Daan Crommelin

Strategies for Model Reduction: Comparing Different Optimal Bases

Abstract Several different ways of constructing optimal bases for efficient dynamical modeling are compared: empirical orthogonal functions (EOFs), optimal persistence patterns (OPPs), and principal interaction patterns (PIPs). Past studies on fluid-dynamical topics have pointed out that EOF-based models can have difficulties reproducing behavior dominated by irregular transitions between different dynamical states. This issue is addressed in a geophysical context, by assessing the ability of these strategies for efficient dynamical modeling to reproduce the chaotic regime transitions in a simple atmosphere model. The atmosphere model is the well-known Charney– DeVore model, a six-dimensional truncation of the equations describing barotropic flow over topography in a β-plane channel geometry. This model is able to generate regime transitions for well-chosen parameter settings. The models based on PIPs are found to be superior to the EOF- and OPP-based models, in spite of some undesirable sensitivities inh...

Subgrid-Scale Parameterization with Conditional Markov Chains

Abstract A new approach is proposed for stochastic parameterization of subgrid-scale processes in models of atmospheric or oceanic circulation. The new approach relies on two key ingredients: first, the unresolved processes are represented by a Markov chain whose properties depend on the state of the resolved model variables; second, the properties of this conditional Markov chain are inferred from data. The parameterization approach is tested by implementing it in the framework of the Lorenz ’96 model. Performance of the parameterization scheme is assessed by inspecting probability distributions, correlation functions, and wave properties, and by carrying out ensemble forecasts. For the Lorenz ’96 model, the parameterization algorithm is shown to give good results with a Markov chain with a few states only and to outperform several other parameterization schemes.

Normal forms for reduced stochastic climate models

The systematic development of reduced low-dimensional stochastic climate models from observations or comprehensive high-dimensional climate models is an important topic for atmospheric low-frequency variability, climate sensitivity, and improved extended range forecasting. Here techniques from applied mathematics are utilized to systematically derive normal forms for reduced stochastic climate models for low-frequency variables. The use of a few Empirical Orthogonal Functions (EOFs) (also known as Principal Component Analysis, Karhunen–Loéve and Proper Orthogonal Decomposition) depending on observational data to span the low-frequency subspace requires the assessment of dyad interactions besides the more familiar triads in the interaction between the low- and high-frequency subspaces of the dynamics. It is shown below that the dyad and multiplicative triad interactions combine with the climatological linear operator interactions to simultaneously produce both strong nonlinear dissipation and Correlated Additive and Multiplicative (CAM) stochastic noise. For a single low-frequency variable the dyad interactions and climatological linear operator alone produce a normal form with CAM noise from advection of the large scales by the small scales and simultaneously strong cubic damping. These normal forms should prove useful for developing systematic strategies for the estimation of stochastic models from climate data. As an illustrative example the one-dimensional normal form is applied below to low-frequency patterns such as the North Atlantic Oscillation (NAO) in a climate model. The results here also illustrate the short comings of a recent linear scalar CAM noise model proposed elsewhere for low-frequency variability.

Regime transitions and heteroclinic connections in a barotropic atmosphere

By interpreting transitions between atmospheric flow regimes as a deterministic rather than a stochastic phenomenon, new insight is gained into the phase-space characteristics of these transitions. The identification of regimes with steady states should be extended with the association of transitions with nearby heteroclinic connections between steady states, as known from the theory of dynamical systems. In the context of a moderately complex barotropic model of the Northern Hemisphere, which possesses regime behavior, steady states are found that correspond with regimes, and heteroclinic connections are approximated using a new algorithm based on adjoint modeling techniques. A 200-yr dataset generated by the model is shown to possess spatial preferences in its transitional behavior that match well with the approximated heteroclinic connections.

Fitting timeseries by continuous-time Markov chains: a quadratic programming approach

Construction of stochastic models that describe the effective dynamics of observables of interest is an useful instrument in various fields of application, such as physics, climate science, and finance. We present a new technique for the construction of such models. From the timeseries of an observable, we construct a discrete-in-time Markov chain and calculate the eigenspectrum of its transition probability (or stochastic) matrix. As a next step we aim to find the generator of a continuous-time Markov chain whose eigenspectrum resembles the observed eigenspectrum as closely as possible, using an appropriate norm. The generator is found by solving a minimization problem: the norm is chosen such that the object function is quadratic and convex, so that the minimization problem can be solved using quadratic programming techniques. The technique is illustrated on various toy problems as well as on datasets stemming from simulations of molecular dynamics and of atmospheric flows.

