This is a quick primer on how to handle MODIS and Landsat NDVI data from Google Earth Engine (GEE) in R. It’s primarily for my Honours student, Hannah Simon, but I thought why not make it public for all to share? I first posted the MODIS section and had Hannah play with the Landsat data as a learning exercise, so full credit to her for the Landsat code.
For a general introduction for many of the functions used and handling spatial data in general see A Primer for Handling Spatial Data in R posted previously.
We start by getting the necessary libraries:
#Get libraries
library(raster) #to handle rasters## Loading required package: splibrary(rgdal) #to handle shapefiles## rgdal: version: 1.3-4, (SVN revision 766)
## Geospatial Data Abstraction Library extensions to R successfully loaded
## Loaded GDAL runtime: GDAL 2.1.3, released 2017/20/01
## Path to GDAL shared files: /Library/Frameworks/R.framework/Versions/3.5/Resources/library/rgdal/gdal
## GDAL binary built with GEOS: FALSE
## Loaded PROJ.4 runtime: Rel. 4.9.3, 15 August 2016, [PJ_VERSION: 493]
## Path to PROJ.4 shared files: /Library/Frameworks/R.framework/Versions/3.5/Resources/library/rgdal/proj
## Linking to sp version: 1.3-1library(reshape) #for data wrangling
library(ggplot2) #for pretty plotting
library(animation) #for silly animations
datwd <- "/Users/jasper/Documents/GIS/CFR/MODISforHannah/"Now let’s start with…
MODIS
#Get data
mNDVI <- stack(paste0(datwd, "2017043_MODIS_v1g_JNP_MOD_NDVI.tif")) #Note that the file has multiple bands so stack() is the easiest way to read all bands
mNDVI #See what the data look like## class : RasterStack
## dimensions : 15, 20, 300, 393 (nrow, ncol, ncell, nlayers)
## resolution : 250, 250 (x, y)
## extent : 361250, 366250, 6242250, 6246000 (xmin, xmax, ymin, ymax)
## coord. ref. : +proj=utm +zone=34 +south +datum=WGS84 +units=m +no_defs +ellps=WGS84 +towgs84=0,0,0
## names : X2017043_//MOD_NDVI.1, X2017043_//MOD_NDVI.2, X2017043_//MOD_NDVI.3, X2017043_//MOD_NDVI.4, X2017043_//MOD_NDVI.5, X2017043_//MOD_NDVI.6, X2017043_//MOD_NDVI.7, X2017043_//MOD_NDVI.8, X2017043_//MOD_NDVI.9, X2017043_//OD_NDVI.10, X2017043_//OD_NDVI.11, X2017043_//OD_NDVI.12, X2017043_//OD_NDVI.13, X2017043_//OD_NDVI.14, X2017043_//OD_NDVI.15, ...
## min values : -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, ...
## max values : 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, ...So we’ve pulled in the full record of NDVI from the “MODIS/MOD13Q1 16-day composite vegetation indices” product from the dawn of time (i.e. the 18th February 2000) until as recent as possible (6th March 2017), but you’ll notice the names in the raster, e.g. X2017043_MODIS_v1g_JNP_MOD_NDVI.1 aren’t very informative, simply numbering the layers from 1 to 393, but not giving us the dates… For some reason I still haven’t worked out how to embed the dates in the .tif raster exported from GEE, so I save them manually from the console into a text file (stone age, I know, sorry…).
metdat <- read.delim(paste0(datwd, "MODIS_metadata.txt"), sep = " ", header = F)
head(metdat)## V1 V2 V3 V4 V5
## 1 0: Image MODIS/MOD13Q1/MOD13Q1_005_2000_02_18 (1 band)
## 2 1: Image MODIS/MOD13Q1/MOD13Q1_005_2000_03_05 (1 band)
## 3 2: Image MODIS/MOD13Q1/MOD13Q1_005_2000_03_21 (1 band)
## 4 3: Image MODIS/MOD13Q1/MOD13Q1_005_2000_04_06 (1 band)
## 5 4: Image MODIS/MOD13Q1/MOD13Q1_005_2000_04_22 (1 band)
## 6 5: Image MODIS/MOD13Q1/MOD13Q1_005_2000_05_08 (1 band)Still not pretty, but getting closer…
date <- substr(metdat[,3], start = 27, stop = 37)
date[1:5]## [1] "2000_02_18" "2000_03_05" "2000_03_21" "2000_04_06" "2000_04_22"With a bit of squint-counting and trial and error we manage to use substr() to extract the dates!
