Learning vertical coordinates via automatic differentiation of a dynamical core
Tim Whittaker, Seth Taylor, Elsa Cardoso-Bihlo, Alejandro Di Luca, Alex Bihlo · 2025
Read Learning vertical coordinates via automatic differentiation of a dynamical core by Tim Whittaker, Seth Taylor, Elsa Cardoso-Bihlo, Alejandro Di Luca, Alex Bihlo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-17877.txt.
Description
Tim Whittaker and four coauthors tackle a grid problem in weather models. Over steep mountains, height based layers bend and can stir up fake winds in the forecast. The usual fixes rest on hand tuned curves that stay fixed once set.
Their idea lets the model learn its own vertical grid inside a solver that can be trained end to end. Exact grid terms come from automatic differentiation, which removes a source of math error. On standard tests the learned grid cut errors by a factor of 1.4 to 2 and erased the fake vertical wind stripes over steep ground. Better grids mean cleaner forecasts near mountains.
Catalog record
| CREATOR | Tim Whittaker, Seth Taylor, Elsa Cardoso-Bihlo, Alejandro Di Luca, Alex Bihlo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-17877