ALBERT

Description: The branch of seismology that deals with strong motion refers to seismic events that are hazardous to society in general.Two aspects drive the development in strong-motion seismology: First comes the societal need to understand the earth-quake hazard and to mitigate the associated risk. While the hazard changed little during human history, the risk in-creases steadily. A growing population—also in the most earthquake-prone regions of the world—and a more and morevulnerable infrastructure contribute to higher exposure to seismic events and higher vulnerability in case an earthquakestruck. The second driver in strong-motion seismology is shared with many other fields: the technological advancement.The available options for processing more and more data is unprecedented in human history and are still not exhausted.Both drivers also pose new challenges as in how to interpret and make use of the data.The scientific question, on the other hand, is clear: What can we learn from the rupture process (the source of earth-quakes), Earth’s structure (the medium through which seismic wave travels), and their interactions (how does an earth-quake affect its surrounding medium)? The question is broad and this thesis can focus only for specific aspects of thisquestion and provide answers for them. To reach the answers, I developed several new algorithms and models, all rootedin the concept of the likelihood function.Seismicity (and population alike) is concentrated along the tectonic plate boundaries. Different earthquake typesoccur at these boundaries and their characteristics in terms of ground shaking are considerably different. It is thereforeimportant to classify earthquakes according to their style of faulting. This classification is the objective of ACE (angularclusterization with expectation-maximization). Founded on the geomechanical principles, ACE provides earthquakeclassifications which can be applied not only for ground-motion related topics but also to study the Earth’s stress field.The development of reliable ground-motion models requires waveform data of high quality. Instrument related errorscan compromise the data quality, however, with large archives of waveform data, the correction for spurious s cannot behandled manually anymore. To alleviate the effect of instrument related data shifts, I developed the integrated combinedbaseline modification (ICBM). This routine is implemented during the data pre-processing and is particularly necessarywhen determining integrated quantities from acceleration records, such as coseismic displacement and radiated seismicenergy.Radiated seismic energy plays a major role in the development of a new type of ground-motion model that uses thesite-dependent energy estimates to model the seismic radiation pattern at lower frequencies of the earthquake amplitudespectrum. This kind of ground-motion model performs better when relating ground motion to earthquake triggeredlandslides, which is demonstrated with the landslides triggered by the 2016MW7.1 Kumamoto earthquake which struckcentral Kyushu (Japan). In this case study, it is also shown that the landslide movement direction is to some extent linkedto the seismic wave polarization.The preferred mathematical model in ground-motion model development is the mixed-effect model. However, themost widely used formalism does not allow data weighting beyond directly related measurement errors and weightsderived from ACE would inadvertently bias the model. To overcome this problem, I derived the model estimators onthe basis of the weighted likelihood. The derivation is exhaustive to allow for any of the currently used model types onthe basis of mixed effects to be augmented with data weighting. This formalism in connection with ACE allows for atransparent model development and also avoids model choices on subjective expert judgment

