EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Biogeomorphic modeling: how to account for subgrid-scale interactions between hydrodynamics and vegetation patches

Olivier Gourgue1,2, Jim van Belzen3, Christian Schwarz4, Tjeerd J. Bouma3, Johan van de Koppel3, and Stijn Temmerman1
Olivier Gourgue et al.
  • 1Ecosystem Management Research Group, University of Antwerp, Belgium (
  • 2Department of Earth and Environment, Boston University, MA, USA
  • 3Department of Estuarine and Delta Systems, and Utrecht University, NIOZ Royal Netherlands Institute for Sea Research, Yerseke, Netherlands
  • 4Coastal Research Group, Department of Physical Geography, Faculty of Geosciences, Utrecht University, Netherlands

Interactions between water flow and patchy vegetation are governing the functioning of many ecosystems, such as river beds, floodplains, wetlands, salt marshes, mangroves and seagrass meadows. However, numerical models that simulate those interactions explicitly, including at the patch-scale (that is, at resolutions below a m²), together with their far-reaching geomorphological and ecological consequences at the landscape-scale (that is, for domain sizes of several km²), are still very computationally demanding. In this communication, we will present a novel efficient technique to incorporate biogeomorphic feedbacks across multiple spatial scales (from below a m2 to several km2) in biogeomorphic models. Our new methodology is based on the mathematical concept of convolution, allowing to spatially refine coarse-resolution (order of meters) hydrodynamic simulations of flow velocity fields around fine-resolution (order of dm) patchy vegetation patterns. We will demonstrate the power of our new method, by comparing our results with reference fine-resolution (order of cm) hydrodynamic model runs, which themselves are calibrated against flume measurements. We will show that our new model approach enables to refine a coarser-resolution hydrodynamic model, by resolving subgrid-scale fine-resolution flow velocity patterns within and around patchy vegetation distributions. With simple example cases, we will show evidence that our novel approach can substantially improve the representation of important processes in current biogeomorphic models, such as subgrid-scale effects on sediment transport and vegetation growth. Finally, we will demonstrate that our convolution method is an important step forward towards more computationally efficient multiscale biogeomorphic modeling, as compared with what is possible to date.

How to cite: Gourgue, O., van Belzen, J., Schwarz, C., Bouma, T. J., van de Koppel, J., and Temmerman, S.: Biogeomorphic modeling: how to account for subgrid-scale interactions between hydrodynamics and vegetation patches, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-10765,, 2020

Display materials

Display file

Comments on the display material

AC: Author Comment | CC: Community Comment | Report abuse

Display material version 1 – uploaded on 11 Apr 2020
  • CC1: Comment on EGU2020-10765, Karin Bryan, 06 May 2020

    Very interesting. Just to confirm, you run low resolution hydro model, apply the correction, and then run a high resolution sediment model seperately? Or you apply the correction to the sediment model, before interpreting the long-time scale evolution pattern like vegetation growth?


  • AC1: Comment on EGU2020-10765, Olivier Gourgue, 06 May 2020

    Hi Karin, thank you for your question.

    Indeed, we first run the low-resolution hydrodynamic model (we call it standard-resolution model), then we apply the correction. The high-resolution model is run separately and serves as reference to evaluate the gain with the correction.

    Sedimentation rates and areas for potential vegetation growth are evaluated a posteriori from the velocity fields. There is no feedback to the hydrodynamics here. That's for another paper :-)

    • CC2: Reply to AC1, Karin Bryan, 06 May 2020

      Very useful. I will read the paper, and try the method. Thank you!


  • CC3: Comment on EGU2020-10765, Pallav Ranjan, 07 May 2020


    It's a really interesting idea to use the convolution product as a way to modify the low resolution velocity signal (field) with a high resolution velocity signal (field). However, I would be concerned about the following :

    Magnitude of the modified flow field and wake interactions: The flow field in case of the mask (single vegetation patch, M) would have higher velocities than that in the original vegetation field (V). This is due to the interaction of adjacent wakes of different vegetation patches in the case of V, which is absent in M. So when the flow fields are convolved, it might be over (under?)-estimating the flow velocities, and missing the effect of the wake interactions on the velocity field. This is evident in figure on slide 2 of the presentation. I would like the authors to comment on the same.


    Also, is there a refence for the mathematical formulation of the convolution process? I am particularly interested in why logarithm of only the masking velocity field (M) was taken? What would happen if we reverse the order of convolution i.e. V**log(M) ?

    • AC3: Reply to CC3, Olivier Gourgue, 07 May 2020

      You're absolutely right, but our objective is not to compete with high-resolution models that can represent accuratly the processes you describe (Slide 2, Figure a). Our objective is to improve current large-scale models that operate at rather coarse resolution due to computational limitations, and are therefore completely unable to represent these fine-scale interactions (Slide 2, Figure b). Our method is a compromise between computational efficiency and accuracy (Slide 2, Figure c) and our main results suggest that the gain is worth the 50-percent extra computational cost (Slide 5).

      Regarding the reference for the mathematical formulation of the convolution method, the full manuscript supporting this study is under review in JAMES. By definition, the convolution is a commutative operation, so log(M) ** V = V ** log(M).

      Thank you for your interest.