@phdthesis{Middelanis2023,
  author    = {Middelanis, Robin},
  title     = {Global response to local extremes—a storyline approach on economic loss propagation from weather extremes},
  doi       = {10.25932/publishup-61112},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-611127},
  school      = {Universit{\"a}t Potsdam},
  pages     = {vii, 237},
  year      = {2023},
  abstract  = {Due to anthropogenic greenhouse gas emissions, Earth's average surface temperature is steadily increasing. As a consequence, many weather extremes are likely to become more frequent and intense. This poses a threat to natural and human systems, with local impacts capable of destroying exposed assets and infrastructure, and disrupting economic and societal activity. Yet, these effects are not locally confined to the directly affected regions, as they can trigger indirect economic repercussions through loss propagation along supply chains. As a result, local extremes yield a potentially global economic response. To build economic resilience and design effective adaptation measures that mitigate adverse socio-economic impacts of ongoing climate change, it is crucial to gain a comprehensive understanding of indirect impacts and the underlying economic mechanisms. Presenting six articles in this thesis, I contribute towards this understanding. To this end, I expand on local impacts under current and future climate, the resulting global economic response, as well as the methods and tools to analyze this response. Starting with a traditional assessment of weather extremes under climate change, the first article investigates extreme snowfall in the Northern Hemisphere until the end of the century. Analyzing an ensemble of global climate model projections reveals an increase of the most extreme snowfall, while mean snowfall decreases. Assessing repercussions beyond local impacts, I employ numerical simulations to compute indirect economic effects from weather extremes with the numerical agent-based shock propagation model Acclimate. This model is used in conjunction with the recently emerged storyline framework, which involves analyzing the impacts of a particular reference extreme event and comparing them to impacts in plausible counterfactual scenarios under various climate or socio-economic conditions. Using this approach, I introduce three primary storylines that shed light on the complex mechanisms underlying economic loss propagation. In the second and third articles of this thesis, I analyze storylines for the historical Hurricanes Sandy (2012) and Harvey (2017) in the USA. For this, I first estimate local economic output losses and then simulate the resulting global economic response with Acclimate. The storyline for Hurricane Sandy thereby focuses on global consumption price anomalies and the resulting changes in consumption. I find that the local economic disruption leads to a global wave-like economic price ripple, with upstream effects propagating in the supplier direction and downstream effects in the buyer direction. Initially, an upstream demand reduction causes consumption price decreases, followed by a downstream supply shortage and increasing prices, before the anomalies decay in a normalization phase. A dominant upstream or downstream effect leads to net consumption gains or losses of a region, respectively. Moreover, I demonstrate that a longer direct economic shock intensifies the downstream effect for many regions, leading to an overall consumption loss. The third article of my thesis builds upon the developed loss estimation method by incorporating projections to future global warming levels. I use these projections to explore how the global production response to Hurricane Harvey would change under further increased global warming. The results show that, while the USA is able to nationally offset direct losses in the reference configuration, other countries have to compensate for increasing shares of counterfactual future losses. This compensation is mainly achieved by large exporting countries, but gradually shifts towards smaller regions. These findings not only highlight the economy's ability to flexibly mitigate disaster losses to a certain extent, but also reveal the vulnerability and economic disadvantage of regions that are exposed to extreme weather events. The storyline in the fourth article of my thesis investigates the interaction between global economic stress and the propagation of losses from weather extremes. I examine indirect impacts of weather extremes — tropical cyclones, heat stress, and river floods — worldwide under two different economic conditions: an unstressed economy and a globally stressed economy, as seen during the Covid-19 pandemic. I demonstrate that the adverse effects of weather extremes on global consumption are strongly amplified when the economy is under stress. Specifically, consumption losses in the USA and China double and triple, respectively, due to the global economy's decreased capacity for disaster loss compensation. An aggravated scarcity intensifies the price response, causing consumption losses to increase. Advancing on the methods and tools used here, the final two articles in my thesis extend the agent-based model Acclimate and formalize the storyline approach. With the model extension described in the fifth article, regional consumers make rational choices on the goods bought such that their utility is maximized under a constrained budget. In an out-of-equilibrium economy, these rational consumers are shown to temporarily increase consumption of certain goods in spite of rising prices. The sixth article of my thesis proposes a formalization of the storyline framework, drawing on multiple studies including storylines presented in this thesis. The proposed guideline defines eight central elements that can be used to construct a storyline. Overall, this thesis contributes towards a better understanding of economic repercussions of weather extremes. It achieves this by providing assessments of local direct impacts, highlighting mechanisms and impacts of loss propagation, and advancing on methods and tools used.},
