If you were building a mapping application that could show only (say) 20 airports on the screen at once at any given zoom level, how would you decide which airports are most important, using only publicly-available data sets? Here are some possibilities:
- Points for being in the list of the top 100 passenger airports.
- Points for having an ICAO code.
- Points for having an IATA code (rarer, so more points than an ICAO code).
- Points for each localizer and glideslope (since they’re unambiguously associated with the airport).
- Points for having a TAF.
- Points for having a METAR.
- Points for each long, paved runway.
These are all easy to measure, but I’m not sure that they capture enough of what makes an airport important for mapping purposes. Really big airports often cluster around urban areas — think of JFK, EWR, and LGA around New York, or LHR, LGW, and LCY around London. These are all busy airports, but they’re very short drives from each other (traffic permitting), so perhaps they don’t have the same kind of importance on a map as the main airport in a smaller country, the only airport serving an isolated community or an island, etc.
I’ve done some experimenting trying to measure isolation: for example, I’ve tried limiting the map to one airport in each 30×30 deg square (world level) or 10×10 deg square (continent level), but the map still ends up with huge clusters of airports in the U.S. and Western Europe and none in most of the rest of the world, and even a 10×10 square means that Toronto’s and Montreal’s main airports won’t show up (same square as JFK and EWR). What would Google do?
Land and Hold Short 
My first thought was to give points if the airport cannot be reached by driving from a different airport, thereby solve the only-airport-on-an-island debacle. But if you have 2 small airports on an island this doesn’t work. I think the best way to choose airports would be some kind of population association algorithm. If you’re considering only airports on the scale of JFK, CYUL, and CYYZ then you can immediately discount any airport with no scheduled air service from a 705 airline. Then ranke airports by the number of people that would use that airport as opposed to another one on the list. To be truely accurate one would have to use an inclusion algorithm that included driving, not merely a circular radius for population inclusion.
US airports report to the FAA their number of operations per year in various categories (airline, military, etc.). Sort the airports in the viewport by each of the categories, so you have say four sorted lists. Then show the top airports by total number of operations, but also include at least the top two from each category.
You have two conflicting goals: make your map as useful as possible (aggregated across all people), and make your map useful to as many people as possible. Operations-per-year optimizes for the first goal, while including the top elements of each individual list optimizes for the second goal.
Given your criteria, I would set the mapping application to count up all the points in the specified region and take the top 20 with the most points. Let the computer decide what’s to get rid of.
Why not let Google decide? When they are clustered too closely to show, pick the one that gets the most Google hits. It would err, perhaps when there was a celebrated crash at the minor member of a pair or triad, but for the most part, people talk/write/blog about the important things.
Aviatrix: That’s a great approach when human intervention is possible; unfortunately, I need to find a way to automate the process, and automatically linking Google results to specific airports would be a bit tricky (not impossible, just tricky) — for example, a search for “airports in Ontario” will return Ontario International in California (ONT) as one of the top hits.