Suitable reduced order models (ROMs) are computationally efficient tools in characterizing key dynamical and statistical features of nature. In this paper, a systematic multiscale stochastic ROM framework is developed for complex systems with strong chaotic or turbulent behavior. The new ROMs are fundamentally different from the traditional Galerkin ROM (G-ROM) or those deterministic ROMs that aim at minimizing the path-wise errors and applying mainly to laminar systems. Here, the new ROM focuses on recovering the large-scale dynamics to the maximum extent while it also exploits cheap but effective conditional linear functions as the closure terms to capture the statistical features of the medium-scale variables and its feedback to the large scales. In addition, physics constraints are incorporated into the new ROM. One unique feature of the resulting ROM is that it facilitates an efficient and accurate scheme for nonlinear data assimilation, the solution of which is provided by closed analytic formulae. Such an analytic solvable data assimilation solution significantly accelerates the computational efficiency and allows the new ROM to avoid many potential numerical and sampling issues in recovering the unobserved states from partial observations. The overall model calibration is efficient and systematic via explicit mathematical formulae. The new ROM framework is applied to complex nonlinear systems, in which the intrinsic turbulent behavior is either triggered by external random forcing or deterministic nonlinearity. It is shown that the new ROM significantly outperforms the G-ROM in both scenarios in terms of reproducing the dynamical and statistical features as well as recovering unobserved states via the associated efficient data assimilation scheme.
Introduction
Modeling complex nonlinear turbulent systems is of practical importance. These complex nonlinear system appear in many areas [1, 2, 3, 4, 5], such as geophysics, climate science, engineering, atmosphere and ocean science, neural science, and material science. They often involve strong nonlinearities and multiscale structures [3, 6, 7, 8]. Energy is transferred across different scales via the intrinsic nonlinearity that triggers regime switching, intermittent instabilities, extreme events and many other chaotic and turbulent features [9, 10, 11, 12, 13, 14, 15, 16]. As a result, the associated probability density functions (PDFs) are usually non-Gaussian with fat tails. Building appropriate mathematical models for complex systems that succeed in characterizing these key features not only advances the understanding of nature, but facilitates effective state estimation and skillful forecast as well. However, the complexity of nature makes it extremely challenging to develop a full order model (FOM) and apply it to forecast many quantities of interest. In fact, due to the lack of physical understanding or the inadequate resolution in computations, a perfect FOM is never available in practice. Even if a nearly perfect FOM is accessible, the model is often of high dimensionality and contains a complicated model structure [17, 18]. As a consequence, there exist many mathematical and computational difficulties in analyzing and simulating such a FOM.
Suitable reduced order models (ROMs) [19, 20, 21, 22, 23] are desirable surrogates for the FOMs to reproduce certain key dynamical and statistical features of nature with a much lower computational cost. Linear regression models are possibly the simplest class of ROMs [24, 25] that can provide certain skill for short-term forecasts but they usually struggle in characterizing the underlying nonlinear physics. Beyond the regression strategies, one systematic and commonly used approach to developing ROMs is to project the starting complex nonlinear system to the leading a few energetic modes in light of the Galerkin proper orthogonal decomposition methods [20] or other empirical basis functions such as the principal interaction patterns [26, 27] and the dynamic mode decomposition [28, 29]. With a careful design of the closure terms (see [30] for a survey) to compensate the truncation error, these ROMs are skillful in resolving certain problems in many areas, including fluids and turbulence [31, 32, 21, 23, 33]. While many traditional ROM strategies focus on seeking for the optimal approximate models with deterministic model structures, stochastic effects have been incorporated into many recent developments of ROMs for complex turbulent systems. The stochasticity is essential in assessing the uncertainty and model error in ROMs. It also plays a pivotal role in facilitating data assimilation and advancing probabilistic forecast, which is vital for predicting non-Gaussian extreme events. The MTV (named after Majda, Timofeyev and Vanden Eijnden) strategy is one of the most straightforward approaches to incorporate stochastic effects in the ROMs [34, 35], the idea of which is to replace the self-interactions among the high frequencies by random noise. The recently proposed nonlinear autoregression (NAR) model emphasizes the random effect in the development of regression-based ROMs and its significance in advancing the statistical forecast [36, 37]. On the other hand, physics-constrained regression models are a set of nonlinear stochastic ROMs [38, 39, 40], which take into account the energy conserving nonlinear interactions in the model development that guarantees the well-posednss of long-term behavior of the system. In addition, stochastic parameterizations are powerful ways to reduce the complexity of the unresolved part of the dynamics that nevertheless provides effective statistical feedbacks to the resolved-scale dynamics [41, 42, 43, 44, 45].
