The Asiana incident is the genesis for the project. If you go to the pprune forums, there is a lot of discussion of approach profiles related to this accident. Through some cute analysis, you can come up with scaled metrics on approach and departure quality. From there, you can do a statistical analysis. So I have a 1090 recorder parked at the local airport recording and sending to a database for further analysis. There is a mix of traffic, the majority of which have ADS-B out. Filtering the data is the tricky bit...
Making a separate reply, but if you are interested in mapping approach paths, Mode C and Mode S (non-ADS-B) data on the 1090 band may still have interest for you. It will be difficult to correlate, however Mode-S contains the ICAO address and altitude info (even at its most basic without GPS), which you could at least map to Flight IDs and the filed flight plan, giving you a basic approach path in the X and Y direction, which you can add semi accurate (100 or 25 foot resolution) information in the Z direction. Even harder to correlate, Mode C data could be used for non-correlatable, but still trackable info to watch a plane land (gather the data, if a Mode-C target went from in-air (>N feet above airport MSL corrected for pressure altitude) to on-ground (<N feet above airport MSL correct for pressure altitude), you could still get approach information for almost any aircraft, not just ADS-B aircraft. Note: Mode C data has 100 foot resolution, and is pretty noisy.
That is ultimately the goal. Having a ubiquitous metric for evaluating pilot performance when subject to different airspace/approaches/airport environments. Right now, I would be happy with reliable interpolation of spacial data. The broadcast packets are temporally inconsistent and the speed/altitude/heading are spread over more than one packet. So a simple interpolation only goes so far. I expect to have to write code to intelligently interpolate based on what aircraft can actually do. So no 3g turns in a transport airplane in the traffic pattern, etc. Once I have a functional model with the "tidy" ADS-B data, then the challenge will be making the process robust enough for mode S/C. It's a plan subject to change...
Very interesting. I guess I had supposed that most pilots on instrument approach would follow the prescribed approach cards, though I guess that Asiana flight really proves that isn't always the case, as well as inclement weather, etc. You mention that most traffic has ADS-B out. Are you listening to non ADS-B Mode S transmissions, or any logging of Mode A/C transmissions? From a single monitoring point, I know that Mode A/C/S transmissions are far less interesting than ADS-B, however I was wonder where your "majority" claim came from. I suppose if you are near a commercial/passenger airport (as opposed to private/GA airport), this will probably be true. I'd be happy to support your efforts if theres anything I can do (I come from an Embedded/Avionics background, including TCAS/Transponder development, as well as some data processing experience), though I can help more with Avionics theory than I can with statistics :). Feel free to send me a PM if you would like to discuss. Best of luck to your efforts, it definitely seems like a cool project!