Using resample vs. aggregate & extend in R to have rasters of matching resolution and extent

Question

I have two rasters of different resolution and extent:

> res(Elevation)
[1] 0.002083333 0.002083333

> res(Ann_precip)
[1] 0.008333333 0.008333333 

> extent(Elevation)
class       : Extent 
xmin        : -15.07722 
xmax        : -7.641806 
ymin        : 7.193611 
ymax        : 12.67694 

> extent(Ann_precip)
class       : Extent 
xmin        : -15.075 
xmax        : -7.641667 
ymin        : 7.191667 
ymax        : 12.675

My question is, in order for these two rasters to have matching resolutions and extents, is it better to:

A) use the raster::aggregate function

> 0.008333333/0.002083333
[1] 4

Elevation_res<-aggregate(Elevation, fact=4, fun=mean)

and the raster::extend function

Elevation_res<-extend(Elevation_res, Ann_precip, values=NA)

(although here I still get different but very similar extents and resolutions):

> res(Elevation_res)
[1] 0.008333333 0.008333333

> res(Ann_precip)
[1] 0.008333333 0.008333333

> res(Elevation_res)==res(Ann_precip)
[1] FALSE FALSE

> extent(Elevation_res)
class       : Extent 
xmin        : -15.07722 
xmax        : -7.635556 
ymin        : 7.193611 
ymax        : 12.67694 

> extent(Ann_precip)
class       : Extent 
xmin        : -15.075 
xmax        : -7.641667 
ymin        : 7.191667 
ymax        : 12.675 

or

b) use the raster::resample function

Elevation_res<-resample(Elevation, Ann_precip, method="bilinear")

> res(Elevation_res)==res(Ann_precip)
[1] TRUE TRUE

> extent(Elevation_res)==extent(Ann_precip)
[1] TRUE

I'm asking this because I've read in Wegmann et al (2016) (p110) (if I understand correctly) that resampling greatly affects pixel values, and that aggregate(),extend() and crop() should be used instead. Since differences in resolution and extent are quite small in my case, can I assume that bias created by resampling would be minimal here?

Answer 1

Check resample function of raster package. When resample is used with 'bilinear method, the output is the same one than aggregate:

if (!skipaggregate) {
    rres <- res(y) / res(x)
    resdif <- max(rres)
    if (resdif > 2) {
        ag <- pmax(1, floor(rres-1))
        if (max(ag) > 1) {
            if (method == 'bilinear') {
                x <- aggregate(x, ag, 'mean')
            } else {  
                x <- aggregate(x, ag, modal)
            }
        }
    }

With an example:

library(raster)

r <- raster(nrow=4,ncol=8)

r2 <- raster(nrow=2,ncol=4)

r <- setValues(r,values = 1:32)

r_agg <- aggregate(r,fact=2,fun=mean)

r_resam <- resample(r,r2,method='bilinear')

values(r_resam) == values(r_agg)
## [1] TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE

values(r_resam)
## [1]  5.5  7.5  9.5 11.5 21.5 23.5 25.5 27.5

But if you use 'ngb' as method, the result is different (method depends of your data, if is categorical you must use 'ngb'):

r_resam2 <- resample(r,r2,method='ngb')

values(r_resam2)
## [1] 10 12 14 16 26 28 30 32

And extend doesn't change resolution, only extent:

r
## class       : RasterLayer 
## dimensions  : 4, 8, 32  (nrow, ncol, ncell)
## resolution  : 45, 45  (x, y)
## extent      : -180, 180, -90, 90  (xmin, xmax, ymin, ymax)
## coord. ref. : +proj=longlat +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0 

r_ext <- extend(r,r2,values=NA)

r_ext
## class       : RasterLayer 
## dimensions  : 4, 8, 32  (nrow, ncol, ncell)
## resolution  : 45, 45  (x, y)
## extent      : -180, 180, -90, 90  (xmin, xmax, ymin, ymax)
## coord. ref. : +proj=longlat +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0

And crop() as extend(), will no change resolution.