Comparing bird calls (1 Viewer)

Hi everyone

I'm a PhD student working on surveying bird populations in broadleaved woodlands. The method that I use involves recording distances to birds heard from points within the woodlands. Using a computer program I can then establish how quickly the surveyors ability to detect birds declines over distance and use this to estimate population density.

This method works well for more common species but for rarer species you don't get enough detections. To get around this you can establish groups of birds with similar levels of "detectability" (assuming no habitat or surveyor effects) and use this to get estimates.

I'm interested in using bioacoustic analysis to compare call intensity and pitch in 40 different species that were recorded in my survey in order to help cluster them into suitable groups.

One of the problems is getting a good reference set of calls to work from, that are recorded under similar conditions so that they can be compared.

I'm interested whether any of you might have any thoughts on this, or any suggestions of ways to compare bird "detectability".

Many thanks

This is much more complicated than it sounds. Many bird species give a variety of calls and songs - some high frequency, some low - so which ones do you group them by? Song or calls? Alarm calls, scolding calls or contact calls - they're all very different, yet very common. Also, there is enormous variability between e.g. one Great Tit song and another. Then you have enormous variability in people's hearing - some cannot hear higher calls (e.g. Goldcrest), some can hear them very well (this varies with age). Then you have the effect of different weather conditions - sound carries much better in still damp conditions than in windy. And then the difference in types of woodland - conifers vs deciduous, March (no leaves) vs May (trees in leaf), thick understorey vs no understorey. So there's 5 major variables that will create massive noise in your results and that you would have to standardise to get anything meaningful. To assume that these have no effects is not really justifiable - they have a very large effect that should not be ignored. Time of day also has an influence in call rate, call type etc, and so does ambient sound (under a flight path, next to a road?). So does the density of other birds - e.g. it's hard to hear a treecreeper contact call during the dawn chorus in a wood full of warblers and tits, but easier at midday in a barren conifer wood. And then there are problems of autocorrelation and replication.

I'm not sure that you can standardise detectability to the degree you're proposing, due to all of these factors and the interaction between them. There are already lots of sampling methods for surveying woodland bird population density, and as long as the method is standardised (eg following a set route for a set period in set weather conditions), then it's not too important what the actual detection distance is. I've been doing some surveying this spring, in woods where I know the actual exact number of certain species, so it's possible to work out detectability from the method I've used (walking a set route, counting birds/min). Detectability is usually quite low - around 50% for the main species, but you would need to calibrate this for every species in every woodland type to get an expected detection rate that was meaningful.

Sorry to be rather negative, and I imagine you have considered most of these points, but you'll need to be really rigorous with these caveats to get it past peer review and your supervisor/external examiner.

Poecile said:

Sorry to be rather negative, and I imagine you have considered most of these points, but you'll need to be really rigorous with these caveats to get it past peer review and your supervisor/external examiner.

Click to expand...

I agree and can't believe that this is a PhD ? Some supervisors are completely clueless so watch the advice you get.

The only way to measure 'detectability' in order to even attempt standardising is surely based more on experimental designs than appraisal of in situ real ones ?

I would recommend that you consider a sound source (playback) of known amplitude that you can 'sample' at various measureable distances in context to different environmental conditions as outlined by Poecile. The playback amplitude can simulate the amplitude of the target species, though as outilined by Poecile different calls will vary depending on their context and the bird's state of excitement. The 747 flying in to Heathrow may have another. Birds do respond to their natural environment as seen in a recent study eg. a Great Tit in a urban environment will sing louder that one in a rural zone.................apparently.

However, this is all academic regarding the species in some ways as it is the aural perception of the humans that will have the most effect on your data. As long as the sound sources are consistent regarding distance and amplitude then it should be okay. Getting a viable 'sample' of human subjects may be difficult however !

How all of this will be any use, I cannot really see, though I can see that such a project may attract funding from organisations looking to enhance their field collection records with regards to population size estimates. Sometimes this,unfortunately, is the most important consideration in PhD projects.

Good luck with the PhD.

Lindsay

PS AFAIK pitch does not vary over distance it is only the amplitude or volume of the call that drops. Harmonics and faint elements may be missing on a sonogram though. This is my experience with my own sound recording. This will have an effect on how they sound to a listner, and presumably their detectability.

Dear Poecile and Griffin

Thanks for your comments! Perhaps I should have not been so simplistic with my description of the methodology.

Distance Sampling is a statistically robust method used to accurately estimate population size (one of the advantages over index counts is that it provides you with confidence limits for your estimates) there is a wealth of papers and books published if you are interested in learning more. The RSPB and BTO used Distance Sampling for the Repeat Woodland Bird Survey.

When I said that I was assuming no habitat, weather or surveyor effects, what I should have said was that I have measured these and will be able to look at their effect on detectability and include them in a model as covariates as necessary. Your point on the type of call is a valid one, counts were conducted during the breeding season so I would expect the majority of records to be for bird songs.

Distance sampling does indeed involve calculating detectability for each individual species. A recent paper (Alldredge et al.) showed that "sharing" information by exploiting similarities in species improved model performance. Although information is shared within a group, detection functions are post-stratified to species level. In this paper they grouped species together by using the rough measure of maximum detection distance as a proxy for detectability, but they also suggested that more quantitative measures of sound pitch, intensity and singing rates would improve species groupings.

So what I'm interested in is information which may help group species together generally. These groupings don't have to be perfect but the more information used to define the groups the better. As you mention I'm interested in properties which will affect aural perception. You mention that Goldcrest calls are higher than other species, so for example I would be interested in grouping together species that sing at a similar level.

