Skip to contents

Summary metrics on a pair of rasters

Source: R/cmp_funs.R
cmp_funs.Rd

Summary functions that return a scalar when passed with two portions of two rasters. Typically passed to CMP.

Usage

dist_euclidean(x, y, ...)

dist_manhattan(x, y, ...)

dist_chebyshev(x, y, ...)

dist_bhat(x, y, bins = 10, ...)

dist_identical(x, y, ...)

dist_different(x, y, ...)

dist_diff(x, y, ...)

cor_pearson(x, y, ...)

cor_pearson_uniform_equals_NA(x, y, ...)

kappa_index(x, y, ...)

diff_rmse(x, y, ...)

diff_mean(x, y, ...)

diff_median(x, y, ...)

diff_var(x, y, ...)

diff_cv(x, y, ...)

p_student(x, y, min_nb = 5, ...)

p_wilcoxon(x, y, min_nb = 5, ...)

kappa_index_cppr(x, y, ...)

dist_euclidean_cppr(x, y, ...)

dist_manhattan_cppr(x, y, ...)

dist_chebyshev_cppr(x, y, ...)

diff_rmse_cppr(x, y, ...)

Arguments

x

a fraction of a SpatRaster raster

y

a fraction of a SpatRaster raster

...

additional arguments

bins

number of bins to use

min_nb

minimal number of observation to calculate a value, otherwise a NA_real is returned

Value

numeric the index of interest

Details

First of all, you can use built-in function, as long as they return a single value such as cor.

  • cor_pearson is pearson correlation coefficient. Note than for uniform window(s) this will return 1, not NA. See below.

  • cor_pearson_uniform_equals_NA is same as above but return NA when sd=0 for one or two windows having uniform values, which is likely to happen for small sample sizes.

  • kappa_index/kappa_index_cppr is Cohen's Kappa

  • dist_identical proportion of identical value, pixel wise

  • dist_different proportion of different value, pixel wise

  • dist_diff (signed) difference, pixel wise

  • dist_euclidean/dist_euclidean_cppr euclidean distance (sqrt of sums of squared diff pixel-wise, divided by the number of valid pixels)

  • dist_manhattan/dist_manhattan_cppr manhattan distance (sum of absolute distances pixel-wise, , divided by the number of valid pixels)

  • dist_chebyshev/dist_chebyshev_cppr Chebyshev's distance (max of absolute distances pixel-wise)

  • dist_bhat Bhattacharyya's distance based on distribution distances calculated using nbins

  • diff_rmse/diff_rmse_cppr root-mean square error pixel-wise

  • diff_mean difference of mean values

  • diff_mean difference of mean values

  • diff_var difference of variance values

  • diff_cv difference of coefficient of variation values

  • p_student p value from a Student t test

  • p_wilcoxon p value from a Wilcoxon rank sum test

Naturally, you can also use built-in function, as long as they return a single value such as: cor,

See also

CMP, mhm_funs

Examples

set.seed(2329) # for reproducibility
x <- sample(1:3, size=49, replace=TRUE, prob=c(0.1, 0.1, 1))  # unbalanced vector
y <- sample(1:3, size=49, replace=TRUE, prob=c(0.1, 0.1, 1))  # another one

# examples runs mostly to test and exemplify
# but in CMP they will run on each pair of focal windows

# built in functions
cor(x, y)
#> [1] -0.2023918
cov(x, y)
#> [1] -0.08418367

# kappa agreement
kappa_index(x, y)
#> [1] -0.0864745
kappa_index_cppr(x, y)
#> [1] -0.0864745

# distance indices
dist_euclidean(x, y)
#> [1] 0.1443075
dist_euclidean_cppr(x, y)
#> [1] 0.1443075

dist_manhattan(x, y)
#> [1] 0.6122449
dist_manhattan_cppr(x, y)
#> [1] 0.6122449

dist_chebyshev(x, y)
#> [1] 2
dist_chebyshev_cppr(x, y)
#> [1] 2

dist_bhat(x, y)
#> [1] 0.01055561

# difference indices
diff_rmse(x, y)
#> [1] 1.010153
diff_rmse_cppr(x, y)
#> [1] 1.010153

diff_mean(x, y)
#> [1] 0.122449
diff_median(x, y)
#> [1] 0
diff_var(x, y)
#> [1] -0.210034
diff_cv(x, y)
#> [1] -0.07160757

# p value from Student t test
p_student(x, y)
#> [1] 0.3572371
p_wilcoxon(x, y)
#> [1] 0.5229461