In my country we always measure parameters of echolocation calls recorded in the open, but also in half-open and dense habitats. I don’t see this in your table, why not?
Answer: Yes, of course bats adapt their signal design strongly to their direct surroundings. Our table does not cover half-open landscapes, but it does describe the situation of each species in a very open or an extremely dense habitat, as experienced by the species. The table, in fact, shows what happens to echolocation calls of each species when pushed to the extreme. It shows values of highest frequency and minimum pulse duration (of course used in very dense environments), but also values of maximum pulse duration and lowest frequency (of course typical in extremely open environments). We don’t mention the word “open” or “dense” because ‘openness’ is impossible to quantify. Instead, we look at the pulses we record and let the bat “tell” us whether it thinks its surroundings are open or cluttered. If its pulses are very short and have a high bandwidth the bat’s surroundings must have been dense, whereas long narrowband pulses tell us the bat probably experienced its surroundings to be open. To make the table we did all we could to use recordings in situations that were so extremely dense and open that there can be no mistake about it. At least, this is what we thought until we saw recordings from France of Myotis bechsteini in a VERY open environment. This was just in time for the website, but this is the whole point to join forces in Europe: if you can make corrections to our table, do let us know! One more thing: pulse intervals in the table are “average” not typical of an extreme environment, but rather of the bat’s ‘typical’ habitat (see below).
I don’t see the typical alternation of the noctule bat: plip plop?
Answer: This is indicated in the table as “low” and “high”, both referring to the frequency of the call type. In the table, we don’t refer directly to habitat type (see above).
I have observed pulse durations much shorter than indicated in the table: “minimum pulse duration”.
Answer: Please look closely at the drawing on page 2 of your pdf. Our minimum pulse duration refers to the shortest pulse duration with the entire bandwidth of the pulse still intact! Buzz pulses are of course much shorter, 400 microseconds in M.daubentonii and even 250 microseconds in M.nattereri. Outside the laboratory, such values are hard to measure. We think our definition of minimum pulse duration has discriminative power, but this needs to be proven with certainty.
Why have you ignored our measurements published in Barbamania-Xtreme?
Answer: I never heard of the journal. I will do my best to read everything and to use it, but this is not my work and I have to do this in the evenings. Please help me a bit where you can by emailing me directly publications you think I should definitely know about
If your table is all about extreme conditions, do you think regular, average habitat recordings have no value?
Answer: No, I think they are very valuable. The table is just to show the LIMITS of each species. Limits are one thing, but of course there is more. All calls, including those from average situations can help us to establish correlations between parameters. Some of these correlations are species specific. Besides the table, this will be yet another means of discriminating species. All this is still under development. Whatever I find out I will put on the web, but the process will be slow.
Why are pulse intervals in the table not close to any limit, but just “average”?
Answer: It would make no sense to give pulse intervals for extremely dense situations. From research we know that vespertilionid bats reduce their pulse intervals when approaching an object to go from the “approach phase” into a “buzz”. In this situation pulse intervals depend more on the proximity of the bat to obstacles or to a target than on its typical characteristics as a species. For the big QCF species (Eptesicus/Nyctalus/Vespertilio) the table gives all intervals, also those typical of very open environments. For the other species I decided to give typical pulse intervals for a normal unhindered flight (I admit this is somewhat vague). Since bats couple call emission with wingbeats, their anatomical differences are most likely to be expressed when flying “optimally”, as much as their wing design demands them to do. The high SDs in the table for Myotis species destroys the myth that some of them have regular rhythms. This is useful information for those people who used this criterion.
In the table you use maximum frequency, but I have been told this parameter is quite unreliable because of atmospheric attenuation or other filtering effects so why do you use it anyway?
Answer: You are right about the strong filtering of high frequencies by propagation, by microphone sensitivity, or by the bat not calling straight into the microphone. These effects exert a strong apparent variability on the maximum frequency, even if in reality the bat kept it constant. We use maximum frequency anyway because it differs strongly between species. Secondly, maximum frequency can be measured with high confidence from a pulse if you first check if the oscillogram has a steep onset. If you see the amplitude attains its average level rapidly, say within 0.5 ms, you are on the safe side in assuming that high frequency filtering has had virtually no effect on your recording. The maximum frequency of such a pulse will therefore be very close to the true maximum frequency the bat emitted.
