A group-kinetic theory with a closure by memory loss for modeling turbulence in the atmosphere and the oceans

Summary The principle of the group-kinetic method is elucidated. This method of renormalization serves as the basis for analyzing the spectral structure of turbulence. The spectral distributions include the Kolmogoroff lawk ^–5/3 for isotropic turbulence, the power lawk ^–1 for shear turbulence, the spectrum for stratified turbulence not in the power law form, the power lawk ^–3 for two-dimensional geostrophic turbulence, and the power lawsk ^–3,k ^–2 andk ^–5 for two-dimensional Rossby wave turbulence with uniform and differential rotations. We discuss a spectrum-dependent modeling in reference to the problems of the universal functions and parameters in the similarity theories for the atmospheric surface layer and the planetary boundary layer. A renormalization-based modeling of atmospheric turbulence is proposed.     

Turbulence decay in stratified and homogeneous marine layers

A spectral approach is applied to shear-induced turbulence in stratified layers. A system of spectral equations for stationary balance of turbulent energy and temperature variances was deduced in the vicinity of the local shear scale L_U = (ε/U_Z³)^1/2. At wavenumbers between the inertial-convective (k^−5/3) and wak turbulence (k⁻³) subranges, additional narrow spectral intervals—‘production’ subranges—may appear (E k⁻¹, E_T k⁻²). The upper boundary of these subranges is determined as L_U, and the lower boundaries as L_R (ε/U_ZN²)^1/2(χ/T_Z²). It is shown that the scale L_U is a unique spectral scale that is uniform up to a constant value for every hydrophysical field. It appears that the spectral scale L_U is equivalent to the Thorpe scale L_Th for the active turbulence model. Therefore, if turbulent patches are generated in a background of permanent mean shear, a linear relation between temperature and mass diffusivities exists. In spectral terms, the fossil turbulence model corresponds to the regime of the Boldgiano-Obukhov buoyancy subrange (E k^−11/5, E_T k^−7/5). During decay the buoyancy subrange is expanded to lower and higher wavenumbers. At lower wavenumbers the buoyancy subrange is bounded by L_** = 3(χ^1/2/N^1/2T_Z), which is equivalent to the Thorpe scale L_Th. In such a transition regime only, when the viscous dissipation rate is removed from the set of main turbulence parameters, the Thorpe scale does not correlate with the buoyancy scale L_N ε^1/2/N^3/2 and fossil turbulence is realized. Oceanic turbulence measurements in the equatorial Pacific near Baker Island confirm the main ideas of the active and fossil turbulence models.     

New Approaches in Two-equation Turbulence Modelling for Atmospheric Applications

The turbulence closure in atmospheric boundary-layer modelling utilizing Reynolds Averaged Navier–Stokes (RANS) equations at mesoscale as well as at local scale is lacking today a common approach. The standard kɛ model, although it has been successful for local scale problems especially in neutral conditions, is deficient for mesoscale flows without modifications. The k–ɛ model is re-examined and a new general approach in developing two-equation turbulence models is proposed with the aim of improving their reliability and consequently their range of applicability. This exercise has led to the replacement of the ɛ-transport equation by the transport equation for the turbulence inverse length scale (wavenumber). The present version of the model is restricted to neutrally stratified flows but applicable to both local scale and mesoscale flows. The model capabilities are demonstrated by application to a series of one-dimensional planetary boundary-layer problems and a two-dimensional flow over a square obstacle. For those applications, the present model gave considerably better results than the standard k–ɛ model.     

Numerical Modelling of the Turbulent Flow Developing Within and Over a 3-D Building Array, Part I: A High-Resolution Reynolds-Averaged Navier—Stokes Approach

A study of the neutrally-stratified flow within and over an array of three-dimensional buildings (cubes) was undertaken using simple Reynolds-averaged Navier—Stokes (RANS) flow models. These models consist of a general solution of the ensemble-averaged, steady-state, three-dimensional Navier—Stokes equations, where the k-ε turbulence model (k is turbulence kinetic energy and ε is viscous dissipation rate) has been used to close the system of equations. Two turbulence closure models were tested, namely, the standard and Kato—Launder k-ε models. The latter model is a modified k-ε model designed specifically to overcome the stagnation point anomaly in flows past a bluff body where the standard k-ε model overpredicts the production of turbulence kinetic energy near the stagnation point. Results of a detailed comparison between a wind-tunnel experiment and the RANS flow model predictions are presented. More specifically, vertical profiles of the predicted mean streamwise velocity, mean vertical velocity, and turbulence kinetic energy at a number of streamwise locations that extend from the impingement zone upstream of the array, through the array interior, to the exit region downstream of the array are presented and compared to those measured in the wind-tunnel experiment. Generally, the numerical predictions show good agreement for the mean flow velocities. The turbulence kinetic energy was underestimated by the two different closure models. After validation, the results of the high-resolution RANS flow model predictions were used to diagnose the dispersive stress, within and above the building array. The importance of dispersive stresses, which arise from point-to-point variations in the mean flow field, relative to the spatially-averaged Reynolds stresses are assessed for the building array.     

