Differences between raster::merge and terra::merge & NA handling

[vvirkki]

Hello,

I am composing global rasters from small parts using ideally merge. I am running terra 1.0.10.

I've noticed some difference between raster and terra: in raster::merge, the parameter overlap controls how NA values should be handled and this seems not to be available in terra::merge. See below for a reprex adapted from terra manual.

There are two issues in total:

• terra::merge lacks support for discarding NA values
• when merging, terra::merge orders the rasters from bottom to top in left-to-right order whereas raster::merge does it right-to-left. In my opinion, terra version is more logical (first raster left is the bottom raster) but a note about this could be added to documentation.

Thanks for superb development with terra; it's incredibly fast and versatile!

# create data
x <- rast(xmin=-110, xmax=-50, ymin=40, ymax=70, ncols=60, nrows=30)
y <- rast(xmin=-80, xmax=-20, ymax=60, ymin=30)
res(y) <- res(x)

# insert values and some NA
vx <- 1:ncell(x)
vx[floor(runif(100, min = 1600, max = 1800))] <- NA
vy <- 1:ncell(y)
vy[floor(runif(100, min = 0, max = 200))] <- NA
values(x) <- vx
values(y) <- vy

plot(c(x,y))

# do merge with raster - discards NA and sets last non-NA value as result by default
raster_merge <- raster::merge(raster::raster(x), raster::raster(y))
plot(raster_merge)

# do merge with terra - y value overrides x value even though y would be NA
terra_merge <- terra::merge(x, y)
plot(terra_merge)

# terra merge behaviour corresponds to raster::merge with overlap = FALSE
rm_no_overlap <- raster::merge(raster::raster(x), raster::raster(y), overlap = FALSE)
plot(rm_no_overlap)

# flipping y and x provides exact same result
rm_no_overlap <- raster::merge(raster::raster(y), raster::raster(x), overlap = FALSE)
plot(rm_no_overlap)

[Rapsodia86]

Hi,
I agree that option to handle NA would be appreciated.
However, I have just explored your example and the expected result can be achieved using terra::mosaic with fun=max.
Of course, I am aware that using other function instead is not a solution.
terra_mosaic <- terra::mosaic(x, y,fun="max")
plot(terra_mosaic)

[rhijmans]

Thank you for reporting this bug.

terra::merge lacks support for discarding NA values

This has already been fixed in the development version available here (currently version 1.1.2)

terra::merge orders the rasters from bottom to top in left-to-right order (raster used the opposite order) ... a note about this could be added to documentation.

terra now behaves like raster --- the idea is the earlier layers are "more important", but I understand why you think the opposite could be more logical. I will keep it as it is now (like raster). The documentation already states that "If objects overlap, the values get priority in the same order as the arguments" .

terra::mosaic

Indeed a good work-around in the CRAN version

[vvirkki]

Thanks for providing the workaround and addressing the comments! I ended up doing my task via data frames, after all, but good to know that the merge/mosaic workflows would be as applicable.

[HassanMasoomi]

I have large number of raster files (with large sizes). The merge function in terra fails when using the code below (with this error: "Error in rc@ptr$merge(opt) : std::bad_alloc"), but it works with the same function in raster package (although very slow).

x <- list(r1, r2)
# add arguments such as filename
x$filename <- 'test.tif'
m <- do.call(merge, x)

[kadyb]

@HassanMasoomi, have you tried using SpatRasterCollection sprc()?

ls <- list(r1, r2)
ras_coll <- sprc(ls)
m <- merge(ras_coll)

Or vrt() on files and then writeRaster()?

[HassanMasoomi]

@kadyb, yes, I did try SpatRasterCollection and it had the same issue. But, haven't tried vrt().