But for R to treat them as dates we need to tell it they are dates with as.Date(), and specify the date format (see ?strptime).
date <- as.Date(date, format = "%Y_%m_%d")
date[1:5]## [1] "2000-02-18" "2000-03-05" "2000-03-21" "2000-04-06" "2000-04-22"Et voilà!
So we now have a stack of MODIS NDVI images through time, and we know their dates. We can embed the dates into the raster like so:
names(mNDVI) <- date
mNDVI #Have another look at the data## class : RasterStack
## dimensions : 15, 20, 300, 393 (nrow, ncol, ncell, nlayers)
## resolution : 250, 250 (x, y)
## extent : 361250, 366250, 6242250, 6246000 (xmin, xmax, ymin, ymax)
## coord. ref. : +proj=utm +zone=34 +south +datum=WGS84 +units=m +no_defs +ellps=WGS84 +towgs84=0,0,0
## names : X2000.02.18, X2000.03.05, X2000.03.21, X2000.04.06, X2000.04.22, X2000.05.08, X2000.05.24, X2000.06.09, X2000.06.25, X2000.07.11, X2000.07.27, X2000.08.12, X2000.08.28, X2000.09.13, X2000.09.29, ...
## min values : -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, ...
## max values : 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, ...Great! But now you’ll notice that the min and max values are rather strange (-32767 to 32767) when we know that NDVI can only range from -1 to 1, and rarely goes below 0. Very strange! Let’s have a look at our first image.
plot(mNDVI[[1]])
0 to 4000? Even stranger!
It turns out the issue is that GEE has exported the data in what appears to be a strange format to reduce the file size. We need to tell R how to interpret it by setting a gain and offset.
NAvalue(mNDVI)=-32767
offs(mNDVI)=0
gain(mNDVI)=.0001
plot(mNDVI[[1]])
Much more reasonable!
Ok, so now we can begin to look at our questions; namely, how does MODIS NDVI vary in space and time?
But first, we should set our focal area.
site <- readOGR(dsn=paste0(datwd, "Jonaskop 120m.kml"), layer="Jonaskop 120m.kml") # A Google Earth KML file## OGR data source with driver: KML
## Source: "/Users/jasper/Documents/GIS/CFR/MODISforHannah/Jonaskop 120m.kml", layer: "Jonaskop 120m.kml"
## with 1 features
## It has 2 fields## Warning in readOGR(dsn = paste0(datwd, "Jonaskop 120m.kml"), layer =
## "Jonaskop 120m.kml"): Z-dimension discardedsite <- spTransform(site, CRS(proj4string(mNDVI))) #Reproject to match the projection of the MODIS data (UTM 34S in this case)
plot(mNDVI[[1]])
plot(site, add=T)
It looks like we’re wrangling a lot more MODIS data than we need to. Let’s crop to out focal area…
mNDVI <- crop(mNDVI, site, snap = "out") #"snap="out"" adds a buffer around the site
plot(mNDVI[[1]])
plot(site, add=T)
…and look at a few images through time.
plot(mNDVI) #Note that R limits it to the first 16 images
Lots of variation in space! What about time?
if(!file.exists("/Users/jasper/GIT/jslingsby.github.io_local/static/img/output/mNDVI.gif")) { # Check if the file exists
saveGIF({
for (i in 1:dim(mNDVI)[3]) plot(mNDVI[[i]],
main=names(mNDVI)[i],
legend.lab="NDVI",
col=rev(terrain.colors(length(c(seq(0,.7,.05),1)))),
breaks=seq(0,.7,.05))},
movie.name = "/Users/jasper/GIT/jslingsby.github.io_local/static/img/output/mNDVI.gif",
ani.width = 600, ani.height = 500,
interval=.25)
}
*NOTE: this writes a .gif to file and does not render in R (or R notebook), so I had to insert it in the Markdown script.