The goal of this paper is to build a systematic multiscale stochastic ROM framework for complex systems with strong chaotic or turbulent behavior. The new ROM framework has several unique features. First, a new strategy of building effective closure terms is incorporated into the framework that includes both deterministic and stochastic components. Fundamentally different from the traditional closure methods, the development of the closures in this new ROM framework exploits the multiscale characteristics of the underlying system to reach a systematic formulation that is mathematically more concise and tractable than most of the traditional nonlinear closures. Nevertheless, appropriate stochastic closure terms are combined with the deterministic ones in this framework that succeeds in providing effective statistical feedbacks between different scales. These closure terms significantly advance the ROM in reproducing both the dynamics and statistics of nature. Second, data assimilation is a necessary procedure in understanding and forecasting turbulent systems [46, 47, 48, 49]. It aims at recovering the unobserved state variables using Bayesian inference, and is the prerequisite of the ensemble forecast for turbulent flows. Different from most of the traditional ROMs that have to rely on ensemble based data assimilation methods for state estimation and the initialization of the forecast, the ROM with the new concise closures facilitates an efficient and accurate scheme for nonlinear data assimilation, the solution of which is provided by closed analytic formulae. Such an analytic solvable data assimilation solution increases the computational efficiency and allows the new ROM to avoid many potential numerical and sampling issues that usually require careful ad hoc tunings. Third, physics constraints can be naturally incorporated into this new ROM framework, which is essential for preserving the basic principles in physics and preventing finite-time blow-up solutions in the statistical forecast [38, 39, 50]. Another desirable feature is that the ROM and the associated closures can be systematically determined via a computationally efficient optimization procedure with no ad hoc tuning. Note that the development of the ROM in this framework does not require to know the exact FOM, which is often not accessible in practice. The framework is in fact amenable to data-driven scenarios. Specifically, the partially available information from the FOM can be used to determine the large-scale structure of the ROM, while the remaining part of the ROM including the closures can be inferred from data. Finally, it is worthwhile to highlight that many well-known complex nonlinear systems in geophysics, neural science, and engineering turbulence have the same structure as those belonging to the new ROM framework [51], which is a further justification of the framework.
The new ROM framework will be applied to complex nonlinear systems that belong to two categories, distinguished by the triggering mechanisms of the turbulent behavior. The first test model is a viscous stochastic Burgers equation [52, 53]. In such a complex system, external random forcing interacts with the deterministic nonlinearity to induce intermittent signals in the form of viscous shocks, which will be dissipated via the viscous term before the appearance of discontinuities. Here, the stochastic forcing triggers turbulent features. The second test example is the quasi-geostrophic (QG) equation [1, 9]. The QG equation is a deterministic system and the turbulent features are induced completely by the nonlinear interactions between the state variables at different spatial scales. In both cases, only a small number of the state variables are involved in the ROM that allows a much lower computational cost to reproduce the key dynamical and statistical features. It is also significant to see the skill of the stochastic closure terms in effectively characterizing the turbulent features triggered completely by the deterministic nonlinearity.
The rest of the paper is organized as follows. § 2 includes the new ROM framework, the associated data assimilation scheme, and the model calibration. § 3 demonstrates the skill of the new ROMs in reproducing the key dynamical and statistical features of both the viscous stochastic Burgers equation and the QG system. The comparison with the state-of-the-art ROMs in terms of both the model simulation and the data assimilation skill is presented in this section as well. The paper is concluded in § 4, together with a discussion of potential extensions of the current framework. All the technical details are included in Appendix.
