Research Collection Journal Article Ensemble flood forecasting considering dominant runoff processes - Part 1: Set-up and application to nested basins (Emme, Switzerland) Author(s): Antonetti, Manuel; Horat, Christoph; Sideris, Ioannis V.; Zappa, Massimiliano Publication Date: 2019-01-07 Permanent Link: https://doi.org/10.3929/ethz-b-000315721 Originally published in: Natural Hazards and Earth System Sciences 19(1), http://doi.org/10.5194/nhess-19-19-2019 Rights / License: Creative Commons Attribution 4.0 International This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use. ETH Library Nat. Hazards Earth Syst. Sci., 19, 19–40, 2019 https://doi.org/10.5194/nhess-19-19-2019 © Author(s) 2019. This work is distributed under the Creative Commons Attribution 4.0 License. Ensemble flood forecasting considering dominant runoff processes – Part 1: Set-up and application to nested basins (Emme, Switzerland) Manuel Antonetti1,2, Christoph Horat1,3, Ioannis V. Sideris4, and Massimiliano Zappa1 1Swiss Federal Institute for Forest, Snow and Landscape Research, Birmensdorf, Switzerland 2University of Zurich, Department of Geography, Zurich, Switzerland 3ETH, Institute for Atmospheric and Climate Science, Zurich, Switzerland 4MeteoSwiss, Swiss Federal Office of Meteorology and Climatology, Locarno, Switzerland Correspondence: Massimiliano Zappa ([email protected]) Received: 25 April 2018 – Discussion started: 2 May 2018 Revised: 31 October 2018 – Accepted: 10 December 2018 – Published: 7 January 2019 Abstract. Flash floods evolve rapidly during and after heavy more, special emphasis was placed on the predictive power precipitation events and represent a potential risk for society. of the new forecasting chains in nested subcatchments when To predict the timing and magnitude of a peak runoff, it is compared with a prediction chain including an original ver- common to couple meteorological and hydrological models sion of the runoff generation module of PREVAH calibrated in a forecasting chain. However, hydrological models rely on for one event. strong simplifying assumptions and hence need to be cali- Results showed a low sensitivity of the predictive power brated. This makes their application difficult in catchments to the amount of expert knowledge included for the mapping where no direct observation of runoff is available. approach. The forecasting chain including a map of runoff To address this gap, a flash-flood forecasting chain is pre- types with high involvement of expert knowledge did not sented based on (i) a nowcasting product which combines guarantee more skill. In the larger basins of the Emme re- radar and rain gauge rainfall data (CombiPrecip); (ii) mete- gion, process-based forecasting chains revealed comparable orological data from state-of-the-art numerical weather pre- skill to a prediction system including a conventional hydro- diction models (COSMO-1, COSMO-E); (iii) operationally logical model. In the small nested subcatchments, although available soil moisture estimations from the PREVAH hy- the process-based forecasting chains outperformed the orig- drological model; and (iv) a process-based runoff genera- inal runoff generation module, no forecasting chain showed tion module with no need for calibration (RGM-PRO). This satisfying skill in the sense that it could be useful for decision last component uses information on the spatial distribution makers. of dominant runoff processes from the so-called maps of Despite the short period available for evaluation, prelimi- runoff types, which can be derived with different mapping nary outcomes of this study show that operational flash-flood approaches with increasing involvement of expert knowl- predictions in ungauged basins can benefit from the use of edge. RGM-PRO is event-based and parametrised a priori information on runoff processes, as no long-term runoff mea- based on the results of sprinkling experiments. surements are needed for calibration. This prediction chain has been evaluated using data from April to September 2016 in the Emme catchment, a medium- sized flash-flood-prone basin in the Swiss Prealps. Two novel forecasting chains were set up with two different maps of runoff types, which allowed sensitivity of the forecast per- formance to the mapping approaches to be analysed. Further- Published by Copernicus Publications on behalf of the European Geosciences Union. 