The only thing that leaps out at me is more something to think about: their climatological average is taken over all years in the dataset. It would be nice to know how sensitive the results are to that: it wouldn't surprise me if it was essentially the same using just the years in the first half, and would preserve the ability to use the second half adls test data for any algorith.s. but I'm not sure it's actually worth a comment in the article.