Probabilistic Flood Forecasting with a Limited-Area Ensemble Prediction System: Selected Case Studies

M. Verbunt Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Search for other papers by M. Verbunt in
Current site
Google Scholar
PubMed
Close
,
A. Walser Swiss Federal Office of Meteorology and Climatology, MeteoSwiss, Zurich, Switzerland

Search for other papers by A. Walser in
Current site
Google Scholar
PubMed
Close
,
J. Gurtz Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Search for other papers by J. Gurtz in
Current site
Google Scholar
PubMed
Close
,
A. Montani Agenzia Regionale Prevenzione e Ambiente dell’Emilia Romagna-Servizio IdroMeterologico, Bologna, Italy

Search for other papers by A. Montani in
Current site
Google Scholar
PubMed
Close
, and
C. Schär Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Search for other papers by C. Schär in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

A high-resolution atmospheric ensemble forecasting system is coupled to a hydrologic model to investigate probabilistic runoff forecasts for the alpine tributaries of the Rhine River basin (34 550 km2). Five-day ensemble forecasts consisting of 51 members, generated with the global ensemble prediction system (EPS) of the European Centre for Medium-Range Weather Forecasts (ECMWF), are downscaled with the limited-area model Lokal Modell (LM). The resulting limited-area ensemble prediction system (LEPS) uses a horizontal grid spacing of 10 km and provides one-hourly output for driving the distributed hydrologic model Precipitation–Runoff–Evapotranspiration–Hydrotope (PREVAH) hydrologic response unit (HRU) with a resolution of 500 × 500 m2 and a time step of 1 h. The hydrologic model component is calibrated for the river catchments considered, which are characterized by highly complex topography, for the period 1997–98 using surface observations, and validated for 1999–2002.

This study explores the feasibility of atmospheric ensemble predictions for runoff forecasting, in comparison with deterministic atmospheric forcing. Detailed analysis is presented for two case studies: the spring 1999 flood event affecting central Europe due to a combination of snowmelt and heavy precipitation, and the November 2002 flood in the Alpine Rhine catchment. For both cases, the deterministic simulations yield forecast failures, while the coupled atmospheric–hydrologic EPS provides appropriate probabilistic forecast guidance with early indications for extreme floods. It is further shown that probabilistic runoff forecasts using a subsample of EPS members, selected by a cluster analysis, properly represent the forecasts using all 51 EPS members, while forecasts from randomly chosen subsamples reveal a reduced spread compared to the representative members. Additional analyses show that the representation of horizontal advection of precipitation in the atmospheric model may be crucial for flood forecasts in alpine catchments.

Corresponding author address: Christoph Schär, Institute for Atmospheric and Climate Science, ETH Zurich, Universitatsstrasse 16, CH-8092 Zurich, Switzerland. Email: schaer@env.ethz.ch

Abstract

A high-resolution atmospheric ensemble forecasting system is coupled to a hydrologic model to investigate probabilistic runoff forecasts for the alpine tributaries of the Rhine River basin (34 550 km2). Five-day ensemble forecasts consisting of 51 members, generated with the global ensemble prediction system (EPS) of the European Centre for Medium-Range Weather Forecasts (ECMWF), are downscaled with the limited-area model Lokal Modell (LM). The resulting limited-area ensemble prediction system (LEPS) uses a horizontal grid spacing of 10 km and provides one-hourly output for driving the distributed hydrologic model Precipitation–Runoff–Evapotranspiration–Hydrotope (PREVAH) hydrologic response unit (HRU) with a resolution of 500 × 500 m2 and a time step of 1 h. The hydrologic model component is calibrated for the river catchments considered, which are characterized by highly complex topography, for the period 1997–98 using surface observations, and validated for 1999–2002.

This study explores the feasibility of atmospheric ensemble predictions for runoff forecasting, in comparison with deterministic atmospheric forcing. Detailed analysis is presented for two case studies: the spring 1999 flood event affecting central Europe due to a combination of snowmelt and heavy precipitation, and the November 2002 flood in the Alpine Rhine catchment. For both cases, the deterministic simulations yield forecast failures, while the coupled atmospheric–hydrologic EPS provides appropriate probabilistic forecast guidance with early indications for extreme floods. It is further shown that probabilistic runoff forecasts using a subsample of EPS members, selected by a cluster analysis, properly represent the forecasts using all 51 EPS members, while forecasts from randomly chosen subsamples reveal a reduced spread compared to the representative members. Additional analyses show that the representation of horizontal advection of precipitation in the atmospheric model may be crucial for flood forecasts in alpine catchments.

Corresponding author address: Christoph Schär, Institute for Atmospheric and Climate Science, ETH Zurich, Universitatsstrasse 16, CH-8092 Zurich, Switzerland. Email: schaer@env.ethz.ch

1. Introduction

The Alpine region is exposed to a high frequency of heavy precipitation events (Frei et al. 2000). Due to the complex topography this region is particularly vulnerable to these events and its consequent effects like floods, landslides, and erosion, which endanger environment, inhabitants, and infrastructure. To mitigate the consequences of such events, it is therefore of utmost importance to produce reliable forecasts with sufficient lead time. Observed and anticipated changes in the frequency of intense precipitation events (e.g., Frei and Schär 2001) and the enhanced economic burdens due to flooding (Bronstert 2002) additionally illustrate the increasing need for appropriate atmospheric and hydrological forecasts.

Recent studies illustrate the potential to improve flood forecasts by coupling numerical weather prediction and hydrologic models (Benoit et al. 2003; Jasper and Kaufmann 2003). Bacchi and Ranzi (2003) concluded that the one-way forcing of hydrological models with mesoscale meteorological models can become a useful short-range (1–2 days) support for flood forecasting at the scale of medium-sized catchments (1000–10 000 km2). However, quantitative precipitation forecasts contain large uncertainties. Some fraction of this uncertainty arises as the atmosphere is a chaotic dynamical system, thus implying intrinsic predictability limitations. This kind of uncertainty would even occur with a perfect model and perfect initial conditions, due to the chaotic nature of the underlying dynamics that intrinsically limits predictability. In the last decade, much research has therefore been devoted to understand predictability and uncertainty in numerical weather forecasts (Palmer 2000) by using ensemble prediction systems (EPS; see Houtekamer et al. 1996; Molteni et al. 1996; Buizza et al. 1999a). Traditionally, this approach has addressed medium-range forecasts (3–10 days), but the need to account for predictability limitations at shorter lead times is becoming apparent. For instance, Walser and Schär (2004) found that even 24-h precipitation forecasts for medium-sized alpine river catchments may on occasion be critically affected by intrinsic predictability limitations, even when the synoptic-scale atmospheric flow is well predictable. Zhu et al. (2002) showed with a simple cost–loss model that for most users the ensemble forecasts offer a higher economic value than the deterministic forecast. Newly developed limited-area ensemble prediction systems (LEPSs) combine the benefits of a probabilistic approach with the high-resolution detail of limited-area model (LAM) integrations and yield considerable improvements compared to global EPSs, in particular concerning the quantitative forecast of intense precipitation and the geographical localization of the regions most likely to be affected by flood events (e.g., Marsigli et al. 2001; Frogner and Iversen 2002). Motivated by these results, hydrological initiatives for operational high-resolution EPSs have emerged recently (Montani et al. 2003; Quiby and Denhard 2003), some of which are coordinated within the International Hydrological Ensemble Prediction Experiment (HEPEX) initiative.