20 M. Antonetti et al.: HEPS using DRPs – Part 1: Set-up 1 Introduction cal model that is run iteratively with increasing amounts of rainfall of a given duration. The FFG provides a value of sus- Flash floods (FFs) arising from the interaction of the at- ceptibility of a basin to a FF and takes the hydrological state mospheric and the hydrological system are characterised by of the system and in particular soil moisture into account. In a runoff peak that develops within time periods that range operational mode, FFG is computed each day. When now- from minutes to hours and may occur during or after intense cast or forecast rainfall depth is higher than FFG, a warning rainfall (Norbiato et al., 2008). They may result in threat- is issued as a flooding is likely. Although this concept is use- ening catastrophes and pose a risk to society, especially on ful, neither the timing nor the magnitude of the event is as- small-scale catchments (of few hundred square kilometres of sessed (Norbiato et al., 2008). As a further approach, Collier size or less) with steep slopes and shallow soils. Since small and Fox(2003) proposed a Flash Flood Susceptibility As- basins react quickly to precipitation there is only little time sessment Procedure (FFSAP), which is similar to what Mani for warnings (Liechti et al., 2013). Furthermore, FFs can be et al.(2012) elaborated for the catchments investigated here accompanied by landslides and mud flows (Collier, 2007). (see Sect.2) and to what is currently deployed in Saxony Impermeable surfaces and saturated soils may accelerate the (eastern Germany) for operational flash-flood early warning rainfall–runoff transition (Norbiato et al., 2008). (Philipp et al., 2016). Mani et al.(2012) developed an ap- FFs are considered to be significant natural hazards and proach for the Swiss Emme basin based on the concept of they are associated with a serious risk to life and destruction “disposition”, defined as the susceptibility of a region to flash of buildings and infrastructure (Collier, 2007; Norbiato et al., floods and debris flow. In their approach, the actual disposi- 2008; Gaume et al., 2009). In Europe, FF occurrence peaks tion is defined by the sum of base and variable disposition, during autumn in Mediterranean and Alpine–Mediterranean whereby the former is inferred from geological properties of areas and during summer in inland continental regions due the catchment and the latter is dependent on time. Whether to pronounced convective activity (Norbiato et al., 2008; a process initiation through heavy precipitation is expected Marchi et al., 2010). The magnitude of the events is in gen- – meaning that the actual disposition reaches a threshold – eral larger in Mediterranean countries than in inner continen- is determined with analyses of rainfall radar data (Panziera tal countries (Gaume et al., 2009; Javelle et al., 2010). Ac- et al., 2016). Although this approach provides the geograph- cording to Gaume et al.(2009), the most severe FF events ical distributions of event-prone areas, it is expensive as it in Europe were the Barcelona flood in Spain (1962) with requires periodic field work to sample the variable disposi- over 400 casualties (Lopez Bustos, 1964), the two floods in tion. In addition, as with the FFG concept, it does not pro- the region of Piedmont in Italy (1968 and 1994) with re- vide detailed information on the magnitude and timing of spectively 72 and 69 victims (Ferro, 2005; Guzzetti et al., an event. Several combinations of meteorological and hy- 2005) and the Aude flood in France (1999) with 35 fatal- drological models were implemented in so-called forecast- ities (Gaume et al., 2004). Economic damages associated ing chains to quantitatively predict peak flows. It was already with such floods were substantial, e.g. EUR 3.3 billion for the examined by, for example, Georgakakos(1986), who imple- Aude flood (Lefrou et al., 2000) and EUR 1.2 billion for the mented a stochastic–dynamic hydrometeorological model. In Garde flood which occurred in 2002 in France (Huet et al., general, a forecasting chain consists of (a) an atmospheric 2003; Delrieu et al., 2005; Braud et al., 2010). In Switzer- model, (b) a hydrological prediction system, (c) a nowcast- land, in June 2007, heavy precipitation caused flooding of ing tool for initial conditions and (d) warnings for end users the river Langeten and landslides in the region of Huttwil, (Zappa et al., 2008, 2011). The advantage of this approach Canton of Bern. This led to three fatalities and damages of is that timing and magnitude of the event can be predicted. CHF 60 million (Liechti, 2008). In July 2014, flooding of the Some examples of forecasting chains are described below, river Emme and landslides were responsible for damages of with a particular focus on the hydrological model. CHF 15 million in Schangnau, Canton of Bern (Andres et al., Rossa et al.(2010) carried out a case study for the 2015). 26 September 2007 Venice FF in the 90 km2 Dese River basin. They implemented a forecasting chain with a semi- 1.1 Current approaches for flash-flood prediction distributed hydrological–hydraulic model that is based on the Green–Ampt approach (Heber Green and Ampt, 1911) for As both meteorological and hydrological conditions are im- infiltration-excess and saturation-excess runoff generation portant for FF prediction, coupled approaches were devel- and the Penman–Monteith equation (Penman, 1948; Mon- oped, for instance, the so-called Flash Flood Guidance (FFG) teith et al., 1965) for evapotranspiration fluxes. As the river concept, which is used to issue warnings in the USA (Car- network of their study area is affected by tides, the coupling penter et al., 1999; Norbiato et al., 2008).