Use at least n = 30 to fulfil the Central Limit Theorem (CLT). The CLT in statistics states that, given a sufficiently large sample size, the sampling distribution of the mean for a variable will approximate a normal distribution regardless of that variable’s distribution in the population, and a sample of at least 30 will do that. But if the distribution is very skewed, the sample size need to be larger. So 100 sounds fine to me :)

https://statisticsbyjim.com/basics/central-limit-theorem/

Although many would recommend power analyses and a couple of other rules (such as 10xN-levels) there is no consensus approach to defining sample size. As Andrew Paul McKenzie Pegman has explained, there are some "magic numbers" of biostatistics (such as 30) that are good enough depending on the situation.

you may approach the issue empirically by starting with 5 replicates (bootstrap), than 10 or 20 and so on and see when your assessment criteria (e.g. AUC) remains stable. You may look at our SDM papers and be surprised on the number of replicates needed.

You can obtain the necessary minimum data by reading the software guide.

Example: MaxEnt is the minimum data required from the points of the presence of n=30.

With regard to the purpose of the study, you can change the settings and use the best result.

I agree with Igor Eloi and Hein Van Gils that the minimum sample size for a study depends on the actual situation. I found N=20 is good enough for some endangered species, since they are habitat specialists and 20 points represent their habitat preference well. Another important question is how many absence points you select for comparison. In my opinion, you’d better select hundreds to thousands points to represent environmental gradient. Balance (presence = absence) is not important here.

Considero que 10 réplicas es un número adecuado.