Reproduce “Wet-Biased Data Pipelines Mislead the Training and Evaluation of Generative Precipitation Nowcasting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Christoph Metzl, Stefan Kapsreiter, Kianusch Vahid Yousefnia, Virginia Poli, Richard Müller, Frank Krüger, Tobias Bölle
The authors test how target-dependent sampling affects precipitation nowcast training and evaluation. Theory predicts wet-biased selection shifts the distribution learned by an ideal generative model. They compare the same frozen diffusion model on uniform and wet-biased test sets: Germany’s signed normalized mean-rank bias changes from −0.065 to +0.260, with the sign flip replicated in Italy. This isolates test-selection effects, not training causation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.