Evelyn Lunasin’s research while affiliated with United States Naval Academy and other places

This study presents a method, along with its algorithmic and computational framework implementation, and performance verification for dynamical system identification. The approach incorporates insights from phase space structures, such as attractors and their basins. By understanding these structures, we have improved training and testing strategies for operator learning and system identification. Our method uses time delay and non-linear maps rather than embeddings, enabling the assessment of algorithmic accuracy and expressibility, particularly in systems exhibiting multiple attractors. This method, along with its associated algorithm and computational framework, offers broad applicability across various scientific and engineering domains, providing a useful tool for data-driven characterization of systems with complex nonlinear system dynamics.

Mesoscale eddies are critical in ocean circulation and the global climate system. Standard eddy identification methods are usually based on deterministic optimal point estimates of the ocean flow field, which produce a single best estimate without accounting for inherent uncertainties in the data. However, large uncertainty exists in estimating the flow field due to the use of noisy, sparse, and indirect observations, as well as turbulent flow models. When this uncertainty is overlooked, the accuracy of eddy identification is significantly affected. This paper presents a general probabilistic eddy identification framework that adapts existing methods to incorporate uncertainty into the diagnostic. The framework begins by sampling an ensemble of ocean realizations from the probabilistic state estimation, which considers uncertainty. Traditional eddy diagnostics are then applied to individual realizations, and the corresponding eddy statistics are aggregated from these diagnostic results. The framework is applied to a scenario mimicking the Beaufort Gyre marginal ice zone, where large uncertainty appears in estimating the ocean field using Lagrangian data assimilation with sparse ice floe trajectories. The probabilistic eddy identification precisely characterizes the contribution from the turbulent fluctuations and provides additional insights in comparison to its deterministic counterpart. The skills in counting the number of eddies and computing the probability of the eddy are both significantly improved under the probabilistic framework. Notably, large biases appear when estimating the eddy lifetime using deterministic methods. In contrast, probabilistic eddy identification not only provides a closer estimate but also quantifies the uncertainty in inferring such a crucial dynamical quantity.

Determining the optimal locations for placing extra observational measurements has practical significance. However, the exact underlying flow field is never known in practice. Significant uncertainty appears when the flow field is inferred from a limited number of existing observations via data assimilation or statistical forecast. In this paper, a new computationally efficient strategy for deploying Lagrangian drifters that highlights the central role of uncertainty is developed. A nonlinear trajectory diagnostic approach that underlines the importance of uncertainty is built to construct a phase portrait map. It consists of both the geometric structure of the underlying flow field and the uncertainty in the estimated state from Lagrangian data assimilation. The drifters are deployed at the maxima of this map and are required to be separated enough. Such a strategy allows the drifters to travel the longest distances to collect both the local and global information of the flow field. It also facilitates the reduction of a significant amount of uncertainty. To characterize the uncertainty, the estimated state is given by a probability density function (PDF). An information metric is then introduced to assess the information gain in such a PDF, which is fundamentally different from the traditional path-wise measurements. The information metric also avoids using the unknown truth to quantify the uncertainty reduction, making the method practical. Mathematical analysis exploiting simple illustrative examples is used to validate the strategy. Numerical simulations based on multiscale turbulent flows are then adopted to demonstrate the advantages of this strategy over some other methods.

Lagrangian descriptors provide a global dynamical picture of the geometric structures for arbitrarily time-dependent flows with broad applications. This paper develops a mathematical framework for computing Lagrangian descriptors when uncertainty appears. The uncertainty originates from estimating the underlying flow field as a natural consequence of data assimilation or statistical forecast. It also appears in the resulting Lagrangian trajectories. The uncertainty in the flow field directly affects the path integration of the crucial nonlinear positive scalar function in computing the Lagrangian descriptor, making it fundamentally different from many other diagnostic methods. Despite being highly nonlinear and non-Gaussian, closed analytic formulae are developed to efficiently compute the expectation of such a scalar function due to the uncertain velocity field by exploiting suitable approximations. A rapid and accurate sampling algorithm is then built to assist the forecast of the probability density function (PDF) of the Lagrangian trajectories. Such a PDF provides the weight to combine the Lagrangian descriptors along different paths. Simple but illustrative examples are designed to show the distinguished behavior of using Lagrangian descriptors in revealing the flow field when uncertainty appears. Uncertainty can either completely erode the coherent structure or barely affect the underlying geometry of the flow field. The method is also applied for eddy identification, indicating that uncertainty has distinct impacts on detecting eddies at different time scales. Finally, when uncertainty is incorporated into the Lagrangian descriptor for inferring the source target, the likelihood criterion provides a very different conclusion from the deterministic methods.

