Soil moisture updating by Ensemble Kalman Filtering in real-time flood forecasting Ju ¨rgen Komma * , Gu ¨nter Blo ¨schl, Christian Reszler Institute of Hydraulic and Water Resources Engineering, Vienna University of Technology, Karlsplatz 13/222, A-1040 Wien, Vienna, Austria Received 22 May 2007; received in revised form 28 April 2008; accepted 5 May 2008 KEYWORDS Data assimilation; Ensemble Kalman Filter; Soil moisture; Distributed rainfall-run- off model Summary The aim of this paper is to examine the benefits of updating soil moisture of a distributed rainfall runoff model in forecasting large floods. The updating method uses Ensemble Kalman Filter concepts and involves an iterative similarity approach that avoids calculation of the Jacobian that relates the states and the observations. The soil moisture is updated based on observed runoff in a real-time mode, and is then used as an initial condition for the flood forecasts. The case study is set in the 622 km 2 Kamp catchment, Austria. The results indicate that the updating procedure indeed improves the forecasts substantially. The mean absolute normalised error of the peak flows of six large floods decreases from 25% to 12% (3 h lead time), and from 25% to 19% (48 h lead time). The Nash-Sutcliffe efficiency of forecasting runoff for these flood events increases from 0.79 to 0.92 (3 h lead time), and from 0.79 to 0.88 (48 h lead time). The flood forecasting system has been in operational use since early 2006. ª 2008 Elsevier B.V. All rights reserved. Introduction Updating methods in real-time flood forecasting have en- joyed wide popularity in the late 1970s and early 1980s with the increasing use of telemetry in the control of water re- source systems (Wood, 1980). While numerous national flood forecasting systems have indeed implemented updat- ing procedures (e.g., Gutknecht, 1991), scientific interest soon ebbed off. The reasons may well be as O’Connell and Clarke (1981, pp. 202–203) noted: ‘‘The above discussion suggests that there are still considerable unsolved estima- tion problems in real-time forecasting, but it is not clear to what extent their solution would result in improved fore- casts. It may be more beneficial to seek a better represen- tation of the spatial variation in rainfall and its effect on streamflow response, and in improving the structure of real-time forecasting models than to expend effort in solv- ing estimation problems. Information on where efforts will be best rewarded can only be obtained by feedback from case studies.’’ Indeed, distributed modelling and use of ra- dar rainfall have been key topics in hydrologic research in the 1990s (e.g., Grayson and Blo ¨schl, 2000). In the mean 0022-1694/$ - see front matter ª 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2008.05.020 * Corresponding author. Tel.: +43 1 58801 22316; fax: +43 1 58801 22399. E-mail address: komma@hydro.tuwien.ac.at (J. Komma). Journal of Hydrology (2008) 357, 228– 242 available at www.sciencedirect.com journal homepage: www.elsevier.com/locate/jhydrol