Background 

The alarming decline in red squirrel populations in the UK can be attributed to the introduction of their North American relatives, grey squirrels. This larger species outcompetes red squirrels through resource exploitation and by acting as vectors of squirrelpox virus. Consequently, the UK population of red squirrels has dwindled, particularly in England, where it is sustained by isolated ‘stronghold’ woodlands.

These strongholds require vigilant monitoring anf management to prevent greys from invading and taking over these areas. Typically, monitoring uses camera traps set up at feeding stations to record visiting squirrels. However, these can be data- and power-intensive, adding to cost and labour, especially across larger sites. These feeding sites can also encourage over reliance. However, the greatest issue with the use of feeders is that it can also create a disease transmission point between reds and greys, as they are prompted to share a food source

This PhD project explores acoustic monitoring as a potential alternative for long-term monitoring of threatened red squirrel populations.

 

Passive acoustic monitoring (PAM)

PAM offers a promising long-term monitoring solution for these stronghold sites. It could substantially reduce the labour intensity of data collection and analysis, while remaining entirely non-invasive to mitigate disease transmission risks. This is done by recording the entire soundscape, and searching within this dataset for the calls of your target organism. Occupancy and potentially abundance can then be estimated from the calls identified. PAM requires a vocally active target species with distinct calls. For grey and red squirrels this is feasible, as their common calls are easily distinguishable from each other. In our field site, a large red squirrel stronghold in the Lake District, PAM has been yielding promising results compared to camera trapping over long periods, both for red squirrel monitoring and detection of invading grey squirrels. The

Automating grey squirrel detection

Arguably, the most significant advantage of using an acoustics-focused approach is the potential for automation, substantially reducing labour requirements for long-term monitoring. Automating aspects of both data collection and analysis requires specialised recording hardware. We have employed ‘guardian’ recorders, designed and built by our project partners, Rainforest Connection. Originally designed to detect illegal logging in tropical rainforests by ‘listening’ for the sound of chainsaws, these recorders have 3G connectivity, allowing for data to be sent remotely to a cloud storage server, with solar panels to keep them charged. Installed in the tree canopy, they reduce the need for site visits when in situ, although during our testing in inclement northern weather, some maintenance has been required. Automation of data analysis can be achieved using AI classifiers trained on large, labelled datasets to identify target species’ vocalisations. Currently, we have trained and are refining a CNN
(convolutional neural network) model capable of automatically extracting red and grey squirrel calls from our data. This eliminates the most time-consuming and specialised part of PAM, which is essential for it to be a tenable approach for most conservation projects. Once our AI classifier is complete, it will be made available open source for use on other recordings. Automated data collection and analysis also offer speed. Rather than waiting weeks before collecting a recorder from the field, with a good connection data could be collected and analysed with little delay, allowing for a much faster response to grey incursions. We aim to have developed a system capable of giving close to instant alerts when a grey squirrel call is picked up by a recorder, allowing for timely management interventions.

 

PhD project carried out by Will Cresswell

Supervised by Prof. Marc Holderied and Dr Phillip Baker

Supported by the Mammal Society, Rainforest Connection and Huawei

 

Any questions contact us at:

di18446@bristol.ac.uk

marc.holderied@bristol.ac.uk