Nowcasting of river discharge and flash floods constitutes a major challenge, partly because operational NWP is not yet capable of predicting convective precipitation events at the sub-km and hourly scale in a useful quality. This leads to unforeseen flash floods resulting in large damages to public property and infrastructure, and potentially loss of life.

Prominent examples in the area of the Geoverbund ABC/J are the destructive flash floods in Wachtberg on 3 July 2010 and on 6 June 2016. The project will develop a novel probabilistic nowcasting framework for river discharge and flash floods in small watersheds (<500 km²). The project focuses on three well instrumented small headwater catchments of the Wachtberg, Ammer and Bode watersheds, prone to flash floods. We will employ QPE, QPN and QPF (Quantitative Precipitation Estimation, Nowcasting and Forecasting, respectively) developed by P1, P2 and P3 in a nowcasting framework for river discharge in order to assess the impact of the improved products on flash flood prediction. One of the original features of the project will be the application of different hydrological models (conceptual and physically-based) with data assimilation (discharge and soil moisture) for flash flood forecasting. We will identify the added value and limitation of each model (and data assimilation method)for this exercise. While conceptual models can benefit from the calibration of their parameters to best fit the main variable of interest (discharge) according to the application (e.g. high flows) and the possibility of quickly running larger ensembles, physically-based models can benefit from the increasing availability of observations and more detailed process and land cover descriptions, which make these models easily transferable to other catchments.

The disastrous mid-July 2021 floods in the Eifel region (see Fig. P4-1a) offered the opportunity to evaluate the new developments in precipitation estimation and prediction within the RealPEP project. In particular, we investigated the added value of improving quantitative precipitation estimates (QPE, cf. P1) and nowcasts (QPN, cf. P2) for hydrological forecasting of such never-before-seen events.

Figure P4-1: (a) Location of the catchment set and (b) their hypsometric curves (Saadi et al., 2023a). Negative elevations are due to open-pit mines in the region.

Over seven catchments (Ahr at Muesch, Ahr at Altenahr, Erft at Bliesheim, Erft at Neubrueck, Kyll at Densborn, Kyll at Kordel, and Rur at Monschau; see Fig. P4-1a), we evaluated the newly-developed QPE products that exploited polarimetric radar variables (specific differential phase and specific attenuation at horizontal polarization) and their vertical gradients, which were produced by the P1 subproject group at a spatial resolution of 1 km2 and a temporal resolution of 5 min (Chen et al., 2023, 2021). Our evaluation framework of the QPE products consisted of two levels (Saadi et al., 2023a).

On the first level, we compared the catchment-scale precipitation depths on 14 July 2021 of each radar-based QPE product with catchment-scale precipitation depths estimated from rain gauges. Catchment-scale precipitation depths were estimated for each catchment at each hour by averaging point-scale estimates using Thiessen polygons. This comparison showed that using the specific attenuation at horizontal polarization combined with a correction of the vertical profile of horizontal reflectivity gave the most concordant estimates of catchment-scale precipitation depths with those from rain gauges (see Fig. P4-2a).


Figure P4-2: (a) Catchment-average, total precipitation depths on 14 July 2021 for each of the seven catchments from rain gauges and seven radar-based QPE products (the meaning of each product is given in Table 2 of Saadi et al., 2023a). (b) Probability (or simulated frequency) of exceeding the historical peak flow (i.e., the highest measured peak flow prior to July 2021) for each catchment. This probability was estimated by counting the number of peak flow estimates (12 from GR4H plus 4 from ParFlowCLM) that surpassed the highest measured peak flow prior to 2021.