Why don’t you use peak frequency to describe the echolocation calls of bats?
Answer: We don’t use it because it is not a good measure. I explain this in detail in the software corner.
Is your measure of QCF the same as peak frequency?
Answer: It will in practice be quite close to peak frequency (but can differ), however, “mathematically” the measures are quite different: the frequency of QCF is the frequency around which the pulse is least steep (lowest modulation rate), whereas peak frequency is the peak of the power spectrum taken over the entire pulse (in scientific publications). In pulses with an FM-tail (final hook), the QCF frequency is exactly at the transition between horizontal slope into hook. QCF frequency may be thought to represent the frequency a bat species manages to amplify best, but this is still a guess for which we have no evidence.
Why is it so important to measure pulse duration from the oscillogram and not from the spectrogram?
Answer: The oscillogram represents the true signal, so its start- and endpoint are exactly as received by the microphone. This is not entirely true for the spectrographic representation. Here, the original signal has been transformed in slices of, for example 512 points (window size), to calculate amplitude at all frequencies. The computer takes a new slice every time (be it overlapping with the previous slice) and does so until it has gone through the entire sequence on your display. The result of this we call spectrogram. If you take a slice of the oscillogram, this means you take a time-slice. The longer this slice (FFT size high), the less precisely temporal information will come through. Since temporal information in a spectrogram is always somewhat degraded, pulse duration is best measured from the oscillogram that has no such degradation.
What window size should I use to analyse pulses of vespertilionid bats?
Answer: This depends on your sampling rate, but assume you use 44.1 or 48kHz as most people do. For short (<4 ms) broad bandwidth pulses I recommend 256 points. For longer pulses 512 points. If you want to measure very long pulses and you are mainly interested to know the exact frequency of the QCF-part, you may want to switch to 1024. However, the higher you choose, the worse the temporal structure of the signal will look (see question above). Note that if you used 96kHz instead of 48kHz to sample your signal all figures above would double as well!
Above you recommend to adapt FFT size to either short broadband (256 points) or to longer QCF calls (512 points), but my professor tells me to always use the same settings on the analysis programme as all elements of our dataset should remain comparable and therefore be measured in the same way. Who is right?
Answer: Your professor is right in saying that all factors you don’t want to measure should be kept constant, which applies to settings on your programme as well. However, the settings you use should also allow you to measure the signal as faithfully as possible. If you insist on using an FFT size of 512 points to measure a dataset of Phyllostomid calls you will certainly miss a lot of details. Some bat species have such a wide range of durations that there is no other option than to adjust FFT size. As long as you indicate precisely what rule you used in changing settings this is fine.
I recently submitted a paper and described the variation in the echolocation of Pertinax utopensis, using over 2000 calls. A reviewer rejected the paper because I got all my calls from 1 individual bat. The reviewer said I should have selected 1 call randomly from at least 20 individuals. However, on this website you select the best calls and of those, the calls with the most extreme values. Is the reviewer wrong or are you making a mistake?
Answer: This is an interesting and long discussion we should maybe put in a forum where people can comment directly. The short answer is that you should proceed as in studies that use few subjects, such as neurobiological studies. You will need more than one subject, but it is OK to do your analyses on one subject at the time. If you want to investigate the colour adaptability of cameleons you don’t select a random video frame per individual with 100 individuals and calculate average colour values. The same is true with bat echolocation so I disagree with the suggestion made by the referee. You do need to select best calls, using clear criteria, since selecting random calls is like selecting random photos from a series of 10 pictures with just 2 in focus. You need to make sure that what you measure is “sharp”. These are some basic comments, but if there is interest we can discuss this in more detail on a forum.
We are planning to conduct a neural network study to discriminate bla bla
Answer: In my opinion average (read very frequent) calls often do not allow identification. This is why our table uses extreme calls (to start with). To put this in statistical terms: extreme calls have more discriminative power than have average calls. Yet, every single neural network study uses training sets with 10x as many average calls than extreme calls. This trains the neural network to base its identifications on the probability of a certain call type in species A. Recordings of a species made in an atypical environment for that species will therefore be frequently be missclassified, even by the most perfect neural network. To use the real power of a neural network it should be trained to “discover” what combination of call parameters is unique for each species. This can hopefully be achieved by making the training sets contain as many average as extreme calls. This way, training on probability of a single parameter per species is eliminated. Please consider this suggestion.