Species Distribution Modeling (SDM) is a field of increasing importance
in ecology1. Several popular applications of
SDMs are understanding climate change effects on
species2, natural reserve
planning3 and invasive species
monitoring4. The sdmbench package solves
several issues related to the development and evaluation of those models
by providing a consistent benchmarking workflow:
- consistent species occurence data acquisition and preprocessing
- consistent environmental data acquisition and preprocessing (both current data and future projections)
- consistent spatial data partitioning
- integration of a wide variety of machine learning models
- graphical user interface for non/semi-technical users
The end result of a sdmbench SDM analysis is to determine the model -
data processing combination that results in the highest predictive power
for the species of interest. Such an analysis is useful to researchers
who want to avoid issues of model selection and evaluation, and want to
rapidly test prototypes of species distribution models.
# add build_vignettes = TRUE if you want the package vignette
devtools::install_github("boyanangelov/sdmbench")There are several additional packages you need to install if you want to
access the complete sdmbench functionality. First Tensorflow. You can
use the keras package to install that (it is installed by the previous
command). Note that this step requires a working Python installation on
your system. Most modern operating systems have Python pre-installed,
but if you are not sure you can check the official
website.
# consult the keras documentation if you want GPU support
keras::install_keras(tensorflow = "default")Additionally you will need MaxEnt. Installation instructions are available here. Note that this requires Java which you can get get from here.
Here are several examples of what you can do with sdmbench.
Downloading and prepare benchmarking data:
library(sdmbench)
#> sdmbench: Tools for benchmarking Species Distribution Models
#> ============================================================
#> For more information visit http://github.com/
#> To start the GUI: run_sdmbench()
benchmarking_data <- get_benchmarking_data("Loxodonta africana", limit = 1200, climate_resolution = 10)
#> [1] "Getting benchmarking data...."
#> [1] "Cleaning benchmarking data...."
#> Assuming 'decimalLatitude' is latitude
#> Assuming 'decimalLongitude' is longitude
#> Assuming 'latitude' is latitude
#> Assuming 'longitude' is longitude
#> Assuming 'latitude' is latitude
#> Assuming 'longitude' is longitude
#> [1] "Done!"
head(benchmarking_data$df_data)
#> bio1 bio2 bio3 bio4 bio5 bio6 bio7 bio8 bio9 bio10 bio11 bio12 bio13
#> 1 181 132 59 3078 283 60 223 187 141 220 141 425 48
#> 2 180 142 58 3336 292 51 241 210 136 222 136 410 50
#> 3 208 115 86 259 275 142 133 208 207 211 206 956 168
#> 4 180 142 58 3336 292 51 241 210 136 222 136 410 50
#> 5 215 129 56 3244 311 84 227 251 168 251 168 673 120
#> 6 236 167 87 691 335 145 190 233 228 244 228 505 133
#> bio14 bio15 bio16 bio17 bio18 bio19 label
#> 1 25 19 122 87 114 87 1
#> 2 21 24 122 73 121 73 1
#> 3 5 58 382 42 178 112 1
#> 4 21 24 122 73 121 73 1
#> 5 10 70 334 34 334 34 1
#> 6 4 96 236 14 189 14 1Preparing data for benchmarking (i.e. add a spatial partitioning method):
data("wrld_simpl", package = "maptools")
benchmarking_data$df_data <- partition_data(dataset_raster = benchmarking_data$raster_data,
dataset = benchmarking_data$df_data,
env = benchmarking_data$raster_data$climate_variables,
method = "block")
learners <- list(mlr::makeLearner("classif.randomForest", predict.type = "prob"),
mlr::makeLearner("classif.logreg", predict.type = "prob"),
mlr::makeLearner("classif.rpart", predict.type = "prob"),
mlr::makeLearner("classif.ksvm", predict.type = "prob"))
benchmarking_data$df_data <- na.omit(benchmarking_data$df_data)Benchmarking machine learning models on parsed species occurence data:
bmr <- benchmark_sdm(benchmarking_data$df_data,
learners = learners,
dataset_type = "block",
sample = FALSE)
best_results <- get_best_model_results(bmr)
best_results
#> # A tibble: 4 x 3
#> # Groups: learner.id [4]
#> learner.id iter auc
#> <fct> <int> <dbl>
#> 1 classif.randomForest 1 0.992
#> 2 classif.logreg 2 0.915
#> 3 classif.rpart 3 0.900
#> 4 classif.ksvm 4 0.963Plot best model results:
bmr_models <- mlr::getBMRModels(bmr)
plot_sdm_map(raster_data = benchmarking_data$raster_data,
bmr_models = bmr_models,
model_id = best_results$learner.id[1],
model_iteration = best_results$iter[1],
map_type = "static") +
raster::plot(wrld_simpl,
add = TRUE,
border = "darkgrey")
#> integer(0)
If you are interested in bringing your own data, rather than using GBIF, you can toggle the checkmark in the sidebar and upload it. The required format is as follows:
| bio1 | bio2 | bio3 | bio4 | bio […] | bio 19 | label |
|---|---|---|---|---|---|---|
where label is 0/1. At the monent custom data is supported only for
General Models, and has no mapping capability.
A good starting point to discover the full package functionality is to
start the GUI with run_sdmbench(). Here are some screenshots:
A thorough introduction to the package is available as a vignette in the package, and online.
# open vignette
vignette("sdmbench")The package tests are in the tests directory, and can be run by using
devtools::test or Ctrl/Cmd + Shift + T within RStudio.
Contributions are welcome and guidelines are stated in
CONTRIBUTING.md.
MIT (LICENSE.md)
1. Elith, J. & Leathwick, J. R. Species Distribution Models: Ecological Explanation and Prediction Across Space and Time. Annu. Rev. Ecol. Evol. Syst. 40, 677–697 (2009).
2. Austin, M. P. & Van Niel, K. P. Improving species distribution models for climate change studies: Variable selection and scale. J. Biogeogr. 38, 1–8 (2011).
3. Guisan, A. et al. Predicting species distributions for conservation decisions. Ecol. Lett. 16, 1424–1435 (2013).
4. Descombes, P. et al. Monitoring and distribution modelling of invasive species along riverine habitats at very high resolution. Biol. Invasions 18, 3665–3679 (2016).