Out of intense complexities… individuals and indices
The number of acoustically active species in a tropical forest is immense, and while we know a lot about what birds make what noises, we know less about what frogs make what noises, and less still about what insects make what noises. Given that the insects are likely to far outnumber the birds and frogs (both in individuals and species), and we know very little about them, it might seem that acoustic monitoring as a concept is premature.
This essay aims to show that is not the case by examining fine and coarse levels of acoustic monitoring through the three lenses of scope, scale and speed.
The use of scope, scale and speed came from Michael Edson of the Smithsonian (Three Urgent Topics for all Museums). Vince Smith and used these as a way of measuring ambition at the Natural History Museum, London, back in 2013: What will a digital natural history museum look like in 10 years time?
These discussions will focus around what I will call “observatory” bioacoustics, that is the deployment of recording devices into the wild and the (often automated) analysis of the acoustic data that is later collected or streamed in real time.
Auditory (OED: a place for hearing): The equivalent noun to observatory for hearing, would probably more appropriate were it not for its potential confusion with the adjective which is frequently used in some fields of bioacoustics. Also, given that so much bioacoustics is now done using visual representations of sound waves the term observatory might be correct in practice, if not logically.
Generally in observatory bioacoustics greater scope, larger scale, and faster speed are all more desirable than lesser scope, smaller scale, and slower speed. The limitations (personnel, costs, hardware, …) of a project will constrain what it is possible, and in many cases this will require a trade-off between scope, scale and speed.
Scope
Observatory acoustics usually takes one of two approaches: targeting a limited number of species — often just one — of research or conservation interest. Machine learning identification tends to target birds or other groups with good data availability (there are orders of magnitude more serious bird enthusiasts than there are bird species — the data are readily available).
Alternatively, broad indices are calculated from recordings and used as a proxy for bioacoustic activity. These are generally based on the assumption that bioacoustic activity makes use of specific frequency bands from the acoustic space (for background on the acoustic space see: Mosaic of the air).
These broad indices are the most coarse bioacoustic measures we have available in observatory studies. Even though new indices are sometimes proposed, I am unaware of anybody with a desire to create an even more coarse measure (or even if such a measure could be biologically meaningful).
At the other end of the coarse-to-fine continuum of scope are many problems that are being actively worked on: calculating species abundance, tracking individual movements, working out what the sound can tell us — is that sound amorous or indicating distress?
Below I have listed how scope within a taxonomic group might increase with further developments. It should be noted that scope can also be expanded by increasing the number of taxa that can be monitored.
- Acoustic indices
- Taxonomic group (e.g. birds)
- Species within taxonomic group (e.g. Atlantic puffin)
- Number of individuals of species
- What species are doing
- Identify individuals
- Identify individuals and what they are doing
Scale
Computing advances in recent decades have facilitated observatory bioacoustics — the difficulties (both practical and financial) involved in trying to undertake such an activity using magnetic tape and analogue visualisation tools were just too high for anything but infrequent and low spatial density data collection. Observatory acoustics has benefitted from the simultaneous decrease in cost and size and increase in capability of digital computers.
I made a similar point in New tools for monitoring biodiversity and environments back in 2013 — the power of the Raspberry Pi platform has increased dramatically since then and been adopted by a number of projects. The Open Hardware AudioMoth has also made low cost and good quality hardware (but not yet network connectivity) to many more people.
There are two straightforward ways of scaling observatories (assuming a standard sensor design) — more sensors (cover a greater area) and higher sensor density (greater resolution within a given area).
A third way is to have more sensors, either moving from mono to stereo recording, or stereo to ambisonic recording (more microphones). Incorporating different types of microphone will all coverage of a greater range of acoustic environments: hydrophones have been used for a long time, there are increasing numbers of researchers interested in the bioacoustics of soil, decaying wood and surfaces.
It should always be remembered that more sensors in the field will mean more resources are needed for long term (archival?) storage and analysis.
Speed
The final consideration here will be of speed — how quickly can the research or policy outputs of an observatory be generated and disseminated? Historically this has been a slow process, recorders were deployed to the field and later the recordings collected manually by the researcher. These were then taken to the lab for analysis, which might itself be a slow process.
Now it is possible — if not always desirable — to stream data in real time (often by mobile data, now perhaps Starlink satellite network?) to a cloud service for storage, analysis and eventual archiving. Although is this more cost effective (or even have a higher data rate) than manually moving hard drives — the so-called sneakernet.
If sufficient network speeds and computational resources are available what conservation and research advances could be possible? Guiding rangers to the gun shots of poachers or chainsaws of illegal loggers? Tracking individuals of an unidentified species by their sound through a forest until researchers can locate them?
While there is a clear need (as a community) for ongoing work to improve scope, scale and speed this does not mean that for a given project a small number of recorders, with data retrieved by hand, and analysed using indices is not the most appropriate solution. It could easily be the most cost effective way to get the information required.
Indices also have the potential to make comparisons between different sites more meaningful — the number of Amazonian endemic waterfowl will differ between Brazil and the tropical forests of Cameroon but the soundscape indices may be broadly similar. The fact that indices reduce the complexities of the soundscape to a single number can be both problematic and useful.
Out of intense complexities, intense simplicities emerge.
Winston S. Churchill
Another consideration in favour of indices is when a substantial proportion of soniferous species are either undescribed by science, or the species has been formally named but not linked to the sound it produces. This is substantially more common in insects than tetrapods.
In conclusion we are still some years away from acoustic observatory systems that can meet the majority of users’ requirements off-the-shelf, but that does not mean what can currently be done is not useful. But as a community we should be thinking about what will happen when we have a better idea of what works well.
How will we preserve the data we collect? How will we make meaningful comparisons over generations of devices, algorithms, and people? How do we move from chasing papers to creating libraries? From research projects to infrastructures?
References
Bennett, W., Chesmore, D. & Baker, E. (2015) Speckled bush cricket data logger — Project Report. Figshare. doi:10.6084/M9.FIGSHARE.1430094.V1.
Heller, K.-G. & Baker, E. (2017) From an old sound recording to a new species in the genus Horatosphaga (Orthoptera: Tettigonioidea: Phaneropterinae: Acrometopini). Zootaxa 4323(3):430. doi:10.11646/zootaxa.4323.3.10.