Description: Der Teil der Seismologie, der sich mit starker Bodenbewegung beschäftigt, bezieht sich auf seismische Ereignisse, dieallgemein ein Gefahrenpotenzial für die Gesellschaft darstellen. Die Seismologie der starken Bodenbewegung wird vonzwei Aspekten angetrieben: An erster Stelle kommt die gesellschaftliche Notwendigkeit, die Erdbebengefährdung zuverstehen und das damit verbundene Risiko zu vermeiden. Während die Gefährdung durch Erdbeben kaum Änderun-gen in der Geschichte der Menschheit unterlag, so wächst das Risiko andererseits kontinuierlich an. Eine wachsendeBevölkerung, insbesondere in den am stärksten von Erdbeben geprägten Regionen der Welt, und eine mehr und mehrstörungsanfällige Infrastruktur, tragen dazu bei, seismische Ereignissen vermehrt ausgesetzt zu sein, bei gleichzeitighöherem Schadenspotenzial. Der zweite Antrieb in der Seismologie wird mit vielen anderen Forschungsfeldern geteilt:der technische Fortschritt. Die verfügbaren Möglichkeiten beim Verarbeiten immer größerer Datenmengen sindbeispiellos in der Geschichte und sind bisher noch nicht erschöpft. Beide Triebfedern stellen aber auch neue Heraus-forderungen dar, inwiefern die Daten zu interpretieren sind und wie man sie nutzbar macht.Andererseits ist die wissenschaftliche Frage klar: Was können wir aus Bruchprozessen (als Erdbebenursachen), demAuf bau der Erde (als Medium, durch welches sich die seismischen Wellen ausbreiten), sowie deren Interaktion (inwiefernbeeinflusst das Beben das umgebende Gesteinsmedium)? Diese Frage ist breit gestellt und diese Abhandlung kann sichletztlich nur auf einige Punkte beziehen und Antworten dazu liefern. Um Antworten zu finden, habe ich mehrere neueAlgorithmen und Modelle entwickelt, die allesamt auf dem Konzept der Likelihood-Funktion beruhen.Seismizität (sowie auch die Bevölkerung) ist stark an den Rändern der tektonischen Platten konzentriert. An denPlattenrändern treten verschiedene Erdbebentypen mit teils erheblich abweichenden Eigenschaften auf. Daher ist esvon Wichtigkeit, Erdbeben nach ihrem Verwerfungstyp zu klassifizieren. Das Ziel von ACE (angular clusterization withexpectation-maximization, zu dt. ungefähr Winkelgruppenbestimmung mit Erwartungswertmaximierung) ist genaudiese Klassifizierung. Auf geomechanischen Prinzipien basierend, können die Erdbebenklassifizierungen mittels ACEnicht nur auf Themen der Bodenbewegungen angewandt werden, sondern auch zur Untersuchung des Spannungsfeldesder Erde herangezogen werden.Der Entwicklung von verlässlichen Bodenbewegungsmodellen bedarf es Wellenformdaten hoher Güte. Instrumentenbezogene Fehler können die Qualität beeinträchtigen, jedoch ist eine manuelle Korrektur großer Datenmengen nichtmehr umsetzbar. Um Instrumentenfehler, die sich in Verschiebungen in den Daten zeigen, zu reduzieren, habe icheine Nulllinienkorrektur entwickelt (ICBM, integrated combined baseline modification, zu dt. integriert kombinierteNulllinienmodifikation). Dieser Algorithmus wird in der Datenvorbereitung eingesetzt und ist insbesondere dannnotwendig, wenn integrierte Größen auf Grundlage von Beschleunigungsdaten bestimmt werden, wie statischer Ver-satz eines Erdbebens als auch abgestrahlte seismische Energie.Abgestrahlte seismische Energie spielt eine herausragende Rolle in der Entwicklung einer neuen Art von Bodenbewe-gungsmodell, welches anstellen von Magnituden stationsabhängige Energieabschätzungen nutzt, um die Erdbebenab-strahlcharakteristik auf tieferen Frequenzen des Erdbebenspektrums zu beschreiben. Diese Art Bodenbewegungsmodellist besser geeignet, wenn Bodenbewegungen in Bezug zu Hangrutschungen, welche durch Erdbeben verursacht wur-den, gesetzt werden. Als Beispiel dienen hier die Hangrutschungen, die 2016 durch das Erdbeben in Zentralkyuschu(Japan) mit einer Momentenmagnitude von 7.1 verursacht wurden. In dieser Fallstudie wird auch aufgezeigt, wie dieBewegungsrichtung der Hangrutschungen zu einem gewissen Grad durch die Ausrichtung des seismischen Wellenfeldesbeeinflusst werden.Das bevorzugte mathematische Modell in der Seismologie zur Beschreibung starker Bodenbewegungen ist das gemis-chte Modell. Jedoch lässt der weitläufig angewendete Formalismus nur die Einbettung von Gewichten in Form vonMessunsicherheiten zu. Gewichte wie sie von ACE erzeugt werden, die in keinem direkten Bezug zur Messgröße stehen,liefern zwangsläufig verzerrte Ergebnisse. Um dieses Problem zu umgehen, habe ich Parameterschätzer auf Basis einergewichteten Likelihood hergeleitet. Die rigorose Herleitung erlaubt sämtliche Arten des gemischten Modells, wie siezur Beschreibung von Bodenbewegungen genutzt werde, mit Datengewichtungen zu kombinieren. Dieser Formalis-mus in Verbindung mit ACE erlaubt die Entwicklung nachvollziehbarer Modelle und vermeidet Entscheidungen aufsubjektiver Expertenmeinung.

Description: The selection of earthquake focal mechanisms (FMs) for stress tensor inversion (STI) is commonly done on a spatial basis, that is, hypocentres. However, this selection approach may include data that are undesired, for example, by mixing events that are caused by different stress tensors when for the STI a single stress tensor is assumed. Due to the significant increase of FM data in the past decades, objective data-driven data selection is feasible, allowing more refined FM catalogues that avoid these issues and provide data weights for the STI routines. We present the application of angular classification with expectation-maximization (ACE) as a tool for data selection. ACE identifies clusters of FM without a priori information. The identified clusters can be used for the classification of the style-of-faulting and as weights of the FM data. We demonstrate that ACE effectively selects data that can be associated with a single stress tensor. Two application examples are given for weighted STI from South America. We use the resulting clusters and weights as a priori information for an STI for these regions and show that uncertainties of the stress tensor estimates are reduced significantly.

