Tools for the assessment of the cumulative impacts of anthropogenic features in ecological studies • oneimpact

oneimpact provides tools for the assessment of cumulative impacts of multiple infrastructure and land use modifications in ecological studies. This includes tools to calculate the zone of influence (ZOI, also called scale of effect in landscape ecology) of anthropogenic variables as well as tools for model fitting, estimation of the effect size and ZOI, making predictions and interpreting fitted relationships, computing model performance and plausibility, and ancillary functions. The functions dealing with spatial data processing can be run in both R and GRASS GIS, using R as an interface.

Context: Why zone of influence and cumulative impacts?

Traditional ecological impact assessments typically focus on the effects of distance to the nearest feature only, which might underestimate impacts where multiple anthropogenic stressors are clustered in space. The cumulative ZOI approach implemented in oneimpact recognizes that:

The impact is a product of both effect size and zone of influence: impacts extend beyond the features themselves — a road, cabin, or wind farm may affect species and processes across the landscapes, not just at its location.
The impact of multiple features might accumulate: impacts of features clustered in space (e.g., “cabin villages” vs. single, isolated cabins) can exceed impacts of isolated features in magnitude and spatial extent.
the ZOI is context-dependent: The spatial extent of an impact varies by species, process, and disturbance type.

For detailed background and comprehensive workflow examples, see the package vignettes, particularly the Comprehensive Workflow Guide.

Scientific foundation

This package implements frameworks from two peer-reviewed studies:

Niebuhr et al. (2023): Estimating the cumulative impact and zone of influence of anthropogenic features on biodiversity. Methods in Ecology and Evolution, 14(6), 1470–1487.
- Introduces the cumulative ZOI approach with mathematical framework and empirical case study
Niebuhr et al. (2026): Ecology-informed machine learning to estimate the zone of influence of multiple disturbances on ecological niche models. Working manuscript.
- Extends framework with penalized regression, ecological constraints, and bootstrap aggregation for fitting complex datasets

Typical use cases

Habitat suitability modeling: how human infrastructure affects species persistence
Landscape permeability: assessing movement probability fragmented by anthropogenic disturbance
Species distribution modeling: Estimating responses to cumulative anthropogenic pressures
Environmental impact assessment: Predicting magnitude (effect size) and extent (ZOI) of proposed development impacts
Conservation planning: Identifying priority areas based on cumulative impacts

Quick Start

library(oneimpact)

# 1. Compute ZOI layers (nearest and cumulative)
zoi <- calc_zoi(
  x = infrastructure_raster,
  radius = c(500, 1000, 2000),
  type = c("linear", "gaussian")
)

# 2. Prepare data with ZOI terms
formula <- add_zoi_formula(
  f = ~slope + vegetation + infrastructureXXX,        # base formula
  zoi_radius = c(500, 1000, 2000),
  type = c("linear", "gaussian"),
  pattern = "XXX",                # placeholder in raster names
  predictor_table = TRUE          # returns both formula and variable table
)$formula

# 3. Set up resampling scheme, random cross-validation
samples <- create_resamples(
  y = your_data$response,
  p = c(0.2, 0.2, 0.2),          # train, validate, test fractions
  times = 100                    # number of resamples
)

# 4. Fit ensemble model
fitted_models <- bag_fit_net_logit(
  f = formula,
  data = your_data,
  samples = samples,
  metric = "AUC",
  method = "Lasso",
  standardize = "internal"
)

# 5. Create bag object
bag <- bag_models(fitted_models, your_data, score_threshold = 0.7)

# 6. Evaluate and visualize
plot_response(bag, predictor = "zoi_500_linear")
variable_importance(bag)
implausibility(bag, response_variable = your_data$response)
impact_map <- bag_predict_spat(bag, newdata = raster_stack)

For complete workflow examples, see the vignettes.

Compute spatial layers representing zones of influence

The first set of functions in oneimpact are aimed at computing the (potential) ZOI of infrastructure or other spatial covariates. This means we use spatial information on where they are located to compute the density of features in space (i.e. the cumulative ZOI) and/or the (decay) distance to the nearest feature (i.e. the ZOI of the nearest), given an expected ZOI radius (i.e. the distance up to which a given feature is expected to affect a certain species or process). These functions do not estimate the ZOI, though (which is context and process dependent); for that see more functions further down.

Here are the main functions in oneimpact to compute spatial layers representing zones of influence.

