patternscaper: Classify Spatial Landscape Patterns Using Neural Networks

Generates artificial landscapes with defined spatial patterns, calculates landscape metrics, and trains neural networks to classify landscape patterns.

The package supports two classification workflows:

• Pixel-based classification with convolutional neural networks using keras3 (see train_pixel_model)

• Metrics-based classification with landscape metrics from landscapemetrics as neural-network inputs (see train_metric_model)

Typical workflow

1. Generate training landscapes with create_landscapes

2. Optionally calculate and evaluate landscape metrics with calculate_metrics and evaluate_metrics

3. Train a classifier with train_pixel_model or train_metric_model

4. Apply the trained model to new landscapes with apply_pixel_model or apply_metric_model

5. Visualize results with plot_classified_landscapes

References

Tietjen, B., Baldauf, S., & Berger, U. (2026). patternscaper: An R package for classifying spatial landscape patterns using neural networks. *Methods in Ecology and Evolution*. In review.

See also

Useful links:

• https://ecomods.github.io/patternscaper/

• https://github.com/ecomods/patternscaper/

• Report bugs at https://github.com/ecomods/patternscaper/issues

Author

Maintainer: Selina Baldauf selina.baldauf@fu-berlin.de (ORCID)

Authors:

• Selina Baldauf selina.baldauf@fu-berlin.de (ORCID)

• Britta Tietjen britta.tietjen@fu-berlin.de (ORCID)

• Uta Berger uta.berger@tu-dresden.de (ORCID)