1 of 1

Deep Bayesian Climate Emulator�A Bayesian Deep Learning Approach to Near‐Term Climate Prediction

1

Scientific Achievement

We present a probabilistic modeling framework for near-term climate forecasting based on deep learning modules. In addition to providing useful measures of predictive uncertainty, Bayesian versions of deep learning models outperform their deterministic counterparts in terms of predictive skill.

Significance and Impact

Following concerted national and international efforts over the past 70 years to model climate, comprehensive climate models have emerged as a powerful tool in helping unravel and better understand the myriad processes underlying climate and climate change. We incorporate Bayesian inference into cutting-edge network architecture, giving current deep learning models uncertainty estimation power.

Technical Approach

• We developed a Bayesian convolutional encoder–decoder deep network for uncertainty quantification.
• Our study considers Stein variational inference for exploring the high-dimensional posterior distribution of the network parameters.

1. Luo, X., Nadiga, B.T., Park, J.H., Ren, Y., Xu, W. and Yoo, S., 2022. A Bayesian Deep Learning Approach to Near‐Term Climate Prediction. Journal of Advances in Modeling Earth Systems, 14(10), p.e2022MS003058.

Figure 1. Prediction results

Figure 2. Predictive uncertainty quantification results