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

Multivariate and spatially calibrated hydrological model for assessing climate change impacts on hydrological processes in West Africa

Moctar Dembélé1, Sander Zwart2, Natalie Ceperley1, Grégoire Mariéthoz1, and Bettina Schaefli1,3
Moctar Dembélé et al.
  • 1Institute of Earth Surface Dynamics (IDYST), University of Lausanne, Lausanne, Switzerland (
  • 2International Water Management Institute, Accra, Ghana
  • 3Now at: Institute of Geography (GIUB), University of Bern, Switzerland

Robust hydrological models are critical for the assessment of climate change impacts on hydrological processes. This study analysis the future evolution of the spatiotemporal dynamics of multiple hydrological processes (i.e. streamflow, soil moisture, evaporation and terrestrial water storage) with the fully distributed mesoscale hydrologic Model (mHM), which is constrained with a novel multivariate calibration approach based on the spatial patterns of satellite remote sensing data (Dembélé et al., 2020). The experiment is done in the large and transboundary Volta River Basin (VRB) in West Africa, which is a hotspot of climate vulnerability. Climate change and land use changes lead to recurrent floods and drought that impact agriculture and affect the lives of the inhabitants.

Based on data availability on the Earth System Grid Federation (ESGF) platform, nine Global Circulation Models (i.e. CanESM2, CNRM-CM5, CSIRO-Mk3-6-0, GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-MR, MIROC5, MPI-ESM-LR and NorESM1-M) available from the CORDEX-Africa initiative and dynamically downscaled with the latest version of the Rossby Centre's regional atmospheric model (RCA4) are selected for this study. Daily datasets of meteorological variables (i.e. precipitation and air temperature) for the medium and high emission scenarios (RCP4.5 and RCP8.5) are bias-corrected and used to force the mHM model for the reference period 1991-2020, and the near- and long-term future periods 2021-2050 and 2051-2080.

The results show contrasting trends among the hydrological processes as well as among the GCMs. The findings reveal uncertainties in the spatial patterns of hydrological processes (e.g. soil moisture and evaporation), which ultimately have implications for flood and drought predictions. This study highlights the importance of plausible spatial patterns for the assessment of climate change impacts on hydrological processes, and thereby provide valuable information with the potential to reduce the climate vulnerability of the local population.



Dembélé, M., Hrachowitz, M., Savenije, H., Mariéthoz, G., & Schaefli, B. (2020). Improving the predictive skill of a distributed hydrological model by calibration on spatial patterns with multiple satellite datasets. Water Resources Research.

How to cite: Dembélé, M., Zwart, S., Ceperley, N., Mariéthoz, G., and Schaefli, B.: Multivariate and spatially calibrated hydrological model for assessing climate change impacts on hydrological processes in West Africa, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-9143,, 2020

Display materials

Display file

Comments on the display material

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

Display material version 1 – uploaded on 06 May 2020
  • CC1: Comment on EGU2020-9143, Jorn Van de Velde, 07 May 2020

    Interesting presentation! Do you have suggestion why the correction of temperature and precipitation by R2D2 performed not as well as expected?

    + Are the results on the bias correction already available in an article? Or do you first want to improve the correction?

    • AC1: Reply to CC1, Moctar Dembélé, 07 May 2020

      Thanks, Jorn!

      I think there are several reasons why the bias correction did not perform as well as expected.

      -The first can be the temporality. R2D2 is a rank resampling approach, which adjust the bias in data by matching rank associations in the reference with those in the projection data. In our case, we adopted a multivariate bias correction approach by jointly correcting 4 variables: rainfall, Tavg, Tmax and Tmin. Therefore, finding the correct rank associations for 4 variables simultaneously is not straighforward, and can lead to poor results. 

      -Secondly, R2D2 requires the user to choose conditioning dimensions (i.e. grid cells) which serve as reference for the rank association. The choice of the conditioning dimensions is not trivial and might also be a source of errors.

      We will readjust those parameters and investigate the outcome. The results are not yet published. We will first try to improve the method.



      • CC2: Reply to AC1, Jorn Van de Velde, 07 May 2020

        Thank you for the response! I didn't notice there were four variables. Temporality is indeed an issue in multivariate methods, and for four variables this could easily be exacerbated. In case you haven't read it already, the preprint by François et al. discusses this as well (and their poster is in session CL3.1). I'm looking forward to read how you deal with temporality in the resulting paper!

        Have you considered bias nonstationarity to be a possible cause? I noticed you read my poster. I also have results on temperature for my paper in preparation, and used a similar setup with MPI-ESM-LR and RCA4 under RCP4.5 conditions, so if you want to discuss that aspect, you can always mail me (



        • AC2: Reply to CC2, Moctar Dembélé, 07 May 2020

          Yes, non-stationarity might be a possible cause. I am planning to test other methods like the dOTC (Robin et al 2019, HESS) that accounts for non-stationarity. I am aware of the great work by François et al., and I am working with Mathieu Vrac (developer of R2D2). I got short of time before EGU, but planning to investigate on the causes that limit the performance of the bias correction.

          I had a look at your poster and added it to my personnal programme. I will check it again as well as your preprint in HESSD, and will contact you for discussions. 



          • CC3: Reply to AC2, Jorn Van de Velde, 08 May 2020

            Good luck in your collaboration with Mathieu Vrac and the use of dOTC. This will probably yield very interesting results, I'm looking forward to it!

            • AC3: Reply to CC3, Moctar Dembélé, 08 May 2020

              Thanks, Jorn!