The effect of idealized water waves on the turbulence structure and kinetic energy budgets in the overlying airflow

The influence of an idealized moving wavy surface on the overlying airflow is investigated using direct numerical simulations (DNS). In the present simulations, the bulk Reynolds number is Re = 8000 (; where U₀ is the forcing velocity of the flow, h the height of the domain and v the kinematic viscosity) and the phase speed of the imposed waves relative to the friction velocity, i.e., the wave age varies from very slow to fast waves. The wave signal is clearly present in the airflow up to at least 0.15λ (where λ is the wave length) and is present up to higher levels for faster waves. In the kinetic energy budgets, pressure transport is mainly of importance for slow waves. For fast waves, viscous transport and turbulent transport dominate near the surface. Kinetic energy budgets for the wave and turbulent perturbations show a non-negligible transport of turbulent kinetic energy directed from turbulence to the wave perturbation in the airflow. The wave-turbulent energy transport depends on the size, tilt, and phase of the wave-induced part of the turbulent Reynolds stresses.According to the DNS data, slow waves are more efficient in generating isotropic turbulence than fast waves.Despite the differences in wave-shape as well as in Reynolds number between the idealized direct numerical simulations and the atmosphere, there are intriguing similarities in the turbulence structure. Important information about the turbulence above waves in the atmosphere can be obtained from DNS—the data must, however, be interpreted with care.     

On the Kinetic Energy Budget of the Unstable Atmospheric Surface Layer

We present a new account of the kinetic energy budget within an unstable atmospheric surface layer (ASL) beneath a convective outer layer. It is based on the structural model of turbulence introduced by McNaughton (Boundary-Layer Meteorology, 112: 199–221, 2004). In this model the turbulence is described as a self-organizing system with a highly organized structure that resists change by instability. This system is driven from above, with both the mean motion and the large-scale convective motions of the outer layer creating shear across the surface layer. The outer convective motions thus modulate the turbulence processes in the surface layer, causing variable downwards fluxes of momentum and kinetic energy. The variable components of the momentum flux sum to zero, but the associated energy divergence is cumulative, increasing both the average kinetic energy of the turbulence in the surface layer and the rate at which that energy is dissipated. The tendency of buoyancy to preferentially enhance the vertical motions is opposed by pressure reaction forces, so pressure production, which is the work done against these reaction forces, exactly equals buoyant production of kinetic energy. The pressure potential energy that is produced is then redistributed throughout the layer through many conversions, back and forth, between pressure potential and kinetic energy with zero sums. These exchanges generally increase the kinetic energy of the turbulence, the rate at which turbulence transfers momentum and the rate at which it dissipates energy, but does not alter its overall structure. In this model the velocity scale for turbulent transport processes in the surface layer is (kzɛ)^1/3 rather than the friction velocity, u_*. Here k is the von Kármán constant, z is observation height, ɛ is the dissipation rate. The model agrees very well with published experimental results, and provides the foundation for the new similarity model of the unstable ASL, replacing the older Monin–Obukhov similarity theory, whose assumptions are no longer tenable.     

Direct numerical simulation of a vigorously heated low Reynolds number convective boundary layer

Computations of the buoyantly unstable Ekman layer are performed at low Reynolds number. The results are obtained by directly solving the three-dimensional time-dependen Navier-Stokes equations with the Boussinesq buoyancy approximation, resolving all relevant scales of motion (no turbulence closure is needed). The flow is capped by a stable temperature inversion and heated from below at a rate that produces an inversion-height to Obukhov-length ratio −z_i/L_* = 32. Temperature and velocity variance profiles are found to agree well with those from an earlier vigorously heated under-resolved computation at higher Reynolds number, and with experimental data of Deardorff and Willis (Boundary-Layer Meteorol., 32: 205–236, 1985). Significant helicity is found in the layer, and helical convection patterns of the scale of the inversion height are observed.     

