> In fact, that is what the researchers did in this case: They applied several statistical models of weather extremes to both models and real data and compared

> In fact, that is what the researchers did in this case: They applied several statistical models of weather extremes to both models and real data and compared them. Every model that failed to predict the actual measured heat wave went out. Wouldn't it be better to select the models which have had the most accurate predictions over as long a period as the required data are available? Mean squared error of temperature, or something. Promoting models primarily for their ability to predict an event that is currently happening seems like a terrible idea.