Description: GPS satellite observations indicate that in the tectonically complex eastern Mediterranean and east African regions microplates rotate counterclockwise with respect to the neighboring African plate. Using 3D numerical models, Glerum relates these observations of crustal deformation to the dynamics of the lithosphere and the underlying mantle that may cause this deformation. Glerum first describes her additions to the ASPECT software necessary for numerically modeling the upper mantle and lithosphere dynamics of convergent and divergent plate boundaries. These additions include the tracking of multiple materials with different physical properties and nonlinear viscous as well as viscoplastic rheologies. The implementations of complex, multi-material rheologies are verified with well-known 2D benchmarks and multi-material viscoplasticity is applied in 3D time-dependent thermomechanical models of oceanic subduction. Subsequently, Glerum uses ASPECT to investigate the sensitivity of horizontal surface motions to individual geodynamic processes in the eastern Mediterranean. Identification of all mantle drivers that should participate in modeling attempts to explain observations of crustal flow is essential to fully exploit the information contained by surface motions about their driving processes. Glerum therefore employs 3D data-driven instantaneous dynamics models of compressible flow including a complete set of possible mantle drivers of surface deformation. The reference instantaneous flow model results indicate that mantle processes can explain a large part of the crustal motion of the Aegean-Anatolian microplate. Subsequent systematic perturbations of model properties with respect to this reference model help estimate the individual contributions of tectonic plate motions, slab pull and trench suction, and density-induced mantle flow interacting with the slab and overlying plates while moderated by the mantle’s bulk viscosity. In order of regional importance, the predicted crustal flow of the Aegean-Anatolian region is most sensitive to slab pull, followed by slab-mantle interaction and basal drag, mantle rheology, and the absolute plate motion reference frame. Lastly, Glerum demonstrates a possible mechanism for the counterclockwise rotation of the Victoria microplate in the East African Rift System, which is in striking contrast to the clockwise motion of the surrounding plates. 3D models of the divergent system show that Victoria’s rotation can be caused by the drag of the African and Somalian plates along the strong edges of the microplate, while the rift segments along inherited lithospheric weaknesses facilitate Victoria’s rotation. The amount of rotation is therefore primarily controlled by the distribution of preexisting stronger regions and the weaker Precambrian mobile belts that surround Victoria. The induced counterclockwise rotation of the microplate leads to a clockwise shift of the local extension direction from E-W to more WNW-ESE along the overlapping rift branches. Comparison of the resulting predicted stress field and tectonic regimes to observations helps to elucidate the interpretation of local stress and strain indicators and to reconcile different opening models used to interpret the East African Rift System.

Description: Voxel representations have been used for years in scientific computation and medical imaging. The main focus of our research is to provide easy access to methods for making large-scale voxel models of built environment for environmental modelling studies while ensuring they are spatially correct, meaning they correctly represent topological and semantic relations among objects. In this article, we present algorithms that generate voxels (volumetric pixels) out of point cloud, curve, or surface objects. The algorithms for voxelization of surfaces and curves are a customization of the topological voxelization approach [1]; we additionally provide an extension of this method for voxelization of point clouds. The developed software has the following advantages: • It provides easy management of connectivity levels in the resulting voxels. • It is not dependant on any external library except for primitive types and constructs; therefore, it is easy to integrate them in any application. • One of the algorithms is implemented in C++ and C for platform independence and efficiency.

Description: Powered by WebGL, some renderers have recently become available for the visualization of point cloud data over the web, for example Plasio or Potree. We have extended Potree to be able to visualize massive point clouds and we have successfully used it with the second national Lidar survey of the Netherlands, AHN2, with 640 billion points. In addition to the visualization, the publicly available service at http://ahn2.pointclouds.nl/ also features a multi-resolution download tool, a geographic name search bar, a measurement toolkit, a 2D orientation map with field of view depiction, a demo mode and the tuning of the visualization parameters. Potree relies on reorganizing the point cloud data into an multi-resolution octree data structure. However, this reorganization is very time consuming for massive data sets. Hence, we have used a divide and conquer approach to decrease the octree creation time. To achieve such performance improvement we divided the entire space into smaller cells, generated an octree for each of them in a distributed manner and then we merged them into a single massive octree. The merging is possible because the extent of all the nodes of the octrees is known and fixed. All the developed tools are free and open-source (FOSS) and they can be used to visualize over the web other massive point clouds.