  language  = {en}
}
@phdthesis{Kraemer2021,
  author    = {Kr{\"a}mer, Kai Hauke},
  title     = {Towards a robust framework for recurrence analysis},
  doi       = {10.25932/publishup-53874},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-538743},
  school      = {Universit{\"a}t Potsdam},
  pages     = {xlii, 217},
  year      = {2021},
  abstract  = {In our daily life, recurrence plays an important role on many spatial and temporal scales and in different contexts. It is the foundation of learning, be it in an evolutionary or in a neural context. It therefore seems natural that recurrence is also a fundamental concept in theoretical dynamical systems science. The way in which states of a system recur or develop in a similar way from similar initial states makes it possible to infer information about the underlying dynamics of the system. The mathematical space in which we define the state of a system (state space) is often high dimensional, especially in complex systems that can also exhibit chaotic dynamics. The recurrence plot (RP) enables us to visualize the recurrences of any high-dimensional systems in a two-dimensional, binary representation. Certain patterns in RPs can be related to physical properties of the underlying system, making the qualitative and quantitative analysis of RPs an integral part of nonlinear systems science. The presented work has a methodological focus and further develops recurrence analysis (RA) by addressing current research questions related to an increasing amount of available data and advances in machine learning techniques. By automatizing a central step in RA, namely the reconstruction of the state space from measured experimental time series, and by investigating the impact of important free parameters this thesis aims to make RA more accessible to researchers outside of physics. The first part of this dissertation is concerned with the reconstruction of the state space from time series. To this end, a novel idea is proposed which automates the reconstruction problem in the sense that there is no need to preprocesse the data or estimate parameters a priori. The key idea is that the goodness of a reconstruction can be evaluated by a suitable objective function and that this function is minimized in the embedding process. In addition, the new method can process multivariate time series input data. This is particularly important because multi-channel sensor-based observations are ubiquitous in many research areas and continue to increase. Building on this, the described minimization problem of the objective function is then processed using a machine learning approach. In the second part technical and methodological aspects of RA are discussed. First, we mathematically justify the idea of setting the most influential free parameter in RA, the recurrence threshold ε, in relation to the distribution of all pairwise distances in the data. This is especially important when comparing different RPs and their quantification statistics and is fundamental to any comparative study. Second, some aspects of recurrence quantification analysis (RQA) are examined. As correction schemes for biased RQA statistics, which are based on diagonal lines, we propose a simple method for dealing with border effects of an RP in RQA and a skeletonization algorithm for RPs. This results in less biased (diagonal line based) RQA statistics for flow-like data. Third, a novel type of RQA characteristic is developed, which can be viewed as a generalized non-linear powerspectrum of high dimensional systems. The spike powerspectrum transforms a spike-train like signal into its frequency domain. When transforming the diagonal line-dependent recurrence rate (τ-RR) of a RP in this way, characteristic periods, which can be seen in the state space representation of the system can be unraveled. This is not the case, when Fourier transforming τ-RR. Finally, RA and RQA are applied to climate science in the third part and neuroscience in the fourth part. To the best of our knowledge, this is the first time RPs and RQA have been used to analyze lake sediment data in a paleoclimate context. Therefore, we first elaborate on the basic formalism and the interpretation of visually visible patterns in RPs in relation to the underlying proxy data. We show that these patterns can be used to classify certain types of variability and transitions in the Potassium record from six short (< 17m) sediment cores collected during the Chew Bahir Drilling Project. Building on this, the long core (∼ m composite) from the same site is analyzed and two types of variability and transitions are identified and compared with ODP Site  wetness index from the eastern