Research
The Computational Geography group develops methods and software for spatio-temporal modelling and applies these in a range of domains from physical geography and environmental sciences. The focus is on three branches of modelling:
Process-based modelling mainly relying on a mechanistic representation (physical or other) of the geographical system that is encoded in a forward simulation model. This branch of modelling is typically employed to improve our understanding of geographical systems (e.g. complex behaviour), for prediction of system state, as well as scenario analysis.
Data-based modelling mainly relying on emperical data (observations) employing geostatistics or AI modelling (machine learning). This branch of modelling is in particular effective in prediction of system states at unvisited (unsampled) locations. An important component of research is the design of optimal sampling schemes.
Hybrid modelling aiming at model identification steered by data as well as process knowledge. This involves data assimilation, which inserts observations into process-based models, as well as machine learning models that incorporate mechanistic understanding of the system represented.
These branches of modelling are supported by and integrated with:
Design, development, and distribution of software frameworks for spatio-temporal modelling.
Examples of applications from geography include soil science, hydrology, health geography, and geo-linguistics.
Process-based modelling
Process-based modelling involves the representation of geographical systems using temporal update functions mimicking real-world processes, through continuous field-based or agent-based approaches. It is employed for estimating future or past states of geographical systems, scenario analysis, and to support understanding of how geographical systems function, e.g. complex behaviour such as emergence and critical transitions. Our group implements process-based models with our own software frameworks, for instance for large scale simulation on computer clusters, as well as existing toolboxes, for instance for computational fluid dynamics modelling.
Projects:
Agent-based Modelling of policy measures to improve diet quality, work package in Exposome-NL.
Conceptual and Agent-based Modelling of Exposure Interventions, work package in Exposome-NL.
The impacts of climate change on spatial inequality in WEF security in South Africa, work package in Spatial inequality in water-energy-food security in South Africa. Implications on public health and the consequences of climate change.
Copernicus Climate Change (C3S) Water Service uses a multi-model hydrological model ensemble to produce consistent global and European scale data.
Computational fluid dynamic (CFD) modal to simulate wind flow over coastal dune.
Water scarcity under droughts and heatwaves: understanding the complex interplay of water quality and sectoral water use, NWO Vidi project of Michelle van Vliet with contributions from our team.
Example publications:
Gargari, S. F., D. Karssenberg, and G. Ruessink (2025). “The influence of foredune geometry on wind flow quantified from computational fluid dynamics simulations”. In: Aeolian Research 74, p. 101001. doi: 10.1016/j.aeolia.2025.101001.
Holler, S., K. R. Hall, B. Rayfield, G. Zapata-Ríos, D. Kübler, O. Conrad, O. Schmitz, C. Bonannella, T. Hengl, J. Böhner, S. Günter, and M. Lippe (2025). “Ubi es, room to roam? Extension of the LPB-RAP model capabilities for potential habitat analysis”. In: Ecological Modelling 501, p. 111005. doi: 10.1016/j.ecolmodel.2024.111005.
Jaarsveld, B. van, N. Wanders, E. H. Sutanudjaja, J. Hoch, B. Droppers, J. Janzing, R. L. P. H. van Beek, and M. F. P. Bierkens (2025). “A first attempt to model global hydrology at hyper-resolution”. In: Earth System Dynamics 16(1), pp. 29–54. doi: 10.5194/esd-16-29-2025.
Sutanudjaja, E. H., R. Van Beek, N. Wanders, Y. Wada, J. H. Bosmans, N. Drost, R. J. Van Der Ent, I. E. De Graaf, J. M. Hoch, K. De Jong, D. Karssenberg, P. López López, S. Peßenteiner, O. Schmitz, M. W. Straatsma, E. Vannametee, D. Wisser, and M. F. Bierkens (2018). “PCR-GLOBWB 2: A 5 arcmin global hydrological and water resources model”. In: Geoscientific Model Development 11(6), pp. 2429–2453. doi: 10.5194/gmd-11-2429-2018.
Karssenberg, D., M. F. P. Bierkens, and M. Rietkerk (2017). “Catastrophic shifts in semiarid vegetation-soil systems may unfold rapidly or slowly”. In: American Naturalist 190(6), E145–E155. doi: 10.1086/694413.
Data-based modelling: machine learning & geostatistics
Data-based modelling involves prediction of spatio-temporal variables by employing machine learning or geostatistics, mainly relying on observations. In our group, we also assess human exposures to environmental variables by combining spatial information of environmental attributes and the spatial context of individual persons.
Projects:
Automated 3D modelling of the subsurface of the Netherlands from heterogeneous data streams.
