont
data$soilm_etobs <- fluxnet[[ sitename ]]$ddf$swc_by_etobs$soilm_from_et
data$soilm_etobs_ob <- fluxnet[[ sitename ]]$ddf$swc_by_etobs$soilm_from_et_orthbucket
## Get observational soil moisture data (variable availability!)
varnams_swc_obs <- c()
if ( is.element( "soilm_obs", varnams_swc ) ){
relevant <- names(fluxnet[[ sitename ]]$ddf$swc_obs)[(is.element( substr(names(fluxnet[[ sitename ]]$ddf$swc_obs), start=1, stop=3), "SWC" ))]
if (length(relevant)>0){
for (iobs in seq(length(relevant))){
varnam <- paste( "soilm_obs_", iobs, sep="" )
data[[ varnam ]] <- fluxnet[[ sitename ]]$ddf$swc_obs[[ relevant[iobs] ]]
varnams_swc_obs <- c( varnams_swc_obs, varnam )
}
# varnams_swc <- c( varnams_swc, "soilm_obs" )
} else {
## no obs. soilm. data found, remove from data list
varnams_swc <- varnams_swc[ -which( "soilm_obs" ) ]
}
}
## Get observational data
# obs <- subset( fluxnet[[ sitename ]]$ddf$obs, select=c( year_dec, gpp_obs2015_GPP_NT_VUT_REF, le_f_mds ) )
obs <- dplyr::select( fluxnet[[ sitename ]]$ddf$obs, year_dec, gpp_obs2015_GPP_NT_VUT_REF, le_f_mds )
## Get input data
inp <- fluxnet[[ sitename ]]$ddf$inp %>% dplyr::select( temp, ppfd, match( fapar_data, names(.) ), vpd, prec )
if (dim(data)[1]==dim(obs)[1] && dim(data)[1]==dim(inp)[1] && data$year_dec[1]==obs$year_dec[1] && data$year_dec[dim(data)[1]]==obs$year_dec[dim(obs)[1]]){
## attach to data
data$gpp_obs <- obs$gpp_obs2015_GPP_NT_VUT_REF
data$et_obs <- obs$le_f_mds
data$wue_obs <- data$gpp_obs / (data$et_obs*1e-6)
data <- cbind( data, inp )
}
##------------------------------------------------
## Remove outliers in WUE
##------------------------------------------------
data$wue_obs <- remove_outliers( data$wue_obs, coef=3.0 )
##------------------------------------------------
## get LUE and remove outliers
##------------------------------------------------
if (!is.null(data$evi)){
data <- data %>% mutate( lue_obs_evi = remove_outliers( gpp_obs / ( ppfd * evi ), coef=3.0 ) )
}
if (!is.null(data$fpar)){
data <- data %>% mutate( lue_obs_fpar = remove_outliers( gpp_obs / ( ppfd * fpar ), coef=3.0 ) )
}
##------------------------------------------------
## normalise soil moisture
##------------------------------------------------
data <- data %>% mutate( soilm_splash150 = soilm_splash150 / 150 )
data <- data %>% mutate( soilm_splash220 = soilm_splash220 / 220 )
data <- data %>% mutate( soilm_swbm = soilm_swbm / 220 )
data <- data %>% mutate( soilm_etobs = soilm_etobs / 220 )
data <- data %>% mutate( soilm_etobs_ob = soilm_etobs_ob / 220 )
for (ivar in varnams_swc_obs){
data[[ ivar ]] <- data[[ ivar ]] / max( data[[ ivar ]] , na.rm=TRUE )
}
for ( isoilm_data in varnams_swc ){
print("---------------------------------------------")
print( paste( "soilm data source:", isoilm_data, "..." ) )
##------------------------------------------------
## Use only days where EVI is above 25% quantile of all days values and temperature is above 5 degrees
##------------------------------------------------
df_nona <- data
# df_nona <- dplyr::filter( df_nona, fapar > quantile( df_nona[[ fapar_data ]], probs=0.25 ) )
df_nona <- dplyr::filter( df_nona, temp > 5.0 )
##------------------------------------------------
## do additional data cleaning, removing NAs, necessary for NN training
##------------------------------------------------
df_nona <- cleandata_nn( df_nona, nam_target )
if ( isoilm_data=="soilm_obs" ){
## use obs soilm data only if of sufficient length
lengths <- apply( subset( df_nona, select=varnams_swc_obs ), 2, function(x) sum(!is.na(x)) )
## drop layer swc obs data if length of data is less than 75% of legth of maximum
idx <- 0
drop_idx <- c()
for (ivar in varnams_swc_obs){
idx <- idx + 1
if (lengths[ivar]<0.75*max(lengths)){
df_nona[[ ivar ]] <- NULL
drop_idx <- c(drop_idx, idx)
}
}
if ( length(drop_idx)>0 ) { varnams_swc_obs <- varnams_swc_obs[-drop_idx] }
## remove NAs in observed soil moisture data
for (ivar in varnams_swc_obs){
