affect the weather

Since our ability to model and predict the weather is improving (and often quite good already), and since computation is continuing to get cheaper, this suggests the following way of taking advantage.

There are things which man could do which might affect the weather. The (not very successful) attempts to do iodide seeding of clouds to cause rain in the 50s are an example. But one can imagine "painting" the tops of clouds to change their reflectivity, or flying planes to have a modest effect on air movement, or other things like that.

So, suppose we run many copies of a weather model in parallel, each modified so that its initial state is the "current" state modified with the effect of some putative change caused by man. If we now run those models, it is possible that one of them will show a state 24-48 hours later which is in some sense "more desirable" than the other models. I use desirable here to mean, say, less rain in an already flooded region, or more rain in a drought area, or something like that.

If the computation is fast enough and reliable enough, one can imagine now performing the action which corresponds to the more desirable outcome. In fact, to account for the time to run the model, you'd want to start it and run it forward for some time, then inject the man-made input, and then run it some more, so that by the time you knew which effect you wanted, the state at which the weather modification was applied was still slightly in the future.

Probably, most modifications would have little or no effect, and it would take some time and experience to learn what things we could do which had a lot of "leverage". Additionally, the weather models would have to be fairly reliable and numerically stable.