Get started

Design

The core of shar are functions to simulate null model data by randomizing either the environmental data (i.e., raster) or the locations of species (i.e., point pattern). The null model data is then used to analyse if significant species-habitat associations are present. Additionally, functions to visualize and analyse the results are available as well as some utility functions. The methods are mainly described in Harms et al. (2001), Plotkin et al. (2000), and Wiegand & Moloney (2014). The methods are not necessary complementary, but are rather different approaches to achieve the same result.

Preprocessing of input data

All functions are designed for discrete habitat classes. Thus, if continuous data is available, this has to be classified to discrete classes first. classify_habitats provides several ways to classify the data, such as the Fisher-Jenks algorithm (Fisher 1958, Jenks & Caspall 1971).

landscape_discrete <- classify_habitats(raster = terra::rast(landscape), n = 5, style = "fisher")

Randomize environmental data

There are two functions to randomize the environmental data: translate_raster() and randomize_raster(). The first function is a torus translation of the raster, shifting the habitat map in all four cardinal directions. This is only possible for rectangular observation areas and results in R (terra::nrow(terra::rast(landscape)) + 1) * (terra::ncol(terra::rast(landscape)) + 1) - 4 randomized raster (based on number of rows and cols). The other function randomizes the environmental data using a random-walk algorithm. Here, the number of randomized raster can be specified using the n_random argument.

All methods require “fully mapped data” in a sense that NA cells of the environmental data are allowed only if these cells cannot accommodate any locations of individuals (e.g., a water body within a forest area). This needs to be reflected in the observation window of the point pattern. For the torus translation method, no NA values are allowed at all.

torus_trans <- translate_raster(raster = landscape_discrete)

random_walk <- randomize_raster(raster = landscape_discrete, n_random = 39)

Randomize location data

To randomize the location data (i.e. the point pattern) either fit_point_process() or reconstruct_pattern() are available. The first fits a Poisson process or a cluster process to the data. The second functions reconstructs the spatial characteristics of the data using pattern reconstruction (Kirkpatrick et al. 1983; Tscheschel & Stoyan 2006). This is advantageous for point patterns not describable by simple point process models. For both function, the number of patterns can be specified by the n_random argument.

gamma_test <- fit_point_process(pattern = species_b, process = "cluster", n_random = 39)

# (this can takes some time)
reconstruction <- reconstruct_pattern(pattern = species_b, n_random = 39, 
                                      e_threshold = 0.05, method = "cluster")

Analyse results

The most important function to analyse results is results_habitat_association(). This function compares the observed data to the null model data and by that is able to show significant species-habitat associations. The function work for both, randomized environmental data or randomized location data.

Please be aware that due to the randomization of the null model data, results might slightly differ between different randomization approaches (e.g., fit_point_process() vs. translate_raster()), and even for repetitions of the same approach. Thus, the exact lo and hi thresholds might be slightly different when re-running the examples. However, the counts of the observed data should be identical, and general results and trends should be similar.

results_habitat_association(pattern = species_a, raster = random_walk)
#> > Input: randomized raster
#> > Quantile thresholds: negative < 0.025 || positive > 0.975
#>   habitat breaks count    lo    hi significance
#> 1       1     NA    35 11.00 30.20     positive
#> 2       2     NA    44 23.70 49.05         n.s.
#> 3       3     NA    36 21.75 49.05         n.s.
#> 4       4     NA     4 24.80 59.05     negative
#> 5       5     NA    73 51.95 77.05         n.s.

results_habitat_association(pattern = reconstruction, raster = landscape_discrete)
#> > Input: randomized pattern
#> > Quantile thresholds: negative < 0.025 || positive > 0.975
#>   habitat breaks count    lo     hi significance
#> 1       1     NA     6  5.90  26.95         n.s.
#> 2       2     NA    18 11.85  51.40         n.s.
#> 3       3     NA    18 16.95  56.25         n.s.
#> 4       4     NA    21 27.70  54.00     negative
#> 5       5     NA   129 39.75 100.10     positive

The data was created that species_a has a negative association to habitat 4 and species_b has a positive association to habitat 5, which is reflected in the results.

Given the characteristics of the method, a positive association to one habitat inevitably leads to a negative association to at least one of the other habitats (and vice versa; Yamada et al. 2006). For example, a high amount of individual points in the positively associated habitat simultaneously mean that less individual points can be present in the other habitats.

Utility functions

There is also the possibility to visualize the randomized data using the plot() function.

plot(reconstruction, ask = FALSE)

col_palette <- c("#440154FF", "#3B528BFF", "#21908CFF", "#5DC863FF", "#FDE725FF")

plot(torus_trans, col = col_palette)

There are many more functions, which can be found here.