A major challenge in geophysical and observational sciences is the representation and quantification of uncertainty in numerical predictions. Uncertainty stems from various sources, most relevantly from incomplete inclusion of all relevant physical mechanisms in the models and uncertainty in the initial and boundary conditions (T. N. Palmer, 2000). Important models for geophysical fluid dynamics, such as the two-dimensional Euler equations, quasi-geostrophic equations or rotating shallow water equations are derived from the three-dimensional Navier-Stokes equations. A sequence of simplifying assumptions is applied in order to reduce the complexity of the model to a more manageable level, while retaining main flow physics (Zeitlin, 2018). Stochastic extensions to these models have also been derived (Holm & Luesink, 2021). These approximate models are nevertheless rich in dynamics and contain a wide range of spatial and temporal scales. Numerically resolving the entire spectrum of scales is often not computationally feasible, meaning that either the complexity of the model should be reduced even further such that the resulting model is simple enough the be solvable numerically, or the complex model is represented on a coarse computational grid and unresolved scales are replaced by a sub-grid model. The latter option may be combined with stochastic forcing, which provides an effective way to represent unresolved scales in numerical simulations (Buizza et al., 1999;Frederiksen & Davies, 1997;Majda et al., 2001). The use of stochasticity as a means to represent the unresolved scales serves to restore some of the missing small-scale