- 1Department of Physical Geography, Faculty of Geosciences, Utrecht University, Netherlands
- 2School of Agricultural, Forest and Food Sciences, Bern University of Applied Sciences, Zollikofen, Switzerland
Increasing a model's spatio-temporal resolution and extent has scientific and operational implications. First, hydrological processes
operating at a smaller spatio-temporal scale may need to be incorporated in the model's description, increasing its computational load. Second,
the data storage requirements of the model will increase. Third, an increase in model size (operations and data) will result in an increase
in memory and runtime requirements. Here, we focus on this latter implication.
To allow models that increase in size to execute they must be able to use additional hardware efficiently: they must scale with hardware. We
have ported two existing hydrological models, PCR-GLOBWB (Sutanudjaja et al. 2018) and PyCatch (Lana-Renault et al. 2013), to the LUE modelling
framework (LUE contributors. 2024), and are conducting scalability experiments to assess how well the models are capable of using
additional hardware.
PCR-GLOBWB simulates hydrology and water resources at a global scale. PyCatch simulates hydrological processes at the catchment scale. Both
models currently use the PCRaster modelling framework (PCRaster contributers. 2024), rasters to represent spatially varying model state,
and discrete time steps for simulating changes in model state over time. PCR-GLOBWB supports being run at continental and global scale at 5
arc-minute spatial resolution, using daily time steps. The PCR-GLOBWB research team aims to support 1km spatial resolution, and even 100m
resolution and hourly time steps after that. PyCatch supports being run at catchment scale at 10m spatial resolution, using hourly time steps.
Its research team aims to support regional scale runs at 5m spatial resolution.
PCRaster supports executing models using a single CPU core. LUE is a successor of PCRaster, capable of using all CPU cores in multiple
computers. It is implemented in C++ and makes use of the HPX standard library for concurrency and parallelism (Kaiser et al. 2024). For model
developers LUE provides language APIs for multiple programming languages, like C, C++, Java, and Python. Currently, the Python API is
ready to be used.
We have developed the lue.pcraster Python sub-package which allows PCR-GLOBWB and PyCatch to be executed with LUE, without having to change
the model code. In our presentation we will show more about LUE and highlight the first results of scalability experiments we are currently
performing for both models. These experiments characterize how well LUE is able to execute models faster by using additional hardware, and how
well LUE is able to use additional hardware to execute models with larger datasets.
References
Kaiser et al. 2024. "STEllAR-GROUP/hpx: HPX: The C++ Standards Library for Parallelism and Concurrency." https://doi.org/10.5281/zenodo.598202
Lana-Renault et al. 2013. "PyCatch: Component Based Hydrological Catchment Modelling." Cuadernos de Investigación Geográfica 39 (2): 315--333. https://doi.org/10.18172/cig.1993
LUE contributors. 2024. "LUE Scientific Database and Environmental Modelling Framework." https://doi.org/10.5281/zenodo.5535686
PCRaster contributers. 2024. "The PCRaster Environmental Modelling Framework." https://pcraster.computationalgeography.org
Sutanudjaja et al. 2018. "PCR-GLOBWB 2: A 5 Arcmin Global Hydrological and Water Resources Model." Geoscientific Model Development 11 (6): 2429--53. https://doi.org/10.5194/gmd-11-2429-2018
How to cite: de Jong, K., Schmitz, O., Sutanudjaja, E., Nussbaum, M., Scheffer, P., and Karssenberg, D.: The LUE modelling framework for scalable hydrological models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9124, https://doi.org/10.5194/egusphere-egu25-9124, 2025.