Looking back to plan ahead – unfolding the natural heritage of Dutch landscapes.
Example publications:
Garzón, S., W. Dabekaussen, F. S. Busschers, E. De Boever, S. Mehrkanoon, and D. Karssenberg (2026). “Assessment of automated stratigraphic interpretations of boreholes with geology-informed metrics”. In: Computers & Geosciences 207, p. 106043. doi: 10.1016/j.cageo.2025.106043.
Aebischer, P., M. Sutter, A. Birkinshaw, M. Nussbaum, and B. Reidy (2024). “Herbage biomass predictions from UAV data using a derived digital terrain model and machine learning”. In: Grass and Forage Science, pp. 530–542. doi: 10.1111/gfs.12694.
Birkinshaw, A., M. Sutter, M. Nussbaum, M. Kreuzer, and B. Reidy (2024). “Suitability of different machine learning algorithms for the classification of the proportion of grassland-based forages at the herd level using mid-infrared spectral information from routine milk control”. In: Journal of Dairy Science 107(12), pp. 10724–10737. doi: 10.3168/jds.2024-25090.
Shen, Y., K. de Hoogh, O. Schmitz, N. Clinton, K. Tuxen-Bettman, J. Brandt, J. H. Christensen, L. M. Frohn, C. Geels, D. Karssenberg, R. Vermeulen, and G. Hoek (2024). “Monthly average air pollution models using geographically weighted regression in Europe from 2000 to 2019”. In: Science of the Total Environment 918. doi: 10.1016/j.scitotenv.2024.170550.
Ntarladima, A.-M., D. Karssenberg, M. Poelman, D. E. Grobbee, M. Lu, O. Schmitz, M. Strak, N. Janssen, G. Hoek, and I. Vaartjes (2022). “Associations between the fast-food environment and diabetes prevalence in the Netherlands: a cross-sectional study”. In: The Lancet Planetary Health 6(1), e29–e39. doi: 10.1016/S2542-5196(21)00298-9.
Hybrid modelling: integrating process-understanding in machine learning
Hybrid modelling involves the use of process-knowledge in machine learning through various approaches including incorporating process-knowledge in designing the loss function, the architecture of the machine learning model itself, or other means such as using variables from process-based models as features in machine learning models. Our group also uses more traditional data-assimilation methodologies for integrating observations and process-based models.
Projects:
Example publications:
Pomarol Moya, O., S. Mehrkanoon, M. Nussbaum, W. W. Immerzeel, and D. Karssenberg (2025). “Machine learning emulators of dynamical systems for understanding ecosystem behaviour”. In: Ecological Modelling 501, p. 110956. doi: 10.1016/j.ecolmodel.2024.110956.
Magni, M., E. H. Sutanudjaja, Y. Shen, and D. Karssenberg (2023). “Global streamflow modelling using process-informed machine learning”. In: Journal of Hydroinformatics 25(5), pp. 1648–1666. doi: 10.2166/hydro.2023.217.
Shen, Y., J. Ruijsch, M. Lu, E. H. Sutanudjaja, and D. Karssenberg (2022). “Random forests-based error-correction of streamflow from a large-scale hydrological model: Using model state variables to estimate error terms”. In: Computers and Geosciences 159. doi: 10.1016/j.cageo.2021.105019.
Verstegen, J. A., D. Karssenberg, F. van der Hilst, and A. P. Faaij (2016). “Detecting systemic change in a land use system by Bayesian data assimilation”. In: Environmental Modelling and Software 75, pp. 424–438. doi: 10.1016/j.envsoft.2015.02.013.
Software framework design and implementation
We design methods and implement software to enable spatio-temporal modelling, in particular for simulation of large, heterogeneous geographical systems:
Representation of fields and agents in simulation models to enable heterogeneous system modelling
Simulation on supercomputers to run models with big data
Numerical solution techniques including PINN
Projects:
LUE Environmental Modelling Software The LUE software enables building and executing simulation models of geographical systems. Model developers can develop models using a syntax that looks very similar to map algebra, in either Python or C++. Given such a model, model developers can simulate real-world geographical systems with a large extent and at high resolutions. LUE models can be executed on small laptops and on large cluster partitions.
Campo, spatial agent-based modelling framework.
PCRaster, spatial simulation model building framework.
GEESE, global human environmental exposure data space, data space in SAGE.
Example publications:
Gargari, S. F., Z. Huang, and S. Dabiri (2024). “An upwind moving least squares approximation to solve convection-dominated problems: An application in mixed discrete least squares meshfree method”. In: Journal of Computational Physics 506. doi: 10.1016/j.jcp.2024.112931.