-
Notifications
You must be signed in to change notification settings - Fork 12
/
WHRC_soilcarbon_10km.R
595 lines (544 loc) · 35.9 KB
/
WHRC_soilcarbon_10km.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
## Derivation of potential soil carbon (https://github.com/whrc/Soil-Carbon-Debt)
## Code by: [email protected]
## Contributions by: J. (Jon) Sanderman (WHRC) and G. (Greg) Fiske (WHRC)
## Cite as: Sanderman, J., Hengl, T., Fiske, G., 2017? "The soil carbon debt of 12,000 years of human land use", sumbitted to PNAS. http://dx.doi.org/10.1073/pnas.1706103114
list.of.packages <- c("raster", "rgdal", "nnet", "plyr", "ROCR", "randomForest", "R.utils", "plyr", "parallel", "psych", "mda", "dismo", "snowfall", "hexbin", "lattice", "ranger", "xgboost", "doParallel", "caret", "plotKML", "GSIF")
new.packages <- list.of.packages[!(list.of.packages %in% installed.packages()[,"Package"])]
if(length(new.packages)) install.packages(new.packages)
## Packages in use:
setwd("/data/WHRC_soilcarbon/model10km")
load(".RData")
library(plyr)
library(aqp)
library(stringr)
library(sp)
library(rgdal)
library(devtools)
#devtools::install_github('dmlc/xgboost')
library(xgboost)
#devtools::install_github("imbs-hl/ranger/ranger-r-package/ranger")
library(ranger)
library(nnet)
library(caret)
library(hexbin)
library(snowfall)
library(utils)
library(plotKML)
library(GSIF)
library(raster)
library(R.utils)
library(doParallel)
library(foreign)
library(tools)
#library(doSNOW)
#library(doMC)
#library(randomForestSRC)
library(parallel)
#library(mxnet)
#options(rf.cores=detectCores(), mc.cores=detectCores())
load("/data/models/equi7t3.rda")
plotKML.env(convert="convert", show.env=FALSE)
if(.Platform$OS.type == "windows"){
gdal.dir <- shortPathName("C:/Program files/GDAL")
gdal_translate <- paste0(gdal.dir, "/gdal_translate.exe")
gdalwarp <- paste0(gdal.dir, "/gdalwarp.exe")
} else {
gdal_translate = "/usr/bin/gdal_translate"
gdalwarp = "/usr/bin/gdalwarp"
gdalbuildvrt = "/usr/bin/gdalbuildvrt"
saga_cmd = "/usr/local/bin/saga_cmd"
}
system("gdal-config --version")
source("/data/models/saveRDS_functions.R")
source("/data/models/extract.equi7t3.R")
source("WHRC_functions.R")
## all processing done on ca 10 km grid
ncols = 4320
nrows = 2160
xllcorner = -180
yllcorner = -90
xurcorner = 180
yurcorner = 90
cellsize = 0.0833333
NODATA_value = -9999
library(maps)
library(maptools)
country.m <- map('world', plot=FALSE, fill=TRUE)
IDs <- sapply(strsplit(country.m$names, ":"), function(x) x[1])
require(maptools)
country <- as(map2SpatialPolygons(country.m, IDs=IDs), "SpatialLines")
## 0. Input data -----------
## Climatic variables (https://crudata.uea.ac.uk/cru/data/hrg/)
## Harris, I., Jones, P.D., Osborn, T.J. and Lister, D.H. (2014), Updated high-resolution grids of monthly climatic observations – the CRU TS3.10 Dataset. Int. J. Climatol., 34: 623–642. doi: 10.1002/joc.3711
## Downloaded from: http://www.ipcc-data.org/observ/clim/cru_ts2_1.html
cl.zip.lst <- list.files(path="./climate", pattern = glob2rx("*.zip$"), full.names = TRUE)
## extract:
sapply(cl.zip.lst, function(x){system(paste("7za e ", x," -r -y"))})
cl.lst <- list.files(pattern=glob2rx("cru_*_*_*-*_*.tif")) ## "cru_*_*_1901-1930_*.tif"
## 324 layers at 50 km resolution
GDALinfo("cru_pre_clim_1961-1990_01.tif")
#r = readGDAL("cru_pre_clim_1961-1990_01.tif")
#plot(raster(r))
GDALinfo("cru_frs_clim_1961-1990_01.tif")
GDALinfo("cru_tmn_clim_1961-1990_01.tif")
GDALinfo("cru_tmx_clim_1961-1990_01.tif")
GDALinfo("cru_tmp_clim_1961-1990_01.tif")
GDALinfo("cru_vap_clim_1961-1990_01.tif")
GDALinfo("cru_wet_clim_1961-1990_01.tif")
GDALinfo("cru_cld_clim_1961-1990_01.tif")
GDALinfo("cru_dtr_clim_1961-1990_01.tif")
