TY - THES A1 - Angwenyi, David T1 - Time-continuous state and parameter estimation with application to hyperbolic SPDEs T1 - Zeitkontinuierliche Zustands- und Parameterschätzung bei Anwendung auf hyperbolische SPDEs N2 - Data assimilation has been an active area of research in recent years, owing to its wide utility. At the core of data assimilation are filtering, prediction, and smoothing procedures. Filtering entails incorporation of measurements' information into the model to gain more insight into a given state governed by a noisy state space model. Most natural laws are governed by time-continuous nonlinear models. For the most part, the knowledge available about a model is incomplete; and hence uncertainties are approximated by means of probabilities. Time-continuous filtering, therefore, holds promise for wider usefulness, for it offers a means of combining noisy measurements with imperfect model to provide more insight on a given state. The solution to time-continuous nonlinear Gaussian filtering problem is provided for by the Kushner-Stratonovich equation. Unfortunately, the Kushner-Stratonovich equation lacks a closed-form solution. Moreover, the numerical approximations based on Taylor expansion above third order are fraught with computational complications. For this reason, numerical methods based on Monte Carlo methods have been resorted to. Chief among these methods are sequential Monte-Carlo methods (or particle filters), for they allow for online assimilation of data. Particle filters are not without challenges: they suffer from particle degeneracy, sample impoverishment, and computational costs arising from resampling. The goal of this thesis is to:— i) Review the derivation of Kushner-Stratonovich equation from first principles and its extant numerical approximation methods, ii) Study the feedback particle filters as a way of avoiding resampling in particle filters, iii) Study joint state and parameter estimation in time-continuous settings, iv) Apply the notions studied to linear hyperbolic stochastic differential equations. The interconnection between Itô integrals and stochastic partial differential equations and those of Stratonovich is introduced in anticipation of feedback particle filters. With these ideas and motivated by the variants of ensemble Kalman-Bucy filters founded on the structure of the innovation process, a feedback particle filter with randomly perturbed innovation is proposed. Moreover, feedback particle filters based on coupling of prediction and analysis measures are proposed. They register a better performance than the bootstrap particle filter at lower ensemble sizes. We study joint state and parameter estimation, both by means of extended state spaces and by use of dual filters. Feedback particle filters seem to perform well in both cases. Finally, we apply joint state and parameter estimation in the advection and wave equation, whose velocity is spatially varying. Two methods are employed: Metropolis Hastings with filter likelihood and a dual filter comprising of Kalman-Bucy filter and ensemble Kalman-Bucy filter. The former performs better than the latter. N2 - Die Datenassimilation war in den letzten Jahren aufgrund ihres breiten Nutzens ein aktives Forschungsgebiet. Im Zentrum der Datenassimilation stehen Filter-, Vorhersage- und Glättungsverfahren. Die Filterung beinhaltet die Einbeziehung von Messinformationen in das Modell, um einen besseren Einblick in einen gegebenen Zustand zu erhalten, der durch ein verrauschtes Zustandsraummodell gesteuert wird. Die meisten Naturgesetze werden von zeitkontinuierlichen nichtlinearen Modellen bestimmt. Das verfügbare Wissen über ein Modell ist größtenteils unvollständig; und daher werden Unsicherheiten mittels Wahrscheinlichkeiten angenähert. Die zeitkontinuierliche Filterung verspricht daher eine größere Nützlichkeit, denn sie bietet die Möglichkeit, verrauschte Messungen mit einem unvollkommenen Modell zu kombinieren, um mehr Einblick in einen bestimmten Zustand zu erhalten. Das Problem der zeitkontinuierlichen nichtlinearen Gaußschen Filterung wird durch die Kushner-Stratonovich-Gleichung gelöst. Leider fehlt der Kushner-Stratonovich-Gleichung eine geschlossene Lösung. Darüber hinaus sind die numerischen Näherungen, die auf der Taylor-Erweiterung über der dritten Ordnung basieren, mit rechnerischen Komplikationen behaftet. Aus diesem Grund wurde auf numerische Methoden zurückgegriffen, die auf Monte-Carlo-Methoden basieren. Die wichtigsten dieser Methoden sind sequentielle Monte-Carlo-Methoden (oder Partikelfilter), da sie die Online-Assimilation von Daten ermöglichen. Partikelfilter sind nicht unproblematisch: Sie leiden unter Partikelentartung, Probenverarmung und Rechenkosten, die sich aus der Neuabtastung ergeben. Das Ziel dieser Arbeit ist es, i) die Ableitung der Kushner-Stratonovich-Gleichung aus den ersten Prinzipien und ihre vorhandenen numerischen Approximationsmethoden zu überprüfen, ii) die Rückkopplungs-Partikelfilter zu untersuchen, um eine Neuabtastung in Partikelfiltern zu vermeiden, iii) Studieren Sie die Zustands- und Parameterschätzung in zeitkontinuierlichen Einstellungen, iv) Wenden Sie die untersuchten Begriffe auf lineare hyperbolische stochastische Differentialgleichungen an. Die Verbindung zwischen Itô Integralen und stochastischen partiellen Differentialgleichungen und denen von Stratonovich wird in Erwartung von Rückkopplungs-Partikelfiltern eingeführt. Mit diesen Ideen und motiviert durch die Varianten von Kalman-Bucy-Filtern, die auf der Struktur des Innovationsprozesses gegründet, wird ein Feedback-Partikelfilter mit zufällig gestörter Innovation vorgeschlagen. Darüber hinaus werden Rückkopplungspartikelfilter basierend auf der Kopplung von Vorhersage- und Analysemaßnahmen vorgeschlagen. Diese Feedback-Partikelfiltern haben eine bessere Leistung als der Bootstrap-Partikelfilter bei niedrigeren Ensemble-Größen. Wir untersuchen gemeinsame Zustands- und Parameterschätzungen, sowohl durch erweiterte Zustandsräume als auch durch Verwendung