df_nona <- df_nona[ which(!is.na(df_nona[[ivar]])), ]
}
}
## get weights for NN training: absorbed light
if (use_weights) {
weights <- df_nona$ppfd * df_nona$evi
} else {
weights <- rep( 1.0, nrow(df_nona) )
}
##------------------------------------------------
## Get NN-predictions and its performance statistics for each soil moisture threshold
##------------------------------------------------
for (isoilm_trh in soilm_threshold){
## subset of supposedly non-soil moisture limited days
if ( isoilm_data=="soilm_obs" ){
## for observational data, do subset w.r.t. soil layer with highest value
tmp <- subset( df_nona, select=varnams_swc_obs )
tmp[ is.na(tmp) ] <- 0.0
vec <- apply( tmp, 1, FUN = max, na.rm=TRUE )
} else {
vec <- df_nona[[ isoilm_data ]]
}
idxs_good <- which( vec > isoilm_trh )
for (ipackage in packages){
if ( length(idxs_good)>30 && (nrow(df_nona) - length(idxs_good))>30 ){
##------------------------------------------------
## Good days GPP (LUE)
##------------------------------------------------
## Get good-days-NN based on subset of data where soil moisture is above threshold
# print("training good-days model ...")
predictors <- c( "ppfd", "temp", "vpd" )
if ( use_fapar ){ predictors <- c( predictors, fapar_data ) }
## Initialise
var_nn_good <- array( NA, dim=c(length(idxs_good),nrep) )
var_nn_pot <- array( NA, dim=c(nrow(df_nona),nrep) )
hidden_best <- NULL
for (irep in 1:nrep){
print(paste("NN repetition", irep))
out_nn_good <- predict_nn(
data = df_nona[ idxs_good, ],
weights = weights[ idxs_good ],
predictors = predictors,
nam_target = nam_target,
do_predict = TRUE,
package = ipackage,
lifesign = "full",
seed = irep,
hidden = hidden_best
)
hidden_best <- out_nn_good$hidden_best
## keep only values
var_nn_good[,irep] <- as.vector( out_nn_good$vals )
##------------------------------------------------
## Bad days GPP (LUE), predicted using the model trained on good days data
##------------------------------------------------
# print("predicting bad days with good-days model ...")
out_nn_pot <- predict_nn(
data = df_nona,
weights = weights,
predictors = predictors,
nam_target = nam_target,
nn = out_nn_good$nn,
do_predict = TRUE,
package = ipackage
)
## keep only values
var_nn_pot[,irep] <- as.vector( out_nn_pot$vals )
}
## Evaluate predictions of good days model
stats_nn_good <- analyse_modobs(
apply( var_nn_good, 1, FUN=mean ) * weights[ idxs_good ],
df_nona[[ nam_target ]][idxs_good],
plot.title=paste( "Good, trh=", isoilm_trh, "data=", isoilm_data ),
do.plot=FALSE
)
## record number of hidden layers of best-performing NN (for quick NN training after this is known)
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$hidden_best_good <- hidden_best
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$hidden_best_good <- hidden_best
## Evaluate predictions of good days model during bad days
stats_nn_pot <- analyse_modobs(
(apply( var_nn_pot, 1, FUN=mean ) * weights)[-idxs_good],
df_nona[[ nam_target ]][-idxs_good],
plot.title=paste( "Bad, trh=", isoilm_trh, "data=", isoilm_data ),
do.plot=FALSE
)
print( paste( "R2 of NN-good with soil moisture threshold ", isoilm_trh, "is", stats_nn_good$rsq ) )
} else {
print(paste("too few data points with soil moisture threshold", isoilm_trh))
idxs_good <- c()
var_nn_good <- NULL
var_nn_pot <- NULL
stats_nn_pot <- NULL
stats_nn_good <- NULL
out_nn_good <- list( nn=NULL )
}
## attach to nn_fuxnet list
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$idxs_good <- idxs_good
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_pot <- var_nn_pot
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_good <- var_nn_good
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$stats_nn_pot <- stats_nn_pot
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$stats_nn_good <- stats_nn_good
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$nn_pot <- out_nn_good$nn # this stores the full output object from last repetition
