Facts about automatic recording and identification systems: the future

2010 marks a new era of bat-detector use. Formerly, bat-detectors used to be the instruments of a small group of people having highly trained ears and / or good sound analysis skills. Now, slowly but surely a new kind of detector is conquering the market. These detectors are made to be left overnight (or even left over several weeks) and to be picked up the next morning with all the bat-passes stored on a memory card. Companies (Ecoobs, Omnibat) have developed software that can automatically identify the species on the recordings, that is, European species. I would like to discuss this development point by point.

Is it really true that leave-alone detectors are such a success in terms of sales?
Yes. It’s not the typical bat sound specialists who buy them though, but mostly companies (windmills) and environmental agencies (consultancy). Many units have been sold in Germany, France and Canada. Agencies love to save on costs associated with field hours and this detector system is exactly what they have always wanted. Processing the files also takes a lot less time this way and to them every hour needs to be paid. Furthermore, they finally have an identification standard (the software) instead of having to rely on the expertise or reputation of the person who conducted the assignment. There are currently 3 models of leave-alone detectors: the Batcorder, Pettersson D500X and the Song Meter SM2. Even more manufacturers may decide to come up with their version in the future. This development seems to last.

Does automatic ID software really work well?
This discussion so far only applies to Europe and to the Ecoobs software. I have no personal experience with their software and comments from users would be most welcome. However, I have asked the programmers about their methods (see Batcorder review). My opinion is that the groundwork of the software is very solid and well thought. The actual performance comes down to details and will improve over time. Most people will agree that the automatic ID performance at least “makes sense” and surely isn’t a complete mess. If all agree on this statement, the discussion is merely about details: “what does it do if Vespertilio enters dense space and uses that strange call type bla bla” This is a discussion comparable to the one bat specialists have on forums or at meetings. Even a staunch opponent of automatic ID will admit that any software should be able to separate Pipistrellus calls from the others, the horseshoe bats and Myotis and do this practically 100% correctly. Isn’t software that does this already very useful to save time? You load the data and the software tells you immediately on which files the potential ‘goodies’ are and which files are filled with the usual 100+ pipistrelle bats. The user can then use his/her time very efficiently to analyse by hand all non-pipistrelle files. If you wanted, you can also check if all the pipistrelle files really are pipistrelle bats. You don’t get any ID forced upon you. To summarise: the unbiased opinion is that very basic automatic ID will already help to save time. How well the software performs under typical species overlap conditions is a discussion for specialists where many facts would need to be proven first. If not, it will remain one word against the other. Main point: the user can always use manual tools to analyse and identify each pulse.

Why do automatic ID software programmers keep their methods secret?
Again, this discussion applies to Ecoobs software. It’s not true that methods and data are kept secret. On the contrary, people from Ecoobs made a lot of effort to gather an enormous database of European bat calls, from dense to open. This they did better than previous researchers had done and it took quite some effort. The batecho table of call parameters was checked and corrected thoroughly by comparing it with their dataset and the table is of course freely available. More call features that Ecoobs discovered will be added and also checked with the Ecoobs database, all freely accessible. You can read a bit about the methodology in the review of the batcorder. Ecoobs is working to publish a Wikipedia site detailing all the ins and outs of the scripts that are used. I would guess that who really wants to have these details right now can go to a friend who has the software and inspect all the programme files to see how things are done.

Will “hand-batting” (using detectors in the hand) disappear and will experts no longer be required (in Europe)?
First of all, stationary detectors don’t find you a colony! Secondly, I could think of many stupid places to put a detector overnight where it would never a record a single bat. In many forests you can already tell where the horseshoe bat will fly, where Eptesicus is likely to make a turn and so on. We’ve developed an eye for it. People with such an eye will always be needed. Automatic ID is working fairly well, but is of course never 100% perfect (maybe it never will be). At this stage, feedback from experienced users is still required. This website is a good opportunity to share your knowledge so the community can profit from it. Some day in the far future, nearly all call features will be known to the community and to software developers. If you insist you only want to use a detector to learn to identify species, you simply book a flight to Thailand and you can happily start all over again. However, if you are a survey-expert in Europe and you count bat species as part of your work, competition will probably force you to shift your focus in the future (see below).

The future of bat-detector work in Europe
It’s 2020. You park your hydrogen-battery powered car silently at the edge of a forest. Last night’s data are quickly transferred to work by using your mobile phone as an interface. Oh, a Bechstein’s with 87% confidence at unit 3, so this is where you have to put a mistnet next time. So, is this age all about the rarities? On the contrary, as we are heading to mass data collection, we will finally have sufficient data to do some proper statistics! What do I mean? What I mean is that present hand-batting results aren’t very impressive in terms of statistical reliability. In the Netherlands, partly for reasons of labour costs, 3 visits to one area are standard. If you detected Eptesicus during one of those visits, it “occurs” in the area. Granted, it surely happened to be there once, but maybe this was the bat’s first time in the area? Who says that 3 visits are sufficient? What about the numbers of pipistrelles one counts? Do they matter? Do they change a lot between nights? In 1990s, some bat workers in the Netherlands already made some efforts to establish how many visits would be required to get a picture of the real number of pipistrelle bats in an area and this number turned out to be over 20! This number has kind of been kept silent by bat-detector users as it was a bit inconveniently high. But what about horseshoe bats? It’s bound to be lower. Maybe, things get much better if data are collected the entire night instead of transects being walked as in the former studies. Obviously, monitoring specialists have since long asked these questions, but the number of field hours required to answer these questions was simply too high to bother. The “invention” of leave-alone detector has changed all this. The typical bat-detector user will simply have to shift his/her focus a little bit to populations, changes in numbers and other statistics. If 1000 respondents of a 10 million population can reveal faithfully who the new president will be, we have to figure out how 100 areas can faithfully tell us the population size of the entire country.