You can check details of bioclimatic variable in following links:

http://www.worldclim.org/formats

http://www.worldclim.org/bioclim

Could you elaborate your question including -

- how you get bio 4 and 14 important?

- Which method did you use for variable selection?

I forgot to mention that I'm using Maxent for modeling and in the part of the "Analysis of variable contributions" I got the percent contribution of each map. The percent contributions are as follows: Bio 4 = 80%, Bio 14 = 6% and the AUC is 0.989

Thanks,

Hi Ali,

BIO14 should be in mm. These are, however, interpolated layers using average montly values. So, they can deviate from real local values. For BIO14 the standard deviation (*100) of the range of annual temperature was used instead a coefficient of variation, because it does not always make sense, e.g. with temperatures between -1 and 1 (which is also stated on the Worldclim website).

Hope this helps.

As stated BIO14 is in mm so the map that you are looking at for BIO14 must not be the original if it only includes 0 and 1 values.

BIO4 is enigmatic. I'm not certain which standard deviation they refer to. I think it is the standard deviation of the mean monthly temperature values. So it is related to the total annual temperature variation. I'd have to look at how it is calculated again because I've forgotten. I think this information can be found in the Hijmans et al. 2005 paper.

It is important also to note that the % contribution to Maxent is potentially variable from one model run to the other. This is because Maxent's "machine learning" algorithms try different combinations of variables and rankings to generate the model that uses the fewest parameters to describe the species distribution in environmental space. Equally good answers can be found multiple by this method especially if you are using all 19 bioclim variables. This is because many of these variables are highly correlated (e.g. precipitation in the driest quarter BIO14 and precipitation in the driest month BIO17). So in one run the % contribution may be high for BIO14 because the model put more weight on that variable but in another run the % contribution may be higher for BIO17. The % contribution values are calculated (I believe) by removing one of the climate variables from the model and recalculating the ROC AUC values and reporting the % change.

This will generally not change your prediction of species ranges but it may mislead your interpretation of the results. I would advise finding another method for assessing how the climate variables are correlated with the species ranges that you are interested in. Though without hypotheses about physiology this may not be all that useful because the limits on distributions are probably more about interactions between climate variables than any individual variable.

If your are looking for contributions of predictors, we usually use PCA or correlation tests to pre-select variables that are finally applied to the SDM. This is also mentioned in the Maxent's manual.

However, as far as I know, MaxEnt uses the L1 regularization to pre select variables. 

I suppose that BIO04 (Temperature Seasonality) means that if there is a high variation then the seasonality is high, but if low, then there is less extreme difference through the year, resulting in no pronounced season, but rather an even and mild climate range? So it is not indicating when the season is, as opposed to an identifiable change through the year.

Dear friend, Here is a link. This report provide a description of bioclimatic predictors.

http://pubs.usgs.gov/ds/691/ds691.pdf

Before running MAXENT have you performed autocorrelation test. First perform autocorrelation test and omit highly autocorrelated variables . Then you can find out only those bioclimatic variables which are be used.

The WorldClim variables are interpolations, not measurements, and therefore as (in)accurate as the bias  in the point location and selection of the meteo-stations underlying the interpolation. Meteo-stations are not randomly distributed and those with free data are not a random subset (e.g. airports are overrepresented). For an example see attached. My guess is that using lat, long and elevation as variable may predict as well as the compound variables of WorldClim.

Secondly, you may pre-select among the bio-climatic variables as various respondents already suggested. For an example see the second attachment.

Article Where the bears roam in Majella National Park, Italy

Article Transferability of species distribution models: The case of ...

Ali: attached are a couple of papers that confirm Prof Van Gils comments.

Very few published papers deal adequately with uncertainties of climate data and therefore projections. I find it disappointing but not altogether surprising that so few scientists want to deal with the messy realities. 

Hi Ali,

     If you read 

 Hijmans, R.J., S.E. Cameron, J.L. Parra, P.G. Jones and A. Jarvis, 2005. Very high resolution interpolated climate surfaces for global land areas. International Journal of Climatology 25: 1965-1978.

you will find the answer.

See also the Data format @  http://www.worldclim.org/formats1 for the unit of data.

     I agree with Peter that there are a lot of uncertainties of climate data. But for species distribution modeling, I will go with  George Box, " all models are wrong, but some are useful ".  In many situations, especially for those species with rare ecological data, SDM is  very useful and provides a good starting point.

Dear Ali

Nice to hear from you again,

As for BIO14, it is in mm and in your study area (Central Zagros?) we can accept that in the hottest and driest month (say, august), there is no precipitation. So the values are Right. You probably should do a correlation test before using Maxent to avoid Auto correlation, but again, it won't change BIO14 issue in your work. Because BIO14 will come out pride! Besides, considering the values of BIO14in your work (0 and 1), it should be the least important variable, as it is in most of dry ecosystems. My best guess is that you may be missing some thing in your modeling procedure. So, do a recheck and see what happens.

Best Wishes

P.S.: I miss you my friend, lets chat: [email protected]

Hi Ali,

You can check and remove multicollinearity using VIF method. This method is described in the following publication. Check electronic supplementary material for R-codes.

Cheers

Article Climate modelling for agroforestry species selection in Yunn...

This new paper is about the details of bioclimatic variable calculation and their impact on species distribution models:

Article The way bioclimatic variables are calculated has impact on p...