However, despite these developments in the numerical weather prediction (NWP), most current hydrologic forecasting systems still produce deterministic forecasts, providing a single estimate without quantifying its uncertainty. Runoff forecasting in Switzerland is still based on deterministic variables of the operational NWP model (FOEN 2005). Only very recently probabilistic approaches are used to quantify uncertainties of flood predictions (Ferraris et al. 2002; Siccardi et al. 2005). This study therefore explores the feasibility of a coupled atmospheric–hydrologic probabilistic system to quantify uncertainty in hydrological flood forecasting in the Alpine tributaries of the Rhine River basin. To account for the intermediate scale of the catchments under consideration, our approach will be based on a limited-area EPS. We will explore the potential of such a system for two flood events and also address aspects of the EPS design, such as the impact of ensemble size on the hydrological forecast spread.

The paper is structured as follows. Section 2 describes the models and the coupling strategy used. A short description of the meteorological situations under consideration is presented in section 3. Results are discussed in section 4, and section 5 presents the conclusions.

2. Methods

The area investigated is the Rhine River basin up to the gauge Rheinfelden (34 550 km2) in central Europe (Fig. 1). The catchment is characterized by complex topography with an altitude range from 262 up to 4225 m above MSL. The catchment has been subdivided into 23 subcatchments for the hydrologic calibration and validation because of its highly complex topography and the presence of lakes (Verbunt et al. 2005).

a. Global EPS

The global ensemble simulations in this study are based upon cycle 26r3 of the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecast System in the operational configuration with T255 horizontal resolution and 40 vertical levels. All 51 ensembles members initialized at 1200 UTC are used. The global ensembles provide initial and boundary conditions for the limited-area ensemble members.

b. Limited-area EPS

The LEPS used in the study is based on the Consortium for Small-scale Modeling (COSMO)-LEPS (Montani et al. 2003). This system has been run in a quasi-operational mode since November 2002 and was designed for the prediction of intense and localized weather events in the short to medium range (48–120 h). The system uses the nonhydrostatic Lokal Modell (LM; Steppeler et al. 2003) of the COSMO over a domain covering central and southern Europe (Fig. 1). The LM is applied on a rotated spherical grid with a grid spacing of 0.09° × 0.09°, equivalent to about 10 × 10 km2, and 32 vertical levels. COSMO-LEPS provides 5-day forecasts with an hourly output interval. In the most recent operational configuration (since June 2004), COSMO-LEPS employs a clustering-selection technique to select 10 EPS members out of a 102-member superensemble generated by joining together two successive EPS sets. The 10 selected EPS members provide both initial and boundary conditions to the LEPS integrations. For the current study, all 51 ensemble members from the most recent global ensemble are downscaled. While the global ensembles account for model uncertainties (by stochastic physics, see Buizza et al. 1999b) and uncertainties in the initial conditions (by singular vector perturbations), the limited-area model merely downscales the global ensemble members without introducing additional small-scale initial perturbations. The latter might be needed to account for growth processes acting on the smaller scales resolved by the LM. The lack of such perturbations might result in a too-small ensemble spread. The introduction of such perturbations for limited-area models is, however, beyond the scope of this paper and currently investigated in other studies (e.g., Hohenegger et al. 2006).

c. Hydrologic model

We use the distributed hydrologic model Precipitation– Runoff–Evapotranspiration–Hydrotype (PREVAH) hydrologic response unit (HRU). It has been developed with the intent of improving the understanding of the spatial and temporal variability of hydrological processes in catchments with complex topography (Gurtz et al. 1999). The spatial discretization of PREVAH relies on the aggregation of gridded spatial information into so-called hydrologic HRUs (Ross et al. 1979). The runoff generation module is based on the conception of the Hydrologiska Byråns Vattenbalansmodell (HBV) model (Bergström and Forsman 1973; Lindström et al. 1997), adapted for spatially distributed applications. The model further contains modules to calculate evapotranspiration, snow- and glacier melt, and soil moisture. Snow- and glacier melt are calculated using a modified temperature-index approach, including potential direct clear sky solar radiation (Hock 1999). For full information on the model physics, structure, interpolation methods, and parameterizations, we refer to Gurtz et al. (2003) and Zappa et al. (2003).

To account for the high heterogeneity in topography, land use, soil characteristics, and microclimatic processes, the hydrologic model is applied with a spatial resolution of 500 × 500 m2 and an hourly time step. The model has successfully been calibrated (period 1997–98) and validated (period 1999–2002) regarding its ability to represent the discharge hydrographs of different runoff regimes and flood peaks (Verbunt et al. 2005). When driven by observations, a set of 656 stations (52 with hourly resolution, 56 which measure twice a day, and 548 rain gauges with daily resolution) were used. Results will be presented for selected catchments (Fig. 1).

d. Coupling strategy

Output variables from the 5-day LEPS forecasts are used as input for the hydrologic model. Unless stated otherwise, this applies to all needed meteorological surface variables, that is, precipitation, 2-m temperature, global radiation, 2-m wind speed, 2-m air humidity, and sunshine duration (derived from cloud cover). Precipitation from the ensemble members is interpolated bilinearly to the hydrological (500 × 500 m2) grids, while for temperature, wind speed, and radiation an altitude-dependent regression function, based on these 500 × 500 m2 grids, is calculated for each hour. Sunshine duration and air humidity are interpolated using an inverse distance weighting method. Since the systematic errors of the LM show complex seasonal and spatial dependencies (Kaufmann 2004; Schubiger 2004) and are quantitatively unknown for the configuration employed, no bias corrections have been applied. To obtain reliable initial conditions for the 5-day hydrological forecasts, the hydrologic model has been driven by surface observations for the two years prior to the events considered.

