These are already predicted pretty well, and a goal of my work is to improve that. A limitation is that to predict well on a computer you need to lay a grid out over the whole world (conceptually) like a giant checkerboard, where each board square represents the average over its contents. Or think of an image being made up of millions of pixels. In traditional weather forecasting each pixel is about 100km wide, so we miss a lot! My approach is something like JPEG for image compression: denser pixels where it helps.
Comments