A Hybrid ABM-PDE Framework for Real-World Infectious Disease Simulations

Kristina Kehrer, Tim O. F. Conrad · 2025

Read A Hybrid ABM-PDE Framework for Real-World Infectious Disease Simulations by Kristina Kehrer, Tim O. F. Conrad (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-08430.txt.

Introduction

Kristina Kehrer and Tim Conrad present, in 2025 work, a hybrid model for the spread of infectious disease that runs far faster than a full person-by-person run. Full agent models track each person, which is true to life but slow at city scale. Their mix couples an agent zone with a smooth flow zone that treats people as density.

People who cross from the agent zone melt into the density, and spare density on the other side spawns new agents with paths drawn from phone data. The states of health run to seven, from well to ill and back. Tested on real movement and case data for the Berlin region, the hybrid matched the core shape of spread with fewer runs and smaller errors. The team offers it for large scenario work.

Catalog record

CREATORKristina Kehrer, Tim O. F. Conrad
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE95 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-08430

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Hybrid ABM-PDE Framework for Real-World Infectious Disease Simulations Kristina Kehrer, Tim O. F. Conrad Abstract This paper presents a hybrid modeling approach that couples an Agent-Based Model (ABM) with a partial differential equation (PDE) model in an epidemic setting to simulate the spatial spread of infectious diseases using a compartmental structure with seven health states. The goal is to reduce the computational complexity of a full-ABM by introducing a coupled ABM-PDE model that offers significantly faster simulations while maintaining comparable accuracy. Our results demonstrate that the hybrid model not only reduces the overall simulation runtime (defined as the number of runs required for stable results multiplied by the duration of a single run) but also achieves smaller errors across both 25% and 100% population samples. The coupling mechanism ensures consistency at the model interface: agents crossing from the ABM into the PDE domain are removed and represented as density contributions, while surplus density in the PDE domain is used to generate agents with plausible trajectories derived from mobile phone data. We evaluate the hybrid model using real-world mobility and infection data for the Berlin-Brandenburg region in Germany, showing that it captures the core epidemiological dynamics while enabling efficient large-scale simulations. These results demonstrate that the proposed ABM-PDE framework provides a robust and computationally efficient alternative to full-scale agent-based simulations, making it suitable for realistic epidemic modeling and scenario analysis. A Preprint Keywords Partial differential equation \cdot epidemic modeling \cdot spatial modeling \cdot coupling approach \cdot diffusion \cdot agent-based model \cdot landscape 1 Introduction Understanding and efficiently simulating complex dynamical systems is a central challenge across many scientific and engineering domains. In epidemiology, this challenge is particularly pronounced due to the need to capture diverse phenomena such as individual behavior, spatial mobility, and population heterogeneity. In this field, various modeling approaches exist, such as agent-based models (ABMs) [ 25 ] , models based on partial differential equations (PDEs) [ 12 ] , and ordinary differential equation (ODE) models [ 37 ] . Each of these approaches offers distinct advantages: ABMs provide detailed individual-level dynamics but are computationally demanding; ODEs are efficient and analytically tractable but lack spatial structure; PDEs incorporate spatial variation, which is critical for modeling heterogeneous regions. Each of these modeling approaches has its place, but choosing the most appropriate one often depends on the characteristics of the domain being studied. In practice, however, many real-world regions exhibit a mix of features that make a single modeling paradigm insufficient. Consider, for instance, the Berlin-Brandenburg region: while Berlin represents a densely populated, relatively homogeneous urban area, Brandenburg is more rural and spatially heterogeneous. Applying the same model across both regions may either introduce unnecessary complexity or overlook important spatial effects. To address this, we adopt a hybrid modeling approach that assigns different models to distinct, non-overlapping subregions – each selected to best capture the local dynamics. This strategy allows us to leverage the respective strengths of each modeling framework while maintaining consistency across the overall