3. Investigated cases

The first week of May 1999 was warmer than climatology, causing enhanced snowmelt, saturated soils, and high lake levels. By 1200 UTC 12 May a pronounced westerly flow was established over Europe, associated with a jet streak reaching from the Gulf of Biskaya to Hungary (Fig. 2a). At 1200 UTC, a cold front progressing from the northwest to the southeast reached the northern edge of the Alps. The most pronounced lifting and induced precipitation occurred over the foothills and northern Alpine slopes in Switzerland. Precipitation persisted from 11 to 15 May, leading to additional snowmelt and large-scale flooding in the Rhine basin, mainly upstream of Rheinfelden. The event was the first in a series of events affecting northern Switzerland, eastern Austria, and southern Germany. Together these caused several casualties and an estimated property damage of U.S. $750 million.

The second investigated event is a flood event on 15 November 2002. Ahead of a deep upper-level through, which stretched from Algeria to the North Sea, a strong southwesterly flow was established over the Alps (Fig. 2b). Pronounced damage associated with several major mudslides occurred in the Rhine/Domat-Ems basin in southeastern Switzerland (catchment 1 in Fig. 1). This catchment is located in an inner-Alpine valley and is protected to the south and partly to the north by mountain chains. Thus, the precipitation leading to the event was carried over the mountains along the Swiss–Italian border (typical pass to peak altitude is 2000 to 3500 m) as spill-over precipitation. The event was associated with exceptionally high precipitation amounts (surface observation of 386 mm in 48 h at station Hinterrhein, as measured by the network of the Federal Office of Meteorology and Climatology) and is estimated to have a return period of more than 500 yr.

4. Results and discussion

This section describes the results of the coupled hydrologic–LEPS forecasting system for the two selected case studies. It illustrates the advantages of probabilistic forecasts compared to deterministic ones, discusses the hydrologic forecast uncertainties, and demonstrates the performance of clustering methods.

a. Impact of precipitation and temperature uncertainty on runoff forecasts

We begin by discussing the driving data and the hydrological model simulations. Figure 3a illustrates the predicted 5-day cumulative precipitation for the May 1999 case study for the Ill catchment. The observed cumulative precipitation lies close to the median of the ensemble, but the spread of the 51 members is considerable. Figure 3c shows the ensemble runoff prediction driven by LEPS forecasts. The uncertainties in the predicted precipitation lead to a large spread in the flood predictions with a maximum runoff of up to 400% of the minimum runoff. The other two panels explore the sensitivity of temperatures on runoff. To this end, the hydrological model was driven by LEPS data for precipitation, but observations for all other variables. For the case under consideration, uncertainties in precipitation amounts are the prime reason for the spread in the hydrological forecast (Figs. 3c,d), despite considerable uncertainties in the temperature forecast (Fig. 3b) and hence in the snow line and snowmelt. Although the predicted 2-m temperature forecasts differ up to 8 K at the end of the 120-h forecasting period, thereby causing small deviations in snowmelt rates, their effect on runoff generation is small in this event since catchment-averaged temperatures lay mostly well above 0°C. The other meteorological input variables (wind speed, air humidity, radiation, and sunshine duration) play a subordinate role in the flood generation algorithm.

In contrast to the spring 1999 event, the November 2002 case study demonstrates that uncertainties in 2-m temperature can lead to large deviations in runoff forecasts (Fig. 4). Although precipitation is again the main source of runoff uncertainty, the two runoff peaks are clearly smaller with the full forecast mode (Fig. 4c) compared to the forecast driven by combined LEPS precipitation and other observed variables (Fig. 4d). This systematic difference can be explained by a systematic underestimation of nighttime surface temperatures (Fig. 4b), which leads to an overestimation of solid, at the expense of liquid, precipitation. Toward the end of the 5-day forecast, some members show a considerable overestimation in runoff in the full forecast mode prediction. This is caused by an overestimation of temperature.

In summary, our simulations confirm that quantitative precipitation and temperature are the key input data for runoff forecasting as, for example, reported by Benoit et al. (2003). Even small differences in 2-m temperatures can substantially affect runoff formation when temperature is close to 0°C. Such forecasts errors arise from uncertainties in the initial conditions and model errors. The results from the two case studies indicate that the runoff ensembles are able to represent the forecast uncertainty to a large extent even though the limited-area ensemble system does not account for model uncertainties.

b. Comparison between deterministic and probabilistic hydrological forecasts

Probabilistic flood forecasts have clear advantages compared to the deterministic forecast for the spring 1999 flood event (Fig. 5). The better performance of probabilistic forecasts compared to one single deterministic simulation was already shown by Atger (2001). While the deterministic control run (initialization at 1200 UTC 9 May) clearly underestimates the observed runoff peak, some of the ensembles members correctly predict the flood peak. Although the ensemble median is close to the deterministic run, a considerable number of ensemble members exceeds the highest-ever measured May discharge at this gauge, indicating a certain probability of an extremely rare event. The runoff ensemble with a reduced lead time (Fig. 5) shows a completely different behavior: The deterministic control run overestimates the flood peak while the ensemble median shows very good agreement with the observed runoff, both in runoff quantity and timing of the flood peak. According to this forecast, the probability to exceed the highest-ever measured May discharge is almost 100%. The change in the deterministic forecast between the two lead times is typical for a chaotic dynamical system, where minor changes in initial conditions may drastically alter the forecast.

This example illustrates the usefulness of probabilistic runoff forecasts. The early indications of an exceptional flood event, as provided by the probabilistic runoff forecasts, are not matched by the deterministic run. For both lead times, the deterministic runs yield considerable forecast failures. The coupled hydrologic–LEPS system provides proper forecast guidance despite large uncertainty. A peculiar feature of the two probabilistic forecasts is the increase in spread with decreasing lead time.

c. Impact of horizontal advection of precipitation in the NWP model

Outcomes from hydrologic models strongly depend on atmospheric inputs such as precipitation. Consequently, biases in NWP models can substantially influence flood prediction. During the course of this study, a newly developed prognostic precipitation scheme became available for the LM. In the old scheme, the fallout of hydrometeors was diagnosed assuming vertical trajectories. In the new prognostic precipitation scheme, the two hydrometeor classes (rain and snow) are advected by the three-dimensional wind field while falling at precipitation-type-dependent fall velocities (see Baldauf and Schulz 2004). This subsection compares the precipitation and runoff forecasts driven by LEPS members using the two precipitation schemes for the November 2002 event.

