Hydrological ensemble prediction systems - Wiley Online Library

16 downloads 176274 Views 78KB Size Report
verification procedures, but there is at present no clear 'best practice' in .... by a HEPS requires new tools to visualize such forecasts and ... Brown JD, Seo D-J.
HYDROLOGICAL PROCESSES Hydrol. Process. 27, 1–4 (2013) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/hyp.9679

Preface Hydrological ensemble prediction systems This Special Issue stems from the recent emergence of a distinct field of researchers and operational forecasters who predict river flow with hydrological ensemble prediction systems (HEPSs). These systems generate an ensemble of river flow forecasts for the same forecast period, providing a probabilistic assessment of future river flows, instead of just relying on a single prediction. Ensembles can be generated in a variety of ways, one of the most popular being the use of numerical weather prediction (NWP) forecasts from ensemble prediction system, which are then used as input to hydrological models (Cloke and Pappenberger, 2009; Cuo et al., 2011) to generate HEPS-based forecasts. This approach was the basis for the development of the European Flood Awareness System (EFAS) (Thielen et al., 2009). Examples of other approaches to ensemble forecasting include consideration of model parameter uncertainty and multimodel approaches, using climatology and initial conditions to generate an ensemble, using known error distributions from past hydrological forecasts to ‘dress’ current predictions, or considering spatial uncertainty in rainfall forecastradar blends. All of these approaches have one important thing in common: they require a clear understanding of how hydrological processes are represented in the modelling, to what extent the uncertainty in this representation can be quantified and how rivers and catchments integrate the meteorological forcing, land uses and landscapes in the rainfall–runoff transformation. The past decade has seen the operational flood forecasting community increasingly using HEPS for their forecasts (Cloke et al., 2009; Wetterhall et al., 2013), driven by the demonstration that HEPS-based forecasts add value and can increase warning lead times (e.g. Pappenberger et al., 2008; He et al., 2009). However, it remains clear that many more quantitative studies of HEPS are required (Cloke and Pappenberger, 2009). Significant challenges to a more widespread use of HEPS by operational forecasters remain. For example, HEPS must receive and process large amounts of data generated by ensemble prediction system weather forecasts (Thielen et al., 2008), and this may not be straightforward for several operational centres. Various techniques to pre-process and post-process forecasts are being explored alongside verification procedures, but there is at present no clear ‘best practice’ in HEPS forecasting. In addition, there are substantial difficulties in understanding how best to base flood warning decisions on probabilistic forecasts (Demeritt et al., 2010; Nobert et al., 2010, Ramos et al., 2010). In order to address these challenges, among others, there is a flurry of high-quality research being carried out, including both ‘blue skies’ science and ‘operational’ Copyright © 2012 John Wiley & Sons, Ltd.

driven model developments. The work presented in this special issue is aligned to the Hydrologic Ensemble Prediction Experiment (HEPEX) initiative, which brings together water professionals from all over the world who are interested in advancing hydrological ensemble predictions and their applications (www.hepex.org) (Thielen et al., 2008). Initiated in 2004, HEPEX has a thriving community of researchers and practitioners across the world, who have organized and attended several workshops and dedicated sessions at international conferences. The research presented in this special issue emerged from a unique concentrated series of four HEPEX-related meetings in 2011: (i) the HEPEX International workshop at UNESCO_IHE in June 2011 on Post-processing and Verification of hydrological ensemble predictions, (ii) the cross-disciplinary session HS4.3/AS4.13/NH1.12, ‘Towards practical applications in ensemble hydro-meteorological forecasting’ at the European Geophysical Union (EGU) General Assembly in Vienna, Austria in April 2011, (iii) the Annual EFAS user meeting in April 2011 and (iv) the international workshop on HEPS in Grenoble in April 2011 that brought together French and Canadian researchers and operational users of HEPS. This Special Issue contains 11 research papers and one scientific briefing, which showcase recent advances in HEPS research. They have been arranged for convenience into three thematic sections: (i) advances in evaluating and incorporating meteorological uncertainty, (ii) postprocessing ensemble forecasts and (iii) communicating uncertainty. In addition the HP-Today invited commentary (Pappenberger and Brown, 2013) discusses why the pursuit of perfection in flood forecasting may not be our best focus. 1. Incorporating meteorological uncertainty Radar data is often used for flash flood guidance and post-event analyses purposes in hydrological forecasting but could also be used to generate ensembles, which can be used to drive a hydrological model. Liechti et al. (2013) compared such method with the traditional approach of using NWP ensembles and an ensemble forecast system based on hydrological parameter uncertainty (this latter driven by interpolated precipitation). The analysis of these three ensemble systems for two nested flash flood prone basins in the southern Swiss Alps is complemented by two deterministic forecasting systems. These deterministic forecasting systems use deterministic hydrological models driven by interpolated precipitation and a single radar estimate of precipitation, respectively. The authors present a comprehensive analysis based on a

