Recently I am using a random forest model to estimate remote sensing surface temperature. assume that I have 15 independent variables and I pick up 5 out of this in every individual turn. so the maximum number of constructed trees would be the combination of 5 out of 15 or much larger than that?

in addition, it is mentioned that the more the better (which means we don't have any maximum values?), and some mentioned that more trees don't necessarily mean the better prediction. Since I am doing my methodology, it could be better to clarify the reason behind the specific number of trees and how it is selected, so if that is possible please refer to a peer-reviewed article so I can cite the findings.