The associated impact of the prognostic precipitation scheme on forecasted precipitation is shown in Fig. 6. Using the old diagnostic scheme, the LM members produce heavy precipitation volumes on the southern side of the Alps, about 20 km to the south of the Rhine/Domat-Ems catchment. Precipitation in the Rhine/Domat-Ems catchment was thus underestimated, and none of the ensemble members detected the flood event (Fig. 7a). With the new prognostic precipitation scheme, substantially higher precipitation amounts reach the catchments on the northern side of the Alps, resulting in much better hydrologic ensemble forecast (Fig. 7). However, the second flood peak is captured only by a few ensemble members, which could be explained by the lead time of more than 100 h, that may limit predictability in this region (see also Fig. 5). The importance of horizontal advection of precipitation is also illustrated by Baldauf and Schulz (2004), who showed that the spatial distribution of precipitation is in much better agreement with observations when the prognostic precipitation scheme is considered.

The diagnostic precipitation scheme largely underestimates spill-over precipitation, which was a dominant contribution to the flood event under consideration. This result suggests that the consideration of horizontal advection of precipitation in NWP models is crucial for reliable flood forecasts, especially in catchments with pronounced topography.

d. Quantification of forecast uncertainty

A thorough validation of a probabilistic forecast does require a large set of cases and the utilization of probabilistic skill scores (e.g., Nurmi 2003). A few cases studies (as currently available to us) cannot be properly validated. However, coupled atmospheric–hydrologic LEPS forecasts enable one to establish geographical maps of probabilities, for instance for the event that a certain flooding threshold (e.g., specified by a return period) is exceeded. The geographical distribution of the early warning signal can then be compared against the observed flood events.

The use of a hydrologic–atmospheric EPS offers a multitude of decision options, based on detailed probability distributions instead of a single deterministic forecast, which does not provide any information about forecast uncertainty. For individual catchments, the information is best displayed as a plume diagram (Fig. 8). The examples shown illustrate that the coupled atmospheric–hydrologic LEPS would have provided early warnings for extreme runoffs, except for the first flood peak at Rheinfelden (see also Fig. 9). The representative members shown will be discussed in the next subsections, which deal with the clustering methodology.

Figure 9 shows results for the May 1999 flood initialized at 1200 UTC 10 May, considering all catchments represented by the hydrologic model. The selected threshold corresponds to 80% of the previously observed record flood event. When deriving probabilities from the distribution of the 51 ensemble members, we assume that all members have the same probability. Results (Fig. 9a) show that for a band along the Alpine topography (from the Aare/Thun, via Reuss/Mellingen, Limmat/Zurich to Rhine/Diepoldsau and Rhine/Neuhausen: see Fig. 1 for catchment names) the ensemble indicates very high probabilities of exceeding these thresholds. For other catchments to the southeast and also for the Emme/Wiler, much lower probabilities for an extreme flood event are predicted. Qualitatively, this pattern of strong floods is confirmed by the observations (Fig. 9b). The coupled modeling system, however, yields low probabilities in the Thur/Andelfingen (12%) and lower Aare catchment (20%) compared to the observed runoff values (Fig. 9b). As a result, the runoff peak at the catchment outlet Rhine/Rheinfelden was not captured (see also Fig. 8).

The use of ensemble forecasts enables one to quantify spatially distributed probabilities for the exceedance of a certain threshold. This is a useful tool in early warning systems, which is about to become popular in weather prediction (Lalaurette 2003; Marsigli et al. 2004; Montani et al. 2003), but is still rarely used for hydrological flood forecasting.

e. Impact of ensemble size on probabilistic runoff forecasts

In an operational environment, a hydrological ensemble forecast with 51 members is hardly feasible due to the large amount of required computer resources. To run such an atmospheric–hydrologic ensemble system in an operational setup, an ensemble reduction technique that selects a few representative members of the global ensemble is necessary. It is, however, not clear to which extent a reduction in ensemble size may induce a loss of information about the spectrum of possible hydrologic forecasts and consequently the performance of the flood predictions. In this subsection, the clustering method provided by Molteni et al. (2001) is applied with 5 and 10 representative members (RMs), respectively. This approach provides a probability for each of the representative ensemble members. Results are compared against the full ensemble and a randomly reduced ensemble. Note that the quantiles of the weighted RM EPS have been computed after creating 51 members, according to their corresponding weighting factors, derived from the cluster populations.

Figure 8 illustrates that the 10 RMs cover the spread of the global ensemble in good agreement. It satisfactorily covers for both case studies and most catchments the same quantile range as the full 51-member ensemble. Figure 10 compares the results in terms of the 20% and 80% quantiles (q20 and q80). The four columns show the full EPS, the RM EPS without and with weighted LEPS members, and results for a randomly sampled EPS of the same size, respectively. Gray and dashed lines show results for the 5- and 10-member ensembles. The randomization in the last column is repeated 100 times, and results are shown for the mean of the 100 realizations. The results show that, overall, the 5 and 10 RMs ensembles properly represent the q20–q80 spread of the total ensemble for the selected catchments. This is in agreement with Montani et al. (2003), who showed that the impact of ensemble size reduction, by using five representative members instead of all 51 members, does not considerably change probability maps for exceeding various precipitation amounts. Our results confirm that the clustering techniques do not reduce the q20–q80 spread compared to the complete ensemble but sometimes even increase the ensemble spread (e.g., Ill/Gisingen and Aare/Thun). A possible disadvantage of the weighted five RMs ensemble is illustrated by the Ill/Gisingen catchment. In this particular case, one representative member accounts for 29 members of the global ensemble, and the derived spread is strongly underestimated. For the weighted 10 RMs ensemble, which is used for the operational forecasts of COSMO-LEPS, such a constricted spread is less likely due to the larger ensemble size.

The comparison between the spread of the RM EPS and the randomly sampled EPS shows that the clustering technique yields a larger spread in all cases. This implies that the clustering technique performs well for the investigated case studies and area. Despite the small sample of cases, however, these results are very encouraging, indicating that the reduction of the ensemble size by clustering is an appropriate path toward reducing the ensemble size (and thus the computational load) without major loss in performance.

