Improved Characterization of Fluxes Across Compartmental Interfaces Gauss Centre for Supercomputing e.V.


Improved Characterization of Fluxes Across Compartmental Interfaces

Principal Investigator:
Clemens Simmer

Institute for Geosciences, University of Bonn

Local Project ID:
chbn29, chbn37

HPC Platform used:

Date published:

This project has the goal to develop a data assimilation framework for coupled atmosphere-land-surface-groundwater models. These coupled models allow potentially a more accurate description of the coupled terrestrial water and energy fluxes, in particular fluxes across compartments, but are affected by large uncertainties related to uncertain input parameters, initial conditions and boundary conditions. Data assimilation can alleviate these limitations and this project is focused in particular on the value of coupled data assimilation which means that observations in one compartment (e.g., subsurface) are used to update states, and possibly also parameters, in another compartment (e.g., land surface). Data assimilation for the coupled model system is normally affected by the limited number of observations for the subsurface which makes validation of new (coupled) data assimilation methods problematic. We circumvent this issue by using a virtual catchment approach. We can extract virtual observations from the virtual reality, to be used in data assimilation, and the virtual reality allows us also to evaluate the impact of data assimilation. Modern data assimilation techniques are ensemble based and running even a single model requires significant computational resources, running a whole ensemble is only possible on an HPC system.

Numerical models of the Earth system compartments are coupled in order to simulate physically consistent water, energy and biogeochemical fluxes in the subsurface-land surface-atmosphere system. Such model systems become increasingly important to analyze and understand the complex processes at boundaries of terrestrial compartments and interdependencies of states across these boundaries. They produce state evolutions which, when run at highest computationally possible resolutions while incorporating as many processes as attainable, may be regarded as a proxy of the real world. Thus, hypotheses regarding the behavior of the coupled terrestrial system can be tested. The coupled model system could include a data assimilation (DA) system which should result in state evolutions of the model system close to given observations, and also non-observed or not accessible state variables and system parameters can be estimated and predictions of future states can be made.

We carry out research with a coupled atmosphere-land-surface-groundwater model including a data assimilation system. We use the Terrestrial Systems Modelling Platform [1], which couples the atmospheric model COSMO, the Community Land Model (CLM) and the groundwater model ParFlow via an external OASIS coupler. DA is performed by the Parallel Data Assimilation Framework PDAF [2], which unlike most other systems operates mainly in the computer memory and minimizes I/O-operations.

Our developments follow two branches. One branch is the virtual catchment, which simulates the coupled atmosphere-land surface-subsurface system at the large catchment scale at a spatial resolution as high as possible, in order to approximate reality as much as possible. The virtual catchment mimics the Neckar catchment in SW-Germany (see Figure 1 for a topographic map of the simulated region). The Neckar catchment is a typical medium sized mid-latitude catchment, which features areas of various land-use, topography and degree of human management. For the atmosphere a resolution of 1.1km is used and currently a resolution of 400m is used for the land-surface and subsurface. Virtual measurements are taken from this virtual catchment by measurement observation operators, and place and time of these measurements, as well as precision are prescribed in the simulation experiments. Extended state evolutions are required to sufficiently sample the potential state space of the system and observation; we simulate the state evolution of the virtual Neckar catchment for the time period 2007 to 2018. Figure 2 shows a snapshot of the virtual catchment with clouds, rain and volumetric soil moisture content of the upper ten soil layers (3m).  The time period of 2007-2015 has been published in the CERA database and is available for download[3]. The main computational challenge of this run was its length and the output quantity. The simulation started on JUQUEEN and migrated later to JUWELS.

Before we can use the virtual observations for DA, we have to make sure that, when compared to real measurements, our system produces statistically comparable results. We do not expect exact matches as the virtual catchment is not expected to reproduce the same random fluctuations as the Neckar catchment, and because some model assumptions are not fully congruent with the real world, especially concerning the treatment of rivers. Figure 3 shows some results from the virtual catchment concerning precipitation and Figure 4 shows the modelled groundwater depth. The precipitation results confirm that our system works as good as any atmospheric model in simulating precipitation. Groundwater depth is according the expectations with deep groundwater levels in the mountains (not in the valleys though) and shallow groundwater in the larger flatlands like the Rhine valley. Given these results we are confident to test our DA system with observations from the virtual catchment.

The second branch is the construction of an ensemble-based DA system, which needs many potential state evolutions simulated in parallel by the model system. Given the large number of model runs that are performed in parallel, the different model runs have to be carried out at a lower spatial resolution than the virtual reality (800m). At the moment, we are finishing the second iteration of the ensemble system consisting of 64 members. This task is far more computationally intensive than even the virtual catchment and we are finishing to run an open-loop ensemble for at least a year to have sufficient data available for analysis. These runs need 256 nodes of JUWELS to run and produce several TB of output per day they are run. HPC is therefore absolutely necessary to run these ensembles. While we do have some first finished fully coupled experiments with data assimilation, the results are not yet ready to be published.

Since running this huge ensemble for every single test of a new DA algorithm would require very large amounts of compute resources, we have developed several smaller experiments that use the same modeling platform but focus on different aspects we want to explore, such as strategies for parameter updates in the subsurface. One system focusing on soil moisture and soil parameter updates does not include the COSMO model and uses prescribed atmospheric forcings instead. Another system seen in Figure 5 uses a higher resolution but with a much smaller domain with just a single hill [4].