}
}
##------------------------------------------------
## Get NNact model, trained at all data, soil moisture included as predictor
##------------------------------------------------
# print("training full GPP model ...")
if ( isoilm_data=="soilm_obs" ){
## include the full list of soil moisture variables
soilm_predictors <- varnams_swc_obs
} else {
soilm_predictors <- isoilm_data
}
predictors <- c( "ppfd", "temp", "vpd", soilm_predictors )
if ( use_fapar ){ predictors <- c( predictors, fapar_data ) }
## Initialise
## NNact model, trained at all data, soil moisture included as predictor
var_nn_act <- array( NA, dim=c(nrow(df_nona),nrep) )
hidden_best <- NULL
for (irep in 1:nrep){
print(paste("NN repetition", irep))
out_nn_act <- predict_nn(
data = df_nona,
weights = weights,
predictors = predictors,
nam_target = nam_target,
do_predict = TRUE,
package = ipackage,
seed = irep,
hidden = hidden_best
)
hidden_best <- out_nn_good$hidden_best
## keep only values
var_nn_act[,irep] <- as.vector( out_nn_act$vals )
}
## get statistics of mod vs. obs of all-days full model
stats_nn_act <- analyse_modobs(
apply( var_nn_act, 1, FUN=mean ) * weights,
df_nona[[ nam_target ]],
plot.title=paste( "All, data=", isoilm_data ),
do.plot=FALSE
)
##------------------------------------------------
## Get NNact-VPDonly model, trained at all data, WITHOUT soil moisture
##------------------------------------------------
predictors <- c( "ppfd", "temp", "vpd" )
if ( use_fapar ){ predictors <- c( predictors, fapar_data ) }
## Initialise
## NNact model, trained at all data, soil moisture included as predictor
var_nn_vpd <- array( NA, dim=c(nrow(df_nona),nrep) )
hidden_best <- NULL
for (irep in 1:nrep){
print(paste("NN repetition", irep))
out_nn_vpd <- predict_nn(
data = df_nona,
weights = weights,
predictors = predictors,
nam_target = nam_target,
do_predict = TRUE,
package = ipackage,
seed = irep,
hidden = hidden_best
)
hidden_best <- out_nn_good$hidden_best
## keep only values
var_nn_vpd[,irep] <- as.vector( out_nn_vpd$vals )
}
## get statistics of mod vs. obs of all-days full model
stats_nn_vpd <- analyse_modobs(
apply( var_nn_vpd, 1, FUN=mean ) * weights,
df_nona[[ nam_target ]],
plot.title=paste( "All, data=", isoilm_data ),
do.plot=FALSE
)
##------------------------------------------------
## attach to nn_fuxnet list
##------------------------------------------------
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_act <- var_nn_act
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_vpd <- var_nn_vpd
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]]$stats_nn_act <- stats_nn_act
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]]$stats_nn_vpd <- stats_nn_vpd
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]] <- df_nona[[ nam_target ]]
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]]$year_dec <- df_nona$year_dec
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]]$hidden_best_all <- hidden_best
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]]$nn_act <- out_nn_act$nn # this stores the full output object from last repetition
profile_nn [[ sitename ]][[ isoilm_data ]][[ ipackage ]]$nn_vpd <- out_nn_vpd$nn # this stores the full output object from last repetition
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$hidden_best_all <- hidden_best
print( paste( "R2 of NN-all is", stats_nn_act$rsq ) )
}
} else {
print("=============================================")
print( paste( "profile already available for site", sitename, "..." ) )
load( outfilnam )
}
for ( isoilm_data in varnams_swc ){
for (ipackage in packages){
##------------------------------------------------
## Reorganise data and evaluate sensitivity to soil moisture threshold
##------------------------------------------------
## re-organise data into new data frames:
## df_good contains all predicted data of good days based on NN trained on "good days"
soilm_thrsh_avl <- as.numeric(substr( ls(profile_nn[[sitename]][[isoilm_data]][[ipackage]]), start=7, stop=10 )[ is.element( substr( ls(profile_nn[[sitename]][[isoilm_data]][[ipackage]]), start=1, stop=6 ), "smtrh_" ) ])
##------------------------------------------------
## Do profiling only if it worked for more than zero soil moisture thresholds
##------------------------------------------------
if (length(soilm_thrsh_avl)>0){
## df_good contains all predicted data of good days based on NN trained on "good days"
df_good <- data.frame()
