Discriminating footprints to improve identification of congeneric invasive Rattus species

Accurate identification of cryptic species is critical for invasive species monitoring. Footprint surveys are often used as an indirect rodent monitoring method, but surveyors can misidentify closely related species. Machine learning techniques can reduce observer errors by enabling species identification through training of statistical algorithms on known footprints and then classifying the footprints of unknown species using the resulting models. Such a tool has important applications for the identification and biosecurity management of invasive rodents.

We conducted a study to test the accuracy of using linear discriminant analyses (LDA) and extreme gradient boosting (XGBoost) to distinguish between footprints of two congeneric invasive rat species in New Zealand, the Pacific rat (Rattus exulans) and ship rat (Rattus rattus). We collected footprints using inked tracking tunnels and extracted geometric profiles of the footprints. We built linear discriminant and XGBoost models on known-species footprints, undertook ten-fold cross-validation, and then applied models to classify footprints of unknown species. The predictive accuracies of the models were all ≥ 90%, with the front foot models (99%) slightly outperforming the hind foot models (94%).

Footprint models provide a reliable tool to distinguish rat species. We discuss potential shortcomings of the models in distinguishing between adult Pacific rats and juvenile ship rats particularly across different populations. We recommend the use of tracking tunnels and footprint models for assessing invasion and reinvasion of congeneric rat species and advocate the application of this technique for identifying and distinguishing among other rodent species. © 2025 Landcare Research New Zealand Limited and The Author(s). Pest Management Science published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.