Whole area

We can see that most of the island is covered by forest, with someagriculture and smaller areas of the other classes. It is alsoreasonably easy to describe these proportions (so-called composition)numerically - we just need to count cells of each category for the wholedata.

We can use thelsp_signature() function for this. Itrequires astars object as the first input and the type ofa signature to calculate -"composition" in this case.There are also several additional arguments, includingthreshold - a share (between 0 and 1) of NA cells to allowsignature calculation andnormalization - decision if theoutput vector should be normalized.

landcover_comp=lsp_signature(landcover,type ="composition",threshold =1,normalization ="none")landcover_comp#> # A tibble: 1 × 3#>      id na_prop signature#> * <int>   <dbl> <list>#> 1     1   0.666 <int [1 × 7]>

The output oflsp_signature(),landcover_comp, has a new classlsp. It is atibble with three columns:

• id - an id of each window (area)
• na_prop - share (0-1) ofNA cells for eachwindow
• signature - a list-column containing with calculatedsignatures

We can take a look at the last column:

landcover_comp$signature#> [[1]]#>           1       2     3    4    5     6      7#> [1,] 862001 8122776 84482 4311 2677 78555 203444

It is a list of signatures. In this case, we have just one signaturedescribing the whole area in the form of a numeric vector. It containshow many cells belong to each land cover category. For example, thereare 8122776 cells of forest, but only 2677 cells of shrubland.

Regular local landscapes

Another approach would be to divide this large area into many regularrectangles (we refer to them as local landscapes) and to calculate asignature in each of them. The previously used signature,"composition" has one important flaw though. It onlydescribes how many cells of each category we have. However, it does notdistinguish an area with the left half of forest and right half ofagriculture from an area with forest mixed with agriculture (think of agreen-yellow checkerboard). Gladly, several more types of signatures doexist. It includes a co-occurrence matrix (type = "coma")."coma" goes to each cell, looks at its value, looks at thevalues of its neighbors and counts how many neighbors of each class ourcentral cell has.

#> downsample set to 11

This time, we set thewindow argument to200, which means that each local landscape will consist of200 by 200 cells. In this example, each cell has a resolution of 300 by300 meters, therefore a local landscape will be 60000 by 60000 meters(60 by 60 kilometers).

landcover_coma=lsp_signature(landcover,type ="coma",window =200)landcover_coma#> # A tibble: 298 × 3#>       id na_prop signature#>  * <int>   <dbl> <list>#>  1     2  0.644  <int [7 × 7]>#>  2     3  0.0992 <int [7 × 7]>#>  3     4  0.145  <int [7 × 7]>#>  4     5  0.602  <int [7 × 7]>#>  5     6  0.775  <int [7 × 7]>#>  6     9  0.796  <int [7 × 7]>#>  7    38  0.255  <int [7 × 7]>#>  8    39  0      <int [7 × 7]>#>  9    40  0      <int [7 × 7]>#> 10    41  0      <int [7 × 7]>#> # ℹ 288 more rows

Now, we have one row per local landscape, where each local landscapeis described by an id (id), the proportion of cells withNAs (na_prop), and a signature(signature). For example, the first signature looks likethis:

landcover_coma$signature[[1]]#>    1     2 3 4 5 6    7#> 1 16    24 0 0 0 0   12#> 2 24 53928 0 0 0 0  255#> 3  0     0 0 0 0 0    0#> 4  0     0 0 0 0 0    0#> 5  0     0 0 0 0 0    0#> 6  0     0 0 0 0 0    0#> 7 12   255 0 0 0 0 1644

It is a matrix where each row and column represent subsequent landcover classes. For example, 141250 times forest cell is next to antherforest cell, and 226 times water cell is next to forest cell. You canlearn more about this signature athttps://jakubnowosad.com/comat/articles/coma.html.Additional signatures are described in theSpatialpatterns’ signatures vignette.

Irregular local landscapes

Themotif package also allows using irregularregions based on the user-provided polygons. It has an example spatialvector dataset,ecoregions.gpkg, which contains terrestrialecoregions for New Guinea fromhttps://ecoregions2017.appspot.com/.

ecoregions=read_sf(system.file("vector/ecoregions.gpkg",package ="motif"))

This dataset has 22 rows, where each row relates to one ecoregion.Each ecoregion is also related to a unique value in theidcolumn.

Now, we need to provide this dataset and its identity column to thewindow argument oflsp_signature().

landcover_coma_e=lsp_signature(landcover,type ="coma",window = ecoregions["id"])landcover_coma_e#> # A tibble: 22 × 3#>       id na_prop signature#>  * <int>   <dbl> <list>#>  1     1 0.114   <int [7 × 7]>#>  2     2 0.0851  <int [7 × 7]>#>  3     3 0.0377  <int [7 × 7]>#>  4     4 0.0914  <int [7 × 7]>#>  5     5 0       <int [7 × 7]>#>  6     6 0       <int [7 × 7]>#>  7     7 0.00285 <int [7 × 7]>#>  8     8 0.0720  <int [7 × 7]>#>  9     9 0.0508  <int [7 × 7]>#> 10    10 0       <int [7 × 7]>#> # ℹ 12 more rows

The output,landcover_coma_e, is also of thelsp class and contains three columns. The first column,id, has the same values as provided in thewindow argument above. The third column,signature, has a spatial signature for each polygon in theecoregions dataset.

landcover_coma_e$signature[[1]]#>     1     2 3  4 5 6    7#> 1 662   752 0  1 0 0   13#> 2 752 70974 0  6 0 0  427#> 3   0     0 0  0 0 0    0#> 4   1     6 0 12 0 0    4#> 5   0     0 0  0 0 0    0#> 6   0     0 0  0 0 0    0#> 7  13   427 0  4 0 0 1560

landcover_proxy=read_stars(system.file("raster/landcover2015.tif",package ="motif"),proxy =TRUE)