New paper: Sequential deep learning models offer only modest gains for upscaling land–atmosphere fluxes. The main bottlenecks are spatial sampling bias, which amplifies regional uncertainties and affects model sensitivity, and feature selection, which has large leverage on results.

Excited about our #AI based runoff reconstruction for #switzerland ranging back to 1962. Lead by @bask0.bsky.social, co-authored by @hydrologywsl.bsky.social , @soniaseneviratne.bsky.social , William Aeberhard, Michael Schirmer. hess.copernicus.org/articles/29/...

Promotional image for the "deep learning in hydrology" session. On the left, shows the following description: "We welcome abstracts related to novel theory development, new methodologies, or practical applications of deep learning in hydrological modeling and process understanding.". On the right, shows an abstract visualization of a brain with some elements of an electrical circuit. There is a waterfall coming out from the middle part of that brain, forming a river that flows towards the front.

Are you working on the intersection of hydrology and deep learning? Consider submitting your abstract to our #EGU25 session on "Deep Learning in Hydrology".

See detailed session description at meetingorganizer.copernicus.org/EGU25/sessio...