for (isoilm_trh in soilm_thrsh_avl){
if (!is.null(profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_good) &&
!is.null( profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_pot)){
var_nn_good <- apply( profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_good, 1, FUN=mean )
idxs_good <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$idxs_good
tmp <- data.frame(
var_obs = profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]][ idxs_good ],
var_nn = var_nn_good,
ratio = var_nn_good
/ profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]][ idxs_good ], # ratio is mod / obs
soilm_threshold = rep( isoilm_trh, length( idxs_good ) )
)
df_good <- rbind( df_good, tmp )
}
}
df_good$goodbad <- rep("good", nrow(df_good))
# ## get available (=tested) soil moisture levels
# soilm_thrsh_avl <- unique( df_good$soilm_threshold )
## df_bad contains all predicted data of bad days based on NN trained on "good days"
df_bad <- data.frame()
for (isoilm_trh in soilm_threshold){
if (!is.null(profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_pot)){
idxs_good <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$idxs_good
var_nn_pot <- apply( profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_pot, 1, FUN=mean )
tmp <- data.frame(
var_obs =profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]][ -idxs_good ],
var_nn =var_nn_pot[ -idxs_good ],
ratio =var_nn_pot[ -idxs_good ]
/ profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]][ -idxs_good ], # ratio is mod / obs
soilm_threshold =rep( isoilm_trh, length( profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]][ -idxs_good ] ) )
)
df_bad <- rbind( df_bad, tmp )
}
}
df_bad$goodbad <- rep("bad", nrow(df_bad))
# ##------------------------------------------------
# ## Determine best split based on ANOVA measure for difference between good and bad days projections of the good-day-trained NN
# ##------------------------------------------------
# anova_by_soilm <- data.frame( soilm_threshold=c(), F_anova=c() )
# for (isoilm_trh in soilm_thrsh_avl){
# df <- rbind( dplyr::filter( df_good, soilm_threshold==isoilm_trh ), dplyr::filter( df_bad, soilm_threshold==isoilm_trh ) )
# df$goodbad <- as.factor( df$goodbad )
# ## get linear model for factor=treatment
# linmod <- lm( ratio ~ goodbad, data = df )
# ## ANOVA
# tmp_anovamod <- anova( linmod )
# ## add row to statistics data frame
# addrow <- data.frame( soilm_threshold=isoilm_trh, F_anova=tmp_anovamod$`F value`[1] )
# anova_by_soilm <- rbind( anova_by_soilm, addrow )
# }
# best_soilm_trh <- anova_by_soilm$soilm_threshold[ which.max(anova_by_soilm$F_anova) ]
# ##------------------------------------------------
# ## Determine 3 best soil moisture cutoffs w.r.t. best split based on Kolmogorov-Smirnov statistics of good-model vs. all-model predicted values of bad days
# ##------------------------------------------------
# ks_by_soilm <- data.frame( soilm_threshold=c(), ks_stat=c() )
# for (isoilm_trh in soilm_thrsh_avl){
# tmp_ks <- ks.test( dplyr::filter( df_bad, soilm_threshold==isoilm_trh )$ratio, dplyr::filter( df_good, soilm_threshold==isoilm_trh )$ratio, alt='greater' )
# ## add row to statistics data frame
# addrow <- data.frame( soilm_threshold=isoilm_trh, ks_stat=tmp_ks$statistic ) #, p_val=tmp_ks$p.value
# ## if mean of bad days is smaller than mean of good days, it obviously doesn't work. Override with dummy (negative ks_stat value)
# if ( median( dplyr::filter( df_bad, soilm_threshold==isoilm_trh )$ratio ) < median( dplyr::filter( df_good, soilm_threshold==isoilm_trh )$ratio ) ){
# addrow$ks_stat <- -9999
# }
# ks_by_soilm <- rbind( ks_by_soilm, addrow )
# }
# ks_by_soilm <- ks_by_soilm[ which(ks_by_soilm$ks_stat!=-9999), ]
# if (nrow(ks_by_soilm)>0){
# list_best_soilm_trh <- ks_by_soilm$soilm_threshold[ order(-ks_by_soilm$ks_stat)][1:min(3,nrow(ks_by_soilm))]
# } else {
# best_soilm_trh <- max( 0.2, min(soilm_thrsh_avl) )
# }
# profile_nn[[ sitename ]]$p_val <- ks_by_soilm$p_val[ which(ks_by_soilm$soilm_threshold==best_soilm_trh) ]
# profile_nn[[ sitename ]]$ks_stat <- ks_by_soilm$ks_stat[ which(ks_by_soilm$soilm_threshold==best_soilm_trh) ]
##------------------------------------------------
## Determine N best soil moisture cutoffs w.r.t. best split based on difference in median of predicted values for dry and moist days from model trained on moist days only
