Statistics: Estimating U.S. population-weighted average temperature from 100+ daily airport station measurements

Question

I have recently signed up for a developer's key for the Census API ( http://www.census.gov/developers/ ) and will be using a Python wrapper class to access the Census database.

I also have access to a data feed for the daily average temperatures & forecasts from 100+ airport stations distributed across the US (These stations are largely representative of the US population since they are located in major cities). With minimal assumptions, what would be the best way to map the entire population of the United States onto the set of 100+ airports, so that I may derive a population weighted average temperature? This would probably entail some kind of distance/climate function. What are some nuances I should consider when doing this?

Answer 1

(1) Sounds like you need something akin to a Voronoi tessellation , but built on zip code regions instead of continuous space. Essentially you need to assign each zip code region to the "nearest" airport, then weight the airport's observations by the fraction of population in all the nearby zip codes. (I'm assuming the census data is organized by zip codes.) I say "nearest" in quote marks because there could be different ways to consider that; eg distance to geographical center of region, distance to population center of region, time to travel from center to airport, probably others. You can probably use a brute force algorithm to assign zip codes to airports: just cycle through all zip codes and find the airport which is "nearest" in the sense you've chosen. That could be slow but you only have to do it once (well, once for each definition of "nearest").

(2) You might get more traction for this question on CrossValidated .