inverting values of rast object instead of subtraction

[atsyplenkov]

I have faced up with strange behaviour of the terra package. I was trying to scale DEM values to the maximum height. While the raster package gives the expected results, the terra package gives me an inverted one. The problem seems to be associated with a subtraction syntax. I solved it by multiplying the rast object by (-1). Please take a look at the example below. Is it a bug or a feature? :-)

library(raster) # CRAN v3.6-14
library(terra)  # CRAN v1.6-53

# load datasets -----------------------------------------------------------
dem <- 
  system.file("ex/elev.tif", package="terra")

dem_terra <- 
  rast(dem)

dem_raster <- 
  raster(dem)

# helper functions --------------------------------------------------------
# function to compare two rast objects
compare_rasts <- 
  function(.rast1, .rast2, .rast3){
    
    rbind(
      .get_global_stats(.rast1),
      .get_global_stats(.rast2),
      .get_global_stats(.rast3)
    )
    
  }

.get_global_stats <- 
  function(.rast){
    
    tibble::tibble(
      rast = names(.rast),
      res = res(.rast)[1],
      mean = global(.rast, "mean", na.rm = T)[1,1],
      med = global(.rast, function(x) median(x, na.rm = T))[1,1],
      sd = global(.rast, function(x) sd(x, na.rm = T))[1,1],
      max = global(.rast, function(x) max(x, na.rm = T))[1,1],
      min = global(.rast, function(x) min(x, na.rm = T))[1,1],
      na = global(.rast, function(x) sum(is.na(x)))[1,1]
    )
    
  }

# function to get maximum values
.max_terra <- 
  function(x){
    
    global(x, function(i) max(i, na.rm = T))[1,1]
    
  }

# scale process -----------------------------------------------------------
# raster approach
norm_raster <- 1 - (dem_raster / (maxValue(dem_raster)))
names(norm_raster) <- "raster"
norm_raster <- rast(norm_raster)

# terra straightforward approach
norm_terra_inverted <- 1 - (dem_terra / (.max_terra(dem_terra)))
names(norm_terra_inverted) <- "terra inverted"

# terra solution...
norm_terra <- 1 + (-1) * (dem_terra / (.max_terra(dem_terra)))
names(norm_terra) <- "terra"

# compare -----------------------------------------------------------------
compare_rasts(norm_terra_inverted, norm_raster, norm_terra)

# A tibble: 3 × 8
  rast               res   mean    med    sd   max    min    na
  <chr>            <dbl>  <dbl>  <dbl> <dbl> <dbl>  <dbl> <int>
1 terra inverted 0.00833 -0.363 -0.391 0.147 0     -0.742  3942
2 raster         0.00833  0.363  0.391 0.147 0.742  0      3942
3 terra          0.00833  0.363  0.391 0.147 0.742  0      3942

[brownag]

I think this is a bug and was fixed in #978

[rhijmans]

With the current CRAN version I see no differences.

library(raster)
#Loading required package: sp
library(terra)
#terra 1.7.3

f <- system.file("ex/elev.tif", package="terra")
dem_terra <- rast(f)
dem_raster <- raster(f)

global_stats <- function(x){
    data.frame(
      rast = names(x),
      res = res(x)[1],
      mean = global(x, "mean", na.rm = T)[1,1],
      med = global(x, function(x) median(x, na.rm = T))[1,1],
      sd = global(x, function(x) sd(x, na.rm = T))[1,1],
      max = global(x, function(x) max(x, na.rm = T))[1,1],
      min = global(x, function(x) min(x, na.rm = T))[1,1],
      na = global(x, function(x) sum(is.na(x)))[1,1]
    )
}

compare_rasts <- function(x, y, z){   
    rbind(
      global_stats(x),
      global_stats(y),
      global_stats(z)
    )
}

norm_raster <- 1 - (dem_raster / (maxValue(dem_raster)))
names(norm_raster) <- "raster"
norm_raster <- rast(norm_raster)

norm_terra_inverted <- 1 - (dem_terra / ( minmax(dem_terra)[2]))
names(norm_terra_inverted) <- "terra inverted"

norm_terra <- 1 + (-1) * (dem_terra / (minmax(dem_terra)[2]))
names(norm_terra) <- "terra"

compare_rasts(norm_terra_inverted, norm_raster, norm_terra)

#            rast         res      mean       med        sd       max min   na
#1 terra inverted 0.008333333 0.3631872 0.3912249 0.1466524 0.7422303   0 3942
#2         raster 0.008333333 0.3631872 0.3912249 0.1466524 0.7422303   0 3942
#3          terra 0.008333333 0.3631872 0.3912249 0.1466524 0.7422303   0 3942

The version you are using was the CRAN version for about a week; unfortunately you got that one.