## Missing value flags are incorrect?
#unlink(paste0("./stacked/", gsub(".tif", "_10km.tif", basename(cl.lst[1]))))
#gdalwarp_clim(cl.lst[1])
## resample to 10 km resolution:
sfInit(parallel=TRUE, cpus=48)
sfExport("gdalwarp", "gdalwarp_clim", "cl.lst", "cellsize", "xllcorner", "yllcorner", "xurcorner", "yurcorner")
out <- sfClusterApplyLB(cl.lst, gdalwarp_clim)
sfStop()
unlink(cl.lst)
## DEM parameters / Geology and landform
## from 250m to 10 km:
des <- read.csv("/data/models/SoilGrids250m_COVS250m.csv")
tcovs <- c(as.character(des$WORLDGRIDS_CODE[c(grep("L??USG5", des$WORLDGRIDS_CODE), grep("???MRG5", des$WORLDGRIDS_CODE), grep("L??USG5", des$WORLDGRIDS_CODE), grep("F??USG5", des$WORLDGRIDS_CODE), grep("C??GLC5", des$WORLDGRIDS_CODE))]), "MNGUSG")
#source("mosaick_function.R")
## 84 layers
#sfInit(parallel=TRUE, cpus=ifelse(length(tcovs)>25,25,length(tcovs)))
#sfExport("equi7t3", "gdalbuildvrt", "gdalwarp", "gdal_translate", "ext", "tcovs", "mosaick.equi7t3", "make_mosaick", "tile.names")
#out <- sfLapply(tcovs, function(x){try( make_mosaick(i="dominant", varn=x, ext=ext, in.path="/data/covs1t", tr=0.00833333, ot="Int16", dstnodata=-32768, tile.names=tile.names) )})
#sfStop()
#geog.lst <- c(list.files(path="./GEOG", pattern=glob2rx("F*_agg_ll.tif$"), full.names =TRUE), list.files(path="./GEOG", pattern=glob2rx("L*_agg_ll.tif$"), full.names =TRUE), list.files(path="./GEOG", pattern=glob2rx("*MRG5_agg_ll.tif$"), full.names =TRUE), "./GEOG/MNGUSG_agg_ll.tif")
geog.lst <- c(list.files(path="/data/stacked250m", pattern="USG5", full.names =TRUE), list.files(path="/data/stacked250m", pattern="MRG5", full.names =TRUE), paste0("/data/stacked250m/S0",3:9,"ESA4.tif"), paste0("/data/stacked250m/S10ESA4.tif"), "./GEOG/MNGUSG_agg_ll.tif", paste0("/data/EarthEnv/MODCF_monthlymean_0",c(1:9),".tif"), paste0("/data/EarthEnv/MODCF_monthlymean_",c(10:12),".tif"), "/data/DAAC/average_soil_and_sedimentary-deposit_thickness.tif")
## 50 layers
## resample to 10 km resolution:
sfInit(parallel=TRUE, type="SOCK", cpus=50)
sfExport("gdalwarp", "geog.lst", "cellsize", "xllcorner", "yllcorner", "xurcorner", "yurcorner")
out <- sfClusterApplyLB(geog.lst, function(x){ if(!file.exists(paste0('./stacked/', gsub(".tif", "_10km.tif", basename(x))))){ system(paste0(gdalwarp, ' ', x, ' ./stacked/', gsub(".tif", "_10km.tif", basename(x)), ' -co \"COMPRESS=DEFLATE\" -r \"average\" -t_srs \"+proj=longlat +datum=WGS84\" -tr ', cellsize, ' ', cellsize, ' -te ', xllcorner,' ', yllcorner, ' ', xurcorner, ' ', yurcorner)) }})
sfStop()
#unlink(geog.lst)
## Intact areas / protected planet:
## 2014, IFL Mapping Team: Greenpeace, University of Maryland, Transparent World, World Resource Institute, WWF Russia. Results/reports can be viewed at www.intactforests.org
## Protected planet WPDA Data (http://www.protectedplanet.net/)
pp.zip.lst <- list.files(path="./intact", pattern = glob2rx("*.zip$"), full.names = TRUE)
sapply(pp.zip.lst, function(x){system(paste("7za x ", x," -r -y"))})
ogrInfo("WDPA_Apr2016-shapefile-polygons.shp", "WDPA_Apr2016-shapefile-polygons")
ogrInfo("ifl_2013.shp", "ifl_2013")
shp.lst = c("WDPA_Apr2016-shapefile-polygons.shp", "ifl_2013.shp", "ifl_2000.shp")
field.lst = c("STATUS_YR","IFL_ID","IFL_ID")
x = sapply(1:length(shp.lst), function(x){rasterize_pol(INPUT=shp.lst[x], FIELD=field.lst[x], cellsize, xllcorner, yllcorner, xurcorner, yurcorner)})
#plot(stack(gsub(".shp", ".sdat", basename(shp.lst[-1]))))
## Terrestrial ecoregions (http://maps.tnc.org/gis_data.html)
download.file("http://maps.tnc.org/files/shp/terr-ecoregions-TNC.zip", "terr-ecoregions-TNC.zip")
system(paste("7za x terr-ecoregions-TNC.zip -r -y"))
#ogrInfo("terr-ecoregions-TNC.shp", "terr-ecoregions-TNC")
ecoregions.db <- read.dbf("tnc_terr_ecoregions.dbf")
str(ecoregions.db)
str(levels(ecoregions.db$WWF_MHTNAM))
ecoregions.db$WWF_i = as.integer(ecoregions.db$WWF_MHTNAM)
write.dbf(ecoregions.db, "tnc_terr_ecoregions.dbf")
rasterize_pol(INPUT="tnc_terr_ecoregions.shp", FIELD="WWF_i", cellsize, xllcorner, yllcorner, xurcorner, yurcorner)
ecoregions_leg = data.frame(Value=1:length(levels(ecoregions.db$WWF_MHTNAM)), Classes=levels(ecoregions.db$WWF_MHTNAM))
write.csv(ecoregions_leg, "ecoregions_leg.csv")
unlink("./stacked/tnc_terr_ecoregions_10km.tif")
system(paste0(gdalwarp, ' tnc_terr_ecoregions.sdat ./stacked/tnc_terr_ecoregions_10km.tif -co \"COMPRESS=DEFLATE\" -r \"near\" -tr ', cellsize, ' ', cellsize, ' -te ', xllcorner,' ', yllcorner, ' ', xurcorner, ' ', yurcorner))
## GAUL country borders:
unlink("./stacked/GAUL_COUNTRIES_10km.tif")
system(paste0(gdalwarp, ' /data/aggregated/GAUL_COUNTRIES_1km.tif ./stacked/GAUL_COUNTRIES_10km.tif -co \"COMPRESS=DEFLATE\" -t_srs \"+proj=longlat +datum=WGS84\" -r \"near\" -tr ', cellsize, ' ', cellsize, ' -te ', xllcorner,' ', yllcorner, ' ', xurcorner, ' ', yurcorner))
## Hyde data set:
system('wget -r -l1 --no-parent ftp://ftp.pbl.nl/hyde/hyde3.2/2016_beta_release/zip/')
hyde.lst <- list.files(path="./ftp.pbl.nl/hyde/hyde3.2/2016_beta_release/zip", pattern = glob2rx("*_lu.zip$"), full.names = TRUE)
## extract files and stack
sapply(hyde.lst, function(x){system(paste("7za e ", x," -r -y"))})
cropland.lst <- list.files(pattern = glob2rx("cropland*.asc$"), full.names = TRUE)
## 74 slices
cropland.tbl <- data.frame(filename=cropland.lst)
cropland.tbl$Year <- strip_year(cropland.tbl$filename, name="cropland")
cropland.tbl <- cropland.tbl[order(cropland.tbl$Year),]
pasture.lst <- list.files(pattern = glob2rx("pasture*.asc$"), full.names = TRUE)
pasture.tbl <- data.frame(filename=pasture.lst)
pasture.tbl$Year <- strip_year(pasture.tbl$filename, name="pasture")
pasture.tbl <- pasture.tbl[order(pasture.tbl$Year),]
## Create animation:
sapply(1:nrow(cropland.tbl), plot_world10km, cropland.tbl)
system(paste0('convert -delay 100 ', paste(gsub(".asc", ".asc.png", cropland.tbl$filename), collapse=" "), ' cropland_historic_Hyde.gif'))
sapply(1:nrow(pasture.tbl), plot_world10km, pasture.tbl)
system(paste0('convert -delay 100 ', paste(gsub(".asc", ".asc.png", pasture.tbl$filename), collapse=" "), ' pasture_historic_Hyde.gif'))
## Resample to 10 km:
asc.lst = list.files(pattern = glob2rx("*.asc$"), full.names = TRUE)
