Bayesian Deep Learning for Convective Initiation Nowcasting Uncertainty Estimation

TL;DR

This paper compares Bayesian deep learning methods to a deterministic ResNet for convective initiation nowcasting, showing Bayesian methods generally improve probabilistic forecasts and uncertainty calibration, with specific ensembles performing best.

Contribution

It evaluates and compares multiple Bayesian deep learning approaches for convective initiation nowcasting, introducing the Bayesian-MOPED ensemble to improve forecast skill and calibration.

Findings

Bayesian methods outperformed the deterministic ResNet in probabilistic forecast skill.

The initial-weights ensemble + MC dropout was the most skillful and well-calibrated.

Bayesian-MOPED ensemble improved forecast skill by constraining hypothesis search.

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.