## criterium: median( NN-good(bad) / obs ) - median( NN-good(good) / obs ) = max!
##------------------------------------------------
n_best <- 5
diff_good_bad <- data.frame( soilm_threshold=c(), diff=c() )
for (isoilm_trh in soilm_thrsh_avl){
## add row to statistics data frame
addrow <- data.frame(
soilm_threshold=isoilm_trh,
diff=median( dplyr::filter( df_bad, soilm_threshold==isoilm_trh )$ratio, na.rm=TRUE ) - median( dplyr::filter( df_good, soilm_threshold==isoilm_trh )$ratio, na.rm=TRUE )
) #, p_val=tmp_ks$p.value
diff_good_bad <- rbind( diff_good_bad, addrow )
}
list_best_soilm_trh <- diff_good_bad$soilm_threshold[ order(-diff_good_bad$diff)][1:min(n_best,nrow(diff_good_bad))]
print("list_best_soilm_trh:")
print( diff_good_bad[ order(-diff_good_bad$diff), ] )
# ##------------------------------------------------
# ## Determine best split based on R-squared of good days model
# ##------------------------------------------------
# ## re-organise statistics into new data frame
# df_stat_good <- data.frame()
# for (isoilm_trh in soilm_thrsh_avl){
# if (!is.null(profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$stats_nn_good)){
# tmp <- data.frame(
# soilm_thrsh_avl =isoilm_trh,
# rsq =profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$stats_nn_good$rsq,
# nse =profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$stats_nn_good$nse,
# rmse =profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$stats_nn_good$rmse
# )
# df_stat_good <- rbind( df_stat_good, tmp )
# }
# }
# ## get soilm_thrsh_avl where best fit is achieved for good-days model (highest NSE)
# sub <- df_stat_good[ is.element( df_stat_good$soilm_thrsh_avl, list_best_soilm_trh ), ]
# best_soilm_trh <- sub$soilm_thrsh_avl[ order(-sub$rsq)][1]
##------------------------------------------------
## Determine best split based on variability of fVAR during good days
##------------------------------------------------
## re-organise statistics into new data frame
df_stat_good <- data.frame()
for (isoilm_trh in soilm_thrsh_avl){
if (!is.null(profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$stats_nn_good)){
idxs_good <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$idxs_good
var_nn_pot <- apply( profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_pot, 1, FUN=mean )
var_nn_act <- apply( profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_act, 1, FUN=mean )
mod <- var_nn_pot[ idxs_good ]
obs <- var_nn_act[ idxs_good ]
## old version
# fvar <- apply( profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_act / profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", isoilm_trh, sep="" ) ]]$var_nn_pot, 1, FUN=mean )
# obs <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]][ idxs_good ]
linmod <- lm( mod ~ profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]][ idxs_good ] )
tmp <- data.frame(
soilm_thrsh_avl = isoilm_trh,
rmse = rmse( obs, mod ),
rsq = summary( linmod )$r.squared
# var = var( fvar[idxs_good] )
)
df_stat_good <- rbind( df_stat_good, tmp )
}
}
## get soilm_thrsh_avl where best fit is achieved for good-days model (lowest variance of fVAR during good days)
sub <- df_stat_good[ is.element( df_stat_good$soilm_thrsh_avl, list_best_soilm_trh ), ]
# best_soilm_trh <- sub$soilm_thrsh_avl[ order( sub$var ) ]
best_soilm_trh <- sub$soilm_thrsh_avl[ order( sub$rmse ) ]
##------------------------------------------------
## Get RMSE-good for this dataset at best soil moisture threshold
##------------------------------------------------
stats_nn_good <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", best_soilm_trh[1], sep="" ) ]]$stats_nn_good
stats_nn_act <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$stats_nn_act
rsq_all_by_smdata <- rbind( rsq_all_by_smdata, data.frame( smdata=isoilm_data, rsq_all=stats_nn_act$rsq ) )
