JOURNAL HIGHLIGHTS BY STEVE RANGER
Researchers have used machine learning tools to analyse inky footprints left by rats in order to help identify different species.
The team said that the techniques could play an important role in the identification and biosecurity management of invasive rodents.
Accurate identification is key to monitoring and managing pests. But direct observation and capturing specimens isn’t always possible, so indirect methods of identification – like hair collection or footprint analysis – can useful for assessing the presence of more elusive animals.
This approach can create additional complications as some indirect methods of identifying pests will require confirmation through DNA analysis. One advantage of footprint surveys is they can confirm the species of a pest by visual inspection alone: however, identifying particular species in this way can be tricky and observer error can be high even among experienced surveyors.
The research from New Zealand, published in SCI’s journal Pest Management Science suggest that machine learning may be an increasingly useful tool here. Advances in machine learning techniques and automatic footprint identification programmes have improved in recent years and the approach has been used for animals ranging from cheetahs and giant pandas to rhinoceroses.
The technique involves luring the animals into the tracking tunnels. Inside the tunnel is a tracking card and ink pad: the animals step on the ink and create footprints on the card as they pass through.
In New Zealand tracking tunnels have become a standard monitoring and management tool for four rodent pests: the Norway rat Rattus norvegicus, ship rat Rattus rattus, Pacific rat Rattus exulans, and house mouse Mus musculus. For example, the country’s Department of Conservation has a network of over 10,000 tracking tunnels across the country.
Distinguishing among the three rat species is hard, but being unable to identify which particular species is present can make it harder to apply species-specific management: different species have varying preferences when it comes to lures for example.
The study explored the feasibility of extracting measurements from footprints of the Pacific rat and ship rat collected from tracking tunnels, and then using machine learning algorithms built on training data and validated on testing data to distinguish between the species.
‘The resulting footprint models will add to the values of tracking tunnels as a biosecurity monitoring tool, allowing identification of rodent pest species and species-specific relative abundance monitoring and management strategies,’ the researchers said.
The researchers obtained footprints known to be of Pacific and ship rats, stored these as digital images and then developed a Python computer program using the OpenCV, NumPy, SciPy, and Matplotlib packages to measure the geometry of each footprint.
They then compared the performances of LDA and XGBoost in classifying Pacific and ship rats, with models built separately for front and hind feet. They also used LDA and XGBoost to build the population models.
To test the practical applicability of the models, the researchers collected tracking cards with unknown footprint identification and used them as classification datasets.
While the predictive accuracy of the hind footprint species model reached 94%, it was lower than that of the front footprint species model - which reached 99%.
The researchers said the use of these footprint models would depend on the purpose of the survey.
‘For example, with an accuracy of 99% (ie 1% error), our front footprint model is useful for identifying Pacific or ship rats that have invaded or reinvaded a rat-free island or for assessing the relative abundance of these two species where they are known as co-existing,’ they said. Where one of the species was not recorded before, the footprint model can be a useful pre-assessment tool but further verification using other monitoring techniques might be necessary, they added.
The researchers said the application of footprint models in tracking tunnel monitoring will greatly improve the practicality and benefit of tracking tunnels as cost-effective biosecurity tools.
‘We recommend the use of tracking tunnels and development of footprint models for assessing invasion or reinvasion of congeneric rat species, or identifying areas where cryptic species, for example, Pacific rats in New Zealand, might exist in low numbers and their presence requires further confirmation,’ they said.
Discriminating footprints to improve identification of congeneric invasive Rattus species
Sze Wing Yiu, Thomas R. Etherington, James C. Russell
Pest Management Science
doi.org/10.1002/ps.70052
Visit Blog