The New Reduced Order Modeling (ROM) Framework
The mathematical framework of the ROM with a conditional linear closure
Many complex turbulent nonlinear systems in nature have the following abstract mathematical structure [6, 1, 9, 47],
dtdu=(L+D)u+B(u,u)+F(t)+σ(t)W˙(t),(1)
where u∈CN is the multidimensional state variable and t is time. In (1), Lu and Du represent linear dispersion and dissipation effects, respectively, where L is skew-symmetric and D is negative-definite. The nonlinear effect is introduced through an energy-conserving quadratic form with u⋅B(u,u)=0. The system is also subject to external forcing effects that contain a deterministic component, F(t), and possibly a stochastic disturbance as well, which is assumed to be a Gaussian random noise, σ(t)W˙(t), for simplicity. The model in (1) can be understood as discretizing a (stochastic) partial differential equation by projecting it onto a given set of basis functions {φ1,φ2,…,φN}, resulting in a high-dimensional system, where the stochastic forcing accounts for the truncation error. Due to the intrinsic turbulent behavior, the system in (1) is highly nonlinear and the state variable u can possess strongly non-Gaussian statistics in both the marginal and joint PDFs.
Decompose the state variable u into u=(vT,wT,zT)T with ⋅T being the vector transpose, where v∈CN1, w∈CN2, z∈CN3, and N1+N2+N3=N. A typical criterion for such a decomposition is based on the rank of the scales for the state variables. In other words, v includes the leading few modes corresponding to the large-scale components of the state variables, w contains the next a few modes describing the medium-scale features, while z involves the remaining modes that are treated as the small-scale variables. Note that different from many traditional ROM development, where the state variables are decomposed into only the large- and small-scale components, a three-scale decomposition is utilized here. Such a new decomposition facilitates a smoother transition from large- to small-scale variables by means of a set of intermediate-scale variables. It also allows the feedback from medium- to large-scale variables to be explicitly incorporated into the ROM, which is often crucial in enhancing the model accuracy in characterizing energy transfer and intermittency. Finally, the choice of N1, N2, and N3 is case dependent but in a typical scenario z accounts for the majority number of the modes.
A natural way to build ROMs with a reduced dimension of the coupled system in (1) is to project u onto its large- and medium-scale components v and w,
where ⋅∗ is the conjugate transpose. It is seen that the linear and the quadratic nonlinear terms in (1) have been explicitly expressed in (2). Therefore, A(⋅) and B(⋅) in (2) are all constant matrices and tensors while the two residual terms include the contribution from the variable z that has been truncated here. The notation in (2) is slightly abused in the sense that the quadratic nonlinear terms between v and w are simply written as v∗Bvw(v)w and v∗Bvw(w)w while the other parts of the associated nonlinear terms, say w∗Bvw(v)v and w∗Bvw(w)v, are omitted in (2) for notation simplicity. Note that if the state variables are all real-valued, then v∗Bvw(v)w and v∗Bvw(w)w are sufficient to characterize the quadratic nonlinear interactions between v and w. Since each complex-valued variable can always be rewritten as two real-valued variables, representing its real and imaginary parts, all the state variables in the following will be regarded as real-valued. Similarly, the notation ⋅∗ hereafter simply denotes the transpose of a real-valued vector. In the development of ROMs, various methods have been proposed to deal with the residual terms. If these residual terms are simply ignored, then the resulting ROM becomes the so-called Garlerkin ROM (G-ROM) [26, 27]. However, it has been shown that the G-ROM often introduces inaccuracy in reproducing even the large-scale dynamics of the FOM. Such a finding implies the importance of building approximate functions to compensate the contribution from the residuals. If these functions depend only on the state variables of the ROM, then they are known as the closures since they allow the ROM to be closed [30, 31, 21, 23, 33]. In most of the existing methods, the closure terms are assumed to be a linear or quadratic function of the entire ROM state variables, which are (v,w) in the setup of (2).