Mediterranean. Type  variability likely reflects the influence of precessional forcing in the lower latitudes at times of maximum values of the long eccentricity cycle ( kyr) of the earth's orbit around the sun, with a tendency towards extreme events. Type  variability appears to be related to the minimum values of this cycle and corresponds to fairly rapid transitions between relatively dry and relatively wet conditions. In contrast, RQA has been applied in the neuroscientific context for almost two decades. In the final part, RQA statistics are used to quantify the complexity in a specific frequency band of multivariate EEG (electroencephalography) data. By analyzing experimental data, it can be shown that the complexity of the signal measured in this way across the sensorimotor cortex decreases as motor tasks are performed. The results are consistent with and comple- ment the well known concepts of motor-related brain processes. We assume that the thus discovered features of neuronal dynamics in the sensorimotor cortex together with the robust RQA methods for identifying and classifying these contribute to the non-invasive EEG-based development of brain-computer interfaces (BCI) for motor control and rehabilitation. The present work is an important step towards a robust analysis of complex systems based on recurrence.},
  language  = {en}
}
@phdthesis{Hesse2018,
  author    = {Hesse, Cornelia},
  title     = {Integrated water quality modelling in meso- to large-scale catchments of the Elbe river basin under climate and land use change},
  doi       = {10.25932/publishup-42295},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-422957},
  school      = {Universit{\"a}t Potsdam},
  pages     = {ix, 217},
  year      = {2018},
  abstract  = {In einer sich {\"a}ndernden Umwelt sind Fließgew{\"a}sser{\"o}kosysteme vielf{\"a}ltigen direkten und indirekten anthropogenen Belastungen ausgesetzt, die die Gew{\"a}sser sowohl in ihrer Menge als auch in ihrer G{\"u}te beeintr{\"a}chtigen k{\"o}nnen. Ein {\"u}berm{\"a}ßiger Eintrag von N{\"a}hrstoffen verursacht etwa Massenentwicklungen von Algen und Sauerstoffdefizite in den Gew{\"a}ssern, was zum Verfehlen der Ziele der Wasserrahmenrichtlinie (WRRL) f{\"u}hren kann. In vielen europ{\"a}ischen Einzugsgebieten und auch dem der Elbe sind solche Probleme zu beobachten. W{\"a}hrend der letzten Jahrzehnte entstanden diverse computergest{\"u}tzte Modelle, die zum Schutz und Management von Wasserressourcen genutzt werden k{\"o}nnen. Sie helfen beim Verstehen der N{\"a}hrstoffprozesse und Belastungspfade in Einzugsgebieten, bei der Absch{\"a}tzung m{\"o}glicher Folgen von Klima- und Landnutzungs{\"a}nderungen f{\"u}r die Wasserk{\"o}rper, sowie bei der Entwicklung eventueller Kompensationsmaßnahmen. Aufgrund der Vielzahl an sich gegenseitig beeinflussenden Prozessen ist die Modellierung der Wasserqualit{\"a}t komplexer und aufw{\"a}ndiger als eine reine hydrologische Modellierung. {\"O}kohydrologische Modelle zur Simulation der Gew{\"a}sserg{\"u}te, einschließlich des Modells SWIM (Soil and Water Integrated Model), bed{\"u}rfen auch h{\"a}ufig noch einer Weiterentwicklung und Verbesserung der Prozessbeschreibungen. Aus diesen {\"U}berlegungen entstand die vorliegende Dissertation, die sich zwei Hauptanliegen widmet: 1) einer Weiterentwicklung des N{\"a}hrstoffmoduls des {\"o}kohydrologischen Modells SWIM f{\"u}r Stickstoff- und Phosphorprozesse, und 2) der Anwendung des Modells SWIM im Elbegebiet zur Unterst{\"u}tzung eines anpassungsf{\"a}higen Wassermanagements im Hinblick auf m{\"o}gliche zuk{\"u}nftige {\"A}nderungen der Umweltbedingungen. Die kumulative Dissertation basiert auf f{\"u}nf wissenschaftlichen Artikeln, die in internationalen Zeitschriften ver{\"o}ffentlicht wurden. Im Zuge der Arbeit wurden verschiedene Modellanpassungen in SWIM vorgenommen, wie etwa ein einfacher Ansatz zur Verbesserung der Simulation der Wasser- und N{\"a}hrstoffverh{\"a}ltnisse in Feuchtgebieten, ein um Ammonium erweiterter Stickstoffkreislauf im Boden, sowie ein Flussprozessmodul, das Umwandlungsprozesse, Sauerstoffverh{\"a}ltnisse und Algenwachstum im Fließgew{\"a}sser simuliert, haupts{\"a}chlich angetrieben von Temperatur und Licht. Auch wenn dieser neue Modellansatz ein sehr komplexes Modell mit einer Vielzahl an neuen Kalibrierungsparametern und steigender Unsicherheit erzeugte, konnten gute Ergebnisse in den Teileinzugsgebieten und dem gesamten Gebiet der Elbe erzielt werden, so dass das Modell zur Absch{\"a}tzung m{\"o}glicher Folgen von Klimavariabilit{\"a}ten und ver{\"a}nderten anthropogenen Einfl{\"u}ssen f{\"u}r die Gew{\"a}sserg{\"u}te genutzt werden konnte. Das neue Fließgew{\"a}ssermodul ist ein wichtiger Beitrag zur Verbesserung der N{\"a}hrstoffmodellierung in SWIM, vor allem f{\"u}r Stoffe, die haupts{\"a}chlich aus Punktquellen in die Gew{\"a}sser gelangen (wie z.B. Phosphat). Der neue Modellansatz verbessert zudem die Anwendbarkeit von SWIM f{\"u}r Fragestellungen im Zusammenhang mit der WRRL, bei der biologische Qualit{\"a}tskomponenten (wie etwa Phytoplankton) eine zentrale Rolle spielen. Die dargestellten Ergebnisse der Wirkungsstudien k{\"o}nnen bei Entscheidungstr{\"a}gern und anderen Akteuren das Verst{\"a}ndnis f{\"u}r zuk{\"u}nftige Herausforderungen im Gew{\"a}ssermanagement erh{\"o}hen und dazu beitragen, ein angepasstes Management f{\"u}r das Elbeeinzugsgebiet zu entwickeln.},
  language  = {en}
}