Eric J. Kostelich

A local ensemble Kalman filter for atmospheric data assimilation

In this paper, we introduce a new, local formulation of the ensemble Kalman Filter approach for atmospheric data assimilation. Our scheme is based on the hypothesis that, when the Earths surface is divided up into local regions of moderate size, vectors of the forecast uncertainties in such regions tend to lie in a subspace of much lower dimension than that of the full atmospheric state vector of such a region. Ensemble Kalman Filters, in general, assume that the analysis resulting from the data assimilation lies in the same subspace as the expected forecast error. Under our hypothesis the dimension of this subspace is low. This implies that operations only on relatively low dimensional matrices are required. Thus, the data analysis is done locally in a manner allowing massively parallel computation to be exploited. The local analyses are then used to construct global states for advancement to the next forecast time. The method, its potential advantages, properties, and implementation requirements are illustrated by numerical experiments on the Lorenz-96 model. It is found that accurate analysis can be achieved at a cost which is very modest compared to that of a full global ensemble Kalman Filter.

Four-dimensional ensemble Kalman filtering

Ensemble Kalman filtering was developed as a way to assimilate observed data to track the current state in a computational model. In this paper we show that the ensemble approach makes possible an additional benefit: the timing of observations, whether they occur at the assimilation time or at some earlier or later time, can be effectively accounted for at low computational expense. In the case of linear dynamics, the technique is equivalent to instantaneously assimilating data as they are measured. The results of numerical tests of the technique on a simple model problem are shown.

Noise reduction: finding the simplest dynamical system consistent with the data

A truck is provided with a torsionally compliant transverse member connecting to transversely spaced side frames. The side frames are suspended on the axles by a spring suspension system of generally cylindrical elastomer elements and a combined damping system. Both springing and damping are generally exponentially variable relative to increases in the load in the railcar. A central pivot pin transfers lateral and longitudinal loads between the truck and the car body without transferring substantial vertical loads. The center pin mounting in the torsionally compliant member is compliant. Vertical loads are transferred between the car body and the truck by omni-directional vertically-incompressible low friction side bearings.

A local ensemble transform Kalman filter data assimilation system for the NCEP global model

The accuracy and computational efficiency of a parallel computer implementation of the Local Ensemble Transform Kalman Filter (LETKF) data assimilation scheme on the model component of the 2004 version of the Global Forecast System (GFS) of the National Centers for Environmental Prediction (NCEP) is investigated. Numerical experiments are carried out at model resolution T62L28. All atmospheric observations that were operationally assimilated by NCEP in 2004, except for satellite radiances, are assimilated with the LETKF. The accuracy of the LETKF analyses is evaluated by comparing it to that of the Spectral Statistical Interpolation (SSI), which was the operational global data assimilation scheme of NCEP in 2004. For the selected set of observations, the LETKF analyses are more accurate than the SSI analyses in the Southern Hemisphere extratropics and are comparably accurate in the Northern Hemisphere extratropics and in the Tropics. The computationalwall-clock times achieved on a Beowulf cluster of 3.6 GHz Xeon processors make our implementation of the LETKF on the NCEP GFS a widely applicable analysis-forecast system, especially for research purposes. For instance, the generation of four daily analyses at the resolution of the NCAR-NCEP reanalysis (T62L28) for a full season (90 d), using 40 processors, takes less than 4 d of wall-clock time.

Assessing a local ensemble Kalman filter: perfect model experiments with the National Centers for Environmental Prediction global model

The accuracy and computational efficiency of the recently proposed local ensemble Kalman filter (LEKF) data assimilation scheme is investigated on a state-of-the-art operational numerical weather prediction model using simulated observations. The model selected for this purpose is the T62 horizontaland 28-level vertical-resolution version of the Global Forecast System (GFS) of the National Center for Environmental Prediction. The performance of the data assimilation system is assessed for different configurations of the LEKF scheme. It is shown that a modest size (40-member) ensemble is sufficient to track the evolution of the atmospheric state with high accuracy. For this ensemble size, the computational time per analysis is less than 9 min on a cluster of PCs. The analyses are extremely accurate in the mid-latitude storm track regions. The largest analysis errors, which are typically much smaller than the observational errors, occur where parametrized physical processes play important roles. Because these are also the regions where model errors are expected to be the largest, limitations of a real-data implementation of the ensemble-based Kalman filter may be easily mistaken for model errors. In light of these results, the importance of testing the ensemble-based Kalman filter data assimilation systems on simulated observations is stressed.