system. Hybrid modeling approaches combining ABMs and PDEs have been employed in other scientific domains, such as cellular biology and oncology; however applications in the context of human-to-human disease transmission remain rare. For instance, Kostre et al. [ 20 ] apply hybrid ABM-PDE models to simulate biochemical processes at the cellular level, modeling interactions between molecules and cells through coupled particle-based simulations and reaction-diffusion PDEs. Particles diffuse within the particle domain. Upon crossing the boundary into the concentration domain, they are eliminated. The concentration at the boundary of the PDE domain is converted into virtual particles and fractional virtual particles, which can jump into the particle domain at a certain rate. Similar ABM-PDE hybrids have been used to investigate viral dynamics in cell populations [ 24 ] , antiviral drug effects [ 16 ] , and treatment strategies in oncology [ 33 ] . Recent studies have also explored machine learning techniques to construct surrogate models that approximate behaviour of detailed ABMs [ 3 ] . These data-driven approaches – often based on neural networks – can replicate complex ABM dynamics with significantly reduced computational cost, enabling more efficient parameter calibration and sensitivity analyses. As such, they offer a promising approach for enabling large-scale simulations. However, a key limitation of these methods lies in their lack of interpretability – an important drawback in epidemiological applications, where understanding the underlying mechanisms is essential for informed public health decision-making. Several studies have explored hybrid epidemic modeling approaches that combine agent-based representations with compartmental formulations (e.g., [ 38 , 31 , 10 ] ). They develop frameworks that integrate agent-based network components with compartmental population-level structures for disease transmission, illustrating both the flexibility and, in certain contexts, the necessity of mixed modeling strategies in epidemiology. A similar principle underlies our Berlin–Brandenburg setting, where the choice of modeling paradigm is guided by structural heterogeneity. In this study, we present and evaluate a hybrid ABM-PDE model using real-world data, specifically applied to the Berlin-Brandenburg region. The model incorporates real-world mobile phone data, representing 25% and 100% of the population, capturing realistic movement patterns crucial for disease dynamics modeling. The model’s evaluation is based on real-world infection data and compared to a high-resolution, full-ABM simulation that serves as a reference. Our hybrid approach enables the exchange of information between the two (sub-)models at each time step, ensuring a dynamic yet consistent interaction between local and global disease dynamics. The resulting hybrid ABM-PDE model provides a faster alternative to a full-ABM model, offering the benefit of quicker parameter fitting due to reduced overall simulation runtime, regardless of whether the 25% or 100% population sample is used. Furthermore, the hybrid model demonstrated better accuracy over short simulation horizons – approximately two months –, when fitting was done for two separate intervals. The underlying ABM is a streamlined version of EpiSim [ 29 ] implemented in C++, and is capable of simulating 100% of the population on a standard computer. Building on these results, our long-term objective is to extend the current hybrid framework into a comprehensive ABM-PDE-ODE coupled model, which will integrate previous advancements in ABM-ODE [ 6 ] and PDE-ODE coupling [ 17 ] . The complete model implementation and the processed infection data from the Robert Koch Institute (RKI) website are available on Zenodo [ 18 ] . Mobility data is available from the authors upon reasonable request. 2 Model Formulation Our proposed hybrid model couples an agent-based model (ABM) and a partial differential equation (PDE) model to simulate the spatial spread of infectious diseases in the Berlin-Brandenburg region. The urban area of Berlin is represented by a PDE model that captures population-level dynamics and health state transitions, while the surrounding rural area of Brandenburg is modeled using an ABM based on real-world individual movement data from mobile phones. The data used for the ABM consists of individuals who have been in Berlin or Brandenburg at least once, with a few individuals from outside the region. For simplicity, we refer to the entire modelling domain as Brandenburg for the ABM contribution of the hybrid model and Berlin-Brandenburg for the full-ABM model. Both the ABM and the PDE model incorporate different epidemiological health states, and are dynamically coupled at each time

Read the complete text