I have seen a couple global studies using WGS84 as geographic reference, with the problem that the area of raster cells is not constant. Which reference system would you use for a global analysis in order to have a constant cell size?

  • Can you post a link to some of these studies? – Pablo Jan 25 '11 at 15:09
  • Studies might not be the right word, what I had in mind was for example the worldclim (www.worldclim.org) data set. – johannes Jan 25 '11 at 15:12
  • Now I get your point. I've crossed with that problem before, I'll post a suggestion in the answer. – Pablo Jan 25 '11 at 15:16

What you're looking for is an equal-area projection, and ideally one which also partially preserves shape. In the past, I've used the USGS projection DSS to help guide these decisions, it'll walk you through a process of choosing a good projection. More generally, tools like Flex Projector and Tissot's Indicatrix should help guide a decision: they provide ways of assessing distortion. There's also an existing question on choosing a projection that may help you.

A favorite of mine for global analysis is Mollweide, which is widely supported in GIS packages and retains shape well enough to remain familiar.

| improve this answer | |

Have a look at the paper "A COMPARISON OF EQUAL-AREA MAP PROJECTIONS FOR REGIONAL AND GLOBAL RASTER DATA" http://carto-research.er.usgs.gov/projection/pdf/nmdrs.usery.prn.pdf which should help answer your question.

| improve this answer | |
  • 1
    Yup, this is a good reference -- created by the same authors who produced the DSS tool I mentioned. – scw Jan 26 '11 at 3:30
  • This reference essentially says that equal-area projections tend not to have equal areas and that some non-equal-area projections are worse than others in this regard. That's worth pointing out to unsophisticated GIS users but readers of this site are likely going to be tempted to say "so what--isn't that obvious?" – whuber Jan 26 '11 at 3:47
  • the point was that there is no "good" projection for this question and that the choice is what is the least "bad" projection. That question requires more domain knowledge than is provided by the questioner so he will need to make his own choice. – Ian Turton Jan 26 '11 at 16:36
  • I have no specific application at the moment for that problem. It is just something I have been thinking about for a while. – johannes Jan 27 '11 at 1:15

It's a good question, because gridding lat-lon usually introduces tremendous distortions and often they can be avoided. However, the answer depends on the analysis. Not all analyses need equal cell sizes. Many times you can weight a grid calculation to compensate for variable cell sizes: the appropriate weighting factor is the cosine of the latitude. Given this possibility, which is usually easy to implement, you might want to weigh additional considerations in the balance, such as the desirability of a conformal projection. This would be especially useful when the analysis involves vector fields and their derivatives, which one would expect in climate models and the like.

| improve this answer | |
  • Thanks for your answer. I like your point about weighting for variable cell size by the cosine of the latitude. Do you have any further references on that? – johannes Jan 27 '11 at 2:24
  • @keymirror No direct references (apart from stuff I have written, which doesn't count ;-). This is just basic spherical trigonometry. However, if you look at formulas for pseudocylindrical global projections, such as the Mollweide, you can see that they do exactly that: they shrink the horizontal lengths by a factor proportional to the cosine of the latitude. – whuber Jan 27 '11 at 3:47

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.