COVID-19: We are shipping within 3-5 business days. Contact sales if you need your order sooner.
I arrived on the island of Rota in mid-August and the new equipment arrived in perfect condition. Some Aga (Mariana Crow) pairs have just begun nesting, and this breeding season is looking like it will be a good one.
This season my study will be comparing the vocalizations of captive-reared young to those of wild-reared young. While I wait for the San Diego Zoo Global team to collect the first eggs and chicks for the captive rear-and-release program (scheduled for early October), I am beginning to collect behavioral observations and recordings of wild Aga with my handheld microphone for later characterizing vocalizations.
I have also begun to test out Kaleidoscope's clustering analysis on recordings made in Aga territories last summer using SM3 and SM4 recorders. I am very pleased with how well Kaleidoscope is performing and is able to find Aga vocalizations even with other background noise in the recordings. It even found vocalizations that I missed when scanning through spectrograms visually!
Very soon I will be recording Aga at their nests with the new external microphones and extra long 50m cables I was awarded, resulting in thousands of hours of audio by the end of the season. Using Kaleidoscope's clustering analysis to find and identify Aga vocalizations recorded at these nests should allow me to analyze a substantially larger dataset compared to what might be possible using other available software. I can't wait till the first wild nest recordings start rolling in!
A lot has happened since my last update! Collections of wild Aga eggs and chicks for the captive rear-and-release program (through San Diego Zoo Global) are now complete. The last of thirteen chicks have hatched, and I have been getting some quality recordings of them during their first months. We have found that the Aga chicks make the tiniest begging "cheeps" even during their first feedings while only a few hours old!
In the wild, we have been steadily finding Aga nests, and I have been getting out the ARUs to some of them. Working with a highly intelligent AND critically endangered species means that extra precautions must be taken when monitoring nests. This is where the external microphones and 50m long cables from my Wildlife Acoustics product grant come in. When we find a nest that I think will be good for recording, I hike in under the cover of darkness and set up the microphone in a tree near the nest, camouflage the cable with leaves and stretch the cable 30-50m away where I attach the ARU to another tree. I also camouflage the ARUs just in case. This is all to keep the Aga from noticing the equipment and if they do see it then hopefully they won't associate it with humans. After this is all set up at night I can then check the batteries and SD cards weekly during the day with a much lower risk of disturbing the nesting pair.
The breeding season is now winding down and we're expecting to find only a few more nests between now and May. I am now focusing more on trying to get more observations and recordings of wild fledglings and adults as they move away from the nests to add to my library of vocalizations to characterize. I have already collected almost one and a half terabytes of audio so far this season! Very soon I will be putting Kaleidoscope to the full test as I begin analyzing this mountain of recordings I have accumulated.
Two Aga fledglings (siblings), showing a few of the vocalizations they are known to make.
Above: Two åga fledglings (siblings), showing a few of the vocalizations they are known to make.
The Aga (Corvus kubaryi) is a critically endangered forest crow endemic to Guam and Rota of the Mariana Islands; less than 200 individuals remain (KRONER & HA 2018). A new captive rear- and-release program, in which hatchlings from wild-collected eggs are puppet-reared in captivity, began on Rota in 2016. While the rearing environment allows for social interaction between nestlings and with visiting wild adult conspecifics, missing are interactions with parents and their offspring- directed vocalizations.
Early studies of ʻAlalā (C. hawaiiensis) captive-rearing suggest that social interactions at young stages, or lack thereof, has profound effects in the interactions of adults, including the ability to breed and rear their own young in captivity (Harvey et al. 2002). Vocalizations are an integral part of avian social interactions, but infrequently provided during captive-rearing and rarely evaluated for long-term effects. Crows’ appropriate use of vocalizations in their many contexts is likely influenced by social and vocal experience across their long developmental period (Brown & Farabaugh 1997). Lack of exposure to parental vocalizations during early development may lead to reduced repertoires and poor behavioral responses in older juveniles. This study is a first step in evaluating the importance of early experience with adult vocalizations on the post-release social success of captive-reared crows.
The objectives of this project were originally stated as: (1) Characterize and archive calls and contexts of nesting-related calls of wild AÌga, (2) Determine if there are stage-specific vocalizations between wild parents and offspring. (3) Document, by comparisons with these normal vocal exchanges and responses, any abnormal vocalizations or use of vocalizations (e.g. contexts and responses) by captive puppet-reared juvenile Aga. In this report I will detail the status and findings, to date, with regard to these objectives.
