Integrating GNSS-Derived Zenith Wet Delay into a Weather Foundation Model Improves Precipitation Forecasting

2026-07-06

AI Summary

This study addresses the persistent systematic underestimation of extreme precipitation in current machine learning weather models and their limited integration of high-precision water vapor observations from Global Navigation Satellite Systems (GNSS). We propose the first direct incorporation of GNSS-derived zenith wet delay (ZWD) into the Aurora foundational weather model. Through multivariate joint training followed by precipitation-specific fine-tuning, our approach significantly enhances short-term forecasting skill for extreme rainfall events. Evaluated at the 99th percentile of precipitation intensity, the method improves the Equitable Threat Score by 8.8% and yields a more realistic representation of the precipitation power spectrum across scales—from the troposphere to planetary wavelengths—thereby effectively mitigating the long-standing issue of extreme precipitation underprediction.

Abstract

Global Navigation Satellite Systems (GNSS), best known for positioning, also serve weather science, as atmospheric water vapour delays their signals. This delay, the Zenith Wet Delay (ZWD), is a direct, all-weather measure of column moisture. Although assimilated into numerical weather prediction for decades, ZWD is not yet used by leading machine learning weather models (MLWM), despite addressing a known deficiency: the underestimation of severe precipitation. Here we present the first integration of GNSS-derived ZWD into Aurora, a state-of-the-art weather foundation model. Our extended Aurora learns ZWD with skill comparable to its pretrained variables. More importantly, including ZWD systematically improves forecasts when fine-tuning for six-hour accumulated precipitation. Gains grow with severity, reaching an 8.8\% increase in Equitable Threat Score at the 99th percentile, while the precipitation power spectrum becomes more realistic at synoptic and planetary scales. Direct GNSS observations therefore encode information that MLWM can exploit for high-impact precipitation.

Problem

• GNSS
• Zenith Wet Delay
• precipitation forecasting
• machine learning weather models
• severe precipitation

Innovation

• GNSS
• Zenith Wet Delay
• weather foundation model
• precipitation forecasting
• machine learning