I think there is many way to attempt your problem ..... Are You familiar with Maximum Entropy Modeling .... please see web page http://homepages.inf.ed.ac.uk/lzhang10/maxent.html . I think Maxent. is a indirect modelling of your problem if you have proper GPS location . If you are familiar with any GIS soft wares like Arc GIS there is direct way to model and mapping , if you have proper GPS points of the distribution .... you can go with spatial interpolation ... in the Spatial Analyst extension of Arc GIS

Both Alex and Alessandro provide good information. In particular, Alessandro is correct in that first it is best to try to identify what factors are correlated with frog occurrence.

If this is not feasible (you don't have any information on the sites other than their location), however, you could simply perform a spatial analysis to interpolate the spatial pattern of occurrence across the landscape. Kriging, and inverse distance weighting (IDW) are two methods for this. However, this won't necessarily provide any more information than you already have (where do frogs occur). These are akin to the spatial interpolations that Alex suggested.

There are a number of methods for generating species-habitat relationships, including MaxEnt as mentioned by Alex. However, MaxEnt uses presence-only data, which is optimal for certain types of surveys and species (riverine fish is a good example). However, if you have presence and absence data, other approaches are more powerful (because they make use of the absence data). These approaches include genearlized linear models (GLM), generalized additive models (GAM), random forests, boosted regression trees (BRT), among others (including BioClim and genetic algorithms mentioned by Alessandro). All of these approaches allow you to determine factors that appear important in determining your frog distribution, and from these responses you can generate a predictive map. All of these are implementable in R, as well as the prediction maps (although I admit I tend to use ArcGIS for the mapping).

I personally have used GAMs (a form of regression) to develop such models (and predictions of habitat suitability across a landscape).

Furey, N.B., Rooker, J.R., 2013. Spatial and temporal shifts in suitable habitat of juvenile southern flounder (Paralichthys lethostigma). Journal of Sea Research 76, 161–169.

I hope this helps.

First what is the scale of your study area that you want to model the population in? Are you talking a small population within an area in the magnitude of 1000Km sq. then the MAXENT and BIOCLIM methods are not the answer for your problem. These methods are valid for large geographical extents where the variation in climatic and environmental conditions are significant to allow for habitats to vary with in them. But if you are addressing a smaller study are (under 1000Km sq.) then you need to measure your own climatic and environmental variables and treat them as covariables. The actual modelling process can be done using Genaeralized additive models in the free statistical package R using the library (mgcv) where you use the values of your species abundance for example as response variable and the values of the other covariables as predictor variables.

ex:

gam(response~ s(pred1) + s(Pred2) + s(Pred3))