## many layers!
sfInit(parallel=TRUE, cpus=48)
sfExport("gdalwarp", "asc.lst", "cellsize", "xllcorner", "yllcorner", "xurcorner", "yurcorner")
out <- sfClusterApplyLB(asc.lst, function(x){ system(paste0(gdalwarp, ' ', x, ' ./stacked/', gsub(".asc", "_10km.tif", basename(x)), ' -co \"COMPRESS=DEFLATE\" -r \"near\" -t_srs \"+proj=longlat +datum=WGS84\" -tr ', cellsize, ' ', cellsize, ' -te ', xllcorner,' ', yllcorner, ' ', xurcorner, ' ', yurcorner)) })
sfStop()
unlink(asc.lst)
## MODIS land cover for years 2000-2015:
mcd.lst <- c(paste0("/data/MCD12Q1/LandCover_", 2001:2013, "001_L1_500m.tif"), paste0("/data/MCD12Q1/LandCover_", 2001:2013, "001_L5_500m.tif"))
#sapply(mcd.lst, function(x){system(paste0(gdalwarp, ' ', x, ' ', basename(x), ' -co \"COMPRESS=DEFLATE\" -r \"near\" -s_srs \"+proj=sinu +lon_0=0 +x_0=0 +y_0=0 +units=m +no_defs\" -t_srs \"+proj=longlat +datum=WGS84\" -tr ', cellsize, ' ', cellsize, ' -te ', xllcorner,' ', yllcorner, ' ', xurcorner, ' ', yurcorner))})
sfInit(parallel=TRUE, cpus=48)
sfExport("gdalwarp", "mcd.lst", "cellsize", "xllcorner", "yllcorner", "xurcorner", "yurcorner")
out <- sfClusterApplyLB(mcd.lst, function(x){ out.file = paste0("./stacked/", gsub("500m", "10km", basename(x))); if(!file.exists(out.file)){ system(paste0(gdalwarp, ' ', x, ' ', out.file, ' -co \"COMPRESS=DEFLATE\" -r \"near\" -s_srs \"+proj=sinu +R=6371007.181 +nadgrids=@null +wktext\" -t_srs \"+proj=longlat +datum=WGS84\" -tr ', cellsize, ' ', cellsize, ' -te ', xllcorner,' ', yllcorner, ' ', xurcorner, ' ', yurcorner))} })
sfStop()
## Historic forest cover
## http://www.unep-wcmc.org/resources-and-data/generalised-original-and-current-forest
shpF.lst = c("./other/ofc_gen.shp", "./other/cfc_gen.shp")
shpF.db.lst <- lapply(gsub(".shp", ".dbf", shpF.lst), read.dbf)
for(i in 1:length(shpF.db.lst)){
shpF.db.lst[[i]]$TYPE_INT = as.integer(shpF.db.lst[[i]]$TYPE)
write.dbf(shpF.db.lst[[i]], gsub(".shp", ".dbf", shpF.lst[[i]]))
}
fieldF.lst = c("TYPE_INT","TYPE_INT")
x = sapply(1:length(shpF.lst), function(x){rasterize_pol(INPUT=shpF.lst[x], FIELD=fieldF.lst[x], cellsize, xllcorner, yllcorner, xurcorner, yurcorner)})
forestcover_leg = data.frame(Value=1:length(levels(shpF.db.lst[[1]]$TYPE)), Classes=levels(shpF.db.lst[[1]]$TYPE))
write.csv(forestcover_leg, "forestcover_leg.csv")
## convert to geotifs:
sapply(list.files(pattern=glob2rx("*.sdat$")), function(x){ system(paste0(gdalwarp, ' ', x, ' ./stacked/', gsub(".sdat", "_10km.tif", basename(x)), ' -co \"COMPRESS=DEFLATE\" -r \"near\" -tr ', cellsize, ' ', cellsize, ' -te ', xllcorner,' ', yllcorner, ' ', xurcorner, ' ', yurcorner)) })
## Potential wetlands GIEMS (http://www.estellus.fr/index.php?static13/giems-d15):
unlink("./stacked/giems_d15_v10_10km.tif")
system(paste0(gdalwarp, ' /data/EarthEnv/giems_d15_v10.tif ./stacked/giems_d15_v10_10km.tif -co \"COMPRESS=DEFLATE\" -t_srs \"+proj=longlat +datum=WGS84\" -r \"average\" -tr ', cellsize, ' ', cellsize, ' -te ', xllcorner,' ', yllcorner, ' ', xurcorner, ' ', yurcorner))
plot(raster("./stacked/giems_d15_v10_10km.tif"), col=SAGA_pal[[1]])
## function to fill in missing values:
source("/data/models/tiler.R")
missing.tifs = list.files(path="./stacked", pattern=glob2rx("MODCF_monthlymean_*_10km.tif"), full.names=TRUE)
#, list.files(path="./stacked", pattern="ESA4", full.names=TRUE))
for(i in 1:length(missing.tifs)){
close.gaps(inputTile=missing.tifs[i], maskTile="landmask_10km.sgrd", outTile=missing.tifs[i], ot="Int16", nodata="32767", nodata_out="-32768", a_srs="+proj=longlat +datum=WGS84 +no_defs +ellps=WGS84 +towgs84=0,0,0", zmin=0, method="stepwise", cpus=56, fix.zmin=FALSE)
}
## Load soil profiles:
#SOCS_profs = readRDS("SOCS_global_profs_December_2016.rds")
OCD_profs = readRDS("OCD_global_profs_March_2017.rds")
## Add points from Remnant / Native DB
#rnDB <- list(site=read.csv("Remnant_native_SOC_DBv1_site.csv"), horizon=read.csv("Remnant_native_SOC_DBv1_horizon.csv"))
#rnDB$site$SOCS = rnDB$site$Reported.100.cm.SOC/10
#str(rnDB)
## IMPORT RASTERS INTO ONE BIG STACK -----------
hyde.cl = c("tot_rice", "tot_rainfed", "tot_irri", "rf_rice", "rf_norice", "rangeland", "pasture", "ir_rice", "grazing", "cropland")
cru.cl = c("cru_pre_clim", "cru_tmp_clim", "cru_tmx_clim","cru_tmn_clim")
## _1901-1930
selP = c("C05GLC5","DEMMRG5","SLPMRG5","TWIMRG5","MNGUSG","MODCF_monthlymean","USG5","ESA4","average_soil","LandCover_2013001","ifl_", sapply(hyde.cl, function(x){paste0(x,900)}), sapply(hyde.cl, function(x){paste0(x,1800)}), sapply(hyde.cl, function(x){paste0(x,1910)}), sapply(hyde.cl, function(x){paste0(x,1960)}), sapply(hyde.cl, function(x){paste0(x,1990)}), sapply(hyde.cl, function(x){paste0(x,1910)}), sapply(hyde.cl, function(x){paste0(x,2016)}), sapply(hyde.cl, function(x){paste0(x,2010)}), sapply(cru.cl, function(x){paste0(x,"_1901-1930")}), sapply(cru.cl, function(x){paste0(x,"_1961-1990")}), "WDPA", "ofc_gen_10km", "cfc_gen_10km", "giems_d15_v10")
