Statistical Week 2025
2-5 September 2025
Wiesbaden, Germany
Conference Agenda
Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
|
Session Overview |
| Session | ||
Ties: Joint DStatG and TIES Session
| ||
| Presentations | ||
4:20pm - 4:45pm
glmSTARMA - An R package for fitting spatio-temporal models based on generalized linear models 1TU Dortmund University, Germany; 2University of Cyprus, Cyprus The analysis of spatio-temporal data plays a crucial role in many research areas. Some examples include the observation of disease cases in different regions in epidemiology or the measurement of various climate variables such as precipitation and temperature at measuring stations. We present a flexible framework for modeling such data that efficiently captures spatial and temporal dependencies. Our approach is based on generalized linear models and allows for the integration of autoregressive dependence structures, the inclusion of covariates as well as different distributions to account for the specific data properties. In addition to methods for parameter estimation, the framework also includes functions for statistical inference. A particular focus is given to a user-friendly implementation. In this talk, we will present the basic functionalities of the R package and illustrate their application using concrete examples. 4:45pm - 5:10pm
Bayesian estimation for virtual experiments in metrology featuring semi-parametric modelling using Gaussian processes Physikalisch-Technische Bundesanstalt, Germany Virtual experiments are a tool commonly used in measurement science to simulate virtual data that mimic real observations. Virtual experiments are particularly useful when real data are sparse and it is impractical to obtain a statistically sufficient sample of real observations needed to conduct a valid data analysis. Constructed by experts, virtual experiments constitute a highly complex representation of a real measurement process where a parametric model is often considered, mapping real physical quantities to virtual observations. However, due to the required accuracy and high resolution of many metrological applications, purely parametric models often do not provide a sufficient representation of the physically meaningful inputs and corresponding outputs of a virtual experiment. For instance, in optical metrology, there can be high-frequency terms representing small imperfections of a machined specimen that cannot be represented by the purely parametric model. Hence, a non-parametric formulation can be employed to improve the representation of a real measurement process. This work proposes a semi-parametric virtual experiment model that operates in conjunction with measurement data and a Bayesian estimation procedure. We introduce a Gaussian process that accompanies our parametric virtual experiment. Heteroscedasticity and correlations between local points, often-ignored factors in data analyses that can have a significant impact on the accuracy of the analysis, are accounted for in the proposed statistical method. We explore estimation techniques to obtain a Bayesian estimator with a quantifiable uncertainty. Once defined, we apply the developed estimation technique to an example from optical metrology where the current approach in literature, a deterministic multi-step optimisation strategy, can be translated to a semi-parametric model with an unknown link function. 5:10pm - 5:35pm
Statistical Modeling of Clustering and Seasonality in Return Times of Midlatitude Cyclones TU Dortmund, Deutschland Many extreme weather events, such as heavy precipitation or storms, tend to appear in temporal clusters and exhibit seasonal fluctuations, two behaviors that the stationary Poisson process cannot capture. In previous research, the fractional Poisson process (FPP) has been used to model occurrences of midlatitude cyclones, enabling the description of temporal clustering. However, since the standard FPP cannot account for seasonal behavior, most existing studies primarily focus on modeling winter cyclones, with less attention given to the other seasons. To address this limitation, we develop a modified approach that incorporates seasonality into the FPP. We also evaluate different estimation methods for the parameters of the Mittag-Leffler distribution, which is the distribution of the return times of an FPP. We propose a new quantile-based estimator for the parameters and compare it with the existing estimation methods. The quantile-based estimator outperforms the widely used log-moments estimator in terms of mean squared error, while also offering shorter computing time and better robustness than the maximum likelihood estimator. We illustrate our approach by modeling the return times of midlatitude cyclones using climate reanalysis data. 5:35pm - 6:00pm
Earth Observation Data and AI for Construction Statistics 1Statistisches Bundesamt, Deutschland; 2Bundesamt für Kartographie und Geodäsie Im Rahmen eines von Eurostat geförderten Projekts (Earth Observation Data and AI for Construction Statistics, EO4ConStat) werden Möglichkeiten geprüft, eine automatisierte Erkennung von Baustellen auf Luftbildern zur Qualitätssicherung in der Bautätigkeitsstatistik zu nutzen. Als Datenquellen für die Detektion von Baustellen mit Verfahren des maschinellen Sehens werden digitale Orthofotos genutzt. Neben der Baustelle selbst, sollen auch vordefinierte Bauphasen automatisch erkannt werden. Die anschließende Auswertung von Satellitendaten soll die zeitliche Einordnung von Baubeginn und -fertigstellung ermöglichen. Hierfür werden Veränderungen in den spektralen Eigenschaften der erkannten Baustellen genutzt. Diese lassen sich durch die Verwendung sogenannter spektraler Indizes ebenfalls maschinell auswerten. Die vorliegenden ersten Ergebnisse sind vielversprechend, zeigen aber auch Herausforderungen bei der Abgrenzung zu optisch ähnlichen aber inhaltlich verschiedenen Flächen auf. Sollten sich die getesteten Verfahren als zuverlässig bestätigen, kann die fernerkundliche Erkennung von Baustellen zu Zwecken der Qualitätssicherung in der Statistik der Baubeginne und Baufertigstellungen eingesetzt werden. Dazu müssen die Ergebnisse der maschinellen Erkennung sorgfältig validiert werden. Auch hierzu werden im Projekt geeignete Methoden entwickelt und getestet. Um einen möglichst repräsentativen Querschnitt verschiedener Baustellentypen, ihrer Umgebungen oder auch Bodenarten abzubilden, wurde als Testgebiet die gesamte Landesfläche von Nordrhein-Westfalen ausgewählt. Der Umfang der hierfür auszuwertenden Daten erhöht einerseits den Aufwand für die Prozessierung, ermöglicht aber andererseits eine Abschätzung zu den notwendigen Ressourcen für eine potentielle Anwendung im größeren Rahmen. Das Projekt ist eine Zusammenarbeit mit dem Bundesamt für Kartographie und Geodäsie (BKG) und dem Deutschen Zentrum für Luft- und Raumfahrt (DLR). | ||