On the second level, we compared the peak flow estimates for 14 July 2021 using two hourly hydrological models: GR4H (Ficchì et al., 2019) and ParFlowCLM (Kollet and Maxwell, 2006; Kuffour et al., 2020; Maxwell, 2013). We fed each model with each of the radar-based QPE products and rain gauges (except for ParFlowCLM, because precipitations from rain gauges were not gridded/spatially interpolated). We chose GR4H and ParFlowCLM to represent highly contrasting modeling approaches. GR4H is a top-down, conceptual model that uses algebraic equations to represent the relationship between temperature, precipitation, and discharge at the catchment scale (i.e., climate inputs are spatially averaged). It contains four catchment-scale parameters estimated by calibration on historical measurements of discharge. For our study, we retained 12 optimal GR4H parameter sets for each catchment to represent the effect of uncertainty in parameter estimation. Conversely, ParFlowCLM is a bottom-up, physics-based model that couples ParFlow and CLM (Common Land Model). CLM estimates the fluxes of infiltration and actual evapotranspiration by explicitly accounting for the spatial heterogeneities of the land surface, and ParFlow solves the 3D Richards equation coupled to the kinematic wave model for overland flow to represent the water dynamics from top of the surface down to bedrock. We used the ParFlowCLM implementation that was designed for the Adapter project (https://adapter-projekt.org/) at the scale of Central Europe with a planar resolution of ~611 m with 15 depth layers having a geometrically increasing thickness down to 60 m below the surface. Each cell has measurable parameters for land use, soil, and subsoil properties. For our study, we ran ParFlowCLM four times for each catchment corresponding to four different values of Manning’s roughness. The comparison of the QPE products for the July 2021 events showed that both models were highly sensitive to the precipitation input. To circumvent the absence of discharge observations for the event, we counted the times the peak flow estimates of GR4H (12 estimates) and ParFlowCLM (4 estimates) surpassed the highest measured peak flow prior to July 2021 for each catchment (“Probability” in Figure P4-2b). In some catchments (Ahr and Kyll catchments), this frequency was highly variable depending on the QPE product used as input precipitation for the models (see Fig. P4-2b), highlighting the importance of accurate precipitation estimates for extreme flooding events. For this event, GR4H and ParFlowCLM produced similar peak flow estimates except for the highly regulated catchments, which underlines the importance of accounting (even implicitly) for the anthropogenic features in the hydrological model structure (GR4H has the ability to learn these effects implicitly through calibration on historical observations, whereas ParFlowCLM does not account for anthropogenic influence on water cycle).

In addition to QPE, we evaluated 3-h long QPN products at 1 km2 resolution computed using two deterministic methods (Lagrangian persistence/advection, and S-PROG; Seed, 2003) and one probabilistic method with 20 members (STEPS; Bowler et al., 2006). To evaluate the added value of each method from a hydrological point of view, we compared the simulated hydrographs by each model run using the QPN (i.e., the forecasted hydrograph) with the one simulated using the QPE (i.e., the hindcasted hydrograph; see Saadi et al., 2023b). To quantify the skill of the QPN products, we compared them to two widely-used benchmarks: the hydrological persistence, and the zero-precipitation nowcasts (ZNC). We found that the three methods yielded highly similar results from a hydrological standpoint, as shown by their Nash-Sutcliffe scores in Figure P4-3 (Nash and Sutcliffe, 1970; Saadi et al., 2023b). In addition, they all improved the skill of the hydrological forecasts. However, the gain in lead time offered by the use of the QPN was variable depending on the hydrological model and the shape of the catchment. The evaluation of the added value of the QPN depended on the methodological choices, namely the benchmark against which the QPN are compared, the evaluation metric, and the chosen accuracy threshold to define the usefulness of the forecasts (Saadi et al., 2023b).



Figure P4-3: Evolution of the Nash-Sutcliffe Efficiency (NSE) of the forecasted hydrographs using the QPN methods and the benchmarks (hydrological persistence Q, zero-precipitation nowcasts ZNC) with respect to lead time. Red dashed lines indicate NSE at 0.9, the chosen threshold under which the forecasts are considered to be “useless”. For GR4H, only the median score from the 12 simulations is shown.