The new ROM to be developed here has several fundamental differences compared with the existing ROMs. First, only the first three terms on the right hand side of (2) are retained in the new ROM framework, which represent the basic dynamics. In other words, the self nonlinear interactions of w are omitted. One motivation for such a simplification is that w is a smaller scale variable compared with v and therefore it is often a faster variable and contains less energy. As a consequence, w∗Bvw(v)w and w∗Bvw(w)w are expected to have a weaker role than the other nonlinear terms in (2). Another important reason is related to the development of an efficient data assimilation scheme for the ROM, which will be discussed in § 2.2. Next, a hybrid closure configuration is introduced to approximate the residual part as well as the contribution of the self nonlinear interactions of w. Here, the closure includes both a set of deterministic functions and a stochastic component. The building block of the deterministic part of the closure in this new ROM framework is very distinct fromthat in the existing methods, which is either linear or quadratic nonlinear in terms of both v and w. Here, a special type of nonlinear functions are adopted, which are fully nonlinear with respect to v but are conditional linear with respect to w. The motivation is again that, due to the relatively faster nature of w, a conditional linear function and stochastic terms can effectively approximate the statistical behavior of w and its feedback to the large-scale component, v. With these new ingredients, the resulting ROM with the closure reads:
where W˙v and W˙w are the standard Gaussian white noises. The noise coefficients σv and σw are constant matrices and for simplicity they are diagonal. The four matrix or vector functions in the closure terms τw(v)(v), τv(v)(v), τw(w)(v), and τv(w)(v) can be arbitrarily nonlinear functions of v. It is worthwhile to highlight that the conditional linear functions of w are not linear functions of v. The conditional linearity means that w is linear with v being fixed. Therefore, the closure in (3) includes the effects from the nonlinear interactions between v and w. Here, for the consistency with the quadratic nonlinearity in the starting model (2), the closure terms $\tau_{w}^{(v)}(\mathbf{v})\mathbf{w} + \tau_{v}^{(v)}(\mathbf{v}) and τw(w)(v)w+τv(w)(v) are assumed to be quadratic, and therefore
References
[1]G. K. Vallis, Atmospheric and oceanic fluid dynamics, Cambridge University Press, 2017.
[2]S. H. Strogatz, Nonlinear dynamics and chaos with student solutions manual: With applications to physics, biology, chemistry, and engineering, CRC press, 2018.
[3]D. C. Wilcox, Multiscale model for turbulent flows, AIAA journal 26 (11) (1988) 1311–1320.
[4]S. A. Sheard, A. Mostashari, Principles of complex systems for systems engineering, Systems Engineering 12 (4) (2009) 295–311.
[5]M. Ghil, S. Childress, Topics in geophysical fluid dynamics: atmospheric dynamics, dynamo theory, and climate dynamics, Springer Science & Business Media, 2012.
[6]A. J. Majda, Introduction to turbulent dynamical systems in complex systems, Springer, 2016.
[7]W.-K. Tao, J.-D. Chern, R. Atlas, D. Randall, M. Khairoutdinov, J.-L. Li, D. E. Waliser, A. Hou, X. Lin, C. Peters-Lidard, et al., A multiscale modeling system: Developments, applications, and critical issues, Bulletin of the American Meteorological Society 90 (4) (2009) 515–534.DOI
[8]A. J. Majda, I. Grooms, New perspectives on superparameterization for geophysical turbulence, Journal of Computational Physics 271 (2014) 60–77.
[9]R. Salmon, Lectures on geophysical fluid dynamics, Oxford University Press, 1998.
[10]H. A. Dijkstra, Nonlinear climate dynamics, Cambridge University Press, 2013.
[11]T. Palmer, A nonlinear dynamical perspective on climate change, Weather 48 (10) (1993) 314–326.
[12]M. Farazmand, T. P. Sapsis, Extreme events: Mechanisms and prediction, Applied Mechanics Reviews 71 (5).
[13]K. E. Trenberth, J. T. Fasullo, T. G. Shepherd, Attribution of climate extreme events, Nature Climate Change 5 (8) (2015) 725–730.
[14]H. Moffatt, Extreme events in turbulent flow, Journal of Fluid Mechanics 914.
[15]A. Majda, Introduction to PDEs and Waves for the Atmosphere and Ocean, Vol. 9, American Mathematical Soc., 2003.DOI
[16]P. Manneville, Y. Pomeau, Intermittency and the Lorenz model, Physics Letters A 75 (1-2) (1979) 1–2.DOI
[17]A. J. Majda, Challenges in climate science and contemporary applied mathematics, Communications on Pure and Applied Mathematics 65 (7) (2012) 920–948.DOI
[18]A. J. Majda, N. Chen, Model error, information barriers, state estimation and prediction in complex multiscale systems, Entropy 20 (9) (2018) 644.DOI
[19]J. S. Hesthaven, G. Rozza, B. Stamm, Certified Reduced Basis Methods for Parametrized Partial Differential Equations, Springer, 2015.
[20]P. Holmes, J. L. Lumley, G. Berkooz, Turbulence, Coherent Structures, Dynamical Systems and Symmetry, Cambridge, 1996.
[21]B. R. Noack, M. Morzynski, G. Tadmor, Reduced-order modelling for flow control, Vol. 528, Springer Science & Business Media, 2011.
[22]A. Quarteroni, A. Manzoni, F. Negri, Reduced Basis Methods for Partial Differential Equations: An Introduction, Vol. 92, Springer, 2015.