The atmospheric boundary layer is as lowermost atmospheric layer heavily influenced by fluxes across the atmosphere-land interface. With an atmospheric only model system, we discovered that wind profile observations in this layer have valuable information for data assimilation [5], especially during nighttime where small-scale processes dominate. We further revised the potential of using observations from the boundary layer to infer states within the soil like the soil moisture. We found in idealized experiments with a coupled COSMO-CLM system that we can directly update the soil moisture based on sparse 2-metre-temperature observations [6]. These results indicate that coupled data assimilation across the atmosphere-land interface is possible.

In the end, all the knowledge gained from these smaller experiments is used to develop a series of experiments that are then run with the fully coupled system. These experiments have been started with the assimilation of soil moisture content in the fully coupled system.

After successful tests, we will transfer and apply the derived methodology to real-world data from the Rur catchment, which is a focus area of the DFG-funded SFB TR32[7].

References and Links

[1] P. Shrestha, M. Sulis, M. Masbou, S. Kollet, and C. Simmer A scale-consistent terrestrial systems modeling platform based on COSMO, CLM, and ParFlow, Monthly Weather Review 142(9), 3466-3483, 2014 doi:10.1175/MWR-D-14-00029.1

[2] W. Kurtz, G. He, S.J. Kollet, R.M. Maxwell, H. Vereecken, and H—J. H. Franssen TerrSysMP-PDAF version 1.0): a modular high-performance data assimilation framework for an integrated land surface-subsurface model, Geoscientific Model Development 9(4), 1341-1360, 2016 doi:10.5194/gmd-9-1341-2016

[3] Schalge, Bernd; Baroni, Gabriele; Haese, Barbara; Erdal, Daniel; Geppert, Gernot; Saavedra, Pablo; Haefliger, Vincent; Vereecken, Harry; Attinger, Sabine; Kunstmann, Harald; Cirpka, Olaf A.; Ament, Felix; Kollet, Stefan; Neuweiler, Insa; Hendricks Franssen, Harrie-Jan; Simmer, Clemens (2020). Virtual catchment simulation based on the Neckar region. World Data Center for Climate (WDCC) at DKRZ.

[4] Daniel Erdal, Gabriele Baroni, Emilio Sánchez-León, Olaf A. Cirpka (2019). The value of simplified models for spin up of complex models with an application to subsurface hydrology. Computers & Geosciences. Volume 126, p. 62-72,doi: 10.1016/j.cageo.2019.01.014.

[5] Finn, T. S., Geppert, G., and Ament, F. (2020): Towards assimilation of wind profile observations in the atmospheric boundary layer with a sub-kilometre-scale ensemble data assimilation system, Tellus A: Dynamic Meteorology and Oceanography, 72:1, 1-14, DOI: 10.1080/16000870.2020.1764307

[6] Finn, T. S., Geppert, G., and Ament, F. (2021): Ensemble-based data assimilation of atmospheric boundary layer observations improves the soil moisture analysis, Hydrol. Earth Syst. Sci. Discuss. [preprint],, in review.

[7] C. Simmer, I. Thiele-Eich, M. Masbou, W. Amelung, S. Crewell, B. Diekkrueger,F. Ewert, H. J. Hendricks Franssen, A. J. Huisman, A. Kemna, N. Klitzsch, S. Kollet,M. Langensiepen, U. Loehnert, M. Rahman, U. Rascher, K. Schneider, J. Schween,Y. Shao, P. Shrestha, M. Stiebler, M. Sulis, J. Vanderborght, H. Vereecken, J. vander Kruk, T. Zerenner, and G. Waldhoff Monitoring and Modeling the Terrestrial Sys-tem from Pores to Catchments - the Transregional Collaborative Research Center onPatterns in the Soil-Vegetation-Atmosphere System Bulletin of the American Meteo-rological Society 96, 17651787, 2015 doi:10.1175/BAMS-D-13-00134.1


Project Team

Sabine Attinger7, Natascha Brandhorst3, Olaf Cirpka4, Tobias Finn1, Barbara Haese2, Harrie-Jan Hendricks Franssen5, Ching Pui Hung5, Stefan Kollet5, Harald Kunstmann6, Shaoning Lv1, Insa Neuweiler3, Emilio Sanchez4, Bernd Schalge1, Lennart Schüler7, Clemens Simmer (PI)1, Bastian Waldowski3

1Institute for Geosciences, University of Bonn (Germany)
2Institute of Geography, University of Augsburg (Germany)
3Institute of Fluid Mechanics and Environmental Physics in Civil Engineering, Leibniz Universität Hannover (Germany)
4Center for Applied Geoscience, Eberhard Karls Universität Tübingen (Germany)
5Institute of Bio- and Geosciences- Agrosphere (IBG-3), Forschungszentrum Jülich (FZJ), Jülich (Germany)
6Institute of Meteorology and Climate Research (IMK-IFU), Karlsruhe Institute of Technology (KIT), Garmisch-Patenkirchen (Germany)
7Computational Hydrosystems, Helmholtz Zentrum für Umweltforschung, Leipzig (Germany)

Scientific Contact

Dr. Bernd Schalge
Institute for Geosciences, University of Bonn
Meckenheimer Allee 176, D-53115 Bonn (Germany)
e-mail: bschalge [@]

Local project ID: chbn29, cbn37

March 2021

Tags: JSC Universität Bonn Environmental Science