REPRESENTING MODEL UNCERTAINTY IN WEATHER AND CLIMATE PREDICTION

TLDR

Weather and climate predictions are uncertain due to uncertain initial conditions and model representations, and ensemble prediction systems estimate the flow‑dependent growth of this uncertainty, requiring representation of all uncertainty sources. The paper discusses methods for representing model uncertainty, focusing on a new paradigm that uses computationally efficient stochastic‑dynamic schemes to model unresolved processes. The authors describe a new paradigm that represents unresolved processes with computationally efficient stochastic‑dynamic schemes. The authors argue that multimodel ensembles outperform single‑model ensembles yet still fail to fully capture model uncertainty.

Abstract

▪ Abstract Weather and climate predictions are uncertain, because both forecast initial conditions and the computational representation of the known equations of motion are uncertain. Ensemble prediction systems provide the means to estimate the flow-dependent growth of uncertainty during a forecast. Sources of uncertainty must therefore be represented in such systems. In this paper, methods used to represent model uncertainty are discussed. It is argued that multimodel and related ensembles are vastly superior to corresponding single-model ensembles, but do not provide a comprehensive representation of model uncertainty. A relatively new paradigm is discussed, whereby unresolved processes are represented by computationally efficient stochastic-dynamic schemes.