Discussion about this post

User's avatar
G. Alex Janevski, PhD's avatar

Do you have any indications of why the predictions differ so much? I'm only passingly familiar with the two datasets, but as an example you noted a north-south trend, but I'd note there's also a roughly similar elevation trend. Precipitation correlates with elevation, and understanding this relationship is a piece of what PRISM does. It also uses elevation to predict/interpolate precipitation. ERA5 doesn't explicitly do that, afaik, and models precipitation based on satellite observations + weather stations? It's a little unclear to me exactly how it works, but it seems like looking at covariates or understanding if there's a geospatial trend in the data (e.g., due to elevation or another factor) which isn't explicitly captured in a model might cause it to underpredict, or, conversely, cause another model to overpredict if it's overfit to that same trend.

Expand full comment
1 more comment...

No posts