It all depends on your question. If your research question aims to predict across unsampled areas, then you should do geographic subsamples. If, on the other hand, you are setting out only to anticipate the distribution of greater densities of points, then you could do a random subsample.

Now, if you are asking about numbers of points for calibration versus evaluation, then one generally wants half and half, except if you have a truly huge number of points available (you do not!). In fact, you are down into the "Pearson Zone," of low numbers of points. See

Pearson, R. G., C. J. Raxworthy, M. Nakamura, and A. T. Peterson. 2007. Predicting species distributions from small numbers of occurrence records: A test case using cryptic geckos in Madagascar. Journal of Biogeography 34:102-117.

This paper presents a low-N test for spatial predictions, and even some software to make it happen. Hope that this helps. All the best, ATP

It all depends on your question. If your research question aims to predict across unsampled areas, then you should do geographic subsamples. If, on the other hand, you are setting out only to anticipate the distribution of greater densities of points, then you could do a random subsample.

Now, if you are asking about numbers of points for calibration versus evaluation, then one generally wants half and half, except if you have a truly huge number of points available (you do not!). In fact, you are down into the "Pearson Zone," of low numbers of points. See

Pearson, R. G., C. J. Raxworthy, M. Nakamura, and A. T. Peterson. 2007. Predicting species distributions from small numbers of occurrence records: A test case using cryptic geckos in Madagascar. Journal of Biogeography 34:102-117.

This paper presents a low-N test for spatial predictions, and even some software to make it happen. Hope that this helps. All the best, ATP

I would use them all because n = 30 is usually the minimum for most experiments :)

Thank you Peterson!!! I have another doubt...My aim is to predict the present potential distribution of a tree based on the climatic conditions prevailing at the occurrence records!!!. I do not have an independent test data set of other species. So from the 23 points, if I use 25% (5 points) as random test sample, would it be fine?. I use the sub-sample method for replication.

Yeah!!! Thanks a lot Peterson!!!

I understood clearly after reading your paper!!! I will proceed with 50% training and 50& test data as total number of samples is low 23.

Thank you

I suggest the reading of this paper: doi: 10.1111/ecog.01509

According to the authors, the minimum number of points required to get a good performance (AUC>0.85) depends also on the prevalence of your species within your study area...But it could be as low as 13 cells occupied.

Cheers