Are you referring to what spatial resolution you would convert a point-based species occurrence data to grid- or raster-based? If so I would say it really depends on two primary factors: 1) What is the coarsest spatial resolution of any covariate or ancillary data you will be using in conjunction with your species data? 2) What is the resolution that makes sense biologically for your species of interest?

I realize this isn't really an answer to your question, but the answer really depends on the trade-offs involved in matching the biologically relevant resolution with the variability captured at that biologically relevant resolution in your other ancillary data.

I agree with Forrest's discussion. The sampling interval should also influence the selection of appropriate resolution in this case.

I agree with Forrest. Some bullet points: The right spatial resolution for Species Distribution Models (SDM) depend also to availables maps where apply your models (maps where derived your regressors to explain the occurrence of species). If work with animals species, normally descriptions of a buffer-m radius area around the sampled-point are made in order to quantify for example land-use composition and structural characteristics of the sampled sites (independent variables). The buffer size must be consistent with species-specific ecology (home-range, etc)

Some considerations to pay attention:

The sampled points must be separated between them uniformely, in order to avoid spatial autocorrelation problems....

Use a multi-spatial scale approach can be a good idea.

When work with land-cover maps the most common are 1:10000 scale. Some studies on biodiversity and for some animal species suggest a range of optimal spatial scale to define better the sampled sites. Look these papers, for example:

Schindler, S., von Wehrden, H., Poirazidis, K., Wrbka, T., Kati, V., 2013. Multiscale performance of landscape metrics as indicators of species richness of plants, insects

and vertebrates. Ecol. Indic. 31, 41–48 http://dx.doi.org/10.1016/j.ecolind.2012.04.012

Or this one

http://www.sciencedirect.com/science/article/pii/S1470160X13002264

Good work!

Federico

I agree with the others. But it is important to remember that the origin of your dataset is also very important to define the scale of your maps. If you are using secondary data (not collected by you or your team) it is very important for you to know the precision with which the data was collected. There is no point on modelling species distribution in a very fine scale if your occurrence points were not collected with that same precision. Many occurrence points from online databases have the coordinates of the municipality the species was registered, and not the actual point in which the species was observed.

Hope I could help.

One final consideration in addition to the very appropriate responses above is what you are doing with the species distribution model. Even if you have very fine scale information it may not be necessary to address region wide management issues, or if you have very course species or covariate information you may need to collect more information to address very fine scale management issues.

I would be very careful with the term 'resolution' here. Resolution is something you associate to continuous (often spatial)data, for example plant species distributions derived from satellite imageries. Since you are obtaining point occurrences of the species the proper term would be 'spatial lag' or 'point density'. As I understood from your question, you are objected to prepare a continuous surface from the species occurrences at points. And then the question of resolution comes into play. The answer to this question is mostly given my Forrest Steven. I would like to add a suggestion of finding out what is the minimum spatial lag where you can describe the spatial pattern of species occurrences with least uncertainty. The answer leads to the theories of geostatistics, though the habitat models somehow cover it. That would probably the best way of choosing resolution. And don't forget about the important spatial covariates when you model spatially any ecological phenomenon. However, as Forrest mentioned it's also important to find out which resolution makes sense biologically.

Hope that helps.

Semi-variograms are useful to look at spatial structure in your data, you can look at the semi-variogram in spatial regression coefficients also to consider the spatial structure of relationships

I agree with Avit and Neil: using geostatistics to determine the minimum spatial distance for your data set (i.e. the ideal) and then take what Forrest said about the spatial resolution of other available data sets and determine a "working" spatial resolution

I read all posted answers and also the reference articles recommend along with answer posted here. There are many ways but none of the answers was really useful for me. Finally I found determining resolution of species distribution is quite easy with 'dismo' package in R, which is also useful in removing duplicate point (pixel) within desired resolution. Integration of package 'raster' and 'sp' is necessary. I am posting this information hoping anybody beginning work in ecological niche modelling or distribution modelling could found this information useful.

My contribution to the discussion are two published SDMs in the same area partly using the same covariates. We empirically decided on the grid size (resolution) using most of the considerations provided in the previous answers. Our highly mobile animal SDMs, including climate covariates are much coarser therefore than the SDMs for plants. For the animal SDM a temporal resolution was selected. Both animal and plant SDMs were resolution-robust as published for the plants, but not for the animal, although we did test a range of resolutions for the latter as well. Attached is the article on the animal SDMs referring to the article on SDMs for (endemic) plants also available on ResearchGate.

Article Where the bears roam in Majella National Park, Italy