Zone of influence (ZOI) decay functions

zoi_functions: a set of decay zone of influence functions to characterize different shapes of the ZOI around infrastructure, parameterized based on the zone of influence radius. The functions implemented so far are: threshold (threshold_decay() or step_decay()), linear decay (linear_decay() or bartlett_decay() or tent_decay()), exponential decay (exp_decay()), or Gaussian decay (gaussian_decay() or half_norm_decay()).
plot_zoi1d(): plot ZOI in 1 dimensional space for multiple points infrastructure, using both the ZOI of the nearest feature and the cumulative ZOI metric.

Compute zones of influence (ZOI)

calc_zoi_nearest(): Calculate the zone of influence from the nearest infrastructure, according to multiple possible decay functions and zones of influence radii.
calc_zoi_cumulative(): Calculate the cumulative zone of influence of multiple features, according to multiple possible decay functions and zones of influence radii.
calc_zoi(): Calculate both the the ZOI of the nearest infrastructure and the cumulative ZOI, at multiple scales or zones of influence radii.

Spatial filters

create_filter(): Create filters or weight matrices for neighborhood analysis, according to different decay functions and parameterized using the zone of influence radius.
save_filter(): Saves filters/weight matrices outside R for use within GRASS GIS modules.

Estimate the cumulative impact and the ZOI of features on a certain species or process

The oneimpact package also allows us to, given a set of potential candidate ZOIs (with possibly different types, shapes, and radii; computed with the functions above), estimate the actual effect and ZOI of the variables on a certain species or process. This is done combining three elements:

Bootstrap aggregation (bagging), a multi-model bootstrap procedure that allows us to estimate the uncertainty in the effect sizes and ZOI radii;
Penalized regression, an approach that allows us to penalize estimated coefficients and possibly remove the least likely covariates from a model, i.e., it allows us to perform model fitting together with variable section;
Nested cross-validation, which allows is to consider hierarchical, spatial, or temporally cross-validation schemes in model and variable/feature selection.

Estimating ZOI - set up analysis

Functions to set up RSF and SSF analyses using ZOI variables:

add_zoi_formula(): Adds ZOI radii to formula
spat_strat(): Prepares data for spatially stratified cross‐validation schemes
explore_blocks_pre() and explore_blocks(): Explore hierarchical blocks before or after sampling or spatial stratification, respectively
create_resamples(): Create samples for fitting, calibrating, and validating models in a bootstrap/baggin procedure.

Estimating ZOI - fit models

Functions to fit RSF and SSF and estimate ZOI using penalized regression

bag_fit_net_clogit(): Fits a a bag of conditional logistic regressions/SSF/iSSF using glmnet. This function is a wrapper around fit_net_clogit() which is the one properly setting up the model fitting, tunning, and validation. It allows the use of different penalization algorithms, including Lasso, Ridge, Adaptive Lasso, and different adaptations from Adaptive Lasso. This function calls the function net_logit() which is the one properly calling glmnet and fitting the model.
bag_fit_net_logit() (and fit_net_logit(), net_logit()): equivalent to the one above, but performing common logistic regression, with no strata.
bag_load_models(): Load a vector of files with the output of fit_net_clogit() or fit_net_logit() and put them on a bag.
bag_models(): Bag a list of loaded/fitted models fitted through fit_net_clogit() or fit_net_logit(). This created an object of class bag with all information for understanding and making prediction from the bag of models.
AUC(), conditionalAUC(), coxnet.deviance(), Cindex(), conditionalSomersD(): functions used for model tunning (selecting penalties) and validation.

Estimating ZOI - interpret and visualize models

Functions to help interpreting parameters and visualizing cumulative impacts from bags of fitted models:

predict(): Prediction of a bag of models to new data.
variable_importance(), plot_importance(): Computes and plots variable importance from a bag of models.
plot_coef(): Plots the coefficients of bags of models.
plot_response(): Plots (partial) response curves from a bag of models.
bag_predict_spat(): Predict bag of models in space.
bag_predict_spat_vars(): Predict reponses of each individual covariate in space according to a bag of models.

Installation

To install the development version of the oneimpact R package, please use:

library(devtools)
devtools::install_github("NINAnor/oneimpact", ref = "HEAD")

Run with Docker

docker run --rm -p 8787:8787 -e PASSWORD=rstudio -v $PWD/myproject:/home/rstudio/myproject ghcr.io/ninanor/oneimpact:main

If you use Compose:

docker compose run rstudio

You can customize docker-compose.yml based on your needs.

oneimpact