All recordings for this study were collected from August 2017 through May 2018. Twenty wild Aga nests were recorded with SM3 and SM4 units at a distance of 15m. When possible, an extended microphone (SMM-A2) was used 15m from the nest while the unit was placed up to 50m from the nest. ARUs were initially scheduled to record from solar sunrise to sunset, daily, at a sample rate of 24000Hz. However, I found that Mariana Crow calls cover a very broad frequency range, and calling can begin before solar sunrise. Therefore, in November 2017 I changed the settings to record at 44100Hz and 48000Hz on the SM4 and SM3, respectively, from civil sunrise to sunset. To mitigate for the increased file sizes at higher sample rates, I also reduced the scheduled recordings to every-other day. Nests varied with regard to stage when recordings began based on when the nest was found but continued until several days post- fledging or failure. Nests were recorded for seven to 67 days. In total, 4160 hours of recordings of nests are archived. Recordings of mobile juveniles and adults were taken with a handheld recorder and microphone opportunistically during regular monitoring and behavioral observations.
Recordings of captive-reared chicks began with the first collection in October 2017. An SM3 ARU recorded 15 chicks housed indoors during feeding sessions from hatch through approximately day 22. Recorders were mounted on a wall or shelf about 1 meter from the feeding table and set to record daily during the first 3 hours after sunrise. After day 22, chicks began spending time in outdoor aviaries. Recordings were made of chicks outdoors until approximately day 60, and again for one week in May 2018 (ages 5-8 months), for a total of 390 hours of recordings. However, the proximity of the aviaries with 2017’s cohort of young Aga to that of aviaries with older Aga, means that individuals of interest were not acoustically isolated. Therefore, further analyses will be necessary to correct for this and analyses of captive-reared Aga beyond day 20 will not be included in this report.
To locate Aga calls in recordings of wild nests, I created a classifier with Kaleidoscope (v4.5). This classifier used approximately 15hrs of training data pulled from ARU recordings made during the non-breeding season in 2016. Nine nests were run with this classifier. Each nest was classified separately, and up to 900 calls were positively identified and labeled from each. I then extracted (using the “save wav” function) 25 calls from each, avoiding calls that occurred within the same hour of the same day, when possible.
Analysis of the captive-reared chicks took a subset of four individuals (two males and two females), that had recordings available at the same four time points from 1 day post-hatch through day 20. These ages were day 1-2, day 7-8, day 13-14, and day 19-20. Fifteen begging vocalizations were selected and measured from each individual at each of the four time points using Raven Pro sound software (v1.5). I included 13 measurements in the time and frequency domains that are considered to be robust to human error (Raven Pro 1.5 manual). I used simple ANOVA and linear regressions to analyze the effects of age, sex, and weight on acoustic measurements.
Statistical analyses were performed in R (v3.5.2). For analyses of acoustic measurements, a Bonferroni correction was applied to all p-values (a= 0.003) to determine level of significance.
With the improvement autonomous recorders and increased data storage capabilities, software such as Kaleidoscope is critical to analyzing the vast amount of data one may collect. In this study, I have amassed over 4000 hours of recordings, which would be nearly impossible to manually sample any appreciable amount. However, building the best classifier can be rather time consuming depending on one’s goals. I found that the best classifier for my needs used training data from non-breeding season recordings which appear to include a wider range of vocalizations and are less biased toward alarm and territorial-type calls than at nests in the breeding season. This classifier had, on average, an 85 percent accuracy rate (up to 0.5 distance from cluster center) in correctly identifying Aga vocalizations. However, I found this to be highly variable between sites, ranging from as low as 55% accuracy at some sites, up to 98% at others. Attempts to further fine-tune this classifier by adding breeding season training data and captive chick recordings, led to classifiers that either had high accuracy but low variation in calls, or good variation but very low accuracy. Additionally, at this time, it appears that Kaleidoscope will not work well for searching nest recordings for young chick vocalizations. Given the very different recording environments, captive chick recordings did not aid in building a classifier that could detect wild chick calls. Even with more explicit training, using manually extracted calls of wild chicks, this software is unlikely to discriminate the quiet calls, with very little structure, against the variety of background noises. Lastly, attempts to use Kaleidoscope to separate Aga call types, as a method to characterize their repertoire, have so far been unsuccessful.
Using the classifier described above, I ran a subset of 9 nests through Kaleidoscope. For each, I manually identified 150-900 calls within the Aga cluster; the variation was due to total hours of recordings per nest and availability of true Aga calls within the cluster. I found that most Aga calls occur during the first hour of the day at 6am, and generally declined through the day, with a slight increase around 4pm (Fig 1). The types of vocalizations used at different stages of nesting may vary, but these analyses are pending a full characterization of this species repertoire. For example, I noticed in at least one nesting pair where a call type that was not present during the first several days of recording emerged, and dominated, immediately following the “failure” of their nest due to egg collection for the captive-rearing program. This will need to be further investigated to determine if it is truly associated with nest-failure, or merely coincidental.
