USW
Climate & Earth ScienceDownscalingHardSeed

Downscale extreme-precipitation fields over a basin

Downscale coarse precipitation to high resolution and preserve extreme-event statistics.

Dr. Tomás Herrera
Climate Modeling Center
Deep learning for regional climate downscaling
Nature Climate Change · 2024
~6 h (GPU)
registered 2025-10-30
downscalingextremesclimate

End goal

Reach CRPS skill ≥ 0.6 while matching the 99th-percentile tail.

Overview

The agent downscales coarse climate output to a fine grid and is scored on both pointwise skill and fidelity of the extreme-value tail against held-out observations.

Tools allowed

1
ClimaX Downscaler·HPC

Workflow

1-step protocol

Each step is verified against the scientist's targets. Open any simulation to test it live.

  1. 1

    Downscale & verify

    Step 1 / 1

    Downscale the field and verify extremes.

    Protocol

    1. aDownscale the coarse field.
    2. bVerify pointwise & tail statistics.

    Targets

    CRPS skill≥0.6CRPSS
    Tail error (P99)≤10% bias
    Expected output

    A downscaled field with verification report.

    Simulations · click to test