The sound of nature

 

 In my continuing efforts to catalogue all the species on my patch, I have splashed out on another piece of surveying equipment. My trail cameras have given me a detailed insight into the mammals using the site — with one obvious exception: bats.

Bats are nocturnal, small and fast, which makes accurate identification difficult. I have seen bats on site and recorded some with my mobile bat detectors, but that depends on me being out in the field every night and staying there long enough to pick up everything. After all, not all bat species fly at the same time of night.

I began by looking at devices sold by Wildlife Acoustics, such as the Song Meter. That got me interested in using sound to identify species. These devices can now help identify bats, small mammals, some insects and birds. Further research led me to the BTO acoustic pipeline, an ambitious programme that analyses sound data and uses machine learning to identify species.

From there, I started exploring the options in earnest, looking at Wildlife Acoustics, Titley and finally Batlogger. After much deliberation over recording types and price, I opted for the Batlogger S2 detector.

The Batlogger S2 is a static logger controlled by an app. The device is small, well-constructed, and easy to use, putting it head and shoulders above the PippyG I had been trying to get to grips with. That said, for the price and with a little perseverance, the PippyG is a good bit of kit — as long as you can understand how to operate it.

The Batlogger app makes setting up the device simple. Strap the unit to a tree with a clear recording area, then activate it using the app. The app works on Android and iOS, connects to the device via Bluetooth, and records its location using the phone’s GPS. You can also schedule when the logger is active. It has an internal SD card, so there is no need to keep swapping cards in and out, and it charges via a lead to your computer, which is also how you download the data.

The data is saved as .wav files, which then need to be analysed. I have used the free version of BatExplorer, but because I am not trained in acoustics or bat identification, I found it difficult to use. It looks like a powerful piece of software, with several ways to present the data, but I needed something a little more approachable.

After experimenting, I opted to use the BTO acoustic pipeline, partly because the analysis is accessible and partly because my data can contribute to wider research, as my bird records on BirdTrack do.

Acoustic analysis is very new to me, but the data produced by the pipeline is clear and approachable. Each automatic identification is given a probability score between 0 and 1. Reading through the literature, the suggested threshold appears to be 0.5, with anything below that disregarded. The statistician in me feels that 0.5 is too low, especially given some of the species suggested in the results. So, for my own analysis, I set two additional thresholds: 0.75 as an amber level, where a species has a good chance of being correct, and 0.9 as a green level, where I felt the identification was acceptable. To help with this, I aggregated the data from the two surveys I ran in June and displayed the results as boxplots. This allowed me to look at the median, mean and interquartile ranges for each record. The final factor I considered was the number of contacts.

So, what species did I find?

After aggregating the data into Chiroptera and other species, for bats at the 0.5 level, 9 species were identified; at the 0.75 level, 8; and at the 0.9 level, 6.

I think it best to use a sensibly high figure as an ID value, so I am only accepting Common Pipistrelle and Soprano Pipistrelle as definitely present on the site, with Brown Long-Eared Bats, Leisler's, and Noctule as possibles. I now plan to review the literature to better understand the ecology of those species at my site. 

The non-bat species, as shown below, offers the most interesting records, but one for which, despite the high probability value, I have the least confidence.

In this analysis, only the Brown Rat fails to reach even the lowest level of acceptance. The other rodent species, like the Wood Mouse and Yellow-necked Mouse, are expected. Wood Mice are widely present on site, and although Yellow-Necked mice have not been identified, they are possible; it's hard to differentiate the two in trail camera footage. 

Harvest Mouse was a very surprising record, and I need to understand their audiology and ecology to determine whether they are likely to be present on the site. There is a section on my patch that is suitable habitat, but it is not in the detector's location.

I know that Bat detectors can detect grasshoppers and crickets, but I would have been sceptical of the presence of two species of Bush-cricket had I not identified a Speckled Bush-cricket nymph only the week before.

Lastly, there are the shrews; they communicate in ultrasound, and so it makes sense they would be detected. Both Common and Pygmy Shrews have been recorded on the site, and Water Shrews are suspected. 

This information gives me some interesting species identifications that warrant closer investigation. I now need to find ways to confirm the presence of Water Shrew and Harvest Mice.

I have only used the logger twice, and I need several more runs and practice with it before I feel more confident. I also really need to understand animal sounds better.

With every new device, my understanding of my patch increases.