PANDEmic MObility Scenarios: Optimizing traffic flows in times of a pandemic

Derivation of all-German transport demand at the person level DEMO is a Germany-wide model of mobility behavior, but it is a) macroscopic, i.e. it provides a description of mobility behavior aggregated over population groups, and b) it does not provide a spatially resolved representation of mobility within transport cells. Therefore, in order to obtain a disaggregated, complete representation of the mobility of all persons living in Germany, the results of the DEMO model are disaggregated in the first project step. The disaggregation of DEMO demand builds on different data. First, the population is transformed into individuals, where statistical information on sociodemographic features like age, gender, employment status, possession of mobility options, etc., is assigned to. Next, data from the OpenStreetMap project is used to distribute the population to residential buildings. This data is also used to determine the disaggregated destinations of trips. Here, the trips taken from DEMO over a day at a time and their purposes are augmented using diaries from the “time use survey”. The result is a disaggregated list of trips including origin and destination, purpose, time and the mode of transport used. The refinement of the mobility mapping by social media and mobile data, which takes place in WP2, is iteratively considered in WP1 to obtain an improvement of the disaggregation methods.

Social Media Analytics and data integration In addition to the analysis of traffic models and traffic demand in work package 1, data from social media that can be queried via public APIs will also be analyzed in work package 2 to obtain an even more accurate picture of mobility in Germany. To this end, various analysis techniques will be made available in a social media monitoring tool. To generate a consistent and comprehensive database, machine learning and text mining techniques will be used to identify location references in public social media posts (georeferencing) even if they are not explicitly specified. It is important to ensure that biases in the data, such as an excessive focus on individual regions or age groups, are largely avoided. This is achieved by matching the acquired contributions with a local reference to data on population density for the purpose of representativeness. Furthermore, voluntary data donations are also used here for validation. The goal is to model actual traffic movements from such data as well as to derive social connections between regions through which mobility can be predicted. To do this, networks of people, places, and posts are first constructed from the collected social media data. These are then embedded in a geographic coordinate system and analyzed using “Spatio Social Network Analysis” and linked to further traffic data to gain further insights into mobility in Germany and influencing factors. The results contribute to the adaptation of traffic models (WP1) and to a more accurate parameterization of simulation models (WP4).

Web interface with visual analytics and interactive visualization The visual analysis of the data is a central part of the project. It shall enable to understand the results of the simulations, to detect non-known patterns and to evaluate possible measures derived from them. The visualization will work on several levels to analyze the infection dynamics, both in a national, as well as in a regional context. For this purpose, several visualization techniques suitable for the different levels will be applied. At the national level, mainly choropleth maps will be used, which, combined with coarse infrastructure maps, will allow to analyze the spread of infection through long-distance travel. At the regional level, fine-grained visualizations will be implemented based on techniques from Krüger (2017). These techniques will be adapted and extended to include application-specific methods. The georeferenced visualizations will be complemented by statistical analysis tools, such as parallel coordinate plots or scatterplot matrices. Automated algorithms will also help users to identify patterns in the data and to visualize them. Interactive elements extend the visualizations to an application that can be used to define, test and analyze different scenarios. Here, an interaction loop for users will be created that allows them to define measures, start simulations based on them, and analyze the results. Based on the results, measures can be adjusted and evaluated. The visual analytics framework will be implemented primarily with web technologies. For this purpose, existing open source libraries will be used. Visualization libraries such as D3.js and amCharts will be at the center. Leaflet will be used in combination with data from OpenStreetMap to implement regional visualizations. The library developed in work package 5 can efficiently mimic the results of the models and thus provide a rapid and cost-effective preview of results.

Epidemiological simulation models with mobility consideration In WP 4, existing epidemiological models will be extended and supplemented with an explicit description of mobility and personal transport. The aim is to investigate and predict the effects of different forms of mobility on the incidence of infections. In this context, both the importance of public transport as a point of infection on the one hand and the importance of mobility as a whole for the spread of the pandemic on the other hand will be investigated in scenario simulations. As a first approximation, infection events can be studied in SEIR-type compartmental models. These models are particularly suitable for simulating larger scales because of their efficiency. However, mobility-specific parameters and mechanisms must be added to the model in any case, especially with respect to the risk of infection in public transport and accompanying measures such as mandatory masking or a limit on maximum occupancy. Compared to differential equation models, agent-based models (ABM) consider a different level of detail. A model is developed that represents individual persons and infection processes between persons as well as associated disease trajectories. Each individual is simulated with specific characteristics such as age and activity, and moves in the model between different nodes of a network representing possible sites of infection. Within these nodes, infection processes can thus be simulated. This type of model allows a very detailed mapping of disease processes, individual characteristics and behavior of persons as well as characteristic properties of a region. As a result, interventions can also be included in comparative detail. However, the high level of detail also means a high demand for computing capacity, so that agent-based models will be developed for pilot cities in a first step. In a second step, these will also be developed for the whole of Germany. Since the size of the system is limiting for certain applications, an abstraction e.g. by means of surrogate models (cf. WP 5) is necessary. Just as for the SEIR models, preliminary work can be used for the development of the ABM. However, a representation of public transport in the model must be newly developed and parameterized.