rmse_good_by_smdata <- rbind( rmse_good_by_smdata, data.frame( smdata=isoilm_data, rmse_good=stats_nn_good$rmse ) )
##------------------------------------------------
## save best soil moisture threshold (for this soil moisture dataset) and available soil moisture data sources
##------------------------------------------------
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$best_soilm_trh <- best_soilm_trh
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$df_stat <- df_stat_good
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$diff_good_bad <- diff_good_bad
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$best_soilm_trh <- best_soilm_trh
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", best_soilm_trh[1], sep="" ) ]]$var_nn_pot <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", best_soilm_trh[1], sep="" ) ]]$var_nn_pot
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", best_soilm_trh[1], sep="" ) ]]$idxs_good <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", best_soilm_trh[1], sep="" ) ]]$idxs_good
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_act <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_act
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_vpd <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_vpd
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$year_dec <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$year_dec
## store full output class from NN training (for Lek profiling)
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$nn_act <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$nn_act
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$nn_vpd <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$nn_vpd
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", best_soilm_trh[1], sep="" ) ]]$nn_pot <- profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ paste( "smtrh_", best_soilm_trh[1], sep="" ) ]]$nn_pot
##------------------------------------------------
## Plot profile
##------------------------------------------------
# library(sm)
# library(vioplot)
# vioplot( dplyr::filter( df_good, soilm_threshold=="0.01" )$ratio, col=c("springgreen"), ylim=c(0,3), las=1, xlab="soil moisture threshold", ylab="ratio of NN-modelled vs. observed GPP during good days", yaxs="i" )
# maxy <- max( quantile( df_bad$ratio, probs=0.90 ), quantile( df_good$ratio, probs=0.90 ) )
# maxy <- min( 10, max( max( df_bad$ratio ), max( df_good$ratio ) ) )
maxy <- min( 10, max( quantile( df_bad$ratio, probs=0.90 ), quantile( df_good$ratio, probs=0.90 ) ) )
if (makepdf) pdf( paste( "fig_nn_fluxnet2015/ratio_vs_threshold/ratio_vs_threshold_", nam_target, char_fapar, "_", isoilm_data, "_", sitename, ".pdf", sep="" ), width=8, height=6 )
bp1 <- boxplot(
ratio ~ soilm_threshold,
data =df_good,
col =c("springgreen"), ylim=c(0,maxy),
# xlim=c(min(soilm_threshold), max(soilm_threshold)+0.025),
las=1,
xlab ="soil moisture threshold",
ylab ="ratio of NN-modelled vs. observed GPP",
yaxs ="i",
at =soilm_thrsh_avl-0.01,
xlim =range(soilm_thrsh_avl)+c(-0.02,0.02),
boxwex = 0.01,
outline=FALSE
)
bp2 <- boxplot(
ratio ~ soilm_threshold,
data =df_bad,
col =c("grey70"), ylim=c(0,maxy), las=1,
xlab ="",
ylab ="",
yaxs ="i",
at =soilm_thrsh_avl+0.01,
# xlim =range(soilm_thrsh_avl)+c(-0.02,0.02),
boxwex = 0.01,
add = TRUE,
axes = FALSE,
outline=FALSE
)
abline( h=1.0, col='red' )
rect( soilm_thrsh_avl-0.02, rep(0, length(soilm_thrsh_avl)), soilm_thrsh_avl+0.02, rep(maxy,length(soilm_thrsh_avl)),
border=NA, col=c(rgb(0,0,0,0.2),rgb(0,0,0,0.0))
)
imax <- which( soilm_thrsh_avl==best_soilm_trh[1] )
if (!is.na(best_soilm_trh[1])){
rect( soilm_thrsh_avl[imax]-0.02, 0.01, soilm_thrsh_avl[imax]+0.02, maxy-0.01,