Measuring intense rotation and dissipation in turbulent flows

Turbulent flows are highly intermittent—for example, they exhibit intense bursts of vorticity and strain. Kolmogorov theory describes such behaviour in the form of energy cascades from large to small spatial and temporal scales, where energy is dissipated as heat. But the causes of high intermittency in turbulence, which show non-gaussian statistics, are not well understood. Such intermittency can be important, for example, for enhancing the mixing of chemicals, by producing sharp drops in local pressure that can induce cavitation (damaging mechanical components and biological organisms), and by causing intense vortices in atmospheric flows. Here we present observations of the three components of velocity and all nine velocity gradients within a small volume, which allow us to determine simultaneously the dissipation (a measure of strain) and enstrophy (a measure of rotational energy) of a turbulent flow. Combining the statistics of all measurements and the evolution of individual bursts, we find that a typical sequence for intense events begins with rapid strain growth, followed by rising vorticity and a final sudden decline in stretching. We suggest two mechanisms which can produce these characteristics, depending whether they are due to the advection of coherent structures through our observed volume or caused locally.

Primary instabilities and bicriticality in flow between counter‐rotating cylinders

The primary instabilities and bicritical curves for flow between counter‐rotating cylinders have been computed numerically from the Navier–Stokes equations assuming axial periodicity. The computations provide values of the Reynolds numbers, wavenumbers, and wave speeds at the primary transition from Couette flow for radius ratios from 0.40–0.98. Particular attention has been focused on the bicritical curves that separate (as the magnitude of counter‐rotation is increased) the transitions from Couette flow to flows with different azimuthal wavenumbers m and m+1. This lays the foundation for further analysis of nonlinear mode interactions and pattern formation occurring along the bicritical curves and serves as a benchmark for experimental studies. Preliminary experimental measurements of transition Reynolds numbers and wave speeds presented here agree well with the computations from the mathematical model.

Nonlinear deterministic modeling of highly varying loads

Typically, the modeling of highly varying, nonlinear loads such as electric arc furnaces has involved stochastic techniques. This paper presents the use of chaotic dynamics to describe the operation of nonlinear loads. Included is a discussion of the Lyapunov exponents, a measure of chaotic behavior. The alternate approach is applied to electric arc furnaces. A tuning mode is described to develop the parameters of a chaotic model. This model is trained to have time and frequency responses that are tuned to match the current from the arc furnace under study. The simulated data are compared to actual arc furnace data to validate the model. This model is used to assess the impact of various highly varying nonlinear loads that exhibit chaos in power systems.

Unstable dimension variability: a source of nonhyperbolicity in chaotic systems

The hyperbolicity or nonhyperbolicity of a chaotic set has profound implications for the dynamics on the set. A familiar mechanism causing nonhyperbolicity is the tangency of the stable and unstable manifolds at points on the chaotic set. Here we investigate a different mechanism that can lead to nonhyperbolicity in typical invertible (respectively noninvertible) maps of dimension 3 (respectively 2) and higher. In particular, we investigate a situation (first considered by Abraham and Smale in 1970 for different purposes) in which the dimension of the unstable (and stable) tangent spaces are not constant over the chaotic set; we call this unstable dimension variability. A simple two-dimensional map that displays behavior typical of this phenomenon is presented and analyzed.

Multi-dimensioned intertwined basin boundaries: Basin structure of the kicked double rotor

Abstract Using numerical computations on a map which describes the time evolution of a particular mechanical system in a four-dimensional phase space (The kicked double rotor), we have found that the boundaries separating basins of attraction can have different properties in different regions and that these different regions can be intertwined on arbitrarily fine scale . In particular, for the double rotor map, if one chooses a restricted region of the phase space and examines the basin boundary in that region, then either one observes that the boundary is a smooth three-dimensional surface or one observes that the boundary is fractal with dimension d ≊ 3.9 , and which of these two possibilities applies depends on the particular phase space region chosen for examination. Furthermore, for any region (no matter how small) for which d ≊ 3.9 , one can choose subregions within it for which d = 3. (Hence d ≊ 3.9 region and d = 3 region are intertwined on arbitrarily fine scale.) Other examples will also be presented and analyzed to show how this situation can arise. These include one-dimensional map cases, a map of the plane, and the Lorenz equations. In one of our one-dimensional map cases the boundary will be fractal everywhere, but the dimension can take on either of two different values both of which lie between 0 and 1. These examples lead us to conjecture that basin boundaries typically can have at most a finite number of possible dimension values. More specifically, let these values be denoted d 1 , d 2 ,…, d N . Choose a volume region of phase space whose interior contains some part of the basin boundary and evaluate the dimension of the boundary in that region. Then our conjecture is that for all typical volume choices, the evaluated dimension within the chosen volume will be one of the values d 1 , d 2 ,…, d N . For example, in our double rotor map it appears that N = 2, and d 1 = 3.0 and d 2 = 3.9.