Reproduce “DySCo: Dynamically consistent data-driven downscaling of extremes in climate projections” 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
S. Stamatelopoulos, M. Wang, I. Lopez-Gomez, L. Zepeda-Nunez, Z. Y. Wan, R. Carver, F. Sha, T. P. Sapsis
DySCo combines data-driven trajectory alignment with probabilistic super-resolution to downscale coarse climate-model projections while preserving their event-by-event relationship to the driving simulation. Across 250 summer simulation-years over the contiguous United States, it improves dynamical alignment over unsupervised generative downscaling, with pressure and humidity anomaly correlations of about 0.73 versus 0.15 for GenFocal, while recovering local spatial patterns and tropical-cyclone landfall distributions.
Key insights
- Approximate trajectory pairing can recover much of the driver-output alignment associated with supervised learning while retaining the flexibility of generative models.
- The method preserves large-scale event structure through debiasing and adds unresolved geographic detail through stochastic super-resolution, including coastline effects and localized temperature patterns.
- Alignment and distributional accuracy can trade off: DySCo improves dynamical consistency and pressure statistics, but other methods perform better for some variables and for the nonlinear heat-index diagnostic.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.