Understanding the impact of model error on data assimilation is an important practical topic. Model error in the subgrid scale is commonly seen in various applications as a natural consequence of stochastic parameterization, coarse-graining model reduction and many other approximations. The goal of this paper is to analyze and understand the impact of such a model error in affecting the nudging data assimilation (DA). To capture realistic features of turbulence, the Navier–Stokes equation (NSE) is utilized as the perfect model that provides the reference solution and creates the observations while a regularized NSE is adopted as the forecast model in DA. We extend the Larios–Pei method and establish our analytical and numerical results taking the simplified Bardina sub-grid scale model as a regularization for the NSE. The proof of the well-posedness and the convergence of the solutions of the DA to the truth, up to a bounded error that depends on a power of the regularization parameter, is shown. On the other hand, the numerical study of the proposed algorithm exploits the computationally more efficient 3D Sabra shell model and its ‘regularized’ counterpart, the characteristics of which mimic the energy cascade of the 3D NSE. While the proposed algorithm has a comparable performance as the Larios-Pei one in the high Reynolds number regime in successfully reproducing the energy spectrum and recovering the path-wise solutions up to the model error in the small scales, the proposed algorithm demonstrates certain improved behavior when the Reynolds number is moderate.

Complex nonlinear turbulent dynamical systems are ubiquitous in many areas of research. Recovering unobserved state variables is an important topic for the data assimilation of turbulent systems. In this article, an efficient continuous-in-time data assimilation scheme is developed, which exploits closed analytic formulas for updating the unobserved state variables. Therefore, it is computationally efficient and accurate. The new data assimilation scheme is combined with a simple reduced order modeling technique that involves a cheap closure approximation and noise inflation. In such a way, many complicated turbulent dynamical systems can satisfy the requirements of the mathematical structures for the proposed efficient data assimilation scheme. The new data assimilation scheme is then applied to the Sabra shell model, which is a conceptual model for nonlinear turbulence. The goal is to recover the unobserved shell velocities across different spatial scales. It is shown that the new data assimilation scheme is skillful in capturing the nonlinear features of turbulence including the intermittency and extreme events in both the chaotic and the turbulent dynamical regimes. It is also shown that the new data assimilation scheme is more accurate and computationally cheaper than the standard ensemble Kalman filter and nudging data assimilation schemes for assimilating the Sabra shell model with partial observations.

Complex nonlinear turbulent dynamical systems are ubiquitous in many areas. Recovering unobserved state variables is an important topic for the data assimilation of turbulent systems. In this article, an efficient continuous in time data assimilation scheme is developed, which exploits closed analytic formulae for updating the unobserved state variables. Therefore, it is computationally efficient and accurate. The new data assimilation scheme is combined with a simple reduced order modeling technique that involves a cheap closure approximation and a noise inflation. In such a way, many complicated turbulent dynamical systems can satisfy the requirements of the mathematical structures for the proposed efficient data assimilation scheme. The new data assimilation scheme is then applied to the Sabra shell model, which is a conceptual model for nonlinear turbulence. The goal is to recover the unobserved shell velocities across different spatial scales. It has been shown that the new data assimilation scheme is skillful in capturing the nonlinear features of turbulence including the intermittency and extreme events in both the chaotic and the turbulent dynamical regimes. It has also been shown that the new data assimilation scheme is more accurate and computationally cheaper than the standard ensemble Kalman filter and nudging data assimilation schemes for assimilating the Sabra shell model.