2

PREFACE

4-year hindcast period as well as individual events. The study demonstrates that the performance of the various methods largely depends on storm characteristics; however, the long-term analysis shows that in general, the ensemble forecasts provide better skill. Liguori and RicoRamirez (2013) assess the performance of hydrological predictions driven by probabilistic hybrid rainfall forecasts, created by merging NWP and radar-based forecasts. They provide a hydrological focus on forecast performance by determining the bias affecting ensemble and deterministic hydrological forecasts based on the probability of exceeding predefined thresholds of discharge at the outlet of the studied catchment. They find that assessing ensemble forecasts in this way can usefully provide and summarize information regarding the quality of the flow forecasts. In addition, it can provide a measure of the uncertainty to be expected from the forecasting process. For their case study, the ensemble forecasts were affected by a larger degree of underestimation than the deterministic forecasts. As HEPSs are becoming more popular in forecasting institutions around the world, simple techniques for assessing forecast uncertainty are welcome. Marty et al. (2013) analyse the performance of an integrated HEPS designed for mid-sized catchments prone to flash flooding. They find that hourly discharge forecasts are conditioned primarily by the accuracy of the meteorological ensemble forecasts at daily/subdaily timesteps. However, as the 6-hourly forecasts correctly reproduce the rainfall temporal dynamics and the daily analogy-based ensemble is less under-dispersive in terms of rainfall amounts, the merging of the two sources substantially increases the performance of the discharge forecasts. 2. Post-processing ensemble forecasts The HEPEX scientific community has shown the importance of post-processing outputs of NWP models, mainly precipitation and temperature, as these are the main meteorological input to hydrological models. Better quality in forecast input is a key component in improving hydrological forecast performance. Indeed, most hydrological models run on finer space and time scales than those currently provided by the resolutions of NWP models. Post-processing of meteorological forecasts (often called pre-processing from the hydrological model point of view) such as interpolation/downscaling and bias correction due to orographic effects are thus often required. Gaborit et al. (2013) compared the often used bilinear interpolation method with other stochastic downscaling methods of precipitation. The latter methods are spread-enhancing in order to compensate for the under-dispersiveness in ensemble forecast systems. A detailed probabilistic analysis of the downscaled results demonstrates that the overall quality of the forecasts are preserved by all post-processing methods but the variance enhancing methods clearly outperformed the bilinear interpolation in variance sensitive scores. The authors perform their case study on a summer storm event over the city of Quebec and point out that a longer comparison with radar data is desirable. Liu et al. (2013) evaluated the predictive skill of post-processed Copyright © 2012 John Wiley & Sons, Ltd.