Description: For more than 40 years, remote sensing satellite missions are globally scanning the earth´s surface. They are ideal instruments for monitoring spatio-temporal changes. A comprehensive analysis of this data has the potential to support solutions to major global change challenges related to climate change, population growth, water scarcity, or loss of biodiversity. However, a comprehensive analysis of these remote sensing data is a challenging task: (a) there is a lack of Big Data-adapted analysis tools, (b) the number of available sensors will steadily increase over the next years and (c) technological advancements allow to measure data at higher spatial, spectral, and temporal resolutions than ever before. These developments create an urgent need to better analyze huge and heterogeneous data volumes in support of e.g. global change research. The interdisciplinary research consortium of the “GeoMultiSens” project focuses on developing an open source, scalable and modular Big Data system that combines data from different sensors and analyzes data in the petabyte range (1015 Byte). The most important modules are: (1) data acquisition, (2) pre-processing and homogenization, (3) storage, (4) analysis, and (5) visual exploration. The data acquisition module enables users to specify a region and time interval of interest, to identify the available remote sensing scenes in different data archives, to assess how these scenes are distributed in space and time, and to decide which scenes to use for a specific analysis. The homogenization module uses novel and state-of-the-art algorithms that combine the selected remote sensing scenes from different sensors into a common data set. The data storage module optimises storage and processing of petabytes of data in a parallel and failure tolerant manner. The core technology of the data storage module is XtreemFS (http://www.xtreemfs.org). The analysis module implements image classification and time series analysis algorithms. The visual exploration module supports users in assessing the analysis results. All modules are adapted to a map-reduce processing scheme to allow a very fast information retrieval and parallel computing within the processing system Flink (http://flink.apache.org). Finally, a Visual Analytics approach integrates the individual modules and provides a visual interface to each step in the analysis pipeline. The Big Data system “GeoMultiSens” will store and process remotely sensed data from space-borne multispectral sensors of high and medium spatial resolution such as Sentinel-2, Landsat 5/7/8, Spot 1-6, ASTER, ALOS AVNIR-2 and RapidEye. Our poster presents the overall scientific concept of the Big Data system “GeoMultiSens” and technical details of the most important modules. We discuss scientific challenges of the Big Data system “GeoMultiSens” and present our ideas to address these scientific challenges.

Description: Landslides are a worldwide natural hazard causing thousands of fatalities and severe monetary losses every year. To predict and thus reduce the landslide risk in the future, a profound knowledge about the past and recent landslide activity is of utmost importance. For this purpose, the records about the landslide activity have to be as complete as possible in time and space, in order to derive spatial and temporal probabilities of landslide occurrence as a crucial prerequisite of landslide hazard and risk assessment. However, for most regions of the world such comprehensive landslide records are not available, because the conventional manual mapping of landslides is an extremely time-consuming and labor-intensive task. This study presents an automated approach for efficient multi-temporal identification of landslides at regional scale based on optical remote sensing time series data. The developed approach allows for retrospective analysis of long-term landslide occurrence and for monitoring recent landslide activity. In case of the long-term analysis, a combined usage of multiple optical sensors is required to achieve best possible temporal data coverage for the longest possible time span. For this study, such a database has been established for a landslide-affected area of 12000 km² in Southern Kyrgyzstan, Central Asia. It consists of about 900 orthorectified multispectral satellite remote sensing datasets acquired by Landsat-(E)TM, SPOT, IRS-1C (LISS3), ASTER and RapidEye during the last 30 years. For monitoring the landslide activity of the last 5 years, high spatial and temporal resolution RapidEye data have been acquired in the frame of the RapidEye Science Archive (RESA) program. The developed approach comprises automated multi-sensor pre-processing and multi-temporal change detection methods allowing spatiotemporal identification of landslides in an object-based form. The change detection builds on the analysis of temporal NDVI-trajectories, representing footprints of vegetation changes over time. Landslide-specific trajectories are characterized by short-term vegetation cover destruction and longer-term revegetation rates resulting from landslide related disturbance and dislocation of the fertile soil cover. In combination with DEM-derivatives the developed approach enables automated identification of landslides of different sizes, shapes and in different stages of development under varying natural conditions. The multi-sensor long-term analysis of a 2500 km² region resulted in the identification of 1583 landslides ranging in size between 50 m² and 2.8 km². The highest overall landslide rates occurred in 2003 and 2004 exceeding the long-term annual average rate of 57 landslides per year by more than five times. For monitoring the recent landslide activity the approach has been applied to the RapidEye time series acquired between 2009 and 2015 for the whole 12000 km² study area. The combination of high spatial resolution (5 m) and frequent data acquisition (up to several days/weeks) of the RapidEye data has allowed for the systematic assessment of the whole variety of landslide processes also including small slope failures, which often represent precursors for subsequent larger and more hazardous landslides. Thus, the approach can provide valuable information in the context of early warning. Currently, the applicability of the approach is investigated for assessing the aftereffects of the disastrous Nepal earthquakes of April and May 2015 that triggered thousands of landslides. First results have shown the general transferability of the approach to the differing natural environment of Nepal and its general suitability to operate within a rapid response system. Together with the newly available Sentinel-2 data, this approach has the potential to be developed into a globally applicable landslide mapper, which will open up new opportunities to analyze spatiotemporal landslide activity over large areas facilitating further development of probabilistic landslide hazard and risk assessments.

