library(raster)
r <- raster('glc2000_v1_1') # http://forobs.jrc.ec.europa.eu/products/glc2000/products/glc2000_v1_1_Grid.zip
extent(r)
# class : Extent
# xmin : -180.0045
# xmax : 179.9955
# ymin : -56.01339
# ymax : 89.99554
ext <- extent(-69,-63,-3,3)
r1 <- crop(r,ext)
#Warning message:
#In .getRat(x, ratvalues, ratnames, rattypes) : NAs introduced by coercion
If I try to crop a smaller area, it works fine.
ext <- extent(-68,-64,-2,2)
r1 <- crop(r,ext) # works fine
This error is preventing me from saving the file with writeRaster, but I have no idea of what's going on.
Another user found the problem in a related question. The RAT (Raster Attribute Table) is being corrupted, depending on which part of the original raster is cropped. Still no idea why this happens.
> ext <- extent(-68,-64,-2,2) # The RAT is copied from the original
> r1 <- crop(r,ext)
> levels(r1)
[[1]]
ID COUNT CLASSNAMES
1 1 12875179 Tree Cover, broadleaved, evergreen
2 2 8688097 Tree Cover, broadleaved, deciduous, closed
3 3 4099003 Tree Cover, broadleaved, deciduous, open
4 4 15080165 Tree Cover, needle-leaved, evergreen
5 5 8054159 Tree Cover, needle-leaved, deciduous
6 6 5606446 Tree Cover, mixed leaf type
7 7 579763 Tree Cover, regularly flooded, fresh water
8 8 115705 Tree Cover, regularly flooded, saline water
9 9 4269938 Mosaic: Tree Cover / Other natural vegetation
10 10 587270 Tree Cover, burnt
11 11 3195387 Shrub Cover, closed-open, evergreen
12 12 15605651 Shrub Cover, closed-open, deciduous
13 13 17560702 Herbaceous Cover, closed-open
14 14 23573022 Sparse herbaceous or sparse shrub cover
15 15 3089962 Regularly flooded shrub and/or herbaceous cover
16 16 21692769 Cultivated and managed areas
17 17 4025653 Mosaic: Cropland / Tree Cover / Other natural vegetation
18 18 3921904 Mosaic: Cropland / Shrub and/or grass cover
19 19 24629888 Bare Areas
20 20 471034157 Water Bodies
21 21 10660085 Snow and Ice
22 22 378999 Artificial surfaces and associated areas
23 23 29056 No Data
> ext <- extent(-69,-63,-3,3) # The RAT is corrupted
> r1 <- crop(r,ext)
> levels(r1)
[[1]]
ID COUNT CLASSNAMES
1 1 8688097 Tree Cover, broadleaved, deciduous, open
2 2 4099003 Tree Cover, needle-leaved, evergreen
3 3 15080165 Tree Cover, needle-leaved, deciduous
4 4 8054159 Tree Cover, mixed leaf type
5 5 5606446 Tree Cover, regularly flooded, fresh water
6 6 579763 Tree Cover, regularly flooded, saline water
7 7 115705 Mosaic
8 8 4269938 Tree Cover / Other natural vegetation
9 9 587270 Tree Cover, burnt
10 10 3195387 Shrub Cover, closed-open, evergreen
11 11 15605651 Shrub Cover, closed-open, deciduous
12 12 17560702 Herbaceous Cover, closed-open
13 13 23573022 Sparse herbaceous or sparse shrub cover
14 14 3089962 Regularly flooded shrub and/or herbaceous cover
15 15 21692769 Cultivated and managed areas
16 16 4025653 Mosaic
17 17 3921904 Cropland / Tree Cover / Other natural vegetation
18 18 24629888 Mosaic
19 19 471034157 Cropland / Shrub and/or grass cover
20 20 10660085 Bare Areas
21 21 378999 Water Bodies
22 22 29056 Snow and Ice
23 23 NA Artificial surfaces and associated areas
24 12875179 NA No Data
The error message indicated something wrong happening with the Raster Attribute Table. The problem is related with special characters on the labels.
In particular, it appears that semicolumns (:) causes somehow a "splitting" of the class names in the cropped raster. Therefore, as soon as both
"Mosaic: Tree Cover / Other natural vegetation"
and
"Mosaic: Cropland / Tree Cover / Other natural vegetation"
pixels were included in the cropped extent, the RAT got corrupted due to multiple Mosaic labels, preventing correct saving of the file.
"Tidying" the class names using for example:
levels(r)[[1]]$CLASSNAMES <- stringr::str_replace(levels(r)[[1]]$CLASSNAMES , ":", "-")
solves the issue.
来源：https://stackoverflow.com/questions/46832976/why-is-crop-sometimes-introducing-nas-on-a-categorical-raster