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

How model selection can determine flood risk estimates – a case study in the Ganges basin using the GLOFRIM framework

Jannis Hoch1, Dirk Eilander2,3, and Hiroaki Ikeuchi4
Jannis Hoch et al.
  • 1Department of Physical Geography, Utrecht University, Utrecht, Netherlands (
  • 2Institute for Environmental Studies, VU Amsterdam, Amsterdam, Netherlands (
  • 3Deltares, Delft, Netherlands (
  • 4Department of Civil Engineering, University of Tokyo, Tokyo (

Fluvial flood events are a major threat to people and infrastructure. To compute flood risk estimates, modelling cascades are often applied. Therein, flood hazard is driven by hydrologic or river routing and floodplain flow processes. As such, model selection within such a cascade can determine how well some of these processes can be simulated. Depending on the selection made, obtained flood maps can vary and, in turn, can have major implications for the analysis of how many people, buildings, economic values and so forth is at risk. Understanding the role of model selection in the flood risk modelling process is thus of great importance.

By means of GLOFRIM 2.0, we coupled the global hydrologic model PCR-GLOBWB with the hydrodynamic models CaMa-Flood and LISFLOOD-FP for the delta region of the Ganges-Brahmaputra basin. Applying the model-coupling framework GLOFRIM facilitates forcing various models with identical boundary conditions and thus transparent and objective inter-comparison of flood models.

While replacing the kinematic wave approximation of the hydrologic model with the local inertia equation of hydrodynamic models does not yield better discharge estimates in the Ganges basin, flood maps obtained with LISFLOOD-FP improved representation of observed flood extent. Compared to downscaled products of PCR-GLOBWB and CaMa-Flood, the critical success index increases by around 50 %.

Combining the obtained flood maps with actual exposure maps gives then a first-order estimate how the selection for one specific model set-ups translates into varying flood risk estimates. The research thus shows how those model selections, deliberately made or not, are an important driver of simulated flood risk. As such, it is detrimental that the various specifics of a model are known to facilitate the optimal model selection for objective-specific modelling requirements.

How to cite: Hoch, J., Eilander, D., and Ikeuchi, H.: How model selection can determine flood risk estimates – a case study in the Ganges basin using the GLOFRIM framework, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-11086,, 2020

Display materials

Display file

Comments on the display material

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

Display material version 2 – uploaded on 04 May 2020
minor content changes (no essential parts touched!)
  • CC1: Comment on EGU2020-11086, Liz Stephens, 04 May 2020

    Hi Jannis,

    Thanks for the presentation I enjoyed reading it, especially the Star Wars font! I think we might need to do a similar analysis for Uganda. Can you tell me if you performed any calibration / optimisation on the models?

    It seems to me that the results for CSI / population affected don't match because the CSI is a measure of flooding across the whole domain, whereas the population estimate is heavily skewed by Dhaka. I wonder if they would better align if you looked at the CSI for just the Dhaka part of the domain?

    I guess an answer to the question you ask at the end is that I think your analysis shows we should be looking at model skill where there is the most exposure, because that will give us the most precise estimate of flood risk.


    • AC1: Reply to CC1, Jannis Hoch, 04 May 2020

      Hi Liz,

      thanks for reading it and glad you liked it - thus far you're the only one mentioning the star wars "hint", but how often does one present on May 4th... :)

      As outlined in the underlying publication (), we did a bit of a calibration. But only for discharge and nothing beyond the friction and river bathymetry scale.

      I would concur with your assumption that if CSI was computed for the strongly populated only, the match between models with highest CSI and closest to exposure benchmark may be probably better. Something to figure out in follow up steps I suppose!

      And I also agree with you last statement. Nevertheless, this is not something you typically see, right? Often the focus is on flood maps over large areas (floodplains typically), but happy to hear otherwise.

      Best, Jannis

      • AC2: Reply to AC1, Jannis Hoch, 04 May 2020

        the link apparently did not work:

        • AC3: Reply to AC2, Jannis Hoch, 04 May 2020

          seems like hyperlinks don't work. the publication i refer to has the DOI 10.5194/nhess-19-1723-2019.

  • CC2: Comment on EGU2020-11086, Maik Renner, 04 May 2020

    Hi Jannis,

    this is great work and I really like the open-source + access approach.

    My first question is a bit offtopic: Do have working examples where the coupling framework was evaluated for a flood forecasting setting?

    Second: How much work is required to switch from one 1D hydrodynamic model (profiles and calibration) to the next?

    Thank you!


    • AC4: Reply to CC2, Jannis Hoch, 05 May 2020

      Hi Maik,

      thanks, the open source part is indeed a core part of the display. Did you check the capsule by any chance?

      We did not use the framework for forecasting. There are, however, ideas to use (parts) of it in forecasting mode. Do you have anything specific in mind? Personally, I am not an expert in forecasting, but would guess using ECMWF data as meteo-forcing should not be a major problem.

      It is not part of the framework to "switch" models. You can, of course, build one model on basis of the data of another model. The framework works more like an interface or wrapper to/around the models, i.e. in a first step you need to set up the models and then you can use the framework. Model set-up is thus not a part of the framework itself but up to the user.

      As said, you can create different models a priori based on the same input data and then apply these different hydrodynamic models and compare/benchmark them, see for example this work:

      Does this answer your questions?


  • CC3: Comment on EGU2020-11086, Maik Renner, 05 May 2020

    Hi Jannis,

    thank you for your reply! I now have a better idea of the potentials of your work!

    Personally, I am looking for open source approaches to perform hydrological- hydraulic river forecasts, including open source models. As an state agency we are spending money to develop and maintain such models. I want to see which open source approaches have the potential to move towards a more transparent and efficient way to do hydrological forecasting.

    The models for which your framework provides links are indeed quite interesting in that direction.



    • AC5: Reply to CC3, Jannis Hoch, 05 May 2020

      Hi Maik,

      good to hear!

      Let me know if you need additional information on the models and/or framework.

      Best, Jannis

Display material version 1 – uploaded on 01 May 2020, no comments