5. Summary and conclusions

The feasibility of a coupled hydrologic–atmospheric limited-area ensemble prediction system (LEPS) for flood prediction in catchments with complex topography has been investigated. The system uses 51 limited-area atmospheric forecasts to drive the hydrological model. Case studies were presented for the Alpine tributaries of the Rhine basin for two severe flood events. The main results of this study are the following.

Comparisons of deterministic and probabilistic flood forecasts show that the use of ensemble forecasts leads to more reliable forecast guidance compared to a single deterministic run. For the investigated cases, the coupled ensemble system provides comparatively wide probability distributions that highlight the possibility of severe flooding, in cases where the corresponding deterministic runs missed the flood events.

The coupled hydrologic–atmospheric LEPS is able to quantify forecast uncertainties. In a single catchment, they are best displayed as plume or quantile diagrams. In more complex areas, the results can be used to estimate spatially distributed probability maps for the exceedance of certain runoff thresholds. It is, therefore, beneficial to exploit the available operational COSMO-LEPS members in operational flood prediction. The availability of multiple decision tools, like quantile ranges and probability maps, can support users in their decision-making processes. It should, however, be kept in mind that long-term precipitation and runoff forecasting remains a difficult task. Results from this study indicate good forecast guidance up to 48-h lead time, which provides sufficient time for appropriate measures in many catchments and floodplains.

The results of this study also show the need for reliable 2-m temperature forecasts when making flood forecasts in alpine regions, where the accurate prediction of the 0° line is of utmost importance. Small biases of the atmospheric system can considerably impact upon the performance of flood predictions. Probabilistic estimates of 2-m temperature forecasts are useful in flood forecasting because they not only cover the uncertainty in the snow line but also the underlying covariance with uncertainties in precipitation forecasts.

We have also assessed the potential of a clustering methodology that selects 5 or 10 representative members from a larger global ensemble in order to reduce the computational cost of the method. It has been shown that the clustering yields a larger forecast spread than randomly sampled members, indicating that the clustering method performs well in the selected catchments.

In one case we have also investigated the role of the wind-driven horizontal transport of falling precipitation in the atmospheric model component. Even at a resolution of 10 km, as employed in the current study, precipitation can be shifted considerably, in particular in the case of snowfall. Neglecting this effect in alpine areas with complex topography may erroneously shift precipitation into a neighboring catchment, where it will contribute to runoff generation in the wrong tributary. Accounting for wind drift of falling precipitation is thus a useful approach toward improving quantitative precipitation forecasts.

As our study was based on only two case studies, a comprehensive validation is thus not feasible. Nevertheless, the results suggest that atmospheric limited-area ensemble prediction systems are a highly promising tool for the hydrological community. Exploiting this potential requires conducting ensemble forecast with the hydrologic model. The coupled system enables new probabilistic methodologies to evaluate and mitigate consequences of extreme flooding events.

Acknowledgments

This research was supported by the National Centres of Competence in Research (NCCR)-Climate and the Swiss National Science Foundation. We are also grateful to the Consortium for Small-scale Modeling (COSMO) for providing the atmospheric data.

REFERENCES

  • Atger, F., 2001: Verification of intense precipitation forecasts from single models and ensemble prediction systems. Nonlinear Processes Geophys., 8 , 401417.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bacchi, B., and Ranzi R. , 2003: Hydrological and meteorological aspects of floods in the Alps: An overview. Hydrol. Earth Syst. Sci., 7 , 785798.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Baldauf, M., and Schulz J. , 2004: Prognostic precipitation in the Lokal Modell (LM) of DWD. COSMO Newsletter, No. 4, Deutscher Wetterdienst, 177–180.

  • Benoit, R., Kouwen N. , Yu W. , Chamberland S. , and Pellerin P. , 2003: Hydrometeorological aspects of the real-time ultrafinescale forecast support during the Special Observing Period of the MAP. Hydrol. Earth Syst. Sci., 7 , 877889.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bergström, S., and Forsman A. , 1973: Development of a conceptual deterministic rainfall–runoff model. Nord. Hydrol., 4 , 147170.

  • Bronstert, A., 2002: Special issue “Advances in flood research”: Preface. J. Hydrol., 267 , 1.

  • Buizza, R., Hollingsworth A. , Lalaurette E. , and Ghelli A. , 1999a: Probabilistic predictions of precipitation using the ECMWF ensemble prediction system. Wea. Forecasting, 14 , 168189.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Buizza, R., Miller M. , and Palmer T. , 1999b: Stochastic representation of model uncertainties in the ECMWF ensemble prediction system. Quart. J. Roy. Meteor. Soc., 125 , 28872908.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ferraris, L., Rudari R. , and Siccardi F. , 2002: The uncertainty in the prediction of flash floods in the northern Mediterranean environment. J. Hydrometeor., 3 , 714727.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • FOEN, cited. 2005: Hydrologische Grundlagen und Daten: Vorhersage und Alarmierung. [Available online at http://www.bafu.admin.ch/hydrologie/01833/.].

  • Frei, C., and Schär C. , 2001: Detection probability of trends in rare events: Theory and application to heavy precipitation in the Alpine region. J. Climate, 14 , 15681584.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Frei, C., Davies H. , Gurtz J. , and Schar C. , 2000: Climate dynamics and extreme precipitation and flood events in Central Europe. Integr. Assess., 1 , 281299.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Frogner, I., and Iversen T. , 2002: High-resolution limited-area ensemble predictions based on low-resolution targeted singular vectors. Quart. J. Roy. Meteor. Soc., 128 , 13211342.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gurtz, J., Baltensweiler A. , and Lang H. , 1999: Spatially distributed hydrotope-based modelling of evapotranspiration and runoff in mountainous basins. Hydrol. Processes, 13 , 27512768.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gurtz, J., Zappa M. , Jasper K. , Lang H. , Verbunt M. , Badoux A. , and Vitvar T. , 2003: A comparative study in modelling runoff and its components in two mountainous catchments. Hydrol. Processes, 17 , 297311.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hock, R., 1999: Distributed temperature-index ice- and snowmelt model including potential direct solar radiation. J. Glaciol., 45 , 101111.

  • Hohenegger, C., Lüthi D. , and Schär C. , 2006: Predictability mysteries in cloud-resolving models. Mon. Wea. Rev., 134 , 20952107.