Stochastic Parameterization of Convective Area Fractions with a Multicloud Model Inferred from Observational Data

AbstractObservational data of rainfall from a rain radar in Darwin, Australia, are combined with data defining the large-scale dynamic and thermodynamic state of the atmosphere around Darwin to develop a multicloud model based on a stochastic method using conditional Markov chains. The authors assign the radar data to clear sky, moderate congestus, strong congestus, deep convective, or stratiform clouds and estimate transition probabilities used by Markov chains that switch between the cloud types and yield cloud-type area fractions. Cross-correlation analysis shows that the mean vertical velocity is an important indicator of deep convection. Further, it is shown that, if conditioned on the mean vertical velocity, the Markov chains produce fractions comparable to the observations. The stochastic nature of the approach turns out to be essential for the correct production of area fractions. The stochastic multicloud model can easily be coupled to existing moist convection parameterization schemes used in ge...

Diffusion Estimation from Multiscale Data by Operator Eigenpairs

In this paper we present a new procedure for the estimation of diffusion processes from discretely sampled data. It is based on the close relation between eigenpairs of the diffusion operator ℒ and those of the conditional expectation operator Pt, a relation stemming from the semigroup structure Pt=exp(tℒ) for t≥0. It allows for estimation without making time discretization errors, an aspect that is particularly advantageous in the case of data with low sampling frequency. After estimating eigenpairs of ℒ via eigenpairs of Pt, we infer the drift and diffusion functions that determine ℒ by fitting ℒ to the estimated eigenpairs using a convex optimization procedure. We present numerical examples in which we apply the procedure to one- and two-dimensional diffusions, reversible as well as nonreversible. In the second part of the paper, we consider estimation of coarse-grained (homogenized) diffusion processes from multiscale data. We show that eigenpairs of the homogenized diffusion operator are asymptotical...

A data-driven multi-cloud model for stochastic parametrization of deep convection

Stochastic subgrid models have been proposed to capture the missing variability and correct systematic medium-term errors in general circulation models. In particular, the poor representation of subgrid-scale deep convection is a persistent problem that stochastic parametrizations are attempting to correct. In this paper, we construct such a subgrid model using data derived from large-eddy simulations (LESs) of deep convection. We use a data-driven stochastic parametrization methodology to construct a stochastic model describing a finite number of cloud states. Our model emulates, in a computationally inexpensive manner, the deep convection-resolving LES. Transitions between the cloud states are modelled with Markov chains. By conditioning the Markov chains on large-scale variables, we obtain a conditional Markov chain, which reproduces the time evolution of the cloud fractions. Furthermore, we show that the variability and spatial distribution of cloud types produced by the Markov chains become more faithful to the LES data when local spatial coupling is introduced in the subgrid Markov chains. Such spatially coupled Markov chains are equivalent to stochastic cellular automata.

Applying a splitting technique to estimate electrical grid reliability

As intermittent renewable energy penetrates electrical power grids more and more, assessing grid reliability is of increasing concern for grid operators. Monte Carlo simulation is a robust and popular technique to estimate indices for grid reliability, but the involved computational intensity may be too high for typical reliability analyses. We show that various reliability indices can be expressed as expectations depending on the rare event probability of a so-called power curtailment, and explain how to extend a Crude Monte Carlo grid reliability analysis with an existing rare event splitting technique. The squared relative error of index estimators can be controlled, whereas orders of magnitude less workload is required than when using an equivalent Crude Monte Carlo method. We show further that a bad choice for the time step size or for the importance function may endanger this squared relative error.

Stochastic Convection Parameterization with Markov Chains in an Intermediate-Complexity GCM

Conditional Markov chain (CMC) models have proven to be promising building blocks for stochastic convection parameterizations. In this paper, it is demonstrated how two different CMC models can be used as mass flux closures in convection parameterizations. More specifically, the CMC models provide a stochastic estimate of the convective area fraction that is directly proportional to the cloud-base mass flux. Since, in one of the models, the number of CMCs decreases with increasing resolution, this approach makes convection parameterizations scale aware and introduces stochastic fluctuations that increase with resolution in a realistic way. Both CMC models are implemented in a GCM of intermediate complexity. It is shown that with the CMC models, trained with observational data, it is possible to improve both the subgrid-scale variability and the autocorrelation function of the cloud-base mass flux as well as the distribution of the daily accumulated precipitation in the tropics. Hovmoller diagrams and wavenumber–frequency diagrams of the equatorial precipitation indicate that, in this specific GCM, convectively coupled equatorial waves are more sensitive to the mean cloud-base mass flux than to stochastic fluctuations. A smaller mean mass flux tends to increase the power of the simulated MJO and to diminish equatorial Kelvin waves.