Big Year Birding - a computational approach? (1 Viewer)

[Nohatch]

Like many of you I have been following Noah Strycker's epic effort to see 5000+ bird species in a single year (https://www.audubon.org/features/birding-without-borders) and will undoubtedly be doing the same with Arjan Dwarshuis' effort in 2016 (http://www.arjandwarshuis.com/#biggestyear). Being a bit of a data geek and avid traveller I've been toying with the idea of a Big Year route planner algorithm. After all we have a pretty good global database of bird distribution and abundances, Google Maps and flight trackers for travel times, etc. This may seem like a hard thing to capture in a computational algorithm, but compared to many scientific data models it would be peanuts. To give you an idea of what it would look like check out Randal Olson's "Optimal road trip across the U.S." (http://www.randalolson.com/2015/03/08/computing-the-optimal-road-trip-across-the-u-s/).

So what kind of input data would be required? In terms of the birds, the Birdlife Data Zone taxonomy and digitized distribution files would be a good start. However, ideally they would be complimented by temporal abundance data. For example, if I visit my local patch in January I'm pretty much guaranteed to see Brant Geese, whereas these would be absent in July. So the kind of data in a system like eBird or BirdTrack would be invaluable. One way to implement it would be to assign a 'likelihood' score for each species at a given place and time. Resolution is a factor and could start off at a fairly coarse grid level (e.g. Maidenhead) and be refined in future, as could the filling in of missing data.
The second important piece of information would be something along the lines of 'travel time within/between' grid squares. Obviously these would score very differently if you were travelling in say Europe or the US, or Borneo. Travel between squares could be easier if based on airline schedule information (ignoring cost...)
Ideally the algorithm would be dynamic and continuously adjust for species already seen - it could even signal when it's time to move on after you've seen x number of species in a certain location. More outlandish additions: "must-see species", avoiding conflict areas, a "green" big year, etc.

Now, if someone could just make it into an app (and I win the lottery) then off we go! Thoughts...?

 

[kitefarrago]

An interesting idea. I wonder whether the data you'd find would be good enough to give you a good stab at deducing your probablitities.

Noah certainly had to decide an overall strategy, but on the ground he is relying on local knowledge to get him to the right spots. I suspect for most parts of the world data at this level of detail just doesn't exist. One could probably generate quite a bit of it by talking to people who have a local patch and can tell you who often they see a given species at a certain time of year along the lines of once every three visiits'.

Certainly a computer could in principle conduct a much more fine-grained exploration than a human being. But if anybody seriously wanted to trial what software can do ti would probably be much easier to start with trying to deal with sometiing like an ABA big year. Lots of data, and the logistics of getting from A to B isn't as bad. One could then compare the results from the program to the strategies employed by successful listers to see whether it suggests anything unexpected.

For a world big year I really don't think the data is there to give your program a decent chance.

Andrea

 

[Paul Chapman]

For me the key variables are diversity within a location (no point wasting travelling time to go to remote spots of low biodiversity), range (no point in concentrating on extra limital species if you can observe them easily in core range at a later stage), timing (a point made by Alan in the other thread) and ease of observation (no point wasting time on the difficult to observe):-

http://www.biodiversitymapping.org/birds.htm

The absolute key will be keeping moving and geographic coverage and I anticipate combining Strycker and Dwarshuis will provide a useful gap analysis. I'll post my (I am sure) imperfect spreadsheet at some point as to what he has seen as a percentage of the various regions' lists.

All the best

 

[Larry Sweetland]

mindblowing stuff Nohatch. bless up bruv :t: This has made me realise it must be possible to generate any kind of data, eg 'what you would have seen if you'd not had a baby before september' or if you'd used the 3rd best guide instead of the 4th, etc

 

[temmie]

While much of this info is maybe hard to translate into useful model input, most of the sites where Noah went and Arjan is going, are well known and reliable sites for many species. Those sites are in the minds of every world birder. I could, from the top of my head, do a big year and see 3000 species without any guides, guidebooks or even road maps. Just going back to the places where I have been before. Easy, because 3000 is not thát much in one year.

But where the model would be useful for (fine-tuning the big year to 6000), is exactly where data-driven complex models are lacking: the details. The model will have an error of let's say plus-minus 1000, or 500 even. That is exactly the number that counts (those last 500), and where the model would struggle.

 

[Brosnabirder]

Super thread thanks. Enjoyed the American road trip link too.