NCEP GFS ensemble precipitation forecasts over the Huai river basin in China. They show that these forecasts can be a valuable resource for places other than the USA. The Ensemble Pre-Processing system version 3 (EPP3) is used to generate individual ensemble members that preserve the space–time correlation of the observational data. Results over the Huai river basin indicate that forecasts thus generated have meaningful predictive skill for the first few days for ensemble daily precipitation forecasts and for lead times up to 14 days for cumulative precipitation forecasts, although the influence of seasonality was significant. In any case, post-processed forecasts are shown to be skilful, performing better than raw or climatological forecasts. The HEPEX community also advocate the postprocessing of the hydrological forecasts themselves as an important alternative or complementary approach to post-processing the raw meteorological ensembles (e.g. Zalachori et al., 2012). In this Special issue, we present a variety of new methods. Raso et al. (2013) present a new methodology to generate a tree from an ensemble forecast known as ‘Information Flow Modelling’, which can be used in multistage stochastic programming for operational water management. Using ensembles directly in stochastic programming is known to overestimate the uncertainty as it does not consider its expected future resolution. Using a tree embeds the ensemble data in multistage stochastic programming intelligently, specifying the moments when uncertainties are resolved. The authors propose a new methodology in which the information flow to the controller is modelled implying the explicit definition of the observations available in the future and their degree of uncertainty. They present an application of their new method to the forecast of discharges of the Salzach river. Brown and Seo (2013) develop linear estimators for the conditional probability that the observed variable does not exceed several thresholds. With the method of Bayesian optimal linear estimation of indicator variables, a discrete approximation of the full conditional probability distribution is derived. The approach is analogous to the method of indicator cokriging (ICK) in geostatistics. This nonparametric technique is used for estimating the conditional probability distribution of the observed streamflow given a vector of predictors formed from a multimodel ensemble of streamflow simulations. In order to better account for the bias conditional upon the observation, the ICK is extended to the conditional-biaspenalized indicator cokriging method (CBP-ICK), and the weights of the predictors can be minimized by a combination of the error variance and the expected square conditional bias. The results show that the CBP-ICK produces unbiased and skilful estimates of the hydrologic uncertainties. Pagano et al. (2013) use ensemble dressing to incorporate uncertainty information on hydrological ensemble members in a HEPS. Empirical ensemble dressing techniques are particularly popular in operational streamflow forecasting systems certainly because of their simplicity and facility of application. In the study proposed by the authors, error distributions are assigned to the ensemble following bias correction and transformation of simulated Hydrol. Process. 27, 1–4 (2013)

3

PREFACE

and observed flow data. A probabilistic forecast is derived from the collective distribution of the dressed ensembles. The ability of the method to provide forecasts of good quality is demonstrated on catchments in SE Australia. The HEPEX led intercomparison experiment for postprocessing techniques for HEPS is reported on by van Andel et al. (2013). They describe the general design of the experiment and the available datasets. Initial results presented during the HEPEX workshop on post-processing and verification in June 2011 are discussed, and follow-up activities are introduced. The hope is to contribute to the fast improvement and applicability of HEPS post-processing techniques. As the authors note, readers are invited to join the post-processing intercomparison experiment (www.hepex.org). 3. Communicating uncertainty Decision making based on probabilistic forecasts issued by a HEPS requires new tools to visualize such forecasts and provide user-focused evaluation of ensemble forecasts. The study by Zappa et al. (2013) provides a novel tool to evaluate ensemble forecasts in the form of a ‘peak-box’ approach. The approach focuses on the magnitude and timing of peak discharges, which is of particular interest to flood management. The proposed method analyses the sharpness and the reliability in timing and magnitude of peak discharges by comparing observed and ensemble forecasts for five catchments in Switzerland. The authors show that this peak-box statistic could be visualized to support the interpretation and verification of operational HEPS. In the examples shown, it can be seen that peak discharge is usually overestimated by forecasts whereas discharge peak timings are captured well by forecasts. Developing visualization tools and forecast products for communicating uncertainty is a very important task if forecasts are to be useful for flood incident management. Pappenberger et al. (2013) provide a study of the perceptions of an expert group of different methods of visualizing probabilistic forecast information. They explore the pros and cons of existing visualization methods and find that experts do not agree on how forecasts should be visualized. The authors discuss future recommendations, including further training for forecasters and a suite of minimum properties that would make forecasts more easily understandable to users. Although much research is now being devoted to forecast visualization and communication techniques, only a few studies have been made so far in current HEPS research to understand how forecasts are perceived, understood and acted upon by those receiving them operationally. Demeritt et al. (2013) address this issue by providing a discussion of the communication, perception and use of EFAS alerts in operational flood management across Europe. They find that at present, although EFAS alerts are seen as useful and welcomed by flood forecasters, there is still hesitancy in responding to EFAS medium-term probabilistic alerts, and some institutional obstacles to be overcome if alerts are to fulfil their potential and lead to improvements in flood incident management. Copyright © 2012 John Wiley & Sons, Ltd.

