HRRRCast: A Data-Driven Emulator for Regional Weather Forecasting at Convection-Allowing Scales
Journal Article
Overview
abstract
Abstract; ; The High-Resolution Rapid Refresh (HRRR) model is a convection-allowing model used in operational weather forecasting across the contiguous United States (CONUS). To provide a computationally efficient alternative, we introduce HRRRCast, a data-driven emulator built with advanced machine learning techniques. HRRRCast includes two architectures: a ResNet-based model (ResHRRR), our main architecture, and a graph neural network–based model (GraphHRRR). ResHRRR utilizes convolutional neural networks enhanced with squeeze-and-excitation blocks and feature-wise linear modulation and supports probabilistic forecasting via the denoising diffusion implicit model (DDIM). To better handle longer lead times, we train a single model to predict multiple lead times (1, 3, and 6 h) and then use a greedy rollout strategy during inference. When evaluated on composite reflectivity over the full CONUS domain using ensembles of 3–10 members, ResHRRR outperforms HRRR forecast at the light rainfall threshold (20 dB; Z; ) and achieves competitive performance at moderate thresholds (30 dB; Z; ). Our work advances the pioneering StormCast model described in Pathak et al. by 1) training on the full CONUS domain, 2) training on multiple lead times to improve long-range performance, 3) using analysis data for training instead of the +1-h postanalysis data inadvertently used in StormCast, and 4) incorporating future Global Forecast System (GFS) weather states as inputs and adding a downscaling component that significantly improves long-lead forecast accuracy. Grid-based, neighborhood-based, and object-based verification metrics confirm improved storm placement, lower-frequency bias, and enhanced success ratios compared with HRRR. Additionally, HRRRCast’s ensemble forecasts maintain sharper spatial detail and reduced blurriness than deterministic baselines, with power spectra more closely matching HRRR analyses. Overall, HRRRCast represents a step toward efficient, data-driven regional weather prediction with competitive accuracy and ensemble capability.; ; ; Significance Statement; ; This study introduces HRRRCast, a data-driven emulator for the High-Resolution Rapid Refresh (HRRR) model, designed for regional, convection-allowing forecasting over the contiguous United States (CONUS) at 6-km resolution. HRRRCast uses a SE-ResNet-based architecture and diffusion modeling for generating probabilistic forecasts. Trained on HRRR analysis data with Global Forecast System (GFS) synoptic input—including future GFS states—it supports multilead time prediction (1, 3, 6 h) in a single model. HRRRCast outperforms HRRR in composite reflectivity skill at 20 dB; Z; and achieves competitive performance at 30 dB; Z; . It produces sharper forecasts, reduced bias, and more reliable ensembles, offering a scalable, cost-effective alternative to physics-based models for regional ensemble forecasting.; ;