Measurements of turbulence and fossil turbulence near ampere seamount

Measurements of temperature and velocity microstructure near and downstream of a shallow seamount are used to compare fossil turbulence versus non-fossil turbulence models for the evolution of turbulence microstructure patches in the stratified ocean. According to non-fossil oceanic turbulence models, all overturn length scales L_T of the microstructure grow and collapse in constant proportion to each other and to the turbulence energy (Oboukov) scale L_O and the inertial buoyancy (Ozmidov) scale of the patches; that is, with L_Trms ≈1.2L_R and viscous dissipation rate ≈ ₀^*. According to the Gibson fossil turbulence model, all microstructure originates from completely active turbulence with ₀ ≈ 3L_T²N³(≈ 28₀^*) and L_{T/√6 ≈ LTrms}, but this rapidly decays into a more persistent active-fossil state with _0F ≈ 30vN², where N is the buoyancy frequency and v is the kinematic viscosity and, without further energy supply, finally reaches a completely fossil turbulence hydrodynamic state of internal wave motions, with _F. The last turbulence eddies, with ≈ _F, vanish at a buoyant-inertial-viscous (fossil Kolmogorov) scale L_KF that is much smaller than the remnant overturn scales L_T for large ₀/_F ratios. These density, temperature, and salinity overturns with L_T ≈ 0.6 L_R0 0.6 L_R persist as turbulence fossils (by retaining the memory of _o) and collapse very slowly. In the near wake below the summit depth of Ampere seamount, a much larger proportion of completely active turbulence patches was found than is usually found in the ocean interior away from sources. Dissipation rates and turbulence activity coefficients of microstructure patches were found to decrease downstream, suggesting that the active turbulence indicated by the patches with A_T 1 was caused by the presence of the seamount as a turbulence source. Therefore, the turbulence and mixing processes of ocean layers far away from turbulence sources probably have been undersampled by microstructure data sets lacking any A_T 1 patches. This is because large fractions of the mixing and viscous dissipation of the patches occur in short-lived active turbulence regimes that are too brief to be detected. Consequently, large underestimates of the true space-time average turbulence fluxes and turbulence and scalar dissipation rates may result if non-fossil turbulence models are assumed in ocean microstructure data interpretation.     

Numerical simulations of pollutant dispersion in an idealized urban area,for different meteorological conditions

In order to estimate the impacts of buildings on air pollution dispersion, numerical simulations are performed over an idealized urban area, modelled as regular rows of large rectangular obstacles. The simulations are evaluated with the results of the Mock Urban Setting Test (MUST), which is a near full-scale experiment conducted in Utah’s West Desert area: it consists of releases of a neutral gas in a field of regularly spaced shipping containers. The numerical simulations are performed with the model Mercure_Saturne, which is a three-dimensional computational fluid dynamics code adapted to atmospheric flow and dispersion simulations. It resolves complex geometries and uses, in this study, a k–∈ closure for the turbulence model. Sensitivity studies focus on how to prescribe the inflow conditions for turbulent kinetic energy. Furthermore, different sets of coefficients available in the literature for the k–∈ closure model are tested. Twenty MUST trials with different meteorological conditions are simulated and detailed analyses are performed for both the dynamical variables and average concentration. Our results show overall good agreement according to statistical comparison parameters, with a fraction of predictions for average concentration within a factor of two of observations of 67.1%. The set of simulations offers several inflow wind directions and allows us to emphasize the impact of elongated buildings, which create a deflection of the plume centerline relative to the upstream wind direction.     

Rapid vertical sampling in stably stratified turbulent shear flows

A new method for obtaining instantaneous vertical profiles of two components of velocity and temperature in thermally stratified turbulent shear flows is presented. In this report, the design and construction of the traversing system will be discussed and results to date will be presented. The method is based on rapid vertical sampling whereby probe sensors are moved vertically at a high speed such that the measurement is approximately instantaneous. The system is designed to collect many measurements for the calculation of statistics such as vertical wave number spectra, mean square vertical gradients, and Thorpe scales. Results are presented for vertical profiles of temperature and compared to vertical profiles measured by single-point Eulerian time averages. The quality of the vertical profiles is found to be good over many profiles. Some comparisons are made between vertical measurements and standard single-point Eulerian measurements for three cases of stably stratified turbulent shear flow in which the initial microscale Reynolds number, Re_λ≈30. In case 1, the mean conditions are characterized by a gradient Richardson number, Ri_g=0.015, for which the flow is “unstable”, meaning the spatially evolving turbulent kinetic energy (E_k) grows. In case 2, Ri_g=0.095, for which the evolving turbulent kinetic energy is almost constant. In case 3, the flow is highly stable, where Ri_g=0.25 and E_k decays with spatial evolution. The measurements indicate anisotropy in the small scales for all cases. In particular, it is found that the ratio grows initially to a maximum and then decays with further evolution. Maximum Thorpe displacements are measured and compared to single-point measures of the vertical scales. It is found that vertical length scales derived from single-point measurements, such as the Ozmidov scale, L_O=(ε/N³)^1/2 and the overturn scale, L_t=θ′/(dT/dz), do not represent well the wide range of overturning scales which are actually present in the turbulence.     

