Stochastic Climate Theory and Modelling

Stochastic methods are a crucial area in contemporary climate research and are increasingly being used in comprehensive weather and climate prediction models as well as reduced order climate models. Stochastic methods are used as subgrid-scale parameterizations as well as for model error representation, uncertainty quantification, data assimilation and ensemble prediction. The need to use stochastic approaches in weather and climate models arises because we still cannot resolve all necessary processes and scales in comprehensive numerical weather and climate prediction models. In many practical applications one is mainly interested in the largest and potentially predictable scales and not necessarily in the small and fast scales. For instance, reduced order models can simulate and predict large scale modes. Statistical mechanics and dynamical systems theory suggest that in reduced order models the impact of unresolved degrees of freedom can be represented by suitable combinations of deterministic and stochastic components and non-Markovian (memory) terms. Stochastic approaches in numerical weather and climate prediction models also lead to the reduction of model biases. Hence, there is a clear need for systematic stochastic approaches in weather and climate modelling. In this review we present evidence for stochastic effects in laboratory experiments. Then we provide an overview of stochastic climate theory from an applied mathematics perspectives. We also survey the current use of stochastic methods in comprehensive weather and climate prediction models and show that stochastic parameterizations have the potential to remedy many of the current biases in these comprehensive models.

💡 Research Summary

This review paper provides a comprehensive synthesis of why stochastic approaches are indispensable for modern weather and climate modeling, and surveys the state‑of‑the‑art developments across laboratory experiments, theoretical foundations, and operational modeling. The authors begin by noting that despite exponential growth in computational power, essential processes such as tropical convection, gravity‑wave drag, and cloud microphysics remain unresolved in global climate models (GCMs) because the required grid spacing is still far beyond practical limits. Traditional deterministic sub‑grid parameterizations assume a one‑to‑one mapping from resolved large‑scale states to the aggregate effect of unresolved scales, an assumption increasingly contradicted by observations showing that a single large‑scale configuration can correspond to many possible small‑scale realizations.

Section 2 presents concrete laboratory evidence from rotating annulus experiments. By driving a two‑layer isothermal fluid with a differentially rotating lid, the authors reproduce large‑scale baroclinic wave modes (wave numbers 1–3) and, crucially, a weak inertia‑gravity wave field that is not directly represented in the quasi‑geostrophic dynamics. When the small‑scale waves are present, the system exhibits stochastic transitions between baroclinic modes (e.g., from wave‑2 to wave‑1) with a finite probability, whereas in their absence the system remains locked in the initial mode. Numerical simulations of a quasi‑geostrophic model confirm that adding a weak white‑noise forcing—intended to mimic the unresolved inertia‑gravity waves—reproduces the observed transitions. This experimental‑numerical synergy demonstrates that small‑scale fluctuations can act as effective stochastic forcing on the large‑scale flow, a phenomenon the authors liken to noise‑induced transitions.

Section 3 builds the theoretical framework. Starting from Hasselmann’s classic two‑time‑scale separation, the full state vector z is decomposed into slow climate components x and fast weather components y. The fast dynamics are assumed to be rapidly mixing, allowing the quadratic self‑interaction B(y,y) to be replaced by a combination of additive white noise and state‑dependent (multiplicative) noise. The resulting stochastic differential equation (SDE) for the slow variables contains: (i) the original deterministic linear and quadratic terms, (ii) a deterministic cubic term that acts as nonlinear damping, (iii) additive noise representing fast‑fast interactions, and (iv) multiplicative noise reflecting fast‑slow coupling. Importantly, because the climate system lacks a clear spectral gap, the reduction inevitably generates non‑Markovian memory kernels that capture the finite autocorrelation time of the fast variables. Rigorous derivations of these results are credited to works by Kurtz, Papanicolaou, Pavliotis & Stuart, among others, and are summarized in accessible reviews (e.g., Givon et al., Pavliotis & Stuart).

Section 4 offers a concise primer on stochastic processes, emphasizing the distinction between Ito and Stratonovich calculus, the role of the Fokker‑Planck equation, and the physical interpretation of multiplicative noise (e.g., larger fluctuations under strong wind conditions). The authors critique early linear‑inverse‑model (LIM) approaches that linearize dynamics and add white noise, noting that such models can only reproduce Gaussian statistics and therefore fail to capture extreme events. Recent advances include the development of nonlinear normal‑form stochastic climate models that respect physical constraints such as global stability, and the use of data‑driven techniques (e.g., multi‑level regression, stochastic averaging, optimal prediction, Markov chains) to estimate model parameters while preserving physical realism.

Section 5 surveys the implementation of stochastic parameterizations in operational forecasting systems. Examples include the Stochastic Kinetic Energy Backscatter Scheme (SKEBS) in the ECMWF Integrated Forecast System, stochastic convection schemes, stochastic gravity‑wave drag, and stochastic surface flux parameterizations. These schemes have been shown to reduce systematic biases, increase ensemble spread to realistic levels, and improve the representation of low‑frequency climate variability such as the North Atlantic Oscillation and ENSO. In data assimilation, stochastic model error representations are incorporated into Ensemble Kalman Filters and Particle Filters, allowing simultaneous treatment of initial‑condition uncertainty and structural model error.

The concluding section outlines future research directions: (1) empirical validation and quantification of non‑Markovian memory kernels, (2) development of scalable Bayesian or variational methods for estimating stochastic model parameters in high‑dimensional state spaces, and (3) integration of machine‑learning‑based, data‑driven components with physically grounded stochastic parameterizations to form hybrid models. The authors argue that stochastic climate theory is not merely an ad‑hoc addition of noise but a systematic framework that captures unresolved scale interactions, memory effects, and non‑Gaussian statistics, thereby offering a pathway to overcome structural limitations of current climate models and substantially enhance predictive skill.