The importance of river hydrodynamics modeling for large scale flood forecasting

Contributed by Ayan Fleischmann and Fernando Fan, members of the LSH Research Group Guest Columnist Team

River hydrodynamics and hydrographs

Satisfactory flood predictions require satisfactory model representativeness of river transport processes. In the past years, the Large Scale Hydrology (LSH) Research Group has carried out many studies regarding the improvement of river hydrodynamics representation in large scale hydrological models (see the list of references at the end of this post) through the implementation of Saint-Venant or Local Inertial flow routing equations in rainfall-runoff models.

Indeed, transport processes are fundamental in defining the basin response (in other words, the basin outflow hydrograph). Hydrographs in steep terrains such as the Taquari-Antas tend to have a steep rising limb due to the non-linear relationship between flood wave celerity and discharge, while systems in flat terrains with well-developed floodplains tend to present attenuated and delayed peaks.

Also, hydrograph skewness is largely affected by how celerity interacts with discharge, so that in events that do not flood overbank, a positive skewed hydrograph tends to occur, while after floodplain inundation, negative skewed ones may happen (Figure 1). Such processes significantly alter peak travel times and the distribution of hydrograph volume, which are main outputs of flood forecasting models.

Sample cases

Most modeling applications of our group use the MGB-IPH hydrological model proposed by Collischonn et al. in 2007, which is a semi-distributed rainfall-runoff hydrological model developed for simulation of large basins. In the next paragraphs, we present some examples of the importance of correctly representing flood routing and river hydrodynamics for good model predictions in such rainfall-runoff models.

Our first example is from Paraná river basin, from the Itaipu Dam forecasting model (Pontes et al. 2015b, in Portuguese). In this study, the aim was to compare the use of the Local Inertial flow routing model using a simple floodplain storage model against the Muskingum-Cunge flood routing model, to accomplish flood forecasts (Figure 2). Both flow routing models are implemented in the MGB-IPH hydrological model. The case study is the Paraná River Basin, delimited until the Yacyretá reservoir (including Itaipu Dam). The assessments showed that MGB-IPH with Local Inertial model (red line) is better than the MGB-IPH using Muskingum-Cunge model (green line), in terms of diffusive effects and attenuation of floods. This is seen in Figure 2 because the red line (Local Inertial routing considering floodplains) is more similar to the blue line (observation) than the green one (Muskingum-Cunge model, without floodplains).

The second example is the Upper Iguaçu Basin. The Upper Iguaçu Basin suffers from frequent flooding, where the city of União da Vitória is located (see Araujo et al., 2014 for a description of flood forecasts in this area). Observed discharges at this city are shown in Figure 3, together with simulated flows with linear Muskingum-Cunge routing method. There is a clear smoothness and peak attenuation on observed hydrographs due to floodplains in upstream areas, which cannot be simulated with such simple routing methods.

The third example is the Amazon river basin. Paiva et al. 2013 used MGB-IPH model to simulate the Amazon river basin, where extensive floodplains exist in the low, flat terrains of the central areas of the basin. Comparison of time delay index between Muskingum-Cunge routing method and hydrodynamics implementation (full Saint-Venant equations and representation of floodplain storage) showed around 60 days of peak time delay for simulations without floodplains (Fig 4.i).

Also, comparison between simulation with hydrodynamic model and Muskingum-Cunge with floodplains (Todini, 2007) indicate that other processes are also important for the definition of hydrograph shape, such as backwater and flood wave diffusion processes. Paiva et al. 2012 used this model to satisfactorily perform ensemble forecasts for the Amazon basin and evaluate the role of initial conditions on flood forecasting (Fig 4.ii). They showed that “uncertainty on initial conditions may play an important role for discharge forecasts even for large lead times (~1 to 3 months) on main Amazonian Rivers”, and that an “Ensemble Streamflow Prediction approach based on a hydrological model forced with historical meteorological data and using optimal initial conditions (e.g. data assimilation) may be feasible for hydrological forecasting even for large lead times (~1 to 3 months)”.

Finally, the fourth example is the Niger river basin model. A coupled hydrologic-hydrodynamic MGB-IPH model was also developed for the Upper Niger Basin (Fleischmann et al., paper in preparation), where a link between flooded areas and soil columns associated with hydrodynamic modeling of large scale floodplain channels allowed a good representation of flooded areas in the Niger Inner Delta as well as discharge and water level time series. Figures 5 and 6 present model results for Diré gauge station, downstream of Niger Inner Delta, with Muskingum-Cunge and Hydrodynamic flood routing methods.

Implications for ensemble flood forecasting predictions

The representation of hydrodynamic processes is necessary for flood forecasting at  many large scale basins. We showed here some examples of watersheds modeled by our group in South America (Paraná, Iguaçu, Amazon) and Africa (Niger) where the correct modelling of floodplains is crucial for a good representation of the system. Flood peak magnitude depends mainly on (i) floodplain attenuation and (ii) acceleration due to non-linear relationship between flood celerity and discharge.

How this relates to ensemble forecasts? Let’s think about ensemble forecasting models in complex basins with floodplains, braided drainage network, or flat relief. Usually forecasts that incorporate more complicated hydrodynamics request longer simulation times and more complex models setup. This way, a number of forecasts, instead of a single one, would take more time than usual to run and the model needs to be very robust to support the various cases that can happen in terms of hydrodynamics. One of the aims of the research carried out by the LSH group in Brazil is to develop robust large scale models and efficient setups to couple with these situations.

Please let us know also about your experience with these situations in the comments!

See here the list of references.