[23]K. Taira, M. S. Hemati, S. L. Brunton, Y. Sun, K. Duraisamy, S. Bagheri, S. T. Dawson, C.-A. Yeh, Modal analysis of fluid flows: Applications and outlook, AIAA journal 58 (3) (2020) 998–1022.arxiv.org/abs/1903.05750
[24]X. Yan, X. Su, Linear regression analysis: theory and computing, World Scientific, 2009.
[25]D. A. Freedman, Statistical models: theory and practice, Cambridge University Press, 2009.
[26]K. Hasselmann, PIPs and POPs: The reduction of complex dynamical systems using principal interaction and oscillation patterns, Journal of Geophysical Research: Atmospheres 93 (D9) (1988) 11015–11021.DOI
[27]F. Kwasniok, The reduction of complex dynamical systems using principal interaction patterns, Physica D: Nonlinear Phenomena 92 (1-2) (1996) 28–60.DOI
[28]C. W. Rowley, I. Mezić, S. Bagheri, P. Schlatter, D. S. Henningson, Spectral analysis of nonlinear flows, Journal of Fluid Mechanics 641 (2009) 115–127.
[29]P. J. Schmid, Dynamic mode decomposition of numerical and experimental data, Journal of Fluid Mechanics 656 (2010) 5–28.
[30]S. E. Ahmed, S. Pawar, O. San, A. Rasheed, T. Iliescu, B. R. Noack, On closures for reduced order models–A spectrum of first-principle to machine-learned avenues, Physics of Fluids 33 (9) (2021) 091301.DOI
[31]K. Carlberg, C. Farhat, J. Cortial, D. Amsallem, The GNAT method for nonlinear model reduction: effective implementation and application to computational fluid dynamics and turbulent flows, Journal of Computational Physics 242 (2013) 623–647.
[32]C. Mou, B. Koc, O. San, L. G. Rebholz, T. Iliescu, Data-driven variational multi-scale reduced order models, Computer Methods in Applied Mechanics and Engineering 373 (2021) 113470.DOI
[33]X. Xie, M. Mohebujjaman, L. G. Rebholz, T. Iliescu, Data-driven filtered reduced order modeling of fluid flows, SIAM Journal on Scientific Computing 40 (3) (2018) B834–B857.
[34]A. J. Majda, I. Timofeyev, E. Vanden Eijnden, A mathematical framework for stochastic climate models, Communications on Pure and Applied Mathematics 54 (8) (2001) 891–974.
[35]A. J. Majda, I. Timofeyev, E. V. Eijnden, Models for stochastic climate prediction, Proceedings of the National Academy of Sciences 96 (26) (1999) 14687–14691.
[36]F. Lu, Data-driven model reduction for stochastic burgers equations, Entropy 22 (12) (2020) 1360.
[37]K. K. Lin, F. Lu, Data-driven model reduction, wiener projections, and the Koopman-Mori-Zwanzig formalism, Journal of Computational Physics 424 (2021) 109864.DOI
[38]A. J. Majda, J. Harlim, Physics constrained nonlinear regression models for time series, Nonlinearity 26 (1) (2012) 201.DOI
[39]J. Harlim, A. Mahdi, A. J. Majda, An ensemble Kalman filter for statistical estimation of physics constrained nonlinear regression models, Journal of Computational Physics 257 (2014) 782–812.DOI
[40]D. Kondrashov, M. D. Chekroun, M. Ghil, Data-driven non-Markovian closure models, Physica D: Nonlinear Phenomena 297 (2015) 33–55.
[41]T. N. Palmer, A nonlinear dynamical perspective on model error: A proposal for non-local stochastic-dynamic parametrization in weather and climate prediction models, Quarterly Journal of the Royal Meteorological Society 127 (572) (2001) 279–304.
[42]A. J. Majda, B. Gershgorin, Improving model fidelity and sensitivity for complex systems through empirical information theory, Proceedings of the National Academy of Sciences 108 (25) (2011) 10044–10049.DOI
[43]D. Crommelin, E. Vanden-Eijnden, Subgrid-scale parameterization with conditional Markov chains, Journal of the Atmospheric Sciences 65 (8) (2008) 2661–2675.DOI
[44]T. J. Phillips, G. L. Potter, D. L. Williamson, R. T. Cederwall, J. S. Boyle, M. Fiorino, J. J. Hnilo, J. G. Olson, S. Xie, J. J. Yio, Evaluating parameterizations in general circulation models: Climate simulation meets weather prediction, Bulletin of the American Meteorological Society 85 (12) (2004) 1903–1916.DOI
[45]M. Branicki, N. Chen, A. J. Majda, Non-Gaussian test models for prediction and state estimation with model errors, Chinese Annals of Mathematics, Series B 34 (1) (2013) 29–64.