## 103 types
tot.tif <- list.files(path="./stacked", pattern = glob2rx("*.tif$"), full.names = TRUE)
## 1236 images
tifP = tot.tif[unlist(sapply(selP, function(x){grep(x, tot.tif)}))]
str(tifP)
## 242 layers
g10kmP = readGDAL(tifP[1])
x = parallel::mclapply(1:length(tifP), function(j){ readGDAL(tifP[j], silent=TRUE)$band1 }, mc.cores=56)
g10kmP@data = data.frame(x)
## replace '-' symbols otherwise reports problems with column names:
names(g10kmP) = gsub("\\-", "\\.", basename(file_path_sans_ext(tifP)))
rm(x)
#names(g10kmP) = basename(file_path_sans_ext(tifP[1]))
#for(j in 2:length(tifP)){
# g10kmP@data[,basename(file_path_sans_ext(tifP[j]))] = readGDAL(tifP[j], silent=TRUE)$band1
#}
#plot(raster(g10kmP["cru_tmp_clim_1901-1930_03_10km"]), col=SAGA_pal[[1]])
#plot(raster(g10kmP["cfc_gen_10km"]), col=SAGA_pal[[1]])
g10kmP$cfc_gen_10km <- ifelse(is.na(g10kmP$cfc_gen_10km), 6, g10kmP$cfc_gen_10km)
g10kmP$ofc_gen_10km <- ifelse(is.na(g10kmP$ofc_gen_10km), 6, g10kmP$ofc_gen_10km)
g10kmP$cfc_gen_10km <- as.factor(g10kmP$cfc_gen_10km)
g10kmP$ofc_gen_10km <- as.factor(g10kmP$ofc_gen_10km)
summary(g10kmP$ofc_gen_10km)
#plot(raster(g10kmP["cfc_gen_10km"]), col=SAGA_pal[[1]])
## 10.6GB object
g10kmP$ifl_2000_10km <- readGDAL("./stacked/ifl_2000_10km.tif")$band1
#g10kmP$cropland2009AD_10km <- readGDAL("./stacked/cropland2009AD_10km.tif")$band1
summary(as.factor(g10kmP$LandCover_2013001_L1_10km))
g10kmP$MASK <- ifelse(is.na(g10kmP$LandCover_2013001_L1_10km) | g10kmP$LandCover_2013001_L1_10km == 15 | g10kmP$LandCover_2013001_L1_10km == 0, NA, 1)
summary(g10kmP$MASK)
#writeGDAL(g10kmP["MASK"], "landmask_10km.tif", type="Byte", mvFlag=255, options = "COMPRESS=DEFLATE")
writeGDAL(g10kmP["MASK"], "landmask_10km.sdat", type="Byte", mvFlag=255, drivername = "SAGA")
g10kmP$s.intact = !is.na(g10kmP$ifl_2000_10km) | (!is.na(g10kmP$'WDPA_Apr2016.shapefile.polygons_10km') & g10kmP$cropland2010AD_10km<10)
summary(g10kmP$s.intact)
## 602,000 pixels in mask
g10mP.mask <- g10kmP["s.intact"]
g10mP.mask <- as(g10mP.mask, "SpatialPixelsDataFrame")
g10mP.mask <- g10mP.mask[g10mP.mask$s.intact==TRUE,]
#plot(raster(g10mP.mask))
#lines(country)
g10mP.mask@data[,1] <- as.numeric(g10mP.mask@data[,1])
writeGDAL(g10mP.mask[1], "intact_mask_10km.tif", type="Byte", mvFlag=255, options="COMPRESS=DEFLATE")
gc(); gc()
## Filter out missing values (latitudes >65 degrees N)
rnd = spsample(g10kmP["MASK"], n=5e3, type = "random")
ovRND = over(y=g10kmP, x=spTransform(rnd, CRS(proj4string(g10kmP))))
ovRND = ovRND[!is.na(ovRND$MASK),]
sel.na <- colSums(sapply(ovRND, function(x){!is.na(x)}))
sel.na[which(sel.na<1000)]
## 1. Historic soil organic carbon stock ----
## modelled using point data
## m.OCS = f ( climate, relief, surface geology, land cover )
## overlay existing profiles and 10km covs
ovM2 <- over(y=g10kmP, x=spTransform(OCD_profs, CRS(proj4string(g10kmP))))
save.image()
#plot(raster(g10kmP["cru_tmn_clim_1901.1930_06_10km"]), col=SAGA_pal[[1]])
## select columns for model building (ONLY COVS representing CURRENT CONDITIONS):
pred.selC = names(g10kmP)[c(grep("MRG5", names(g10kmP)), grep("ESA4", names(g10kmP)), grep("MODCF_monthlymean", names(g10kmP)), grep("USG", names(g10kmP)), grep("2010AD", names(g10kmP)), grep("clim_1961", names(g10kmP)), grep("cfc_gen", names(g10kmP)), grep("giems_d15", names(g10kmP)))]
#pred.selO = names(g10kmP)[c(grep("MRG5", names(g10kmP)), grep("USG", names(g10kmP)), grep("900AD", names(g10kmP)), grep("clim_1901", names(g10kmP)), grep("ofc_gen", names(g10kmP)))]
## without climate as dynamic variable
str(pred.selC)
## 106 in total
#ovMC2 <- cbind(as.data.frame(SOCS_profs[c("SOURCEID","dSOCS_30cm","dSOCS_100cm","dSOCS_200cm")]), ovM2[,pred.selC])
ovMC2 <- cbind(as.data.frame(OCD_profs[c("SOURCEID","DEPTH.f","OCDENS")]), ovM2[,pred.selC])
names(ovMC2) = gsub("2010AD", "", gsub("1961.1990_", "", names(ovMC2)))
## Replace factors with indicators:
in1km = data.frame(model.matrix(~cfc_gen_10km-1, ovMC2))
ovMC2 = cbind(ovMC2, in1km)
#str(ovMC2)
predT.sel = names(ovMC2)[grep(pattern="_10km", names(ovMC2))]
## For simulated points make sure that OCS is 0 for HYDE = 0:
sel.sim = grep("SIM_", paste(ovMC2$SOURCEID))
#View(ovMC2[sel.sim,(c("SOURCEID","OCDENS","grazing_10km", "cropland_10km"))])
ovMC2[sel.sim,"OCDENS"] = ifelse(ovMC2[sel.sim,"grazing_10km"]==0&ovMC2[sel.sim,"cropland_10km"]==0,0,NA)
saveRDS.gz(ovMC2, file="regMatrix_OCD_global_profs.rds")
