Of course, there is almost no discussion about methodology and data sets, which is just about all there is when it comes to these modeling studies. The closest they come is saying that the paper preceeded the pacnorwest heat wave.<p>It seems like the closest you could hope to come to scientific method in these studies is to try to divide the world's geography into a set of geographical pairs that could be considered climatically similar but distant enough from one another to be decorrelated in terms of weather systems. For example, maybe Mediterranean climates on the Mediterranean paired with comparable regions in coastal California. Then, use a random selection of one region from each pair in all the set of pairs to be used as training data, with the other component of each pair for testing.<p>Maybe this is already done? I recognize that drawing such equivalency is impossible, but it seems like it might be an improvement nonetheless.