[46]G. Evensen, Data assimilation: the ensemble Kalman filter, Springer Science & Business Media, 2009.
[47]E. Kalnay, Atmospheric modeling, data assimilation and predictability, Cambridge University Press, 2003.
[48]K. Law, A. Stuart, K. Zygalakis, Data assimilation, Cham, Switzerland: Springer 214.
[49]A. J. Majda, J. Harlim, Filtering complex turbulent systems, Cambridge University Press, 2012.
[50]A. J. Majda, Y. Yuan, Fundamental limitations of ad hoc linear and quadratic multi-level regression models for physical systems, Discrete and Continuous Dynamical Systems B 17 (4) (2012) 1333–1363.
[51]N. Chen, A. Majda, Conditional Gaussian systems for multiscale nonlinear stochastic systems: Prediction, state estimation and uncertainty quantification, Entropy 20 (7) (2018) 509.DOI
[52]W. E, K. Khanin, A. Mazel, Y. Sinai, Invariant measures for Burgers equation with stochastic forcing, Annals of Mathematics (2000) 877–960.
[53]W. E, K. Khanin, A. Mazel, Y. Sinai, Probability distribution functions for the random forced Burgers equation, Physical Review Letters 78 (10) (1997) 1904.
[54]N. Chen, A. J. Majda, Filtering nonlinear turbulent dynamical systems through conditional Gaussian statistics, Monthly Weather Review 144 (12) (2016) 4885–4917.DOI
[55]N. Chen, A. J. Majda, D. Giannakis, Predicting the cloud patterns of the Madden-Julian oscillation through a low-order nonlinear stochastic model, Geophysical Research Letters 41 (15) (2014) 5612–5619.
[56]N. Chen, A. J. Majda, Predicting the real-time multivariate Madden–Julian oscillation index through a low-order nonlinear stochastic model, Monthly Weather Review 143 (6) (2015) 2148–2169.DOI
[57]N. Chen, A. J. Majda, X. T. Tong, Information barriers for noisy Lagrangian tracers in filtering random incompressible flows, Nonlinearity 27 (9) (2014) 2133.DOI
[58]N. Chen, A. J. Majda, X. T. Tong, Noisy Lagrangian tracers for filtering random rotating compressible flows, Journal of Nonlinear Science 25 (3) (2015) 451–488.DOI
[59]M. Branicki, A. J. Majda, Dynamic stochastic superresolution of sparsely observed turbulent systems, Journal of Computational Physics 241 (2013) 333–363.DOI
[60]S. R. Keating, A. J. Majda, K. S. Smith, New methods for estimating ocean eddy heat transport using satellite altimetry, Monthly Weather Review 140 (5) (2012) 1703–1722.DOI
[61]M. Kaercher, S. Boyaval, M. A. Grepl, K. Veroy, Reduced basis approximation and a posteriori error bounds for 4D-Var data assimilation, Optim. Eng. (2018) 1–33.
[62]Y. Maday, A. T. Patera, J. D. Penn, M. Yano, A parameterized-background data-weak approach to variational data assimilation: formulation, analysis, and application to acoustics, Int. J. Num. Meth. Engng. 102 (5) (2015) 933–965.DOI
[63]S. Pagani, A. Manzoni, A. Quarteroni, Efficient state/parameter estimation in nonlinear unsteady PDEs by a reduced basis ensemble Kalman filter, SIAM-ASA J. Uncertain. 5 (1) (2017) 890–921.DOI
[64]A. A. Popov, C. Mou, T. Iliescu, A. Sandu, A multifidelity ensemble Kalman filter with reduced order control variates, SIAM J. Sci. Comput. 43 (2) (2021) A1134–A1162.