You could make it a lot prettier, but still pretty fancy. Not very useful for the print edition though. How about just a simple line graph? this requires extracting the data to a data.frame.
dat <- as.data.frame(mNDVI)
dat[1:10,1:10] #look at the data## X2000.02.18 X2000.03.05 X2000.03.21 X2000.04.06 X2000.04.22 X2000.05.08
## 1 0.2889 0.3280 0.3723 0.3743 0.3979 0.3902
## 2 0.3127 0.3745 0.3850 0.3872 0.3680 0.4681
## 3 0.3205 0.3508 0.3909 0.3878 0.3647 0.4488
## 4 0.3256 0.3323 0.3823 0.4001 0.3542 0.4050
## 5 0.3206 0.3841 0.3934 0.3817 0.3864 0.3987
## 6 0.2866 0.3567 0.3723 0.3757 0.3799 0.4278
## 7 0.2866 0.3567 0.3832 0.3808 0.3510 0.4681
## 8 0.3340 0.3709 0.4215 0.3787 0.4191 0.4681
## 9 0.3205 0.4144 0.4215 0.4099 0.3900 0.4615
## 10 0.3256 0.4278 0.4363 0.4085 0.4115 0.4615
## X2000.05.24 X2000.06.09 X2000.06.25 X2000.07.11
## 1 0.4644 0.4777 0.4730 0.4220
## 2 0.4704 0.4777 0.4730 0.4289
## 3 0.4698 0.4574 0.4340 0.4532
## 4 0.4596 0.4574 0.4636 0.4228
## 5 0.4522 0.4193 0.4636 0.4517
## 6 0.4493 0.4538 0.4091 0.4160
## 7 0.4043 0.4913 0.4730 0.4359
## 8 0.4885 0.4913 0.4616 0.4300
## 9 0.4318 0.4589 0.4908 0.4363
## 10 0.4212 0.4413 0.4697 0.4489So each date is a column and each cell in space is a row in our data frame. This isn’t ideal for plotting so let’s reshape the data into long format.
rownames(dat) <- paste0("A",1:20) #Add dummy column labels
dat <- as.data.frame(t(dat)) #transpose data.frame
dat$Date <- as.Date(rownames(dat), format = "X%Y.%m.%d")
dat <- melt(dat, id=c("Date"))
head(dat)## Date variable value
## 1 2000-02-18 A1 0.2889
## 2 2000-03-05 A1 0.3280
## 3 2000-03-21 A1 0.3723
## 4 2000-04-06 A1 0.3743
## 5 2000-04-22 A1 0.3979
## 6 2000-05-08 A1 0.3902Now let’s make some pretty pictures!
ggplot(data = dat, aes(x = Date, y = value, colour = variable)) + geom_line()
Pretty neat! It looks like our site burnt in the summer of 2000/2001. Now let’s split them up.
ggplot(data = dat, aes(x = Date, y = value, colour = variable)) + geom_line() + facet_wrap(~variable)
Well there’s definitely clear seasonality and, while subtle, there are differences among the cells. What do we see with…
Landsat
First we read in the full record of data for Landsat 8 for our area of interest.
lsNDVI <- stack(paste0(datwd,"Landsat/2017045_v1_LC8_L1T_TOA_Jonaskop_daily__2013-04-15-2017-03-25.tif")) #note that I have used LC8 (LandSat 8)
lsNDVI## class : RasterStack
## dimensions : 114, 157, 17898, 154 (nrow, ncol, ncell, nlayers)
## resolution : 30, 30 (x, y)
## extent : 361320, 366030, 6242370, 6245790 (xmin, xmax, ymin, ymax)
## coord. ref. : +proj=utm +zone=34 +south +datum=WGS84 +units=m +no_defs +ellps=WGS84 +towgs84=0,0,0
## names : X2017045_//17.03.25.1, X2017045_//17.03.25.2, X2017045_//17.03.25.3, X2017045_//17.03.25.4, X2017045_//17.03.25.5, X2017045_//17.03.25.6, X2017045_//17.03.25.7, X2017045_//17.03.25.8, X2017045_//17.03.25.9, X2017045_//7.03.25.10, X2017045_//7.03.25.11, X2017045_//7.03.25.12, X2017045_//7.03.25.13, X2017045_//7.03.25.14, X2017045_//7.03.25.15, ...
