Probabilistic deep learning for seasonal forecasting

Abstract:

The path toward realizing the potential of seasonal forecasting and its socioeconomic benefits depends heavily on improving general circulation model-supported dynamical forecast systems. To accelerate forecast improvement, it is crucial to set up forecast benchmarks, and clarify forecast limitations posed by initialization errors, model formulation deficiencies, and inherent climate variability. With limited observations to support forecast diagnosis, these tasks prove challenging. Here, we develop a deep learning-powered variational inference methodology, drawing on a wealth of existing climate simulations to enhance seasonal forecast capability and forecast diagnosis. By leveraging complex physical relationships encoded in climate simulations, our probabilistic forecast model demonstrates favorable performance compared to state-of-the-art dynamical forecast systems in global seasonal forecast of precipitation and temperature. We apply this probabilistic forecast methodology to quantify the impacts of initialization errors and model formulation deficiencies in dynamical seasonal forecasts. We introduce a saliency analysis approach to efficiently identify the key predictors that influence seasonal variability. Furthermore, by explicitly modeling uncertainty using Variational Bayes, we give a more definitive answer to how the El Nino Southern Oscillation, a key mode of variability in the climate system, modulates global seasonal predictability. Our work contributes to the following two aspects. First, we demonstrate that probabilistic machine learning, in particular, deep learning-based variational inference, is a powerful tool for leveraging the rich information from climate simulations to inform seasonal forecast and forecast uncertainty. Second, we provide efficient approaches for verifying and diagnosing the ever-complicated dynamical forecast systems, pinpointing clear paths toward forecast improvement. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344.