Mind the Residual Gap: Probabilistic Downscaling under Real-World Bias

Short summary

Researchers introduce ReMatch, a method for improving probabilistic downscaling—predicting high-resolution climate fields from coarse inputs. It uses optimal transport to align training and test-time residual distributions, fixing a fundamental bias issue in traditional mean-residual models. Tests on synthetic and real-world wind data show substantial improvements in calibration versus state-of-the-art baselines.

• •Addresses residual target misspecification problem in probabilistic downscaling for climate modeling
• •Uses optimal transport in PCA space to align training and test distributions
• •Achieves better calibration and reduces under-dispersion on real-world wind field data

Generated with AI, which can make mistakes.