unlink("mrf.OCD_2010.rds")
unlink("mgb.OCD_2010.rds")
library(xgboost)
library(ranger)
library(caret)
## Fit models / sub-sample to speed up model fitting ----
Nsub <- 1e4
formulaString.OCD <- as.formula(paste0('OCDENS ~ DEPTH.f + ', paste(predT.sel[-which(predT.sel=="cfc_gen_10km")], collapse="+")))
## Initiate cluster
require(parallel)
cl <- makeCluster(56)
registerDoParallel(cl)
cat("Results of model fitting 'randomForest / XGBoost':\n\n", file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"))
cat("\n", file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"), append=TRUE)
cat(paste("Variable:", all.vars(formulaString.OCD)[1]), file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"), append=TRUE)
cat("\n", file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"), append=TRUE)
LOC_ID <- ovMC2$SOURCEID
## Caret training settings (reduce number of combinations to speed up):
ctrl <- trainControl(method="repeatedcv", number=3, repeats=1)
gb.tuneGrid <- expand.grid(eta = c(0.3,0.4,0.5), nrounds = c(50,100,150), max_depth = 2:3, gamma = 0, colsample_bytree = 0.8, min_child_weight = 1)
rf.tuneGrid <- expand.grid(mtry = seq(10,60,by=5))
out.rf <- paste0("mrf.OCD_2010.rds")
if(!file.exists(out.rf)){
dfs <- ovMC2[,all.vars(formulaString.OCD)]
sel <- complete.cases(dfs)
dfs <- dfs[sel,]
if(nrow(dfs)<Nsub){Nsub=nrow(dfs)}
## optimize mtry parameter:
if(!file.exists(gsub("mrf","t.mrf",out.rf))){
t.mrfX <- caret::train(formulaString.OCD, data=dfs[sample.int(nrow(dfs), Nsub),], method="ranger", trControl=ctrl, tuneGrid=rf.tuneGrid)
saveRDS.gz(t.mrfX, file=gsub("mrf","t.mrf",out.rf))
} else {
t.mrfX <- readRDS.gz(gsub("mrf","t.mrf",out.rf))
}
## fit RF model using 'ranger' (fully parallelized)
mrfX <- ranger(formulaString.OCD, data=dfs, importance="impurity", write.forest=TRUE, mtry=t.mrfX$bestTune$mtry, num.trees=85)
saveRDS.gz(mrfX, file=out.rf)
## Top 15 covariates:
sink(file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"), append=TRUE, type="output")
print(mrfX)
cat("\n Variable importance:\n", file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"), append=TRUE)
xl <- as.list(ranger::importance(mrfX))
print(t(data.frame(xl[order(unlist(xl), decreasing=TRUE)[1:40]])))
## save fitting success vectors:
fit.df <- data.frame(LOC_ID=LOC_ID[sel], observed=dfs[,1], predicted=predictions(mrfX))
unlink(paste0("RF_fit_OCD.csv.gz"))
write.csv(fit.df, paste0("RF_fit_OCD.csv"))
gzip(paste0("RF_fit_OCD.csv"))
mg.out = paste0("mgb.OCD_2010.rds")
if(!file.exists(mg.out)){
## fit XGBoost model (uses all points):
mgbX <- caret::train(formulaString.OCD, data=dfs, method="xgbTree", trControl=ctrl, tuneGrid=gb.tuneGrid)
saveRDS.gz(mgbX, file=mg.out)
}
importance_matrix <- xgb.importance(mgbX$coefnames, model=mgbX$finalModel)
cat("\n", file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"), append=TRUE)
print(mgbX)
cat("\n XGBoost variable importance:\n", file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"), append=TRUE)
print(importance_matrix[1:40,])
cat("--------------------------------------\n", file=paste0("Potential_OCD_resultsFit_LandUse_only.txt"), append=TRUE)
sink()
}
stopCluster(cl);
closeAllConnections()
#rm(mrfX); rm(mgbX)
mrfX
## 55%
mgbX
## 45%
## Predict OCD values using current and past climate/land cover/land use ----
cfc.levs = levels(ovMC2$cfc_gen_10km)
DepthI = c(0,30,100,200)
unlink(list.files(pattern=glob2rx(paste0("OCD_", DepthI, "cm_year_*_10km.tif$"))))
#mrfX = readRDS.gz(file="mrf.OCD_2010.rds")
#mgbX = readRDS.gz(file="mgb.OCD_2010.rds")
periods = c("NoLU","900AD","1800AD","1910AD","1960AD","1990AD","2010AD","2016AD")
ofc.lst = c("ofc_gen","ofc_gen","ofc_gen","cfc_gen","cfc_gen","cfc_gen","cfc_gen","cfc_gen")
## run in loop - TAKES CA 30 MINS
for(j in 1:length(periods)){
pred.selC = c(names(g10kmP)[c(grep("MRG5", names(g10kmP)), grep("ESA4", names(g10kmP)), grep("MODCF_monthlymean", names(g10kmP)), grep("USG", names(g10kmP)), grep(periods[j], names(g10kmP)), grep("clim_1961", names(g10kmP)), grep(ofc.lst[j], names(g10kmP)), grep("giems_d15", names(g10kmP)))], "MASK")
g10kmP_current <- g10kmP[pred.selC]
names(g10kmP_current) = gsub(periods[j], "", gsub("1961.1990_", "", names(g10kmP_current)))
## Special case = reduce all Land use to 0
if(periods[j]=="NoLU"){
g10kmP_current@data[,c("tot_rice_10km","tot_rainfed_10km","tot_irri_10km", "rf_rice_10km","rf_norice_10km","rangeland_10km","pasture_10km","ir_rice_10km","grazing_10km","cropland_10km")] = 0
} else {
## Filter out all LU values smaller than 3% (HYDE is often not that precise):
for(i in c("tot_rice_10km","tot_rainfed_10km","tot_irri_10km", "rf_rice_10km","rf_norice_10km","rangeland_10km","pasture_10km","ir_rice_10km","grazing_10km","cropland_10km")){
