Depending on the density and selection of the meteorological stations in your/around your research area underlying the WorldClim extrapolations, your model accuracy (AUC) will be lower or much lower than necessary. You will need more (WorldClim) variables to reach an acceptable AUC. Your results will not be very transferable outside your test area. My prediction (hypothesis) only. The test would be to include climate proxies other than extrapolations of double biased point data. Such data could include DEM (elevation), vegetation cover types ( take care of the number of categories) and/or NDVI that are all available in primary values per pixel/grid/ cell.

looking forward to your test results

Article Fine resolution distribution modelling of endemics in Majell...

Article Where the bears roam in Majella National Park, Italy

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

Hi Joshua,

Worldclim is one of the best options to download data for current, past and future conditions. It has many advantages. You can get data at different resolutions from all the world. I don't think you will have problems justifying why you are using it. If the highest resolution of Worldclim isn't good enough for your purposes, you can always resample to a higher resolution, e.g. using bilinear or cubic interpolation.

Hi Joshua,

What resolution are you looking for? I agree with Dr. Van Gils that downscaling will lead to a certain amount of error, so you have to make sure that you acknowledge it (however, any climate data, downscaled or not, already has some error attached to it, so nothing will be perfect). Depending on your goals, the lack of transferability to another region might not be important. I would definitely add DEM to the maps you use in MaxEnt, as I found that it is usually an important one when running a Jackknife test. I personally have downscaled climate data (using bicubic interpolation) to use with MaxEnt, and I have seen enough articles doing the same to accept it as a reasonable method. Just be aware of the possible error it might create and make sure to discuss it in your interpretations. 

As for the use of AUC to determine the best combination of MaxEnt parameters, I found Beck et al. 2014 interesting as it shows that a high AUC might not be the best measure to determine accuracy of predictions. 

Finally, with WorldClim, it is easy to just put all the bioclim variables in MaxEnt and trust the outputs. However, this can lead to problematic predictions. I would recommend reading Merow et al. 2013 to understand the impact of variables and evaluate how to determine the best variables to include in your model.

Beck, J., Böller, M., Erhardt, A., & Schwanghart, W. (2014). Spatial bias in the GBIF database and its effect on modeling species' geographic distributions. Ecological Informatics, 19, 10-15. doi:10.1016/j.ecoinf.2013.11.002

Merow, C., Smith, M. J., & Silander, J. A. (2013). A practical guide to maxent for modeling species’ distributions: What it does, and why inputs and settings matter. Ecography, 36(10), 1058-1069. doi:10.1111/j.1600-0587.2013.07872.x

It also depends on what organism you are looking at and where it lives. WorldClim extrapolations are based on measurements at open sites, while a forest animal, particularly a small one, may experience very different temperatures and humidities. See Brett Scheffers' new paper in Biotropica, Extreme thermal heterogeneity in structurally complex tropical rain forests.

Dear Joshua,

If you need the 19 bioclimatic variables at a resolution finer than the 1 km × 1 km grid provided by WorldClim, you can generate these data in several ways. Here I recommend you one of them that worked for me (using R statistical environment and regression kriging), but I'm sure that you should take the previously mentioned advices about uncertainty.

1) Download the raw monthly climate rasters from WorldClim. Obtain Digital Elevation Model (DEM) at resolution fine enough.

2) Open them with functions raster() or stack() of package 'raster', or with readGDAL() of the package 'rgdal'. Alternatively you can use function getData() of package 'raster'.

3) Convert them to SpatialPointsDataFrame or SpatialGridDataFrame object of package 'sp'.

4) Be sure that you have two additional columns containing latitudinal/longitudinal data (e.g. in WGS-84 system).

5) Center/Standardize your data, including climate, elevation and coordinates.[(x-mean)/sd]

6) Iterate through the monthly climate variables, do the following steps in each iterations:

6a) Bulid a linear model with formula: climate ~ elevation + lat + lon; if it's significant, go to 6b)

6b) Calculate experimental variogram with the same formula and then fit a variogram model using functions vgm(), variogram() and fit.variogram() of package 'gstat'. Set initial nugget to be 0 and initial psill to be mean(variogram_experimental$gamma). Go to step 6c) if the fitted variogram is not singular or the plotted variogram seems to be OK [attr(variogram_fitted, "singular") == TRUE]

6c) Run kriging. It may take a while.. (couple of hours if your stady area is large enough). It is slower but much more sophisticated then bilinear interpolation or inverse distance weighted (IDW).

result

Hi, Ákos Bede-Fazekas! Can you refer a paper for this method?

Hello Patrick,

the online tutorial on spatial-analyst.net that you can find in my previous post is the best guide I know. If you need paper about research that did tha same steps, you may be interested in two of my articles (attached and linked to this post). The "Vulnerability of natural habitats. Research study" contains a lot of references to kriging and regression kriging.

Anyway there must exist several articles that fit better to your request, so I hope others give you more relevant answer to your question.

Have a nice week,

Ákos

http://nater.mfgi.hu/sites/nater.mfgi.hu/files/files/Natural_Habitats_EN.pdf

Article Implementation and application of multiple potential natural...

Thank you Ákos! I tried to understand your workflow but I am not shure if I understand every point. For example, why is centering/decentering the data necessary. This is not explained in the Tutorial. 

Thank you for the paper and hve a nice week too!

Patrick

Dear Patrick,

I'm not sure if all the followers of this question are interested in the details... so it is better if you write me your e-mail address privately and I can send you R script of my previous work (which is not readable and is commented in Hungarian, however, it works...).

I usually standardize data when the used method deals with several series in the same time and the series have different scale. I'm not sure if standardization is necessary for all the steps of the regression + semivariogram fitting + kriging procedure, but see the linked forum topics for some reasons.

HTH,

Ákos

https://stats.stackexchange.com/questions/7112/when-and-how-to-use-standardized-explanatory-variables-in-linear-regression

https://www.researchgate.net/post/Can_anyone_help_with_interpretation_kriging_prediction_and_variance_maps

Dear Akos

I want to downscale GRACE satellite data (111km by 111km) to the resulotion of TRMM (25km by 25km).

can you please suggest me some way how to do it R?

Dear Shoaib,

I think we should not mess up the original question-answers thread with this new question. Write a new question or contact me privately.

(Anyway, I don't know what GRACE satellite data are. But I guess that the interpolation you should do is similar to that of the WorldClim data.)

Ákos

Dear Joshua A. Spiers , RE "some people are against the use of variables from Worldclim": I think bioclimateic variables from WorldClim are good predictors if you use them in a clever way. There are some pitfalls that you may keep in mind. E.g. a new source of uncertainty has been discovered that is related to the selection of the specific month/quarter (e.g. wettest quarter in case of bio8). This uncertainty still holds for downscaled data. Please refer to this paper:

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