Description: The remote sensing analyses in the BMBF-SPACES collaborative project Geoarchives - Signals of Climate and Landscape Change preserved in Southern African Geoarchives - focuses on the use of recent and upcoming Earth Observation Tools for the study of climate and land use changes and its impact on the ecosystem. It aims at demonstrating the potential of recently available advanced optical remote sensing imagery with its extended spectral coverage and temporal resolution for the identification and mapping of sediment features associated with paleo-environmental archives as well as their recent dynamic. In this study we focus on the analyses of two ecosystems of major interest, the Kalahari salt pans as well as the lagoons at Namibia’s west coast, that present high dynamic caused by combined hydrological and surface processes linked to climatic events. Multitemporal remote sensing techniques allow us to derive the recent surface dynamic of the salt pans and also provide opportunities to get a detailed understanding of the spatiotemporal development of the coastal lagoons. Furthermore spaceborne hyperspectral analysis can give insight to the current surface mineralogy of the salt pans on a physical basis and provide the intra pan distribution of evaporites. The soils and sediments of the Kalahari salt pans such as the Omongwa pan are a potentially significant storage of global carbon and also function as an important terrestrial climate archive. Thus far the surface distribution of evaporites have been only assessed mono-temporally and on a coarse regional scale, but the dynamic of the salt pans, especially the formation of evaporites, is still uncertain and poorly understood. For the salt pan analyses a change detection is applied using the Iterative-reweighted Multivariate Alteration Detection (IR-MAD) method to identify and investigate surface changes based on a Landsat time-series covering the period 1984-2015. Furthermore the current spatial distribution of evaporites is obtained using of EO-1 Hyperion hyperspectral imagery linked with geochemical field data. Results reveal a highly heterogeneous dynamic of the pan surface, which seems to be associated with varying surface crust types, halite or gypsum dominated. The lagoons at Namibia’s west coast such as of Sandwich Harbour and Walvis Bay, are important habitats and also serve as a natural barrier to protect shipping and ports on an otherwise inhospitable coastline. Several studies have shown that these lagoons are highly dynamic and are known to have altered their shape in historical time. These changes occur due to sediment transport forced by aeolian processes or either by longshore or cross-shore drifts. A profound understanding of the spatiotemporal variations in the sand spits is of high relevance. In the lagoon environment the Landsat time-series is used to separate sand spits from open water. This way, changes in morphology of the sand spit are identified over time. The results reveal the presence of long-term and short-term changes as well as the presence of stable parts in the sand spits. These findings are linked to temporal patterns of forcing processes such as wind, tidal and ocean current data.

Description: The Singular Value Decomposition (SVD) is a mathematical procedure with multiple applications in the geosciences. For instance, it is used in dimensionality reduction and as a support operator for various analytical tasks applicable to spatio-temporal data. Performing SVD analyses on large datasets, however, can be computationally costly, time consuming, and sometimes practically infeasible. However, techniques exist to arrive at the same output, or at a close approximation, which requires far less effort. This article examines several such techniques in relation to the inherent scale of the structure within the data. When the values of a dataset vary slowly, e.g., in a spatial field of temperature over a country, there is autocorrelation and the field contains large scale structure. Datasets do not need a high resolution to describe such fields and their analysis can benefit from alternative SVD techniques based on rank deficiency, coarsening, or matrix factorization approaches. We use both simulated Gaussian Random Fields with various levels of autocorrelation and real-world geospatial datasets to illustrate our study while examining the accuracy of various SVD techniques. As the main result, this article provides researchers with a decision tree indicating which technique to use when and predicting the resulting level of accuracy based on the dataset’s structure scale.