Predicting Species Abundance from Environmental Suitability

Published:

Problem

Conservation and land-use planning require maps of abundance, not just where a species can occur. Traditional surveys are expensive and spatially limited; presence-only data are abundant but lack counts.

Goal: Build a reproducible, data-driven workflow that predicts continuous abundance from environmental suitability, using ensemble machine learning to combine multiple niche modelling algorithms and link them to observed counts.

Approach

1) Data acquisition & wrangling

  • Integrated large-scale presence-only data, count surveys, and high-resolution climate/topography layers.
  • Cleaned, standardised, and processed data for 23 focal species across tropical systems.
  • Automated feature extraction and alignment across spatial grids.

2) Suitability modelling (multi-algorithm ensemble)

  • Trained individual models using:
    • Surface Range Envelope (SRE)
    • Classification Tree Analysis (CTA)
    • Random Forest (RF)
    • Multivariate Adaptive Regression Spline (MARS)
    • Flexible Discriminant Analysis (FDA)
    • MaxEnt
    • Generalised Additive Models (GAM)
    • Generalised Boosted Regression Models (GBM)
    • Artificial Neural Networks (ANN)
  • Combined predictions into an ensemble suitability surface for each species.

3) Linking suitability to abundance

  • Modelled observed abundance as a flexible function of suitability (tested multiple link functions).
  • Accounted for sampling effort and detectability.
  • Validated with spatial cross-validation to avoid overfitting.

4) Delivery

  • Produced spatially explicit abundance maps with uncertainty bands.
  • Exported gridded rasters and tabular summaries for stakeholders.

Stack

  • Advanced statistical modelling: generalised linear models (GLM), generalised additive models (GAM), boosted regression, multivariate adaptive regression splines.
  • Machine learning: tree-based ensembles, discriminant analysis, MaxEnt, artificial neural networks.
  • Data workflows: large-scale data wrangling, geospatial processing, exploratory analysis, visualisation, and fully reproducible pipelines.
  • Implementation: all modelling, analysis, and visualisation conducted in R with version control.

Results

  • Strong suitability–abundance relationships across species.
  • Ensemble models outperformed single algorithms in predictive accuracy and calibration.
  • Abundance maps successfully prioritised high-density areas.
  • Robustness confirmed via sensitivity analysis across link functions and validation folds.

Impact

  • Provided conservation managers with high-resolution, actionable maps to target monitoring and intervention.
  • Demonstrated that ensemble ML applied to presence-only data can yield reliable abundance estimates — a transferable approach for other taxa and regions.

Links & Resources

Role

  • Led study design and workflow development.
  • Implemented ML modelling and spatial validation.
  • Created reproducible scripts, figures, and outputs.
  • Wrote manuscript and coordinated co-author contributions.