Figure 1: Vocalizations per hour at each of nine AÌga nests. Values represent percent of vocalizations during each hour, totaling 100% for each nest. As the hour bars represent the sum of percent for each site in that hour, some hours exceed 100%.
In an analysis of a subset of four captive-reared AÌga, I found a general decrease in frequency (Hz) measurements, such as peak frequency and first-quartile frequency, as age increased. This was not unexpected since body size was also increasing. However, the rate of change differed between males and females at different ages. Using the peak frequency (Hz) measurement as an example, females changed most significantly from day 1 to day 7, where changes from day 7 to day 13 and 13 to 19 were not significant (p= 0.32 and 0.97, respectively) (Fig 2). Conversely, males changed most significantly between days 7 and 13 (p
Figure 2: Peak frequency (Hz) measurements of two male and two female captive-reared AÌga at four time-points in development. While frequency generally decreases with age, males and females see the highest rate of change at different ages.
Unfortunately, due to permit requirements, microphones at nests were not able to be as close as I would have preferred in order to get high quality recordings of young, wild AÌga chicks. Therefore, direct comparisons of fine-scale acoustic measurements between captive-reared and wild-reared chicks may not be possible due the attenuation of sound in these very different settings. Based on personal observations, however, chick differences in the broad types of calls are unlikely to show at nestling ages (
Wild AÌga juveniles are dependent on their parents for eight months, on average (Morton et al. 1999). During this time, most vocalizations appear to fall into three categories: begging, food receiving “gobbles”, and contact calls (personal obs.). It seems that most wild AÌga do not begin using/practicing a larger repertoire until they become more independent (personal obs). Anecdotally, captive chicks actually seemed to make more types of calls by 5-8 months of age, which may represent more “vocal play”, or practicing, as seen in wild sub-adults; possibly a result of their being artificially more “independent”. Wild chicks are often still dependent at this age, so most of their vocalizations still fall into the three categories described above. However, this will need to be quantified and it will be important to follow-up on captive-reared AÌga post- release to understand how their adult repertoire develops compared to wild-reared counterparts.
Perhaps the most important next step is to characterize the full AÌga repertoire. This will allow for a better understanding of potential age, nest-stage and/or season specific call types and make comparisons between wild and captive reared AÌga. My preferred method is to use a method of unsupervised clustering to accomplish this, as many AÌga vocalizations seems graded between call types, as in some primate repertoires, as opposed discrete song-types in other songbirds, making human classification quite subjective (Wadewitz et al. 2015).
Next, I will compare the repertoire of the juvenile (>35 days old) captive-reared AÌga to the wild juveniles. Because of the nature of the recordings at the aviaries, it will not be possible to analyze the captive juveniles individually, rather I will need to use them as a group. Although, I may be able to separate 2017’s cohort from the other captive AÌga by creating an amplitude cut- off, since the microphones were nearest to the 2017 cohort.
Finally, I will integrate what is gained from this project with the active AÌga recovery program. When manually scanning recordings, I was able to detect some nestlings as young as 2 days old. Since most young nestling vocalizations only occur during feeding, sub-sampling these recordings could reveal natural feeding rates, better than the normally brief visual observations that happen in the field, which could help improve captive-rearing methods. I will be providing the captive-rearing program with recordings from nests to play to captive-reared chicks and will work with the captive-rearing team to gather new recordings of the next cohort.
Boughman JW, Moss CF. 2003. Social sounds: vocal learning and development of mammal and bird calls. Pages 138—224 in A. M. Simmons, R. . Fay, and A. N. Popper, editors. Acoustic CommunicationâSpringer Handbook of Auditory Research. Springer- Verlag, New York.
Brown ED, Farabaugh SM. 1997. What birds with complex social relationships can tell us about vocal learning: Vocal sharing in avian groups. Pages 98—127 in C. Snowdon and M. Hausberger, editors. Social influences on vocal development. Cambridge University Press, Cambridge, U.K.
Harvey NC, Farabaugh SM, Druker BB. 2002. Effects of early rearing experience on adult behavior and nesting in captive Hawaiian crows (Corvus hawaiiensis). Zoo Biology 21:59—75.
KRONER A, HA RR. 2018. An update of the breeding population status of the critically endangered Mariana Crow Corvus kubaryi on Rota, Northern Mariana Islands 2013—2014.Bird Conservation International 28:416—422.
Morton JM, Plentovich S, Sharp T. 1999. Reproduction and juvenile dispersal of Mariana Crows on Rota 1996-1999.
Wadewitz P, Hammerschmidt K, Battaglia D, Witt A, Wolf F, Fischer J. 2015. Characterizing vocal repertoires - Hard vs. Soft classification approaches. PLoS ONE 10:1—16.