  • Houtekamer, P., Lefaivre L. , Derome J. , Ritchie J. , and Houtekamer H. , 1996: A system simulation approach to ensemble prediction. Mon. Wea. Rev., 124 , 12251242.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Jasper, K., and Kaufmann P. , 2003: Coupled runoff as validation tools for atmospheric models at the regional scale. Quart. J. Roy. Meteor. Soc., 129 , 673693.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kaufmann, P., 2004: Verification of aLMo runs with European SYNOP and GPS data. COSMO Newsletter, No. 4, Deutscher Wetterdienst, 67–71.

  • Kleinn, J., Frei C. , Gurtz J. , Lüthi D. , Vidale P. , and Schär C. , 2005: Hydrologic simulations in the Rhine basin driven by a regional climate model. J. Geophys. Res., 110 .D04102, doi:10.1029/2004JD005143.

    • Search Google Scholar
    • Export Citation
  • Lalaurette, F., 2003: Early detection of abnormal weather conditions using a probabilistic extreme forecast index. Quart. J. Roy. Meteor. Soc., 129 , 30373057.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lindström, G., Johansson B. , Persson M. , Gardelin M. , and Bergström S. , 1997: Development and test of the distributed HBV-96 hydrological model. J. Hydrol., 201 , 272288.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Marsigli, C., Montani A. , Nerozzi F. , Paccagnelli T. , Tibaldi S. , Molteni F. , and Buizza R. , 2001: A strategy for high-resolution ensemble prediction. Part II: Limited-area experiments in four Alpine flood events. Quart. J. Roy. Meteor. Soc., 127 , 20952115.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Marsigli, C., Montani A. , Nerozzi F. , and Paccagnelli T. , 2004: Probabilistic high-resolution forecast of heavy precipitation over central Europe. Nat. Hazards Earth Syst. Sci., 4 , 315322.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Molteni, F., Buizza R. , Palmer T. , and Petroliagis T. , 1996: The ECMWF ensemble prediction system: Methodology and validation. Quart. J. Roy. Meteor. Soc., 122 , 73119.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Molteni, F., Buizza R. , Marsigli C. , Montani A. , Nerozzi F. , and Paccagnelli T. , 2001: A strategy for high-resolution ensemble prediction. I: Definition of representative members and global-model experiments. Quart. J. Roy. Meteor. Soc., 127 , 20692094.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Montani, A., Marsigli C. , Nerozzi F. , Paccagnella T. , Tibaldi S. , and Buizza R. , 2003: The Soverato flood in southern Italy: Performance of global and limited-area ensemble forecasts. Nonlinear Processes Geophys., 10 , 261274.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Nurmi, P., 2003: Recommendations on the verification of local weather forecasts. ECMWF Tech. Memo. 430, Reading, United Kingdom, 19 pp.

  • Palmer, T., 2000: Predicting uncertainty in forecasts of weather and climate. Rep. Prog. Phys., 63 , 71116.

  • Quiby, J., and Denhard M. , 2003: SRNWP-DWD poor-man ensemble prediction system: The PEPS Project. Eumetnet Newsletter, No. 8, 9–12. [Available online at http://www.eumetnet.eu.org.].

  • Ross, B., Contractor D. , and Shanholtz V. , 1979: A finite-element model of overland and channel flow for assessing the hydrologic impact of land-use change. J. Hydrol., 41 , 1130.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schubiger, F., 2004: High resolution verification of daily cycle over Switzerland. COSMO Newsletter, No. 4, Deutscher Wetterdienst, 63–66.

  • Siccardi, F., Boni G. , Ferraris L. , and Rudari R. , 2005: A hydrometeorological approach for probabilistic flood forecast. J. Geophys. Res., 110 .D05101, doi:10.1029/2004JD005314.

    • Search Google Scholar
    • Export Citation
  • Steppeler, J., Doms G. , Schättler U. , Bitzer H. , Gassmann A. , Damrath U. , and Gregoric G. , 2003: Meso-gamma scale forecasts using the nonhydrostatic model LM. Meteor. Atmos. Phys., 82 , 7596.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Verbunt, M., Zappa M. , Gurtz J. , and Kaufmann P. , 2005: Verification of a coupled hydrometeorological modelling approach for Alpine tributaries in the Rhine basin. J. Hydrol., 324 , 224238.

    • Search Google Scholar
    • Export Citation
  • Walser, A., and Schär C. , 2004: Convection-resolving precipitation forecasting and its predictability in Alpine river catchments. J. Hydrol., 288 , 5773.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zappa, M., Pos F. , Strasser U. , Warmerdam P. , and Gurtz J. , 2003: Seasonal water balance of an Alpine catchment as evaluated by different methods for spatially distributed snowmelt modelling. Nord. Hydrol., 34 , 179202.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zhu, Y., Toth Z. , Wobus R. , Richardson D. , and Mylne K. , 2002: The economic value of ensemble-based weather forecasts. Bull. Amer. Meteor. Soc., 83 , 7383.

    • Crossref
    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The coupling strategy for the atmospheric–hydrologic EPS system and some selected catchments: 1: Rhine/Domat-Ems 2: Ill/Gisingen, 3: Rhine/Diepoldsau, 4: Rhine/Neuhausen, 5: Thur/Andelfingen, 6: Rhine/Rekingen, 7: Aare/Thun, 8: Aare/Hagneck, 9: Emme/Wiler, 10: Aare/Brugg, 11: Reuss/Luzern, 12: Aare/Untersiggenthal, 13: Rhine/Rheinfelden; globe from Kleinn et al. (2005).

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 2.
Fig. 2.

Overviews of the synoptic situation for the two case studies in terms of geopotential height (m) at 500 hPa at 1200 UTC from ECMWF analysis for (a) 12 May 1999 and (b) 15 Nov 2002.

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 3.
Fig. 3.

Evolution of ensemble members (gray lines) for the Ill/Gisingen catchment (1281 km2) during the spring 1999 flood: (a) catchment-averaged cumulated precipitation, (b) surface temperature, and (c) runoff. (d) The hydrological model is driven by forecasted precipitation, but observations for all other variables (in particular surface temperature). The solid black line shows observations; the dashed line the ensemble median.

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 4.
Fig. 4.

As in Fig. 3, but for the Rhine/Domat-Ems catchment (3229 km2) during the November 2002 flood.

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 5.
Fig. 5.

Comparison between deterministic and probabilistic runoff predictions for two initialization times for the Reuss catchment at gauge Luzern (2251 km2): (left) 1200 UTC 9 May and (right) 1200 UTC 10 May.

