If I buffer hundreds of unprojected WGS84 points across the US by say, 3 miles, will I get accurate areas returned (3 x pi x r) for each of the circles created? I suspect not. If not, what would be some best practice to minimize error for each buffer.

Basically, I'd like to minimize error across the US (perhaps including Hawaii) and processing time (in ArcGIS, for instance).

For context, my goal is to extract demographics (also in unprojected geography) for each point lying within a constant distance of each point for the entire nation. My fear is that using unprojected data for this task means that if I'm comparing the demographic info captured by each buffer, I'm not comparing apples to apples.

  • 2
    You mention ArcGIS. It can project the data. Is there some reason that this option might not be available to you? – whuber Jan 23 '12 at 14:18
  • 1
    I agree with @whuber. Why don't you project, buffer, then unproject? – Stephen Quan Jan 23 '12 at 17:29

In ArcGIS 10, running Buffer on multipoints will produce "geodesic" buffers that represent the true distance on an oblate spheroid. See this help topic for more information.

That being said, the difference in the demographic info you'd get for a geodesic buffer isn't going to be THAT different than what you'd get using a buffer in UTM or North American Albers Equal Area Conic, assuming you're using a demographic base layer like Census Block Groups.

I've done extensive work using Census geography, so please feel free to ask followups if you have questions.

  • 2
    You're right that the individual results would not be very different, but on a national scale--or for any region extensive from north to south--you would be introducing a systematic confounder which, for some studies, might invalidate the results. On the average, this will result in systematic changes in counts and densities that vary by about a factor of 7:5 (=cos(25)/cos(50)) within the continental US. Some studies may be able to tolerate that kind of systematic error, but many others will not, especially because the error is not random. – whuber Jan 23 '12 at 14:16
  • 1
    That's true. I guess I was contextualizing the question with the Census GIS work I've done previously, which showed no statistically significant difference in ACS demographics when using geodesic buffers vs. Albers buffers. But these were only ever ~100 mile point buffers. – dmahr Jan 23 '12 at 14:41
  • Great! Thanks for your responses. This is the area of the topic I wanted to head towards. I guess I'm looking for options on how to run such a study across the nation (again, including HI). For instance, I could see writing code to center a projection at each point for the ultimate amount of standardization in buffer size, but I've always wondered how much variance from the ideal 3piR would Albers Equal Area Conic get. I'd like to get a good sense, because the easiest thing to do given how the data is stored is to run it as is, from it's compressed, unprojected format. – angtay Feb 5 '12 at 5:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.