Uncertainty Quantification for Data-Driven Weather Models

Abstract Artificial intelligence (AI)-based data-driven weather forecasting models have experienced rapid progress over the last years. Recent studies, with models trained on reanalysis data, achieve impressive results and demonstrate substantial improvements over state-of-the-art physics-based numerical weather prediction models across a range of variables and evaluation metrics. Beyond improved predictions, the main advantages of data-driven weather models are their substantially lower computational costs and the faster generation of forecasts once a model has been trained. However, most efforts in data-driven weather forecasting have been limited to deterministic, point-valued predictions, making it impossible to quantify forecast uncertainties, which is crucial in research and for optimal decision-making in applications. Our overarching aim is to systematically study and compare uncertainty quantification methods to generate probabilistic weather forecasts from a state-of-the-art deterministic data-driven weather model, Pangu-Weather. Specifically, we compare approaches for quantifying forecast uncertainty based on generating ensemble forecasts via perturbations to the initial conditions, with the use of statistical and machine learning methods for post hoc uncertainty quantification. In a case study on medium-range forecasts of selected weather variables over Europe, the probabilistic forecasts obtained by using the Pangu-Weather model in concert with uncertainty quantification methods show promising results and provide improvements over ensemble forecasts from the physics-based ensemble weather model of the European Centre for Medium-Range Weather Forecasts for lead times of up to 5 days. Significance Statement Weather forecasts have long relied on physics-based numerical weather prediction (NWP) models, which simulate atmospheric processes using complex equations. However, a groundbreaking shift is underway. Data-driven artificial intelligence (AI) models are now outperforming traditional NWP systems. These AI models predict future weather states based solely on historical data, offering a faster generation of forecasts at significantly reduced computational costs. However, a critical limitation persists: In contrast to NWP models which are often used to generate ensemble predictions, most AI weather models provide deterministic predictions only and thus lack the ability to quantify forecast uncertainties. Our research addresses this gap by systematically exploring uncertainty quantification methods to generate probabilistic weather forecasts from a state-of-the-art deterministic data-driven model, Pangu-Weather.