[65]R. Ștefănescu, A. Sandu, I. M. Navon, POD/DEIM reduced-order strategies for efficient four dimensional variational data assimilation, J. Comput. Phys. 295 (2015) 569–595.DOI
[66]D. Xiao, J. Du, F. Fang, C. C. Pain, J. Li, Parameterised non-intrusive reduced order methods for ensemble Kalman filter data assimilation, Comput. & Fluids 177 (2018) 69–77.DOI
[67]C. Zerfas, L. G. Rebhol z, M. Schneier, T. Iliescu, Continuous data assimilation reduced order models of fluid flow, Comput. Meth. Appl. Mech. Eng. 357 (2019) 112596.
[68]W. Lahoz, B. Khattatov, R. Ménard, Data assimilation and information, in: Data Assimilation, Springer, 2010, pp. 3–12.
[69]J. S. Whitaker, T. M. Hamill, Ensemble data assimilation without perturbed observations, Monthly weather review 130 (7) (2002) 1913–1924.
[70]B. R. Hunt, E. J. Kostelich, I. Szunyogh, Efficient data assimilation for spatiotemporal chaos: A local ensemble transform kalman filter, Physica D: Nonlinear Phenomena 230 (1-2) (2007) 112–126.DOI
[71]P. L. Houtekamer, F. Zhang, Review of the ensemble kalman filter for atmospheric data assimilation, Monthly Weather Review 144 (12) (2016) 4489–4532.
[72]R. S. Liptser, A. N. Shiryaev, Statistics of random processes II: Applications, Vol. 6, Springer Science & Business Media, 2013.
[73]R. E. Kalman, R. S. Bucy, New results in linear filtering and prediction theory, Journal of Basic Engineering 83 (1) (1961) 95–108.DOI
[74]S. Saha, S. Moorthi, H.-L. Pan, X. Wu, J. Wang, S. Nadiga, P. Tripp, R. Kistler, J. Woollen, D. Behringer, et al., The ncep climate forecast system reanalysis, Bulletin of the American Meteorological Society 91 (8) (2010) 1015–1058.DOI
[75]N. Chen, Learning nonlinear turbulent dynamics from partial observations via analytically solvable conditional statistics, Journal of Computational Physics 418 (2020) 109635.DOI
[76]S. Boyd, L. Vandenberghe, Convex optimization, Cambridge University Press, 2004.
[77]K. Bergemann, S. Reich, An ensemble Kalman-Bucy filter for continuous data assimilation, Meteorologische Zeitschrift 21 (2012) 213–219.
[78]R. J. Greatbatch, B. T. Nadiga, Four-gyre circulation in a barotropic model with double-gyre wind forcing, Journal of Physical Oceanography 30 (6) (2000) 1461–1471.doi.org/10.1175/1520-0485(2000)030<1461:fgciab>2.0.co;2
[79]C. Mou, H. Liu, D. R. Wells, T. Iliescu, Data-driven correction reduced order models for the quasi-geostrophic equations: A numerical investigation, International Journal of Computational Fluid Dynamics 34 (2) (2020) 147–159.DOI
[80]O. San, T. Iliescu, A stabilized proper orthogonal decomposition reduced-order model for large scale quasigeostrophic ocean circulation, Advances in Computational Mathematics (2015) 1289–1319.DOI
[81]O. San, A. E. Staples, Z. Wang, T. Iliescu, Approximate deconvolution large eddy simulation of a barotropic ocean circulation model, Ocean Modelling 40 (2011) 120–132.DOI
[82]S. Kullback, R. A. Leibler, On information and sufficiency, The Annals of Mathematical Statistics 22 (1) (1951) 79–86.
[83]S. Kullback, Letter to the editor: The kullback-leibler distance, American Statistician.
[84]R. Kleeman, Information theory and dynamical system predictability, Entropy 13 (3) (2011) 612–649.
[85]A. Majda, R. V. Abramov, M. J. Grote, Information theory and stochastics for multiscale nonlinear systems, Vol. 25, American Mathematical Soc., 2005.DOI
[86]S. L. Brunton, J. L. Proctor, J. N. Kutz, Discovering governing equations from data by sparse identification of nonlinear dynamical systems, Proceedings of the National Academy of Sciences 113 (15) (2016) 3932–3937.DOI
[87]J. Sun, E. M. Bollt, Causation entropy identifies indirect influences, dominance of neighbors and anticipatory couplings, Physica D: Nonlinear Phenomena 267 (2014) 49–57.DOI
[88]J. Elinger, Information theoretic causality measures for parameter estimation and system identification, Ph.D. thesis, Georgia Institute of Technology (2020).