g10kmP_current@data[,i] <- ifelse(g10kmP_current@data[,i]<3, 0, g10kmP_current@data[,i])
}
}
for(k in DepthI){
out.tif = paste0("OCD_", k, "cm_year_", periods[j], "_10km.tif")
if(!file.exists(out.tif)){
## predict OCD
pr.OCD_c <- predict_e(mrfX, mgbX, newdata=g10kmP_current, cfc.levs=cfc.levs, depth=k)
writeGDAL(pr.OCD_c[1], out.tif, type="Int16", mvFlag=-9999, options="COMPRESS=DEFLATE")
}
}
}
## Derive cumulative SOCS for 0-2 m ----
tif.lst <- lapply(periods, function(x){paste0("./OCD/OCD_",c(0,30,100,200),"cm_year_",x,"_10km.tif")})
source("WHRC_functions.R")
sfInit(parallel=TRUE, cpus=length(periods))
sfLibrary(raster)
sfLibrary(rgdal)
sfExport("sum_SOCS", "tif.lst", "periods")
missing.lst <- sfLapply(1:length(periods), function(i){sum_SOCS(tif.lst[[i]], year=periods[i])})
sfStop()
## for NoLU and 2010:
sum_SOCS(tif.lst[[1]], year=periods[1], depth.sel=c(30,100,200))
sum_SOCS(tif.lst[[7]], year=periods[7], depth.sel=c(30,100,200))
## Clean-up coastline offset pixels ----
## All maps should refer to the same land mask:
OCS.lst = list.files("./SOCS", pattern=glob2rx("*10km.tif$"), full.names = TRUE)
for(k in c("900AD","1800AD","1910AD","1960AD","1990AD","2010AD")) { landmask_fix(OCS.lst, s.year = k) }
## And also mask out 0 values in noLandUse:
for(i in c("30cm","100cm","200cm")){
s = raster::stack(c(paste0("./SOCS/SOCS_0_",i,"_year_NoLU_10km.tif"), "./OCD/landmask_10km.tif"))
s = as(s, "SpatialGridDataFrame")
s@data[,"fix"] = ifelse(is.na(s$landmask_10km), NA, s@data[,1])
writeGDAL(s["fix"], paste0("./SOCS/SOCS_0_",i,"_year_NoLU_10km.tif"), type="Int16", mvFlag=-32767, options="COMPRESS=DEFLATE")
}
## plot difference:
SOC_10km = stack(c("./SOCS/SOCS_0_200cm_year_NoLU_10km.tif","./SOCS/SOCS_0_200cm_year_2016AD_10km.tif"))
SOC_10km = as(SOC_10km, "SpatialGridDataFrame")
#plot(stack(SOC_10km))
SOC_10km$dif = (SOC_10km@data[,1]-SOC_10km@data[,2])
plot(SOC_10km["dif"], zlim=c(-50,200), col=SAGA_pal[[1]])
writeGDAL(SOC_10km["dif"], "SOCS_0_200cm_year_difference_10km.tif", options = c("COMPRESS=DEFLATE"))
unlink("SOCS_0_200cm_year_difference_10km_xy.tif")
system('gdalwarp SOCS_0_200cm_year_difference_10km.tif SOCS_0_200cm_year_difference_10km_xy.tif -t_srs \"+proj=robin +lon_0=0 +x_0=0 +y_0=0 +ellps=WGS84 +datum=WGS84 +units=m +no_defs\" -te -16810131 -6625155 16810131 8343004 -co \"COMPRESS=DEFLATE\"')
system('gdalwarp ./stacked/cropland2016AD_10km.tif cropland2016AD_10km_xy.tif -t_srs \"+proj=robin +lon_0=0 +x_0=0 +y_0=0 +ellps=WGS84 +datum=WGS84 +units=m +no_defs\" -te -16810131 -6625155 16810131 8343004 -co \"COMPRESS=DEFLATE\"')
system('gdalwarp ./stacked/pasture2016AD_10km.tif pasture2016AD_10km_xy.tif -t_srs \"+proj=robin +lon_0=0 +x_0=0 +y_0=0 +ellps=WGS84 +datum=WGS84 +units=m +no_defs\" -te -16810131 -6625155 16810131 8343004 -co \"COMPRESS=DEFLATE\"')
download.file(url="http://www.naturalearthdata.com/http//www.naturalearthdata.com/download/110m/cultural/ne_110m_admin_0_countries.zip", "ne_110m_admin_0_countries.zip", "auto")
unzip("ne_110m_admin_0_countries.zip")
world <- readOGR(".", "ne_110m_admin_0_countries")
world <- spTransform(world, CRS("+proj=robin +lon_0=0 +x_0=0 +y_0=0 +ellps=WGS84 +datum=WGS84 +units=m +no_defs"))
world@bbox
image(raster("SOCS_0_200cm_year_difference_10km_xy.tif"), zlim=c(-20,120), col=SAGA_pal[[1]])
lines(as(world, "SpatialLines"))
## Plot variable importance:
xl = as.list(ranger::importance(mrfX))
xl = t(data.frame(xl[order(unlist(xl), decreasing=TRUE)[2:26]]))
pdf(file = "Fig_RF_importance_plot_200cm.pdf", width = 7, height = 7.5)
par(mar=c(2.5,9,2.5,0.5), oma=c(1,1,1,1))
plot(x=rev(xl)/max(xl)*100, y=1:25, pch = 19, col="blue", xlab="Importance (%)", xlim=c(0,105), ylim=c(0,26), yaxp=c(0,25,25), xaxs="i", yaxs="i", cex=1.4, yaxt="n", ylab="", main="SOC density model importance plot", cex.main=1)
abline(h=1:25, lty=2, col="grey")
#axis(2, at=1:25, labels=rev(attr(xl, "dimnames")[[1]]), las=2)
axis(2, at=1:25, labels=rev(c("Max. temp. September", "Max. temp. August", "MCF October", "MCF December", "Elevation", "MCF September", "MCF January", "MCF November", "Max. temp. October", "MCF June", "Wetness Index", "MCF May", "MCF July", "MCF February", "MCF March", "MCF August", "MCF April", "Snow prob. March", "Precipitation October", expression(bold("Grazing (HYDE)")), "Max. temp. April", "Hillands class", expression(bold("Cropland (HYDE)")), "Max. temp. February", expression(bold("Total rainfed (HYDE)")))), las=2)
dev.off()
## Plot 1 to 1 relationships ----
library(scales)
library(hexbin)
pfun <- function(x,y, ...){
panel.hexbinplot(x,y, ...)