Application of a Large-Eddy Simulation Database to Optimisation of First-Order Closures for Neutral and Stably Stratified Boundary Layers

Large-eddy simulation (LES) is a well-established numerical technique, resolving the most energetic turbulent fluctuations in the planetary boundary layer. By averaging these fluctuations, high-quality profiles of mean quantities and turbulence statistics can be obtained in experiments with well-defined initial and boundary conditions. Hence, LES data can be beneficial for assessment and optimisation of turbulence closure schemes. A database of 80 LES runs (DATABASE64) for neutral and stably stratified planetary boundary layers (PBLs) is applied in this study to optimize first-order turbulence closure (FOC). Approximations for the mixing length scale and stability correction functions have been made to minimise a relative root-mean-square error over the entire database. New stability functions have correct asymptotes describing regimes of strong and weak mixing found in theoretical approaches, atmospheric observations and LES. The correct asymptotes exclude the need for a critical Richardson number in the FOC formulation. Further, we analysed the FOC quality as functions of the integral PBL stability and the vertical model resolution. We show that the FOC is never perfect because the turbulence in the upper half of the PBL is not generated by the local vertical gradients. Accordingly, the parameterised and LES-based fluxes decorrelate in the upper PBL. With this imperfection in mind, we show that there is no systematic quality deterioration of the FOC in the strongly stable PBL provided that the vertical model resolution is better than 10 levels within the PBL. In agreement with previous studies, we found that the quality improves slowly with the vertical resolution refinement, though it is generally wise not to overstretch the mesh in the lowest 500 m of the atmosphere where the observed, simulated and theoretically predicted stably stratified PBL is mostly located. The submission to a special issue of the “Boundary-Layer Meteorology” devoted to the NATO advanced research workshop “Atmospheric Boundary Layers: Modelling and Applications for Environmental Security”.     

True gravity in ocean dynamics Part 1 Ekman transport

The direction normal to the Earth spherical (or ellipsoidal) surface is not vertical (called deflected vertical) since the vertical direction is along the true gravity g (= ig_λ+ jg_φ+ kg_z). Here, (λ, φ, z) are (longitude, latitude, depth), and (i, j, k) are the corresponding unit vectors. The spherical (or ellipsoidal) surfaces are not horizontal surfaces (called deflected-horizontal surfaces). The most important body force g (true gravity) has been greatly simplified without justification in oceanography to the standard gravity (-g₀k) with g₀ = 9.81 m/s². Impact of such simplification on ocean dynamics is investigated in this paper using the Ekman layer model. In the classical Ekman layer dynamic equation, the standard gravity (-g₀k) is replaced by the true gravity g(λ, φ, z) with a constant eddy viscosity and a depth-dependent-only density ρ(z) represented by an e-folding near-inertial buoyancy frequency. New Ekman spiral and in turn new formulae for the Ekman transport are obtained for ocean with and without bottom. With the gravity data from the global static gravity model EIGEN-6C4 and the surface wind stress data from the Comprehensive Ocean-Atmosphere Data Set (COADS), large difference is found in the Ekman transport using the true gravity and standard gravity.     

Application of the E-ε closure model to simulations of mesoscale topographic effects

A mesoscale Planetary Boundary Layer (PBL) model with a simple turbulence closure scheme based on the turbulence kinetic energy (TKE) equation and the dissipation () equation is used to simulate atmospheric flow over mesoscale topography. Comparative studies with different parameterizations suggest that with a proper closure assumption for turbulence dissipation, the E-model can simulate the circulation induced by the mesoscale topography with results similar to those obtained using the E- model. On the other hand, the first-order closure using O'Brien's cubic interpolation for eddy diffusivities (K) generally produces much larger K profiles in the stable and the unstable regions, which is believed to be due to the overprediction of the height of the PBL. All models with the TKE equation yield quite similar ensemble mean fields, which are found to be little sensitive to the closure assumption for turbulence dissipation, though their predicted magnitudes of TKE and K may differ appreciably. A discussion on the diurnal evolution of the mesoscale topography-induced circulation and the spatial variations of the turbulence fluxes in the surface layer is also given based on the E- model results.     