von Doppelfiltern. Rückkopplungs-Partikelfilter scheinen in beiden Fällen gut zu funktionieren. Schließlich wenden wir eine gemeinsame Zustands- und Parameterschätzung in der Advektions-und Wellengleichung an, deren Geschwindigkeit räumlich variiert. Es werden zwei Verfahren verwendet: Metropolis-Hastings mit Filterwahrscheinlichkeit und ein Doppelfilter bestehend aus Kalman-Bucy-Filter und Ensemble-Kalman-Bucy-Filter. Ersteres schneidet besser ab als letzteres. KW - state estimation KW - filtering KW - parameter estimation KW - Zustandsschätzung KW - Filterung KW - Parameter Schätzung Y1 - 2019 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-436542 ER - TY - GEN A1 - Nüsken, Nikolas A1 - Reich, Sebastian A1 - Rozdeba, Paul J. T1 - State and parameter estimation from observed signal increments T2 - Postprints der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe N2 - The success of the ensemble Kalman filter has triggered a strong interest in expanding its scope beyond classical state estimation problems. In this paper, we focus on continuous-time data assimilation where the model and measurement errors are correlated and both states and parameters need to be identified. Such scenarios arise from noisy and partial observations of Lagrangian particles which move under a stochastic velocity field involving unknown parameters. We take an appropriate class of McKean–Vlasov equations as the starting point to derive ensemble Kalman–Bucy filter algorithms for combined state and parameter estimation. We demonstrate their performance through a series of increasingly complex multi-scale model systems. T3 - Zweitveröffentlichungen der Universität Potsdam : Mathematisch-Naturwissenschaftliche Reihe - 916 KW - parameter estimation KW - continuous-time data assimilation KW - ensemble Kalman filter KW - correlated noise KW - multi-scale diffusion processes Y1 - 2020 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-442609 SN - 1866-8372 IS - 916 ER - TY - JOUR A1 - Nüsken, Nikolas A1 - Reich, Sebastian A1 - Rozdeba, Paul J. T1 - State and parameter estimation from observed signal increments JF - Entropy : an international and interdisciplinary journal of entropy and information studies N2 - The success of the ensemble Kalman filter has triggered a strong interest in expanding its scope beyond classical state estimation problems. In this paper, we focus on continuous-time data assimilation where the model and measurement errors are correlated and both states and parameters need to be identified. Such scenarios arise from noisy and partial observations of Lagrangian particles which move under a stochastic velocity field involving unknown parameters. We take an appropriate class of McKean-Vlasov equations as the starting point to derive ensemble Kalman-Bucy filter algorithms for combined state and parameter estimation. We demonstrate their performance through a series of increasingly complex multi-scale model systems. KW - parameter estimation KW - continuous-time data assimilation KW - ensemble Kalman filter KW - correlated noise KW - multi-scale diffusion processes Y1 - 2019 U6 - https://doi.org/10.3390/e21050505 SN - 1099-4300 VL - 21 IS - 5 PB - MDPI CY - Basel ER - TY - JOUR A1 - Rosenbaum, Benjamin A1 - Raatz, Michael A1 - Weithoff, Guntram A1 - Fussmann, Gregor F. A1 - Gaedke, Ursula T1 - Estimating parameters from multiple time series of population dynamics using bayesian inference JF - Frontiers in ecology and evolution N2 - Empirical time series of interacting entities, e.g., species abundances, are highly useful to study ecological mechanisms. Mathematical models are valuable tools to further elucidate those mechanisms and underlying processes. However, obtaining an agreement between model predictions and experimental observations remains a demanding task. As models always abstract from reality one parameter often summarizes several properties. Parameter measurements are performed in additional experiments independent of the ones delivering the time series. Transferring these parameter values to different settings may result in incorrect parametrizations. On top of that, the properties of organisms and thus the respective parameter values may vary considerably. These issues limit the use of a priori model parametrizations. In this study, we present a method suited for a direct estimation of model parameters and their variability from experimental time series data. We combine numerical simulations of a continuous-time dynamical population model with Bayesian inference, using a hierarchical framework that allows for variability of individual parameters. The method is applied to a comprehensive set of time series from a laboratory predator-prey system that features both steady states and cyclic population dynamics. Our model predictions are able to reproduce both steady states and cyclic dynamics of the data. Additionally to the direct estimates of the parameter values, the Bayesian approach also provides their uncertainties. We found that fitting cyclic population dynamics, which contain more information on the process rates than steady states, yields more precise parameter estimates. We detected significant variability among parameters of different time series and identified the variation in the maximum growth rate of the prey as a source for the transition from steady states to cyclic dynamics. By lending more flexibility to the model, our approach facilitates parametrizations and shows more easily which patterns in time series can be explained also by simple models. Applying Bayesian inference and dynamical population models in conjunction may help to quantify the profound variability in organismal properties in nature. KW - Bayesian inference KW - chemostat experiments KW - ordinary differential equation KW - parameter estimation KW - population dynamics KW - predator prey KW - time series analysis KW - trait variability Y1 - 2019 U6 - https://doi.org/10.3389/fevo.2018.00234 SN - 2296-701X VL - 6 PB - Frontiers Research Foundation CY - Lausanne ER -