panel.loess(x, y, ..., col = "black",lty=1,lw=2,span=1/18)
}
pal = R_pal[["bpy_colors"]][1:18]
#par(mfcol=c(1,2), oma=c(1,1,1,1))
p3 = hexbinplot(ovMC2$OCDENS~ovMC2$cru_tmx_clim_10_10km, colramp=colorRampPalette(pal), ylab="Organic carbon density (kg/cubic-m)", xlab="Grazing (HYDE)", type="g", ylim=c(0,120), lwd=1, lcex=8, inner=.2, cex.labels=.8, xbins=30, asp=1, colorcut=c(0,0.01,0.03,0.07,0.15,0.25,0.5,0.75,1)) ## , panel=pfun
p4 = hexbinplot(ovMC2$OCDENS~ovMC2$DEMMRG5_agg_ll_10km, colramp=colorRampPalette(pal), ylab="Organic carbon density (kg/cubic-m)", xlab="Grazing (HYDE)", type="g", ylim=c(0,120), lwd=1, lcex=8, inner=.2, cex.labels=.8, xbins=30, asp=1, colorcut=c(0,0.01,0.03,0.07,0.15,0.25,0.5,0.75,1))
p1 = hexbinplot(ovMC2$OCDENS~ovMC2$grazing_10km, colramp=colorRampPalette(pal), ylab="Organic carbon density (kg/cubic-m)", xlab="Grazing (HYDE)", type="g", ylim=c(0,120), xlim=c(0,100), lwd=1, lcex=8, inner=.2, cex.labels=.8, xbins=30, asp=1, colorcut=c(0,0.01,0.03,0.07,0.15,0.25,0.5,0.75,1))
p2 = hexbinplot(ovMC2$OCDENS~ovMC2$cropland_10km, colramp=colorRampPalette(pal), ylab="", xlab="Cropland (HYDE)", type="g", ylim=c(0,120), xlim=c(0,100), lwd=1, lcex=8, inner=.2, cex.labels=.8, xbins=30, asp=1, colorcut=c(0,0.01,0.03,0.07,0.15,0.25,0.5,0.75,1))
library(gridExtra)
pdf(file = "Fig_hexbinplots_correlations_HYDE.pdf", width = 8, height = 4, pointsize=14)
par(oma=c(1,1,1,1))
#do.call(grid.arrange, c(list(p3,p4,p1,p2), ncol=2))
do.call(grid.arrange, c(list(p1,p2), ncol=2))
dev.off()
library(leaflet)
library(htmlwidgets)
unlink("SOCS_0_200cm_year_difference_10km_plt.tif")
system(paste0(gdalwarp, ' SOCS_0_200cm_year_difference_10km.tif SOCS_0_200cm_year_difference_10km_plt.tif -co \"COMPRESS=DEFLATE\" -overwrite -s_srs EPSG:4326 -t_srs EPSG:3857 -tr 10000 10000 -multi -of GTiff -te -20037508 -7706358 20022492 15506358')) ## \"+init=epsg:3857\"
unlink("SOCS_0_200cm_year_NoLU_10km_plt.tif")
system(paste0(gdalwarp, ' SOCS_0_200cm_year_NoLU_10km.tif SOCS_0_200cm_year_NoLU_10km_plt.tif -co \"COMPRESS=DEFLATE\" -overwrite -s_srs EPSG:4326 -t_srs EPSG:3857 -tr 10000 10000 -multi -of GTiff -te -20037508 -7706358 20022492 15506358'))
system(paste0(gdalwarp, ' SOCS_0_200cm_year_2016AD_10km.tif SOCS_0_200cm_year_2016AD_10km_plt.tif -co \"COMPRESS=DEFLATE\" -overwrite -s_srs EPSG:4326 -t_srs EPSG:3857 -tr 10000 10000 -multi -of GTiff -te -20037508 -7706358 20022492 15506358'))
#plot(raster("OCS_2m_10km.tif"))
r = readGDAL("SOCS_0_200cm_year_difference_10km_plt.tif")
r$band1 <- ifelse(r$band1< -27, -27, ifelse(r$band1>85, 85, r$band1))
#r$band1 <- ifelse(r$band1<0, 0, ifelse(r$band1>450, 450, r$band1))
summary(r$band1)
r = raster(r)
pal <- colorNumeric(SAGA_pal[[1]], values(r), na.color = "transparent")
m1 <- leaflet() %>% addTiles() %>% addRasterImage(r, colors=pal, opacity=0.6, project=FALSE, maxBytes = 6 * 1024 * 1024) %>% addLegend(pal=pal, values=values(r), title="SOCS 0--200 cm in t/ha (difference)")
saveWidget(m1, file="SOCS_0_200cm_year_difference_10km.html")
r2 = readGDAL("SOCS_0_200cm_year_NoLU_10km_plt.tif")
summary(r2$band1)
r2$band1 <- ifelse(r2$band1<0, 0, ifelse(r2$band1>680, 680, r2$band1))
r2 = raster(r2)
pal2 <- colorNumeric(SAGA_pal[[1]][5:20], values(r2), na.color = "transparent")
m2 <- leaflet() %>% addTiles() %>% addRasterImage(r2, colors=pal2, opacity=0.6, project=FALSE, maxBytes = 6 * 1024 * 1024) %>% addLegend(pal=pal2, values=values(r2), title="SOCS 0---200 cm in t/ha (no LU)")
saveWidget(m2, file="SOCS_0_200cm_year_NoLU_10km.html")
r3 = readGDAL("SOCS_0_200cm_year_2016AD_10km_plt.tif")
r3$band1 <- ifelse(r3$band1<0, 0, ifelse(r3$band1>680, 680, r3$band1))
r3 = raster(r3)
pal3 <- colorNumeric(SAGA_pal[[1]][5:20], values(r3), na.color = "transparent")
m3 <- leaflet() %>% addTiles() %>% addRasterImage(r3, colors=pal3, opacity=0.6, project=FALSE, maxBytes = 6 * 1024 * 1024) %>% addLegend(pal=pal3, values=values(r3), title="SOCS 0---200 cm in t/ha (2016AD)")
saveWidget(m3, file="SOCS_0_200cm_year_2016AD_10km.html")
save.image()
## comparison histograms:
ca.sg = stack(list("OCS_100cm_historic_10km_xy.tif","OCS_100cm_current_10km_xy.tif"))