## min values : -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, -32768, ...
## max values : 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, 32767, ...Then the metadata with dates.
metdat2 <- read.delim(paste0(datwd,"Landsat/LC8_L1T_TOA_metadata.txt"), sep = " ", header = F)
head(metdat2) #first few rows of metdat## V1 V2
## 1 0: 2013-04-15
## 2 1: 2013-04-22
## 3 2: 2013-05-17
## 4 3: 2013-05-24
## 5 4: 2013-06-09
## 6 5: 2013-06-18tail(metdat2) #last few rows of metdat (our data goes up to 25th march, 2017)## V1 V2
## 149 148: 2017-02-12
## 150 149: 2017-02-21
## 151 150: 2017-02-28
## 152 151: 2017-03-09
## 153 152: 2017-03-16
## 154 153: 2017-03-25Extract the dates, embed into the raster and have a look at the first scene.
date2 <- substr(metdat2[,2], start = 0, stop = 10) #will extract YYYY_MM_DD
names(lsNDVI) <- date2 #assign the date to the 'names' of each NDVI value
plot(lsNDVI[[1]]) 
Looks good, but the scale’s all wrong again, so we need to fix the gain and offset
NAvalue(lsNDVI)=0
offs(lsNDVI)=-1.5
gain(lsNDVI)=.001
plot(lsNDVI[[1]])
plot(site, add=T)
Better! But still suspiscious… Note that I hacked the offset at -1.5 (it’s usually 2) I’ll need to recheck what the gain and offset values should be and fix this later…
Now let’s zoom in on our focal site
lsNDVI <- crop(lsNDVI, site, snap = "out") #"snap="out"" adds a buffer around the site
plot(lsNDVI[[1]])
plot(site, add=T)
Now let’s have a look at the variation in both space and time.
plot(lsNDVI)
Lots of errors because of NA’s! It looks like some scenes are empty, while others are dominated by cloud (we applied a cloud masking algorithm in GEE). We’ll need to get rid of these. Note that this isn’t a problem with the MODIS data, because the MODIS data are an 8-day composite based on 2 scenes per day (i.e the choice of 16 scenes to make up one complete one). We can easily exclude the blank Landsat scenes by identifying those with NA values only (if the minimum or maximum value is NA, then there are no values other than NA).
metdat2 <- metdat2[-which(is.na(minValue(lsNDVI))),] #Drop empty scenes from metadata
lsNDVI <- lsNDVI[[-which(is.na(minValue(lsNDVI)))]] #Drop empty scenes from rasterStackNow we can look at variation through time using an animation as we did before. Note that in the previous plots the colour range varies. For the animation we need to set a fixed colour range so that a particular shade of green or yellow represents the same value across scenes. For the Landsat data these range from 0.255 to 0.555
if(!file.exists("/Users/jasper/GIT/jslingsby.github.io_local/static/img/output/lsNDVI.gif")) { # Check if the file exists
saveGIF({
for (i in 1:dim(lsNDVI)[3]) plot(lsNDVI[[i]],
main=names(lsNDVI)[i],
legend.lab="NDVI",
col=rev(terrain.colors(length(c(seq(min(minValue(lsNDVI)),max(maxValue(lsNDVI)),.01),1)))),
breaks=seq(min(minValue(lsNDVI)),max(maxValue(lsNDVI)),.01))},
movie.name = "/Users/jasper/GIT/jslingsby.github.io_local/static/img/output/lsNDVI.gif",
ani.width = 600, ani.height = 500,
interval=.25)
}
Looks like there’s generally lower NDVI in the lower righthand corner parhaps?
