Abstract
<title>Abstract</title> <p>Long-term planning and short-term operational models of bulk power systems require the characterization of weather-based uncertainties in electricity supply and demand. Resource adequacy, defined as the ability of a bulk power system to fully meet demand under all but the rarest and most extreme operating conditions, is typically measured by subjecting models to past observations (weather reanalysis), outputs from forward-looking global climate models, or both. While employing larger sample sizes can improve uncertainty characterization in stationary systems, in non-stationary systems (such as climate) this can lead to statistical bias in risk assessment, resulting in lowered reliability or imprudent capital investments. To address this, we introduce a universal, open-source synthetic weather data augmentation technique capable of producing meteorological ensembles of unlimited size for stress-testing models across the contiguous United States. The generated synthetic data preserves the statistical and temporal characteristics of empirical datasets by maintaining cross-correlations among weather variables and capturing empirical pattern distributions, while significantly improving the characterization of tail-risk events. We demonstrate our approach using the U.S. Eastern Interconnection, one of the world's largest bulk power systems. Our results indicate that synthetic maximum net loads exceed recent historical records (2004-2022) by an average of 28.5% and full historical records (1941-2022) by 21.5%. Our analysis of three major balancing authorities reveals hidden capacity shortfalls of at least 2.1 GW in PJM, 4.0 GW in NYIS, and 6.6 GW in MISO that cannot be detected from recent historical records, representing billions of dollars in capital investments needed to protect millions of households during extreme hot and cold events. Furthermore, our simulated locational marginal prices highlight plausible, high-cost operating conditions significantly more extreme than those found in empirical records. By uncovering critical vulnerabilities that remain obscured when relying solely on historical datasets, this framework provides a practical means of leveraging large sample sizes to inform robust system design and long-term resource adequacy.</p>