HourGlass: A probabilistic data-driven temporal downscaler for global and regional weather forecasting
- 1Norwegian Meteorological Institute , Oslo , Norway
Abstract
Many forecast applications require high frequency temporal resolution, yet most state-of-the-art data-driven weather forecasting systems operate at 6-hourly resolution. Although direct hourly forecasting is possible, it suffers from error accumulation and temporal inconsistency. We introduce HourGlass, a probabilistic data-driven temporal downscaling method that reconstructs the evolution between forecast states. HourGlass is trained using variants of the continuous ranked probability score (CRPS) preserving small-scale spatial variability while encouraging temporal consistency. Unlike existing deterministic temporal downscaling approaches, which tend to produce overly smooth fields, HourGlass generates realistic probabilistic forecasts. Training on forecast trajectories rather than reanalysis or analysis data also avoids the temporal inconsistencies present in datasets used by previous methods. We evaluate HourGlass in two settings: AIFS-HourGlass, applied globally to ECMWF's AIFS-Single and AIFS-ENS forecast systems, and Bris-HourGlass, applied regionally to MET Norway's high-resolution stretched-grid ensemble model, Bris. Verification against observations shows that both models retain the skill of their underlying forecasting systems while producing temporally coherent hourly forecasts with realistic small-scale variability. Case studies demonstrate physically consistent evolution during rapidly developing weather events, including extratropical cyclones and organised convection. Hourly precipitation remains challenging: HourGlass improves the spatial realism of precipitation fields but still underestimates the most intense extremes, a common limitation of data-driven weather forecasting models. These results demonstrate that HourGlass effectively bridges the gap between 6-hourly data-driven forecasts and the hourly products required for operational regional and global forecasting.
Related
- Graph-based probabilistic earthquake clustering and forecasting in Sumatra
- Advancing Near-Term Water Quality Forecasting with Explainable Machine Learning and Probabilistic Uncertainty Quantification
- SDECast: Probabilistic Weather Forecasting in Continuous Time with Neural SDEs
- Njord: A Probabilistic Graph Neural Network for Ensemble Ocean Forecasting
- Predictability-Guided Multiscale Probabilistic Forecasting of Wind Direction under Extreme Shear
BibTeX
@misc{ingstad2026hourglass,
title = {{HourGlass: A probabilistic data-driven temporal downscaler for global and regional weather forecasting}},
author = {Magnus Sikora Ingstad and Mariana Clare and Olav Ersland and Vera Gahlen and Håvard Homleid Haugen and Ophélia Miralles and Even Marius Nordhagen and Thomas N. Nipen and Ivar A. Seierstad and John Bjørnar Bremnes and Michael Maier‐Gerber and Zied Ben Bouallègue and Harrison Cook and Christian Lessig and Gert Mertes and Cathal O'Brien and Florian Pinault and Ana Prieto Nemesio and Matthew Chantry},
howpublished = {arXiv (Cornell University)},
year = {2026},
month = jul,
doi = {10.48550/arxiv.2607.11457},
url = {https://arxiv.org/abs/2607.11457},
note = {CC BY},
}
RIS
TY - UNPB
TI - HourGlass: A probabilistic data-driven temporal downscaler for global and regional weather forecasting
AU - Magnus Sikora Ingstad
AU - Mariana Clare
AU - Olav Ersland
AU - Vera Gahlen
AU - Håvard Homleid Haugen
AU - Ophélia Miralles
AU - Even Marius Nordhagen
AU - Thomas N. Nipen
AU - Ivar A. Seierstad
AU - John Bjørnar Bremnes
AU - Michael Maier‐Gerber
AU - Zied Ben Bouallègue
AU - Harrison Cook
AU - Christian Lessig
AU - Gert Mertes
AU - Cathal O'Brien
AU - Florian Pinault
AU - Ana Prieto Nemesio
AU - Matthew Chantry
PY - 2026
DA - 2026/07/13
PB - arXiv (Cornell University)
DO - 10.48550/arxiv.2607.11457
UR - https://arxiv.org/abs/2607.11457
AB - Many forecast applications require high frequency temporal resolution, yet most state-of-the-art data-driven weather forecasting systems operate at 6-hourly resolution. Although direct hourly forecasting is possible, it suffers from error accumulation and temporal inconsistency. We introduce HourGlass, a probabilistic data-driven temporal downscaling method that reconstructs the evolution between forecast states. HourGlass is trained using variants of the continuous ranked probability score (CRPS) preserving small-scale spatial variability while encouraging temporal consistency. Unlike existing deterministic temporal downscaling approaches, which tend to produce overly smooth fields, HourGlass generates realistic probabilistic forecasts. Training on forecast trajectories rather than reanalysis or analysis data also avoids the temporal inconsistencies present in datasets used by previous methods. We evaluate HourGlass in two settings: AIFS-HourGlass, applied globally to ECMWF's AIFS-Single and AIFS-ENS forecast systems, and Bris-HourGlass, applied regionally to MET Norway's high-resolution stretched-grid ensemble model, Bris. Verification against observations shows that both models retain the skill of their underlying forecasting systems while producing temporally coherent hourly forecasts with realistic small-scale variability. Case studies demonstrate physically consistent evolution during rapidly developing weather events, including extratropical cyclones and organised convection. Hourly precipitation remains challenging: HourGlass improves the spatial realism of precipitation fields but still underestimates the most intense extremes, a common limitation of data-driven weather forecasting models. These results demonstrate that HourGlass effectively bridges the gap between 6-hourly data-driven forecasts and the hourly products required for operational regional and global forecasting.
SN - 2331-8422
ER -