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 6.
Fig. 6.

Comparison between the precipitation fields of the ensemble member with maximum runoff in the run (left) using the old diagnostic scheme and the same member (right) using the prognostic precipitation scheme.

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 7.
Fig. 7.

The impact of prognostic precipitation on flood prediction in the Rhine/Domat-Ems catchment (3229 km2) during the November 2002 flood. The runoff forecast using (left) the old diagnostic scheme and (right) the newly developed prognostic scheme.

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 8.
Fig. 8.

Several interquantile ranges together with the 10 representative members for different catchment areas and different case studies: (a: top left) Aare/Hagneck (5128 km2), (b: top right) Reuss/Luzern (2251 km2), (c: bottom left) Rhine/Rheinfelden (34 550 km2), and (d: bottom right) Rhine/Domat-Ems (3229 km2).

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 9.
Fig. 9.

Spatially distributed probability map, (left) indicating the probability of exceeding the 80% level of the maximum measured May discharge for each investigated catchment for the spring 1999 flood event compared to (right) the ratio actual measured maximum/80% level of the (till 1999) maximum measured May discharge.

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Fig. 10.
Fig. 10.

The comparison of the 20%–80% quantile spread for the complete ensemble (1st column), the 5 unweighted representative members (2nd column), the 5 weighted representative members (3rd column), and the average of 100 × 5 random members (4th column) for the 1ll catchment (first row), the Reuss/Luzern (second row), the Aare/Thun catchment (third row), and the Rhine/Domat-Ems catchment (fourth row).

Citation: Journal of Hydrometeorology 8, 4; 10.1175/JHM594.1

Save
  • Atger, F., 2001: Verification of intense precipitation forecasts from single models and ensemble prediction systems. Nonlinear Processes Geophys., 8 , 401417.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bacchi, B., and Ranzi R. , 2003: Hydrological and meteorological aspects of floods in the Alps: An overview. Hydrol. Earth Syst. Sci., 7 , 785798.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Baldauf, M., and Schulz J. , 2004: Prognostic precipitation in the Lokal Modell (LM) of DWD. COSMO Newsletter, No. 4, Deutscher Wetterdienst, 177–180.

  • Benoit, R., Kouwen N. , Yu W. , Chamberland S. , and Pellerin P. , 2003: Hydrometeorological aspects of the real-time ultrafinescale forecast support during the Special Observing Period of the MAP. Hydrol. Earth Syst. Sci., 7 , 877889.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bergström, S., and Forsman A. , 1973: Development of a conceptual deterministic rainfall–runoff model. Nord. Hydrol., 4 , 147170.

  • Bronstert, A., 2002: Special issue “Advances in flood research”: Preface. J. Hydrol., 267 , 1.

  • Buizza, R., Hollingsworth A. , Lalaurette E. , and Ghelli A. , 1999a: Probabilistic predictions of precipitation using the ECMWF ensemble prediction system. Wea. Forecasting, 14 , 168189.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Buizza, R., Miller M. , and Palmer T. , 1999b: Stochastic representation of model uncertainties in the ECMWF ensemble prediction system. Quart. J. Roy. Meteor. Soc., 125 , 28872908.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ferraris, L., Rudari R. , and Siccardi F. , 2002: The uncertainty in the prediction of flash floods in the northern Mediterranean environment. J. Hydrometeor., 3 , 714727.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • FOEN, cited. 2005: Hydrologische Grundlagen und Daten: Vorhersage und Alarmierung. [Available online at http://www.bafu.admin.ch/hydrologie/01833/.].

  • Frei, C., and Schär C. , 2001: Detection probability of trends in rare events: Theory and application to heavy precipitation in the Alpine region. J. Climate, 14 , 15681584.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Frei, C., Davies H. , Gurtz J. , and Schar C. , 2000: Climate dynamics and extreme precipitation and flood events in Central Europe. Integr. Assess., 1 , 281299.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Frogner, I., and Iversen T. , 2002: High-resolution limited-area ensemble predictions based on low-resolution targeted singular vectors. Quart. J. Roy. Meteor. Soc., 128 , 13211342.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gurtz, J., Baltensweiler A. , and Lang H. , 1999: Spatially distributed hydrotope-based modelling of evapotranspiration and runoff in mountainous basins. Hydrol. Processes, 13 , 27512768.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gurtz, J., Zappa M. , Jasper K. , Lang H. , Verbunt M. , Badoux A. , and Vitvar T. , 2003: A comparative study in modelling runoff and its components in two mountainous catchments. Hydrol. Processes, 17 , 297311.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hock, R., 1999: Distributed temperature-index ice- and snowmelt model including potential direct solar radiation. J. Glaciol., 45 , 101111.

  • Hohenegger, C., Lüthi D. , and Schär C. , 2006: Predictability mysteries in cloud-resolving models. Mon. Wea. Rev., 134 , 20952107.

  • Houtekamer, P., Lefaivre L. , Derome J. , Ritchie J. , and Houtekamer H. , 1996: A system simulation approach to ensemble prediction. Mon. Wea. Rev., 124 , 12251242.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Jasper, K., and Kaufmann P. , 2003: Coupled runoff as validation tools for atmospheric models at the regional scale. Quart. J. Roy. Meteor. Soc., 129 , 673693.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kaufmann, P., 2004: Verification of aLMo runs with European SYNOP and GPS data. COSMO Newsletter, No. 4, Deutscher Wetterdienst, 67–71.

  • Kleinn, J., Frei C. , Gurtz J. , Lüthi D. , Vidale P. , and Schär C. , 2005: Hydrologic simulations in the Rhine basin driven by a regional climate model. J. Geophys. Res., 110 .D04102, doi:10.1029/2004JD005143.

