Species distribution modeling with machine learning

Highlights

• This is a contribution to a worldwide collaborative study about structural uncertainty in species distribution models (SDMs).
  
• We predicted the distributions of the species Prionailurus bengalensis and Zamia Prasina.
  
• We adopted an ensemble approach that averages predictions from four decision tree-based classifiers: Random Forests, Extra Trees, XGBoost, and LightGBM.

Data collection

We retrieved species occurrence data from GBIF and environmental predictors from CHELSA.

Locations of P. bengalensis presence retrieved from GBIF data and locations of pseudo-absence generated in random.

Decision tree-based classifiers

Our method was adapted from Daniel Furman’s tutorial on species distribution modeling with python. We implemented regularization for each classifier.

Predicted probability of presence of P. bengalensis using Random Forests.

Predicted probability of presence of P. bengalensis using Extra Trees.

Predicted probability of presence of P. bengalensis using XGBoost.

Predicted probability of presence of P. bengalensis using LightGBM.

The output

Predicted distribution of P. bengalensis by averaging the probabilities of presence from the four classifiers.