Way of topic but interesting to me.
You know that ageing fish by looking at scales takes a long time, whilst weighing them takes seconds. I developed a program where say 80 fish where aged and 1000 weighed. The 80 where used to construct a normal distribution for age and weight, then the 1000 where put through the filter so that say a fish weighing a kilo might add 0.95 to the 4 year old counter and 0.05 to the 5 year old. Since no one will believe 9.63 fish are ten years old a kludge at the end to make them all integers and the total the same. Cutting edge stuff in my day, I take it packaged software now.
I was modelling ground water, same problems, you know only the rain that fell in a few places, evaporation and transiparation no data. The ground water level was known in a few observation wells, and the knowledge of what is under the ground is incomplete. You make the model behave the same as all known data, then next year tweek it again when you have more data. You can actually discover geological features at this stage, i.e. the only possible way that could happen is if there is something impervious 'clay' just there. I would have a fair degree of confidence that any ones ground water model would be about 95% accurate in predicting what if's.