Interventions, optimization and artificial intelligence The focus of work package 5 is the derivation of an optimized set of interventions to maintain mobility while reducing infection dynamics. For this purpose, this work package will develop more efficient surrogate models using artificial intelligence, which will enable interactive elaboration of different sets of measures. In a first step, possible measures to curb the infection dynamics will be discussed and defined. After the explicit modeling of mobility behavior in the epidemiological models has been designed in work package 4, a mapping of these measures to the modeled mobility behavior will be designed and a corresponding implementation will be performed in a step beyond. In parallel to the above steps, a market analysis of available optimization tools will be conducted and an evaluation will be made regarding their applicability to the problems described in this project. Based on the simulation results of the expert models developed in work package 4, surrogate artificial intelligence models are trained. In a feedback loop of optimization and simulation, both expert and surrogate models are further refined and adapted to possible new developments. By optimizing both models, parameter studies are performed so that an evaluation can be made for the predefined measures. Finally, an optimized catalog of measures is generated from this and then optimized again. Finally, the cost-efficient surrogate models are further optimized so that they can be coupled with the web interface developed in work package 3. The coupling is implemented via a REST API, which the interface can access after user action.

• W. Koslow, M. J. Kühn, S. Binder, M. Klitz, D. Abele, A. Basermann, M. Meyer-Hermann, Appropriate relaxation of non-pharmaceutical interventions minimizes the risk of a resurgence in SARS-CoV-2 infections in spite of the Delta variant (2022). PLoS Comput Biol 18(5): e1010054. Link Preprint
   
• M. J. Kühn, D. Abele, S. Binder, K. Rack, M. Klitz, J. Kleinert, J. Gilg, L. Spataro, W. Koslow, M. Siggel, M. Meyer-Hermann, A. Basermann. Regional opening strategies with commuter testing and containment of new SARS-CoV-2 variants (2022). BMC Infectious Diseases 22:333. Link Preprint.
   
• M. Zeumer, J. Gilg, P. K. Betz, A. Gerndt, VVAFER—Versatile visual analytics framework for exploration and research. Electronic Imaging 35 (2023): 1-9. Link.
   
• S. Yosmaoglu, D. Diallo, T. Hecking, A. Schengen Refining agent-based travel demand models using social media data, Transport Research Arena, Lisbon, 2022 (2023)
   
• P. Kaur Betz, J. Stoll , V. Grappendorf, J. Gilg, M. Zeumer , M. Klitz, L. Spataro, A. Klein, L. Rothenhäusler, H. Bohnacker, H. Krämer M. Meyer-Hermann, S. Somogyi, A. Gerndt and M. J. Kühn ESID: Exploring the Design and Development of a Visual Analytics Tool for Epidemiological Emergencies. IEEE VIS Workshop on Visualization for Pandemic and Emergency Responses (Vis4PandEmRes) (2023). Link. Preprint
   
• J. L. Malkus, M. L. Díaz, A. Schengen, T. Mocanu, and M. J. Kühn An OpenStreetMap-based approach for generating capacity-restricted POIs for activity-based travel demand modeling. (2024).
   
• H. Zunker, R. Schmieding, D. Kerkmann, A. Schengen, S. Diexer, R. Mikolajczyk, M. Meyer-Hermann, and M. J. Kühn Novel travel time aware metapopulation models: A combination with multi-layer waning immunity to assess late-phase epidemic and endemic scenarios (2024). Preprint
   

• S. Yosmaoglu, Refining agent-based travel demand models using social media data, Transport Research Arena, Lisbon, 2022
• D. Diaoule, Gradual Network Sparsification and Georeferencing for Location-Aware Event Detection in Microblogging Services, Complex Networks Conference, Palermo, Italy, 2022.
• M. Zeumer, VVAFER — Versatile visual analytics framework for exploration and research, Electronic Imaging Symposium, USA, 2023
• M. J. Kühn, PANDEMOS: PANdemic MObility Scenarios, First National Conference on Infectious Disease Modeling, Berlin, 2023
• M. J. Kühn, An overview on mobility-aware models for infectious disease dynamics and non-pharmaceutical interventions, Colloquium of work group Modeling in Medical Science, Halle (Saale), 2023
• H. Zunker, Mobility processes and transmission on travel and commute in hybrid graph-ODE models, Colloquium of work group Modeling in Medical Science, Halle (Saale), 2023
• H. Zunker, Travel time aware metapopulation models for estimating transmissions and nonpharmaceutical intervention effects, Complex Networks Conference, Menton Riviera, France, 2023.
• J. L. Malkus, An OpenStreetMap-based approach for generating capacity-restricted POIs for activity-based travel demand modeling. 15th International Conference on Ambient Systems, Networks and Technologies (ANT), Hasselt, Belgium, 2024.

• Martin Kühn - Mathematical modeling of infectious diseases, University of Bonn (2022/2023)
• Martin Kühn - Mathematical modeling of infectious diseases, University of Bonn (2023/2024)

• 2022/03/22: Prof Kai Nagel (TU Berlin) - Predicting the effects of COVID-19 related interventions
• 2022/06/02: Dr. Pascal Klamser (Humboldt University of Berlin) - Inferring the country specific import risk of diseases from the World-Aviation-Network
• 2022/09/16: Dr. Lorenzo Contento (University of Bonn) - Integrative modelling of reported case numbers and seroprevalence reveals time-dependent test efficiency and infectious contacts
• 2022/10/05: Massimo Achterberg (TU Delft) - Network-based models to describe the spread of the COVID-19 epidemic
• 2022/11/08: Prof. Viola Priesemann (Max Planck Institute for Dynamics and Self-Organization) - Spreading dynamics: from neural information flow to COVID-19

Project lead

Martin Kühn (Contact: Martin.Kuehn@DLR.de)

Funder information

Project description at Ministry for Digital and Transport.

Association

Associated consortium with the Modeling Network for Severe Infectious Diseases.

Partners

The infection dynamics models are available in: https://github.com/DLR-SC/memilio