border='red', col=NA
)
}
title( sitename )
if (makepdf) dev.off()
##------------------------------------------------
## Plot R-squared versus soil moisture threshold
##------------------------------------------------
if (!is.na(best_soilm_trh[1])){
if (makepdf) pdf( paste( "fig_nn_fluxnet2015/plot_rsq_vs_smtrh/plot_rsq_vs_smtrh_", nam_target, char_fapar, "_", isoilm_data, "_", sitename, ".pdf", sep="" ) )
par( las=1, yaxs="i", xaxs="i" )
plot( df_stat_good$soilm_thrsh_avl, df_stat_good$rsq, pch=16, xlab="soil moisture threshold (rel. units)", ylab=expression(paste("R"^2)), xlim=c(0,1), ylim=c(0,1), col='red' )
abline( v=df_stat_good$soilm_thrsh_avl, col="grey70")
abline( v=best_soilm_trh[1], col='green' )
if (makepdf) dev.off()
}
##------------------------------------------------
## Analyse modelled vs. observed on good days
##------------------------------------------------
if (makepdf){
plot.fil <- NA
} else {
plot.fil <- paste( "fig_nn_fluxnet2015/modobs_gooddays/profile/modobs_gooddays_", nam_target, char_fapar, "_", isoilm_data, "_", sitename, ".pdf", sep="" )
}
if (!is.na(best_soilm_trh[1])){
out <- analyse_modobs(
dplyr::filter( df_good, soilm_threshold==best_soilm_trh[1] )$var_nn, dplyr::filter( df_good, soilm_threshold==best_soilm_trh[1] )$var_obs,
plot.xlab = expression(paste("observed GPP (gC m"^{-2}, " d"^{-1}, ")")),
plot.ylab = expression(paste("modelled GPP (gC m"^{-2}, " d"^{-1}, ")")),
plot.title = paste( sitename, ", good NN,", nam_target ), plot.col=rgb(0,0,0,0.3),
plot.fil = plot.fil
)
}
##------------------------------------------------
## Analyse modelled vs. observed on all days (full model)
##------------------------------------------------
if (makepdf){
plot.fil <- NA
} else {
plot.dir <- "fig_nn_fluxnet2015/modobs_alldays/profile/"
if (!file.exists(plot.dir)) system( paste( "mkdir -p", plot.dir ) )
plot.fil <- paste( plot.dir, "modobs_alldays_", nam_target, char_fapar, "_", isoilm_data, "_", sitename, ".pdf", sep="" )
}
out <- analyse_modobs(
apply( profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$var_nn_act, 1, FUN=mean ),
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]][[ nam_target ]],
plot.xlab = expression(paste("observed GPP (gC m"^{-2}, " d"^{-1}, ")")),
plot.ylab = expression(paste("modelled GPP (gC m"^{-2}, " d"^{-1}, ")")),
plot.title = paste( sitename, ", full NN,", nam_target ), plot.col=rgb(0,0,0,0.3),
plot.fil = plot.fil,
do.plot = FALSE
)
} else {
##------------------------------------------------
## save best soil moisture threshold (for this soil moisture dataset) and available soil moisture data sources
##------------------------------------------------
profile_nn[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$best_soilm_trh <- NA
profile_nn_light[[ sitename ]][[ isoilm_data ]][[ ipackage ]]$best_soilm_trh <- NA
}
}
}
profile_nn[[ sitename ]]$varnams_swc <- varnams_swc
profile_nn[[ sitename ]]$varnams_swc_obs <- varnams_swc_obs
profile_nn_light[[ sitename ]]$varnams_swc <- varnams_swc
profile_nn_light[[ sitename ]]$varnams_swc_obs <- varnams_swc_obs
##------------------------------------------------
## Determine best soil moisture data w.r.t. RMSE of good days (given best soil moisture threshold)
##------------------------------------------------
profile_nn [[ sitename ]]$best_smdata <- as.character( rmse_good_by_smdata$smdata[ order( rmse_good_by_smdata$rmse_good ) ] )
profile_nn_light[[ sitename ]]$best_smdata <- as.character( rmse_good_by_smdata$smdata[ order( rmse_good_by_smdata$rmse_good ) ] )
##------------------------------------------------
## Save per site
##------------------------------------------------
print( paste( "saving profile file:", outfilnam ) )
save( profile_nn, file=outfilnam )
save( profile_nn_light, file=outfilnam_light )
}