As we all know, physics is a science of estimation and approximation. For example, if one were to see a tree a distance away, one should be able to estimate its height without any empirical evidence. On such a macro level it may seem appropriate. My question today is therefore this: how do we cultivate the ability to estimate objects with minimal deviation from the known, calculated quantity?
What comes to mind is dimensional analysis. Starting from the units of the quantity one wants to calculate (e.g. the velocity of a planet around a star), one tries to jumble up all relevant quantities and find the right combination of input parameters.