Model of convective heat exchange due to isolated thermals in the atmospheric boundary layer

A numerical model of convective heat transfer due to isolated thermals in the atmospheric boundary layer is used to describe the temperature profile transformation in undisturbed conditions as a result of intensive dry free convection. Based on some assumptions, the heat transfer Equation (2) is transformed to the form (14) in which the coefficients and the function F are expressed by (d/dz)(ln ) and by parameters of thermals. Equation (14) has been solved numerically with the help of Equation (15) obtained from the statics equation because of Equation (8). The size distribution function f(z, r, t) of the thermals is discrete (Table I), according to Vulf'son (1961). On Figures 1 and 2 are plotted successive temperature profiles for a ground inversion, transformed due to free convection and turbulence (Figures 1a and 2a), and due to turbulence only (Figures 1b and 2b). The profiles are computed from Equation 14 (Figures 1a and 2a) and Equation 16 (Figures 1b and 2b) for k _z= 1 m² s^–1 (Figure 1) and k _z= 10 m² s^–1 (Figure 2). On Figure 3 the real temperature profiles in Sofia for June 22nd 1976 are compared with the profiles computed using the real initial profile for 4.30 h local time. Good qualitative agreement can be seen.     

Buoyancy and The Sensible Heat Flux Budget Within Dense Canopies

In contrast to atmospheric surface-layer (ASL) turbulence, a linear relationship between turbulent heat fluxes (F_T) and vertical gradients of mean air temperature within canopies is frustrated by numerous factors, including local variation in heat sources and sinks and large-scale eddy motion whose signature is often linked with the ejection-sweep cycle. Furthermore, how atmospheric stability modifies such a relationship remains poorly understood, especially in stable canopy flows. To date, no explicit model exists for relating F_T to the mean air temperature gradient, buoyancy, and the statistical properties of the ejection-sweep cycle within the canopy volume. Using third-order cumulant expansion methods (CEM) and the heat flux budget equation, a “diagnostic” analytical relationship that links ejections and sweeps and the sensible heat flux for a wide range of atmospheric stability classes is derived. Closure model assumptions that relate scalar dissipation rates with sensible heat flux, and the validity of CEM in linking ejections and sweeps with the triple scalar-velocity correlations, were tested for a mixed hardwood forest in Lavarone, Italy. We showed that when the heat sources (S_T) and F_T have the same sign (i.e. the canopy is heating and sensible heat flux is positive), sweeps dominate the sensible heat flux. Conversely, if S_T and F_T are opposite in sign, standard gradient-diffusion closure model predict that ejections must dominate the sensible heat flux.     

A Hierarchy of Energy- and Flux-Budget (EFB) Turbulence Closure Models for Stably-Stratified Geophysical Flows

Here we advance the physical background of the energy- and flux-budget turbulence closures based on the budget equations for the turbulent kinetic and potential energies and turbulent fluxes of momentum and buoyancy, and a new relaxation equation for the turbulent dissipation time scale. The closure is designed for stratified geophysical flows from neutral to very stable and accounts for the Earth’s rotation. In accordance with modern experimental evidence, the closure implies the maintaining of turbulence by the velocity shear at any gradient Richardson number Ri, and distinguishes between the two principally different regimes: “strong turbulence” at ${Ri \ll 1}$ typical of boundary-layer flows and characterized by the practically constant turbulent Prandtl number Pr _T; and “weak turbulence” at Ri > 1 typical of the free atmosphere or deep ocean, where Pr _T asymptotically linearly increases with increasing Ri (which implies very strong suppression of the heat transfer compared to the momentum transfer). For use in different applications, the closure is formulated at different levels of complexity, from the local algebraic model relevant to the steady-state regime of turbulence to a hierarchy of non-local closures including simpler down-gradient models, presented in terms of the eddy viscosity and eddy conductivity, and a general non-gradient model based on prognostic equations for all the basic parameters of turbulence including turbulent fluxes.     