ca.sg = as(ca.sg, "SpatialGridDataFrame")
str(ca.sg@data)
#lm(ca.sg$OCS_100cm_historic_10km~ca.sg$OCS_100cm_current_10km, ca.sg@data[!is.na(ca.sg$OCS_100cm_historic_10km),])
with(ca.sg@data[sample.int(length(ca.sg$OCS_100cm_historic_10km_xy),20000),], psych::scatter.hist(log1p(OCS_100cm_historic_10km_xy),log1p(OCS_100cm_current_10km_xy), xlab="historic", ylab="current", pch=19, title="Organic carbon stock (0--100 cm)", col=alpha("lightblue", 0.4), cex=1.5))
## difference in current and historic OCS total:
ca.sg = as(ca.sg, "SpatialPixelsDataFrame")
ratioOCS = ca.sg@data$OCS_100cm_historic_10km_xy/ca.sg@data$OCS_100cm_current_10km_xy*100
hist(ratioOCS[ratioOCS<300 & ratioOCS>0 & !is.na(ratioOCS)], breaks=seq(0,300,by=10), col="grey", xlim=c(0,300), xlab="Difference in percent", main="Historic / current SOCS ratio")
quantile(ratioOCS[ratioOCS<300 & ratioOCS>0 & !is.na(ratioOCS)], c(.1,.9))
mean(ratioOCS[ratioOCS<450 & ratioOCS>0 & !is.na(ratioOCS)])
## Prediction error for RF ----
## based on an empirical solution explained in: https://github.com/imbs-hl/ranger/issues/136
## TAKES 30 mins
resid.OCDENS = predict_cv_resid(formulaString.OCD, data=ovMC2, nfold=4)
#xyplot(Predicted~Observed, resid.OCDENS, asp=1, par.settings=list(plot.symbol = list(col=alpha("black", 0.6), fill=alpha("red", 0.6), pch=21, cex=0.9)), xlab="measured", ylab="predicted (ranger)")
ovMC2$resid = NA
ovMC2[complete.cases(ovMC2[,all.vars(formulaString.OCD)]),"resid"] = abs(resid.OCDENS$Observed - resid.OCDENS$Predicted)
## model absolute residuals as function of covariates:
var.fm2 = as.formula(paste("resid ~ ", paste0(all.vars(formulaString.OCD)[-1], collapse = "+")))
var.OCDENS.rf <- ranger(var.fm2, data = ovMC2[complete.cases(ovMC2[,all.vars(var.fm2)]),], write.forest = TRUE, num.trees=85, mtry = mrfX$mtry)
var.OCDENS.rf
## R-square = 35%
## predict uncertainty:
for(d in c(0,30,100,200)){
pred.selC = c(names(g10kmP)[c(grep("MRG5", names(g10kmP)), grep("ESA4", names(g10kmP)), grep("MODCF_monthlymean", names(g10kmP)), grep("USG", names(g10kmP)), grep("2010AD", names(g10kmP)), grep("clim_1961", names(g10kmP)), grep(ofc.lst[j], names(g10kmP)), grep("giems_d15", names(g10kmP)))], "MASK")
g10kmP_current <- g10kmP[pred.selC]
names(g10kmP_current) = gsub("2010AD", "", gsub("1961.1990_", "", names(g10kmP_current)))
pr.OCD_var <- predict_e(var.OCDENS.rf, newdata=g10kmP_current, cfc.levs=cfc.levs, depth=d)
writeGDAL(pr.OCD_var[1], paste0("abs_error_OCDENS_",d,"cm_2010AD_10km_ll.tif"), type="Int16", mvFlag=-9999, options="COMPRESS=DEFLATE")
}
## All covariates:
#system("7za a SOCS_global_Covariates_10km.7z ./stacked/*.tif")
## Derive total soil organic carbon stock in Pg ----
## Convert maps to Equal area projection (http://geoawesomeness.com/best-map-projection/) e.g. Sinusoidal or the Eckert IV projection:
system('gdalwarp ./SOCS/SOCS_0_100cm_year_2010AD_10km.tif ./SOCS/SOCS_0_100cm_year_2010AD_10km_sin.tif -t_srs \"+proj=sinu +lon_0=0 +x_0=0 +y_0=0 +units=m +no_defs\" -tr 10000 10000 -co \"COMPRESS=DEFLATE\" -overwrite')
grid10km.sin = readGDAL("./SOCS/SOCS_0_100cm_year_2010AD_10km_sin.tif")
summary(grid10km.sin$band1)
## Total stock in Pg:
round(sum(grid10km.sin$band1*1e4^2/1e4, na.rm=TRUE)/1e9)
## [1] 1998
system('gdalwarp ./SOCS/SOCS_0_30cm_year_2010AD_10km.tif ./SOCS/SOCS_0_30cm_year_2010AD_10km_sin.tif -t_srs \"+proj=sinu +lon_0=0 +x_0=0 +y_0=0 +units=m +no_defs\" -tr 10000 10000 -co \"COMPRESS=DEFLATE\" -overwrite')
grid10km.sin$ocs30cm = readGDAL("./SOCS/SOCS_0_30cm_year_2010AD_10km_sin.tif")$band1
round(sum(grid10km.sin$ocs30cm*1e4^2/1e4, na.rm=TRUE)/1e9)
## [1] 868
## Check the Eckert projection:
system('gdalwarp ./SOCS/SOCS_0_100cm_year_2010AD_10km.tif ./SOCS/SOCS_0_100cm_year_2010AD_10km_gp.tif -t_srs \"+proj=eck4 +lon_0=0 +x_0=0 +y_0=0 +ellps=WGS84 +datum=WGS84 +units=m +no_defs\" -tr 10000 10000 -co \"COMPRESS=DEFLATE\" -overwrite')
#plot(raster("./SOCS/SOCS_0_100cm_year_2010AD_10km_gp.tif"))
grid10km.gp = readGDAL("./SOCS/SOCS_0_100cm_year_2010AD_10km_gp.tif")
## Total stock in Pg:
round(sum(grid10km.gp$band1*1e4^2/1e4, na.rm=TRUE)/1e9)
## [1] 1999