Bayesian Deep Learning for Convective Initiation Nowcasting Uncertainty Estimation

2025-07-22

AI Summary

This study addresses insufficient uncertainty quantification in 0–1 hour convective initiation (CI) nowcasting, leveraging GOES-16 infrared imagery. We systematically evaluate five Bayesian deep learning approaches—including weight ensembles with Monte Carlo (MC) dropout, MOPED-constrained priors, model priors, and empirical Bayes—against a deterministic ResNet baseline. Our key contributions are: (i) a novel weight ensemble framework jointly incorporating MC dropout, which significantly improves probabilistic forecast skill while achieving superior uncertainty calibration; and (ii) integration of the MOPED strategy to constrain the hypothesis space, enhancing generalization under complex meteorological conditions and long-horizon robustness. Results demonstrate that most Bayesian methods outperform the ResNet baseline; notably, the weight ensemble + MC dropout method achieves the best overall performance, effectively disentangling forecast errors and reliably identifying high-uncertainty samples.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Intelligent Robots: State Estimation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

This study evaluated the probability and uncertainty forecasts of five recently proposed Bayesian deep learning methods relative to a deterministic residual neural network (ResNet) baseline for 0-1 h convective initiation (CI) nowcasting using GOES-16 satellite infrared observations. Uncertainty was assessed by how well probabilistic forecasts were calibrated and how well uncertainty separated forecasts with large and small errors. Most of the Bayesian deep learning methods produced probabilistic forecasts that outperformed the deterministic ResNet, with one, the initial-weights ensemble + Monte Carlo (MC) dropout, an ensemble of deterministic ResNets with different initial weights to start training and dropout activated during inference, producing the most skillful and well-calibrated forecasts. The initial-weights ensemble + MC dropout benefited from generating multiple solutions that more thoroughly sampled the hypothesis space. The Bayesian ResNet ensemble was the only one that performed worse than the deterministic ResNet at longer lead times, likely due to the challenge of optimizing a larger number of parameters. To address this issue, the Bayesian-MOPED (MOdel Priors with Empirical Bayes using Deep neural network) ResNet ensemble was adopted, and it enhanced forecast skill by constraining the hypothesis search near the deterministic ResNet hypothesis. All Bayesian methods demonstrated well-calibrated uncertainty and effectively separated cases with large and small errors. In case studies, the initial-weights ensemble + MC dropout demonstrated better forecast skill than the Bayesian-MOPED ensemble and the deterministic ResNet on selected CI events in clear-sky regions. However, the initial-weights ensemble + MC dropout exhibited poorer generalization in clear-sky and anvil cloud regions without CI occurrence compared to the deterministic ResNet and Bayesian-MOPED ensemble.

Problem

• Evaluate Bayesian deep learning for convective initiation nowcasting uncertainty
• Compare probabilistic forecasts to deterministic ResNet baseline
• Assess calibration and error separation in uncertainty estimation

Innovation

• Initial-weights ensemble with MC dropout
• Bayesian-MOPED ResNet ensemble
• Probabilistic forecasts outperform deterministic ResNet