The next steps include the extension of the evaluation framework to a large sample of events in order to discriminate the QPN depending on the type of the event and the hydromorphological properties of the catchment. In addition, a hydrological evaluation of the added value of quantitative precipitation forecasts (QPF, cf. P3) in comparison with QPN is planned. Finally, better estimation of the initial hydric state of the catchment using data assimilation of soil moisture and river discharge is considered as a future step.

References

Bowler, N.E., Pierce, C.E., Seed, A.W., 2006: STEPS: A probabilistic precipitation forecasting scheme which merges an extrapolation nowcast with downscaled NWP. Q. J. R. Meteorol. Soc. 132, 2127–2155. https://doi.org/10.1256/qj.04.100.

Chen, J.-Y., Reinoso-Rondinel, R., Trömel, S., Simmer, C., Ryzhkov, A., 2023: A Radar-Based Quantitative Precipitation Estimation Algorithm to Overcome the Impact of Vertical Gradients of Warm-Rain Precipitation: The Flood in Western Germany on 14 July 2021. J. Hydrometeorol. 24, 521–536. https://doi.org/10.1175/JHM-D-22-0111.1.

Chen, J.-Y., Trömel, S., Ryzhkov, A., Simmer, C., 2021: Assessing the Benefits of Specific Attenuation for Quantitative Precipitation Estimation with a C-Band Radar Network. J. Hydrometeorol. 22, 2617–2631. https://doi.org/10.1175/JHM-D-20-0299.1.

Ficchì, A., Perrin, C., Andréassian, V., 2019: Hydrological modelling at multiple sub-daily time steps: Model improvement via flux-matching. J. Hydrol. 575, 1308–1327. https://doi.org/10.1016/j.jhydrol.2019.05.084.

Kollet, S.J., Maxwell, R.M., 2006: Integrated surface–groundwater flow modeling: A free-surface overland flow boundary condition in a parallel groundwater flow model. Adv. Water Resour. 29, 945–958. https://doi.org/10.1016/j.advwatres.2005.08.006.

Kuffour, B.N.O., Engdahl, N.B., Woodward, C.S., Condon, L.E., Kollet, S., Maxwell, R.M., 2020: Simulating coupled surface–subsurface flows with ParFlow v3.5.0: capabilities, applications, and ongoing development of an open-source, massively parallel, integrated hydrologic model. Geosci. Model Dev. 13, 1373–1397. https://doi.org/10.5194/gmd-13-1373-2020.

Maxwell, R.M., 2013: A terrain-following grid transform and preconditioner for parallel, large-scale, integrated hydrologic modeling. Adv. Water Resour. 53, 109–117. https://doi.org/10.1016/j.advwatres.2012.10.001.

Nash, J.E., Sutcliffe, J.V., 1970: River flow forecasting through conceptual models part I — A discussion of principles. J. Hydrol. 10, 282–290. https://doi.org/10.1016/0022-1694(70)90255-6.

Saadi, M., Furusho-Percot, C., Belleflamme, A., Chen, J.-Y., Trömel, S., Kollet, S., 2023a. How uncertain are precipitation and peak flow estimates for the July 2021 flooding event? Nat. Hazards Earth Syst. Sci. 23, 159–177. https://doi.org/10.5194/nhess-23-159-2023.

Saadi, M., Furusho-Percot, C., Belleflamme, A., Trömel, S., Kollet, S., Reinoso-Rondinel, R., 2023b: Comparison of three radar-based precipitation nowcasts for the extreme July 2021 flooding event in Germany. J. Hydrometeorol. (Online). https://doi.org/10.1175/JHM-D-22-0121.1.

Seed, A.W., 2003: A Dynamic and Spatial Scaling Approach to Advection Forecasting. J. Appl. Meteorol. Climatol. 42, 381–388. https://doi.org/10.1175/1520-0450(2003)042<0381:ADASSA>2.0.CO;2.