Let’s look at some time-series graphs. First the data wrangling, then plotting; following just about the same formula as for the MODIS data.
ldat <- as.data.frame(lsNDVI)
#ldat[1:10,1:10] #look at the data
rownames(ldat) <- paste0(1:nrow(ldat)) #dummy column labels
ldat <- as.data.frame(t(ldat)) #transpose data.frame
ldat$Date <- as.Date(rownames(ldat), format = "X%Y.%m.%d")
ldat <- melt(ldat, id=c("Date"))
colnames(ldat)<- c("Date","Pixel.Number","NDVI.Value")
#head(ldat)
NDVIplot2 <- ggplot(data = ldat, aes(x = Date, y = NDVI.Value, colour = Pixel.Number)) + geom_line() + theme(legend.position="none") #remove legend as plot space cannot accommodate it - there are >800 pixels!
NDVIplot2 
So we still pick up some nice seasonality! Note this is “zoom in” on just the last ~4 years. It is a bit messy with all >800 pixels though, so let’s select just a few.
Converting a rasterStack into a data.frame fills the dataframe with raster cell values by row from left to right and numbers them. We know there are 28 rows and 31 columns (= 868 pixels), so we can use this to sample cells of interest.
This is how the grid is numbered:
matrix(1:868, 28, 31, byrow = T)## [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10] [,11] [,12] [,13]
## [1,] 1 2 3 4 5 6 7 8 9 10 11 12 13
## [2,] 32 33 34 35 36 37 38 39 40 41 42 43 44
## [3,] 63 64 65 66 67 68 69 70 71 72 73 74 75
## [4,] 94 95 96 97 98 99 100 101 102 103 104 105 106
## [5,] 125 126 127 128 129 130 131 132 133 134 135 136 137
## [6,] 156 157 158 159 160 161 162 163 164 165 166 167 168
## [7,] 187 188 189 190 191 192 193 194 195 196 197 198 199
## [8,] 218 219 220 221 222 223 224 225 226 227 228 229 230
## [9,] 249 250 251 252 253 254 255 256 257 258 259 260 261
## [10,] 280 281 282 283 284 285 286 287 288 289 290 291 292
## [11,] 311 312 313 314 315 316 317 318 319 320 321 322 323
## [12,] 342 343 344 345 346 347 348 349 350 351 352 353 354
## [13,] 373 374 375 376 377 378 379 380 381 382 383 384 385
## [14,] 404 405 406 407 408 409 410 411 412 413 414 415 416
## [15,] 435 436 437 438 439 440 441 442 443 444 445 446 447
## [16,] 466 467 468 469 470 471 472 473 474 475 476 477 478
## [17,] 497 498 499 500 501 502 503 504 505 506 507 508 509
## [18,] 528 529 530 531 532 533 534 535 536 537 538 539 540
## [19,] 559 560 561 562 563 564 565 566 567 568 569 570 571
## [20,] 590 591 592 593 594 595 596 597 598 599 600 601 602
## [21,] 621 622 623 624 625 626 627 628 629 630 631 632 633
## [22,] 652 653 654 655 656 657 658 659 660 661 662 663 664
## [23,] 683 684 685 686 687 688 689 690 691 692 693 694 695
## [24,] 714 715 716 717 718 719 720 721 722 723 724 725 726
## [25,] 745 746 747 748 749 750 751 752 753 754 755 756 757
## [26,] 776 777 778 779 780 781 782 783 784 785 786 787 788
## [27,] 807 808 809 810 811 812 813 814 815 816 817 818 819
## [28,] 838 839 840 841 842 843 844 845 846 847 848 849 850
## [,14] [,15] [,16] [,17] [,18] [,19] [,20] [,21] [,22] [,23] [,24]
## [1,] 14 15 16 17 18 19 20 21 22 23 24
## [2,] 45 46 47 48 49 50 51 52 53 54 55
## [3,] 76 77 78 79 80 81 82 83 84 85 86
## [4,] 107 108 109 