    • Search Google Scholar
    • Export Citation
  • Lalaurette, F., 2003: Early detection of abnormal weather conditions using a probabilistic extreme forecast index. Quart. J. Roy. Meteor. Soc., 129 , 30373057.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lindström, G., Johansson B. , Persson M. , Gardelin M. , and Bergström S. , 1997: Development and test of the distributed HBV-96 hydrological model. J. Hydrol., 201 , 272288.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Marsigli, C., Montani A. , Nerozzi F. , Paccagnelli T. , Tibaldi S. , Molteni F. , and Buizza R. , 2001: A strategy for high-resolution ensemble prediction. Part II: Limited-area experiments in four Alpine flood events. Quart. J. Roy. Meteor. Soc., 127 , 20952115.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Marsigli, C., Montani A. , Nerozzi F. , and Paccagnelli T. , 2004: Probabilistic high-resolution forecast of heavy precipitation over central Europe. Nat. Hazards Earth Syst. Sci., 4 , 315322.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Molteni, F., Buizza R. , Palmer T. , and Petroliagis T. , 1996: The ECMWF ensemble prediction system: Methodology and validation. Quart. J. Roy. Meteor. Soc., 122 , 73119.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Molteni, F., Buizza R. , Marsigli C. , Montani A. , Nerozzi F. , and Paccagnelli T. , 2001: A strategy for high-resolution ensemble prediction. I: Definition of representative members and global-model experiments. Quart. J. Roy. Meteor. Soc., 127 , 20692094.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Montani, A., Marsigli C. , Nerozzi F. , Paccagnella T. , Tibaldi S. , and Buizza R. , 2003: The Soverato flood in southern Italy: Performance of global and limited-area ensemble forecasts. Nonlinear Processes Geophys., 10 , 261274.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Nurmi, P., 2003: Recommendations on the verification of local weather forecasts. ECMWF Tech. Memo. 430, Reading, United Kingdom, 19 pp.

  • Palmer, T., 2000: Predicting uncertainty in forecasts of weather and climate. Rep. Prog. Phys., 63 , 71116.

  • Quiby, J., and Denhard M. , 2003: SRNWP-DWD poor-man ensemble prediction system: The PEPS Project. Eumetnet Newsletter, No. 8, 9–12. [Available online at http://www.eumetnet.eu.org.].

  • Ross, B., Contractor D. , and Shanholtz V. , 1979: A finite-element model of overland and channel flow for assessing the hydrologic impact of land-use change. J. Hydrol., 41 , 1130.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Schubiger, F., 2004: High resolution verification of daily cycle over Switzerland. COSMO Newsletter, No. 4, Deutscher Wetterdienst, 63–66.

  • Siccardi, F., Boni G. , Ferraris L. , and Rudari R. , 2005: A hydrometeorological approach for probabilistic flood forecast. J. Geophys. Res., 110 .D05101, doi:10.1029/2004JD005314.

    • Search Google Scholar
    • Export Citation
  • Steppeler, J., Doms G. , Schättler U. , Bitzer H. , Gassmann A. , Damrath U. , and Gregoric G. , 2003: Meso-gamma scale forecasts using the nonhydrostatic model LM. Meteor. Atmos. Phys., 82 , 7596.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Verbunt, M., Zappa M. , Gurtz J. , and Kaufmann P. , 2005: Verification of a coupled hydrometeorological modelling approach for Alpine tributaries in the Rhine basin. J. Hydrol., 324 , 224238.

    • Search Google Scholar
    • Export Citation
  • Walser, A., and Schär C. , 2004: Convection-resolving precipitation forecasting and its predictability in Alpine river catchments. J. Hydrol., 288 , 5773.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zappa, M., Pos F. , Strasser U. , Warmerdam P. , and Gurtz J. , 2003: Seasonal water balance of an Alpine catchment as evaluated by different methods for spatially distributed snowmelt modelling. Nord. Hydrol., 34 , 179202.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Zhu, Y., Toth Z. , Wobus R. , Richardson D. , and Mylne K. , 2002: The economic value of ensemble-based weather forecasts. Bull. Amer. Meteor. Soc., 83 , 7383.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    The coupling strategy for the atmospheric–hydrologic EPS system and some selected catchments: 1: Rhine/Domat-Ems 2: Ill/Gisingen, 3: Rhine/Diepoldsau, 4: Rhine/Neuhausen, 5: Thur/Andelfingen, 6: Rhine/Rekingen, 7: Aare/Thun, 8: Aare/Hagneck, 9: Emme/Wiler, 10: Aare/Brugg, 11: Reuss/Luzern, 12: Aare/Untersiggenthal, 13: Rhine/Rheinfelden; globe from Kleinn et al. (2005).

  • Fig. 2.

    Overviews of the synoptic situation for the two case studies in terms of geopotential height (m) at 500 hPa at 1200 UTC from ECMWF analysis for (a) 12 May 1999 and (b) 15 Nov 2002.

  • Fig. 3.

    Evolution of ensemble members (gray lines) for the Ill/Gisingen catchment (1281 km2) during the spring 1999 flood: (a) catchment-averaged cumulated precipitation, (b) surface temperature, and (c) runoff. (d) The hydrological model is driven by forecasted precipitation, but observations for all other variables (in particular surface temperature). The solid black line shows observations; the dashed line the ensemble median.

  • Fig. 4.

    As in Fig. 3, but for the Rhine/Domat-Ems catchment (3229 km2) during the November 2002 flood.

  • Fig. 5.

    Comparison between deterministic and probabilistic runoff predictions for two initialization times for the Reuss catchment at gauge Luzern (2251 km2): (left) 1200 UTC 9 May and (right) 1200 UTC 10 May.

  • Fig. 6.

    Comparison between the precipitation fields of the ensemble member with maximum runoff in the run (left) using the old diagnostic scheme and the same member (right) using the prognostic precipitation scheme.

  • Fig. 7.

    The impact of prognostic precipitation on flood prediction in the Rhine/Domat-Ems catchment (3229 km2) during the November 2002 flood. The runoff forecast using (left) the old diagnostic scheme and (right) the newly developed prognostic scheme.

  • Fig. 8.

    Several interquantile ranges together with the 10 representative members for different catchment areas and different case studies: (a: top left) Aare/Hagneck (5128 km2), (b: top right) Reuss/Luzern (2251 km2), (c: bottom left) Rhine/Rheinfelden (34 550 km2), and (d: bottom right) Rhine/Domat-Ems (3229 km2).

  • Fig. 9.

    Spatially distributed probability map, (left) indicating the probability of exceeding the 80% level of the maximum measured May discharge for each investigated catchment for the spring 1999 flood event compared to (right) the ratio actual measured maximum/80% level of the (till 1999) maximum measured May discharge.

  • Fig. 10.

    The comparison of the 20%–80% quantile spread for the complete ensemble (1st column), the 5 unweighted representative members (2nd column), the 5 weighted representative members (3rd column), and the average of 100 × 5 random members (4th column) for the 1ll catchment (first row), the Reuss/Luzern (second row), the Aare/Thun catchment (third row), and the Rhine/Domat-Ems catchment (fourth row).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 572 172 12
PDF Downloads 334 61 9