Listening to the Woods

Using bioacoustics to explore sapsucker woods in space and time - By Dimitri Ponirakis.    

One of the advantages of working at the Cornell Laboratory of Ornithology is being able to walk outside of the office and into the woods where one can spend a quiet hour contemplating nature and putting the world to rights, at least in your head.

Many visitors each year visit the lab and walk around the woods to enjoy nature and to go bird-watching. There are excellent trails with boardwalks over wetland areas and views across the ponds. But to a more than casual observer, how do we know what is happening in the woods beyond the trails and beyond the line of site or the well trained ear of the expert birder. What happens at night when no one is around or in the depths of winter? How can we continuously track the activities of birds and other animals across space and time and gain insights into the behavior and lives and the environment in which they inhabit. What are the patterns across the woods how do they change over a day, a week, the seasons and over the years?

Until a few years ago, these questions would have been hard or almost impossible to answer without excessive effort and prohibitive cost. However with advances in hardware and software technology, these questions are now within reach of being answered.

The Cornell Bioacoustics Research Program has been listening to and interpreting the sounds of nature across the globe for more than two decades has now turned it’s sites (ears?) to its own back yard, Sapsucker Woods in Ithaca, NY.

An array of 30 of autonomous acoustic recording units (SWIFTs) have been deployed in the woods over the past 2 years from January 2017 up to the present day.

Two years of nearly continuous sound recordings across 30 units amounts to what is described as big data. In terms of storage, the sound files take up 1.7 TB on their own. In terms of time it add up to over 1/2 million hours of recording. Using software developed in-house, we are able to take this data and process it to generate spectrograms (visualizations of sound in frequency and time) and noise level measurements that can give us an insight into the acoustic environment across the woods.

Variations in Space and Time…

One year of spectrograms for all 30 SWIFT units.

Looking at the figure above you can see the big picture (big sound?) Each row represents a location in the wood. It is obvious looking here that many of the sounds occur across all units, weather events such as rain showers and wind storms will occur over the whole area. Some units are louder than others (warmer colors; reds and yellows are louder and cooler colors; blues and greens are quieter).

Diurnal plot of sound in the range of human hearing over the period of a year from a cabled microphone located next to the main pond. Hour of day reads from left to right and time of year reads from bottom to top.

The above diurnal plot highlights variation in the sound environment at one location next to Sapsucker Woods pond. The daytime is generally louder than the night time across the whole year. It tends to be louder during the spring months as migrating birds move through the region and birds are singing to attract mates and establish nesting territories. Later towards the summer, chorusing frogs can be detected as noisy periods after sunset. Straight lines in the image indicate the presence of man made or anthropogenic noise. The horizontal banding especially visible during the daytime shows the difference between daytime levels on weekdays which are much louder compared to weekend days. The vertical line of noise at around 6 am is noise from scheduled jets taking off and landing at Ithaca Airport just to the north of the woods.

Color-map of computer modeled estimates of man-made noise from road traffic and aircraft sources across Sapsucker Woods. (Redder areas are louder and yellow areas are relatively less affected by man-made noise from these sources.)

The figure above shows us that the woods are not the pristine sounds of nature that we might like them to be. The closer to the road, and airport you are, the more you will notice the sounds of cars and aircraft. This does however present an opportunity for research and we are indeed looking at how noise from morning flights out of Ithaca Airport might be having an effect on birds singing in the dawn chorus.

Noise from a jet plane taking off in a southeasterly direction in the frequency of vocalization of chorusing birds (2 to 4 kilohertz)

The previous figure was a computer prediction of what the noise from the airport and roads would be like, the above figure is an actual measurement of a jet plane taking off in a southeasterly direction. The red areas show high noise levels that follow parallel to the flight path. The noise levels are raised by about 15 dB above background levels in the frequency of vocalization of chorusing birds (2 to 4 kilohertz).

Animation of hourly noise levels measured across Sapsucker Woods over a period of half a year.

We can create animation from a sequence of spatial maps to visualize how the sound fluctuates across the woods over time. The above figure is an animation of average hourly sound levels in the range of human hearing over the period of half a year from late winter to late summer. Noticeable features are the dawn chorus, where the woods light up around sunrise, the ever present noise from road traffic along the northern edge. Pockets of noise around the ponds where there is chorusing from frogs and very vocal Canada geese.

BirdNET: Using Artificial Intelligence to identify species

BirdNET is a software tool that can be used to detect species vocalizations in the sound recordings using Deep Learning Algorithms

In addition to measuring the overall sound levels in different frequencies, we are able to automatically detect vocalizations of different species within the recordings using software developed using AI. The software known as BirdNET was developed by The Cornell Bioacoustics Research Program in collaboration with the Chemnitz University of Technology in Chemnitz Germany.

Running BirdNET on the Sapsucker Woods recordings allows us to spatially and temporally map the presence of most of the vocalizing birds within the woods. We can then begin to record and analyze individual species behavior, when do they start their dawn chorus, when they migrate in and out of the woods. We can combine and enhance other modes of observation such as eBird an app that allows birders to record their observations online via a mobile phone site or web interface. Also with BirdCast which uses NEXRAD weather radar to detect bird migrations in real-time.

Applying BirdNET to the SSW data-set, we can count the number of vocalizing species in any given hour. The figure above shows the species vocalization richness for a single day in 2017. It is apparent that most species vocalizations occur in the daytime, reaching peaks around sunrise and sunset. Also some locations appear to me more species rich than others. The more species rich locations tend to be at sites located nearer to ponds and other wetland areas within the woods.

Hourly detection counts for 27 species for one week in spring 2018.

We can track the vocalization rate of multiple species across time. In the above figure, 27 species were tracked over a period of a week in March 2018. Vocalizations for most species occur in the daytime only, with the exception of Canada geese, which appear to vocalize continuously. In addition to the Canada geese, other more vocal species include the Red-winged Blackbird and the Common Grackle. At the other end of the vocal spectrum species such as the White-breasted Nuthatch and the Purple Finch only appear to vocalize intermittently in the daytime, it might also be the case that the sound levels of these species are quieter and are consequently less likely to be detected.

Count of species vocalizations recorded on Cornell Bioacoustic Research Program’s SWIFT Recording units, detected using BirdNET.

The above animation shows the distribution of detections of vocalizations for a number of species vocalizing over a period of three weeks in November 2017. Here we can start to map species distribution.

The above figure is an interactive plot of the dawn chorus around the woods on July 15th 2017.  To hear the sounds at a given SWIFT location (denoted by the numbered yellow squares), move the mouse over the unit and click and hold down the mouse button to hear the sounds. A spectrogram will appear below that will also list the bird species present in that recording.

Lessons learned, and challenges ahead…

Sapsucker Woods

Here we have scratched the surface of what can be learned using the latest software and hardware developments to gain insights into not just Sapsucker Woods, but all around the globe. It is important that we are able to rapidly uptake and utilize these emerging technologies, as the challenges that the natural world is facing are growing rapidly too.

Artificial Intelligence promises a brighter future for bioacoustics research