Perhaps this is opening too large a can of worms and I should use maximum detection distances as they did in the Alldredge paper!

Thanks again for your comments and concerns.

Most of the detections you get with distance sampling are not songs, so you do need to consider the different calls types. An example would be Great Tit. They are very vocal, and their songs are designed to carry over distance (as are most birds) , so they have a wide frequency range. But when you're sampling you usually pick up Great Tits by call, which may be a low frequency churr, or a high frequency contact call. Many contact calls for many species are high frequency, for short-range communication. So, as you can see, there is no clear way of categorising species, as they all have a variety of calls at a variety of frequencies, for a variety of purposes. And all of these are detected in sampling (although the ranges of detection will of course vary - song carries further than contact calls).

So song is the easy bit, in that virtually all species have songs that slide up and down the frequencies precisely so that they carry. But you need to decide how you will treat the other calls. You may have to have several groups for each species. So, contact calls woudlbe one grouping, and that would contain tits, crests, finches and creepers at least, all of which communicate over short distances with high frequency 'sip' calls. You'd then need another category for scolding calls, conflict calls, and other contact calls.

If you think of Robin, it has a song which covers a fairly wide frequency, then a contact call (a soft 'tic'), then a scolding call (a harsher, lower 'tac'), then a predator alarm call (a very high 'see'). How do you categorise that? You van't ignore the calls and just focus on the song, as that's not how it works in the field.

You're not going to be able to have a category called 'high frequency birds' and another called 'low frequency birds', as all have a variety of common calls at all of these ranges.

Phdstudent said:

Distance Sampling is a statistically robust method used to accurately estimate population....

Click to expand...

Accurately estimate ? !

Seriously, I know what you are trying to say

Without really wanting to delve into papers on distance sampling ( please forgive me that ) can you describe how you measure the distance from an observer (or listener) to the target bird in situ. If it is a calling chiffchaff early in the season they have a habit of moving around considerable distances as they call. A bird initially picked up calling at an actual distance of 100m away could be 150 metres away by the time you measure it. How do you even know it is the same bird ? I guess being a chiffcaff you would hear it >200m so maybe it doesn't matter !

I am also bothered by the 'observer' effect. Do all of your human 'samples' get a hearing test to determine a full range of hearing at various frequencies. Some people I know can't hear a Goldcrest for example. Can they tell a Garden Warbler song from a Blackcap ? Or is it simply the detectability of 'calls' call that you are getting them to do and not worrying about if they can identify them or not ?

Lindsay

griffin said:

Accurately estimate ? !
I am also bothered by the 'observer' effect. Do all of your human 'samples' get a hearing test to determine a full range of hearing at various frequencies. Some people I know can't hear a Goldcrest for example. Can they tell a Garden Warbler song from a Blackcap ? Or is it simply the detectability of 'calls' call that you are getting them to do and not worrying about if they can identify them or not ?

Lindsay

Click to expand...

That is a serious flaw. From my experience, observer effect is great. Their skills vary enormously, and so does their ability to pick up birds. As an example, 3 very experienced ornithologists did a sampling survey in a wood, each doing the sampling 3 times and taking an maximum and a mean. The oldest one consistently got less birds, and the youngest got most. This is undoubtedly down to hearing deterioration with age. The youngest also got more birds of 2 species that they have particular expertise in, as they were able to hear (and identify) subtle contact calls that others either missed or couldn't distinguish. The only way around it is to have one observer do as much as possible, or possibly calibrate the results to an observer based on known detection rates.

When you've got casual or short-contract fieldworkers doing these surveys (eg RSPB), you can get real problems. The Repeat Woodland Bird Survey by RSPB in one of my survey woods reported a pair of Willow Tits (colour-ringed), despite my knowing that WT hadn't been seen for 5 years and every colour-ringed bird was a Marsh Tit. So you can imagine the problems with Garden Warbler/Blackcap, calling WillowChiff, 'tacking' anything...

One of the important assumptions of distance sampling is that birds are located at their original location. To ensure we don't violate this assumption we use a snap-shot technique, whereby at a precise 'moment' the locations of birds who are known to be present are recorded (and distance measured using laser rangefinders). The thing is not all birds that are present need to be detected, and in the analysis you often truncate the data to exclude those distances that are further away and therefore likely to be inaccurate. What is important is that all birds actually at the point are detected (though because this is a small area this isn't generally a problem). Birds reacting to the surveyor is another issue but can again be corrected for in the analysis.

Accuracy of distances is tricky - especially when detections are aural and your in a complex habitat such as a woodland, this can be mitigated by 'binning' the data - i.e so long as the birds are accurately placed within distance groups then there isn't a problem.

As for observer error, yes this is a problem. You can include observer as a covariate to try and tease out some of the problems. The problem with using humans is that they are all inclined to human error, you can mitigate this by using experienced surveyors and hopefully having a sample size which will mean that errors don't have a significant impact on the final result.

How important observer error is depends on the aims of the study. If you are aiming for a complete census of a woodland then it would be very important, however, the reality is that this is rarely the goal because it is impractical and not necessarily useful. If you are surveying for a rare species, where it is incredibly important that you know exactly how many birds there are then again it is important. When you are the RSPB conducting a large survey over a number of woods, the odd pair of Marsh Tits misidentified is unlikely to affect the overall conclusions of the study.

As for the use of the word 'estimate' the feeling in ecology is that unless you have a closed population you can only ever 'estimate' population size. Communities are inherently dynamic with births, deaths, immigration, emigration and though they are more stable at certain times of year, how can you be 'sure' in a purely scientific sense that you have, counted, ringed or captured all the individuals of a population?

But I digress! Thank you both for your thoughts, I should probably get on with some work...