Informing operational £ood management with ensemble predictions: lessons from Sweden S. Nobert, D. Demeritt and H. Cloke Department of Geography, King’s College London, Strand, WC2R 2LS, London, UK Correspondence: S´ ebastien Nobert, Department of Geography, King’s College London, Strand, WC2R 2LS, London, UK Email: sebastien.nobert@kcl.ac.uk DOI:10.1111/j.1753-318X.2009.01056.x Key words civil protection; flood risk communication; ensemble forecasting; Europe; PREVIEW project; uncertainty. Abstract This paper highlights some communicative and institutional challenges to using ensemble prediction systems (EPS) in operational flood forecasting, warning, and civil protection. Focusing in particular on the Swedish experience, as part of the PREVIEW FP6 project, of applying EPS to operational flood forecasting, the paper draws on a wider set of site visits, interviews, and participant observation with flood forecasting centres and civil protection authorities (CPAs) in Sweden and 15 other European states to reflect on the comparative success of Sweden in enabling CPAs to make operational use of EPS for flood risk management. From that experience, the paper identifies four broader lessons for other countries interested in developing the operational capacity to make, communicate, and use EPS for flood forecasting and civil protection. We conclude that effective training and clear communication of EPS, while clearly necessary, are by no means sufficient to ensure effective use of EPS. Attention must also be given to overcoming the institutional obstacles to their use and to identifying operational choices for which EPS is seen to add value rather than uncertainty to operational decision making by CPAs. Introduction Uncertainty is a fundamental challenge in operational flood forecasting. With the twin aims of improving forecast skill, particularly over the medium term of 2–10 days ahead, and quantifying forecast uncertainty, hydrologists have begun to experiment with applying Ensemble Prediction Systems (EPS) to flood forecasting (e.g. Pappenberger et al., 2005; Bartholmes et al., 2009; He et al., 2009). EPS was first developed by meteorologists in the early 1990s to cope with the inevitable uncertainties in numerical weather prediction about initial conditions and parameterization of complex, often stochastic, atmospheric processes (Molteni et al., 1996). EPS is now well-established operationally among many of the leading weather forecasting agencies internationally, includ- ing the European Centre for Medium Range Weather Fore- casting (ECMWF), the US National Weather Service, and the UK Met Office, among others (Park et al., 2008). Compared with deterministic rainfall forecasts, EPS often exhibits great- er skill over the medium range (Richardson, 2000) and, crucially for flood forecasting applications where public safety concerns must be balanced against the dangers of crying wolf (Barnes et al., 2007; Demeritt et al., 2007), they ‘predict not only the most likely outcome but also the probability of occurrence of extreme and rare events’ (Buizza, 2008, p. 36). Building on the successful application of EPS to weather forecasting (Buizza et al., 2005) and to climate prediction (Collins and Knight, 2007), there is now a large and growing body of research exploring ways of using EPS to drive flood forecasting systems (see review by Cloke and Pappenberger, 2009; Cloke et al., 2009). Cloke and Pappenberger (2009) note a number of key technical challenges, including prop- erly capturing and cascading the full spectrum of uncertain- ties through the coupled meteorological–hydrological model, adequately representing extreme events for which, by definition, there is little data for validation, and develop- ing robust skill scoring techniques to measure the relation- ship between the statistical spread of the ensemble population and the phase space of the physical system being represented. Many of these scientific challenges to applying EPS to operational flood forecasting are being addressed internationally through Hydrological Ensemble Prediction Experiment, being conducted under the auspices of the World Meteorological Organisation (Thielen et al., 2008). But these outstanding scientific challenges are not the only reason that operational flood forecasters have lagged behind meteorologists in taking up EPS. There are also a number of institutional and communicative challenges to applying EPS to operational flood forecasting (cf. Morss et al., 2005; Demeritt et al., 2007; Faulkner et al., 2007). J Flood Risk Management (2010) c  2010 The Authors Journal Compilation c  2010 Blackwell Publishing Ltd