REFERENCES van Andel SJ, Weerts A, Schaake J, Bogner K. 2013. Post-processing hydrological ensemble predictions intercomparison experiment. Hydrological Processes HEPS Special Issue 27: 158–161. Brown JD, Seo D-J. 2013. Evaluation of a nonparametric post-processor for bias correction and uncertainty estimation of hydrologic predictions. Hydrological Processes HEPS Special Issue 27: 83–105. Cloke HL, Pappenberger F. 2009. Ensemble flood forecasting: a review. Journal of Hydrology 375: 613–626. DOI: 10.1016/j.jhydrol.2009.06.005 Cloke HL, Thielen J, Pappenberger F, Nobert S, Salamon P, Buizza R, Bálint G, Edlund C, Koistinen A, de Saint-Aubin C, Viel C, Sprokkereef E. 2009. Progress in the implementation of Hydrological Ensemble Prediction Systems (HEPS) in Europe for operational flood forecasting. ECMWF Newsletter No. 121 – Autumn 2009. Cuo L, Pagano TC, Wang QJ. 2011. A review of quantitative precipitation forecasts and their use in short- to medium-range streamflow forecasting. Journal of Hydrometeor 12: 713–728. Demeritt D, Nobert S, Cloke HL, Pappenberger F. 2010. Challenges in communicating and using ensembles in operational flood forecasting. Meteorological Applications 17(2): 209–222. DOI: 10.1002/met.194 Demeritt D, Nobert S, Cloke HL, Pappenberger F. 2013. The European Flood Alert System and the communication, perception and use of ensemble predictions for operational flood risk management. Hydrological Processes HEPS Special Issue 27: 147–157. Gaborit E, Anctil F, Fortin V, Pelletier G. 2013. On the reliability of spatially disaggregated global ensemble rainfall forecasts. Hydrological Processes HEPS Special Issue 27: 45–56. He HY, Cloke HL, Wetterhall F, Pappenberger F, Freer J, Wilson M. 2009. Tracking the uncertainty in flood alerts driven by grand ensemble weather predictions, Meteorological Applications 16(1): 91–101. Liechti K, Zappa M, Fundel F, Germann U. 2013. Probabilistic evaluation of ensemble discharge nowcasts in two nested Alpine basins prone to flash floods. Hydrological Processes HEPS Special Issue 27: 5–17. Liguori S, Rico-Ramirez. 2013. A practical approach to the assessment of probabilitistic flow predictions. Hydrological Processes HEPS Special Issue 27: 18–32. Liu Y, Duan Q, Zhao L, Ye A, Tao Y, Miao C, Mu X, Schaake JC. 2013. Evaluating the predictive skill of post-processed NCEP GFS ensemble precipitation forecasts in China’s Huai river basin. Hydrological Processes HEPS Special Issue 27: 57–74. Marty R, Zin I, Obled Ch. 2013. Sensitivity of hydrological ensemble forecasts to different sources and temporal resolutions of probabilistic quantitiative precipitation forecasts: flash flood case studies in the Cevannes–Vivarais region (Southern France). Hydrological Processes HEPS Special Issue 27: 33–44. Nobert S, Demeritt D, Cloke HL. 2010. Informing operational flood management with ensemble predictions: lessons from Sweden. Journal of Flood Risk Management 3: 72–79. Pagano TC, Shrestha DL, Wang QJ, Robertson D, Hapuarachchi P. 2013. Ensemble dressing for hydrological applications. Hydrological Processes HEPS Special Issue 27: 106–116. Pappenberger F, Brown JD. 2013. HP today: on the pursuit of (im) perfection in flood forecasting. Hydrological Processes HEPS Special Issue 27: 162–163. Pappenberger F, Bartholmes J, Thielen J, Cloke HL, de Roo A, Buizza R. 2008. New dimensions in early flood warning across the globe using GRAND ensembles. Geophysical Research Letters 35(10): Art No. L10404. Pappenberger F, Stephens E, Thielen J, Salamon P, Demeritt D, van Andel SJ, Wetterhall F, Alfieri L. 2013. Visualizing probabilistic flood forecast information: expert preferences and perceptions of best practice in uncertainty communication. Hydrological Processes HEPS Special Issue 27: 132–146. Ramos MH, Mathevet T, Thielen J, Pappenberger F. 2010. Communicating uncertainty in hydro-meteorological forecasts: mission impossible? Meteorological Applications 17: 223–235. DOI: 10.1002/met.202 Raso L, van de Giesen N, Stive P, Schwanenberg D, van Overloop PJ. 2013. Tree structure generation from ensemble forecasts for real time control. Hydrological Processes HEPS Special Issue 27: 75–82. Thielen J, Schaake J, Hartman R, Buizza R. 2008. Aims, challenges and progress of the hydrological ensemble prediction experiment (HEPEX) following the third HEPEX workshop held in Stresa 27 to 29 June 2007. Atmospheric Science Letters 9: 29–35. Thielen J, Bartholmes J, Ramos M-H, de Roo A. 2009. The European Flood Alert System – Part 1: Concept and development. Hydrology and Earth System Science 13: 125–140. DOI: 10.5194/hess-13-125-2009 Wetterhall F, Pappenberger F, Cloke HL, Thielen-del Pozo J, Balabanova S, Da nhelka J, Vogelbacher A, Salamon P, Carrasco I, Cabrera-Tordera AJ, Hydrol. Process. 27, 1–4 (2013)

