Signature & sensitivity-based comparison of conceptual and process oriented models GR4H, MARINE and SMASH on French Mediterranean flash floods

Faced with the major challenges of floods and droughts forecasting, especially with the ongoing climate change and potential intensification of the hydrological cycle, advanced modeling tools are needed to perform effective predictions. Nevertheless, hydrological models, regardless of their complexity, encounter difficulties in accurately and reliably predicting quantities of interest such as river discharge or soil saturation dynamics and its spatial variability. Because of physical processes complexity and their limited observability, of the absence of an easily exploitable "first-principle", hydrological modeling remains a difficult task involving emprism and the internal fluxes are generally tinged with large uncertainties. Moreover, multiple model and parameter combinations can lead to comparable performances in discharge simulation at locations where models are evaluated (unicity problem, so called equifinality in hydrology).

This contribution investigates flash flood modeling with models of different complexities: lumped GR4H (Perrin et al. 2003, Mathevet 2005) or distributed  SMASH (Jay-Allemand et al. 2020) conceptual models, process oriented distributed MARINE model (Roux et al. 2011). Considering two flash flood prone catchments (the Gardon at Anduze and the Ardèche at Vogüé, France) a methodology consisting in model global sensitivity analysis, calibration and hydrological signatures analysis is used. Model robustness and accuracy is analyzed in the light of model response surfaces, parameter sensitivity rankings and functionning points found with the different models and global calibration algorithms. Next, event performances and flow signatures are analyzed for contrasted events, but also simulated soil moisture evolutions (or equivalently available “soil” storage) compared to root zone soil moisture from the operational SIM hydro-meteorological model (Habets et al. 2008). This analysis is aimed at understanding how each model simulates the catchment behaviour: what are the differences between the simulated dynamics and how this understanding can be used to improve the relevance of the models. Finally, this study paves the way for extended model hypothesis testing and intercomparison in the light of multi-sourced signatures, for future improvements of vertical and lateral flow components of the SMASH* platform along with its variational calibration and assimilation algorithm.

References:

• Habets F., A. Boone, J.L Champeaux, et al. (2008)) : The SAFRAN-ISBA-MODCOU hydrometeorological model applied over France, Journal of Geophysical Research 113, D06113 (2008) 18

• Jay-Allemand M., P. Javelle, I. Gejadze, et al., On the potential of variational calibration for a fully distributed hydrological model: application on a mediterranean catchment. HESS, pages 1–24, 2019

• Mathevet, T., 2005. Which lumped rainfall-runoff models for the hourly time-step? Empirical development and comparison of models on a large sample of catchments. PhD Thesis. ENGREF, Cemagref (Irstea), Paris, France, pp. 463.

• Roux H., D. Labat , P.-A. Garambois, M.-M. Maubourguet, J. Chorda, D. Dartus, A physically-based parsimonious hydrological model for flash floods in mediterranean catchments. NHESS, 11(9):2567–2582, 2011.

• Perrin C., C. Michel, V. Andréassian, Improvement of a parsimonious model for streamflow simulation. Journal of hydrology, 279(1-4):275–289, 2003

*SMASH : Spatially-distributed Modelling and ASsimilation for Hydrology, platform developped by INRAE-Hydris corp., operationally applied  in the french flashflood forecast system VigicruesFlash - see presentation by J. Demargne et al.).