I want to use maxent to make some future distribution predictions. I downloaded future climate data from worldclim. In worldclim's page there are three variables (max temp, min temp and precipitation) as well as a fourth titled "bioclimatic variables" which mentions that it contains the other three and some other new ones. My question is: if i download all four and use them for my analysis, will maxent produce reliable results? Or would the duplicate variables be a problem?

  • 1
    $\begingroup$ When you say "maxent" and "distribution", what do you mean exactly? For example there is specific ecological software called "MaxEnt". But that's also a generic name for maximum entropy methods, which are fairly diverse. $\endgroup$ Feb 21, 2023 at 21:53
  • 1
    $\begingroup$ Looking at the data source (which I believe to be this), it does not seem to line up with what you describe. There are more than 3 variables in each of the 4 datasets except the monthly values which has 3, but is presumably a bunch of time series. I am not sure that you will be able to meaningfully combine the time series data with e.g. bioclimatic point estimates for each location (?). You might think to yourself, "would it be handy to use a dataset that someone already preprocessed to be useful for biological modeling?" $\endgroup$ Feb 21, 2023 at 22:10

1 Answer 1


Ok I understand better now so I'll write an answer.

Look at the bioclimatic data site. You can download them from the main page, but the suggestion to generate them from the time series data using an R package (dismo). The bioclimatic variables are thus various linear combinations of the 3 variable time series input.

Not having run this myself, I speculate, but it doesn't make sense to me that the bioclimatic variables are also time series, they appear to be summaries across the time series. So they're not exactly identical in information content, but they're close enough that I'm not sure it's worth the effort of figuring out how to munge the monthly data into a format that works with the summaries.

Someone already went to a lot of trouble to think about the biologically relevant representation of this data (e.g. what ANUCLIM does). I recommend relying on that work, or at the least looking at their arguments and deciding whether you agree enough to use their summaries.

It doesn't particularly matter what software you use, I'd argue (probably MaxEnt the species distribution package, but you haven't clarified). Most methods are going to be doing similar things with this data. They are going to find the variables with the best explanation and weigth them more heavily. Throwing in a couple extra correlated variables might get you a little margin, but it is unlikely to increase explanatory power.

If you want a formal methodology to compare the two, you could use something like AIC or BIC to compare the models formally with the 2 variable sets. But it hardly seems worth the effort of the data munging.


I am assuming that you are not talking about adding ~identical variables also into your model, e.g. files here, but are instead talking about the less identical monthly data here. I'm pretty sure this is right, but I wanted to clarify.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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