There is no definite way of doing such tasks. Off the top of my head, nearest neighbor and area averaging methods are easy to apply. Basically, it is about control of error propagation produced by sampling process. I might be able to deliver more detailed answers if you have specific cases to deal with.