4

PREFACE

Corzo-Toscano M, Garcia-Padilla M, Garcia-Sanchez RJ, Ardilouze C, Jurela S, Terek B, Csik A, Casey J, Stankūnavicius G, Ceres V, Sprokkereef E, Stam J, Anghel E, Vladikovic D, Alionte Eklund C, Hjerdt N, Holmberg F, Nilsson J, Nyström K, Djerv H, Susnik M, Hazlinger M, Holubecka M. 2013. Forecasters priorities for improving probabilistic flood forecasts, Submitted to Hydrology and Earth System Sciences-Opinion December 2012. Zalachori I, Ramos M-H, Garçon R, Mathevet T, Gailhard J. 2012. Statistical processing of forecasts for hydrological ensemble prediction: a comparative study of different bias correction strategies. Advances in Science and Research 8: 135–141. Zappa M, Fundel F, Jaun S. 2013. A ‘Peak-Box’ approach for supporting interpretation and verification of operational ensemble peak-flow forecasts. Hydrological Processes HEPS special issue 27: 117–131.

Guest Editors Hannah L. Cloke Department of Geography and Environmental Science, Department of Meteorology, University of Reading, Whiteknights, Reading, RG6 6DW, UK

Copyright © 2012 John Wiley & Sons, Ltd.

Florian Pappenberger European Centre for Medium-range Weather Forecasts, Shinfield Park, Reading, RG2 9AX, UK Schalk Jan van Andel UNESCO-IHE Institute for Water Education, Delft, The Netherlands John Schaake Consultant, Annapolis, MD, USA Jutta Thielen Joint Research Centre of the European Commission, Ispra, Italy Maria-Helena Ramos Irstea/Cemagref, Hydrology Research Group, Antony, France

Hydrol. Process. 27, 1–4 (2013)