110 111 112 113 114 115 116 117
## [5,] 138 139 140 141 142 143 144 145 146 147 148
## [6,] 169 170 171 172 173 174 175 176 177 178 179
## [7,] 200 201 202 203 204 205 206 207 208 209 210
## [8,] 231 232 233 234 235 236 237 238 239 240 241
## [9,] 262 263 264 265 266 267 268 269 270 271 272
## [10,] 293 294 295 296 297 298 299 300 301 302 303
## [11,] 324 325 326 327 328 329 330 331 332 333 334
## [12,] 355 356 357 358 359 360 361 362 363 364 365
## [13,] 386 387 388 389 390 391 392 393 394 395 396
## [14,] 417 418 419 420 421 422 423 424 425 426 427
## [15,] 448 449 450 451 452 453 454 455 456 457 458
## [16,] 479 480 481 482 483 484 485 486 487 488 489
## [17,] 510 511 512 513 514 515 516 517 518 519 520
## [18,] 541 542 543 544 545 546 547 548 549 550 551
## [19,] 572 573 574 575 576 577 578 579 580 581 582
## [20,] 603 604 605 606 607 608 609 610 611 612 613
## [21,] 634 635 636 637 638 639 640 641 642 643 644
## [22,] 665 666 667 668 669 670 671 672 673 674 675
## [23,] 696 697 698 699 700 701 702 703 704 705 706
## [24,] 727 728 729 730 731 732 733 734 735 736 737
## [25,] 758 759 760 761 762 763 764 765 766 767 768
## [26,] 789 790 791 792 793 794 795 796 797 798 799
## [27,] 820 821 822 823 824 825 826 827 828 829 830
## [28,] 851 852 853 854 855 856 857 858 859 860 861
## [,25] [,26] [,27] [,28] [,29] [,30] [,31]
## [1,] 25 26 27 28 29 30 31
## [2,] 56 57 58 59 60 61 62
## [3,] 87 88 89 90 91 92 93
## [4,] 118 119 120 121 122 123 124
## [5,] 149 150 151 152 153 154 155
## [6,] 180 181 182 183 184 185 186
## [7,] 211 212 213 214 215 216 217
## [8,] 242 243 244 245 246 247 248
## [9,] 273 274 275 276 277 278 279
## [10,] 304 305 306 307 308 309 310
## [11,] 335 336 337 338 339 340 341
## [12,] 366 367 368 369 370 371 372
## [13,] 397 398 399 400 401 402 403
## [14,] 428 429 430 431 432 433 434
## [15,] 459 460 461 462 463 464 465
## [16,] 490 491 492 493 494 495 496
## [17,] 521 522 523 524 525 526 527
## [18,] 552 553 554 555 556 557 558
## [19,] 583 584 585 586 587 588 589
## [20,] 614 615 616 617 618 619 620
## [21,] 645 646 647 648 649 650 651
## [22,] 676 677 678 679 680 681 682
## [23,] 707 708 709 710 711 712 713
## [24,] 738 739 740 741 742 743 744
## [25,] 769 770 771 772 773 774 775
## [26,] 800 801 802 803 804 805 806
## [27,] 831 832 833 834 835 836 837
## [28,] 862 863 864 865 866 867 868So let’s take a few from the bottom right (where the NDVI is generally low):
lowc <- c(614, 617, 707, 710)
highc <- c(128, 131, 221, 224)
lowdat <- subset(ldat, Pixel.Number%in%lowc)
highdat <- subset(ldat, Pixel.Number%in%highc)
NDVIplot_low <- ggplot(data = lowdat, aes(x = Date, y = NDVI.Value, colour = Pixel.Number)) + geom_line() #+ theme(legend.position="none") #remove legend as plot space cannot accommodate it - there are >800 pixels!
NDVIplot_low 
And a few from the top left (where the NDVI is generally high):
NDVIplot_high <- ggplot(data = highdat, aes(x = Date, y = NDVI.Value, colour = Pixel.Number)) + geom_line() #+ theme(legend.position="none") #remove legend as plot space cannot accommodate it - there are >800 pixels!
NDVIplot_high