A Comprehensive Modelling Approach for the Neutral Atmospheric Boundary Layer: Consistent Inflow Conditions, Wall Function and Turbulence Model

We report on a novel approach for the Reynolds-averaged Navier-Stokes (RANS) modelling of the neutral atmospheric boundary layer (ABL), using the standard k-ek-{\varepsilon} turbulence model. A new inlet condition for turbulent kinetic energy is analytically derived from the solution of the k-ek-{\varepsilon} model transport equations, resulting in a consistent set of fully developed inlet conditions for the neutral ABL. A modification of the standard k-ek-{\varepsilon} model is also employed to ensure consistency between the inlet conditions and the turbulence model. In particular, the turbulence model constant C _μ is generalized as a location-dependent parameter, and a source term is introduced in the transport equation for the turbulent dissipation rate. The application of the proposed methodology to cases involving obstacles in the flow is made possible through the implementation of an algorithm, which automatically switches the turbulence model formulation when going from the region where the ABL is undisturbed to the region directly affected by the building. Finally, the model is completed with a slightly modified version of the Richards and Hoxey rough-wall boundary condition. The methodology is implemented and tested in the commercial code Ansys Fluent 12.1. Results are presented for a neutral boundary layer over flat terrain and for the flow around a single building immersed in an ABL.     

Probabilistic Model for Concentration Fluctuations in Compact-Source Plumes in an Urban Environment

A comprehensive model for the prediction of concentration fluctuations in plumes dispersing in the complex and highly disturbed wind flows in an urban environment is formulated. The mean flow and turbulence fields in the urban area are obtained using a Reynolds-averaged Navier-Stokes (RANS) flow model, while the standard k-ϵ turbulence model (k is the turbulence kinetic energy and ϵ is the viscous dissipation rate) is used to close the model. The RANS model provides a specification of the velocity statistics of the highly disturbed wind flow in the urban area, required for the solution of the transport equations for the mean concentration and concentration variance (both of which are formulated in the Eulerian framework). A physically-based formulation for the scalar dissipation time scale t _d , required for the closure of the transport equation for , is presented. This formulation relates t _d to an inner time scale corresponding to “internal” concentration fluctuation associated with relative dispersion, rather than an outer time scale associated with the entire portion of the fluctuation spectrum. The two lowest-order moments of concentration ( and ) are used to determine the parameters of a pre-chosen functional form for the concentration probability density function (clipped-gamma distribution). Results of detailed comparisons between a water-channel experiment of flow and dispersion in an idealized obstacle array and the model predictions for mean flow, turbulence kinetic energy, mean concentration, concentration variance, and concentration probability density function are presented.     

The temporal evolution of a geostrophic flow in a rotating stratified basin

A laboratory study in a rotating stratified basin examines the instability and long time evolution of the geostrophic double gyre introduced by the baroclinic adjustment to an initial basin-scale step height discontinuity in the density interface of a two-layer fluid. The dimensionless parameters that are important in determining the observed response are the Burger number S=R/R₀ (where R is the baroclinic Rossby radius of deformation and R₀ is the basin radius) and the initial forcing amplitude (H₁ is the upper layer depth). Experimental observations and a numerical approach, using contour dynamics, are used to identify the mechanisms that result in the dominance of nonlinear behaviour in the long time evolution, τ>2⁻¹ (where τ is time scaled by the inertial period T_I=2π/f). When the influence of rotation is moderate (0.25≤S≤1), the instability mechanism is associated with the finite amplitude potential vorticity (PV) perturbation introduced when the double gyre is established. On the other hand, when the influence of rotation is strong (S≤0.1), baroclinic instability contributes to the nonlinear behaviour. Regardless of the mechanism, nonlinearity acts to transfer energy from the geostrophic double gyre to smaller scales associated with an eddy field. In the lower layer, Ekman damping is pronounced, resulting in the dissipation of the eddy field after only 40T_I. In the upper layer, where dissipative effects are weak, the eddy field evolves until it reaches a symmetric distribution of potential vorticity within the domain consisting of cyclonic and anticyclonic eddy pairs, after approximately 100T_I. The functional dependence of the characteristic eddy lengthscale L_E on S is consistent with previous laboratory studies on continuously forced geostrophic turbulence. The cyclonic and anticyclonic eddy pairs are maintained until viscous effects eventually dissipate all motion in the upper layer after approximately 800T_I. The outcomes of this study are considered in terms of their contribution to the understanding of the energy pathways and transport processes associated with basin-scale motions in large stratified lakes.