The use of multidimensional Langevin processes for stochastic uncertainty quantification in the NOAA Unified Forecast System (UFS)

<p>Modern numerical weather prediction (NWP) model forecasts for various applications require not only high-quality deterministic forecasts, but also information about forecast uncertainty.&#160; An ensemble forecast is commonly used to provide an estimation of forecast uncertainty.&#160; Since a great deal of the forecast uncertainty comes from dynamical processes not resolved or explicitly represented by NWP models, there is a need to correctly quantify and simulate NWP model uncertainty for an ensemble forecast to be useful and reliable.</p><p>We present an overview of a theoretical framework for simulating the uncertainty in unresolved physics in the NOAA Unified Forecast System (UFS).&#160; This framework is derived from the connection in mathematical physics between the Mori-Zwanzig formalism and multidimensional Langevin processes.&#160; It follows the correspondence principle, a philosophical guideline for new theory development, such that it can be shown that the previously implemented stochastic uncertainty quantification schemes in the UFS are particular cases of this framework.&#160; We will show an example of how we have used this framework to develop a new process-level stochastic uncertainty quantification scheme in the UFS.&#160; We will also present a preliminary performance comparison of these previously-implemented schemes with the newly-developed process-level scheme in the UFS ensemble predictions on short, medium and sub-seasonal time scales.</p>