adm - Abundance-based species distribution models

---

Overview

This package aims to support the construction of Abundance-based species distribution models, including data preparation, model fitting, prediction, and model exploration. The package offers several modeling approaches (i.e., algorithms) that users can fine-tune and customize. Models can be predicted in geographic space and explored regarding performance and response curves. Because modeling workflows in adm are constructed based on a combination of distinct functions and simple outputs, adm can be easily integrated into other packages.

Structure of adm

adm functions are grouped in three categories: modeling, post-modeling, and miscellaneous tools

i) modeling

Functions to tune, fit, and validate models with nine different algorithms, with a suite of possible model-specific hyperparameters

Fit and validate models without hyperparameters tuning

• fit_abund_cnn() Fit and validate Convolutional Neural Network Model

• fit_abund_dnn() Fit and validate Deep Neural Network model

• fit_abund_gam() Fit and validate Generalized Additive Models

• fit_abund_gbm() Fit and validate Generalized Boosted Regression models

• fit_abund_glm() Fit and validate Generalized Linear Models

• fit_abund_net() Fit and validate Artificial Neural Network models

• fit_abund_raf() Fit and validate Random Forests models

• fit_abund_svm() Fit and validate Support Vector Machine models

• fit_abund_xgb() Fit and validate Extreme Gradient Boosting models

Fit and validate models with hyperparameters tuning

• tune_abund_cnn() Fit and validate Convolutional Neural Network with exploration of hyper-parameters that optimize performance

• tune_abund_dnn() Fit and validate Deep Neural Network model with exploration of hyper-parameters that optimize performance

• tune_abund_gam() Fit and validate Generalized Additive Models with exploration of hyper-parameters that optimize performance

• tune_abund_gbm() Fit and validate Generalized Boosted Regression models with exploration of hyper-parameters that optimize performance

• tune_abund_glm() Fit and validate Generalized Linear Models with exploration of hyper-parameters that optimize performance

• tune_abund_net() Fit and validate Shallow Neural Networks models with exploration of hyper-parameters that optimize performance

• tune_abund_raf() Fit and validate Random Forest models with exploration of hyperparameters that optimize performance

• tune_abund_svm() Fit and validate Support Vector Machine models with exploration of hyper-parameters that optimize performance

• tune_abund_xgb() Fit and validate Extreme Gradient Boosting models with exploration of hyper-parameters that optimize performance

Modeling evaluation

• adm_eval() Calculate different model performance metrics

ii) post-modeling

Functions to predict abundance across space and construct partial dependence plots to explore the relationships between abundance and environmental predictors

• adm_predict() Spatial predictions from individual and ensemble models

• p_abund_bpdp() Bivariate partial dependence plots for abundance-based distribution models

• p_abund_pdp() Partial dependent plots for abundance-based distribution models

• data_abund_bpdp() Calculate data to construct bivariate partial dependence plots

• data_abund_pdp() Calculate data to construct partial dependence plots

iii) miscellaneous tools

Extra functions to support the modeling workflow, including data handling, transformations, and hyperparameter selection.

• adm_extract() Extract values from a spatial raster based on x and y coordinates

• adm_summarize() Merge model performance tables

• adm_transform() Performs data transformation on a variable based on the specified method.

• balance_dataset() Balance database at a given absence-presence ratio

• cnn_make_samples() Creates sample data for Convolutional Neural Network

• croppin_hood() Crop rasters around a point (for Convolutional Neural Networks)

• family_selector() Select probability distributions for GAM and GLM

• generate_arch_list() Generate architecture list for Deep Neural Network and Convolutional Neural Network

• generate_cnn_architecture() Generate architectures for Convolutional Neural Network

• generate_dnn_architecture() Generate architectures for Deep Neural Network

• model_selection() Best hyper-parameters selection

• res_calculate() Calculate the output resolution of a layer

• select_arch_list() Select architectures for Convolutional Neural Network or Deep Neural Network

Installation

You can install the development version of adm from github

# For Windows and Mac OS operating systems
remotes::install_github("sjevelazco/adm")

Package website

See the package website (https://sjevelazco.github.io/adm/) for functions explanation and vignettes.

Package citation

de Oliveira Junior A.C., Velazco S.J.E. (2025). adm: an R package for constructing abundance-based species distribution models. Methods in Ecology and Evolution. https://doi.org/10.1111/2041-210X.70074