In boyanangelov/sdmbench: Benchmark Species Distribution Models

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "README-"
)

sdmbench

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.

Installation

# 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.

Examples

Here are several examples of what you can do with sdmbench. Downloading and prepare benchmarking data:

library(sdmbench)

benchmarking_data <- get_benchmarking_data("Loxodonta africana", limit = 1200, climate_resolution = 10)
head(benchmarking_data$df_data)

Preparing 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

Plot 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")

Using custom data

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 moment 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:

Vignette

A thorough introduction to the package is available as a vignette in the package, and online.

# open vignette
vignette("sdmbench")

Tests

The package tests are in the tests directory, and can be run by using devtools::test or Ctrl/Cmd + Shift + T within RStudio.

Contributors

Contributions are welcome and guidelines are stated in CONTRIBUTING.md.

License

MIT (LICENSE.md)

References

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).

boyanangelov/sdmbench documentation built on Dec. 14, 2020, 1:08 a.m.