H52A-01 INVITED
Global uncertainty assessment in hydrological forecasting by means of statistical analysis of forecast errors
It is well known that uncertainty assessment in hydrological forecasting is a topical issue. Already in 1905 W.E. Cooke, who was issuing daily weather forecasts in Australia, stated: "It seems to me that the condition of confidence or otherwise form a very important part of the prediction, and ought to find expression". Uncertainty assessment in hydrology involves the analysis of multiple sources of error. The contribution of these latter to the formation of the global uncertainty cannot be quantified independently, unless (a) one is willing to introduce subjective assumptions about the nature of the individual error components or (2) independent observations are available for estimating input error, model error, parameter error and state error. An alternative approach, that is applied in this study and still requires the introduction of some assumptions, is to quantify the global hydrological uncertainty in an integrated way, without attempting to quantify each independent contribution. This methodology can be applied in situations characterized by limited data availability and therefore is gaining increasing attention by end users. This work aims to propose a statistically based approach for assessing the global uncertainty in hydrological forecasting, by building a statistical model for the forecast error xt,d, where t is the forecast time and d is the lead time. Accordingly, the probability distribution of xt,d is inferred through a non linear multiple regression, depending on an arbitrary number of selected conditioning variables. These include the current forecast issued by the hydrological model, the past forecast error and internal state variables of the model. The final goal is to indirectly relate the forecast error to the sources of uncertainty, through a probabilistic link with the conditioning variables. Any statistical model is based on assumptions whose fulfilment is to be checked in order to assure the validity of the underlying theory. Statistical testing for the proposed approach will be discussed in detail. Particular focus will be given to the hypothesis of stationarity of the forecast error, in view of the key role that stationarity plays in hydrological modeling in general. Applications are presented in validation mode, that refer to a rainfall-runoff model applied to an Italian river basin. http://www.costruzioni-idrauliche.ing.unibo.it/people/alberto/
H52A-02 INVITED
Complete representation of uncertainty in hydrological forecasts
Probabilistic hydrological forecasts are routinely produced by running a hydrological model up to the start of the forecast period with historical station data to estimate basin initial conditions, and then running the model into the future with an ensemble of weather and climate forecasts. The uncertainty in streamflow forecasts depends on uncertainties in estimates of basin initial conditions, uncertainty in the weather and climate forecasts, and uncertainties in the hydrological model itself. Typically, uncertainty in weather and climate forecasts is the only source of uncertainty considered in operational hydrological forecasts. This paper will describe attempts to provide a complete representation of uncertainty in hydrological forecasts. · We will describe methods to quantify uncertainties in basin initial conditions that stem from uncertainties in interpolated historical station data, and we will describe applications of ensemble data assimilation methods to reduce uncertainties in basin initial conditions. · We will discuss the application and limitations of new methods that we have developed to use multi-model ensembles to quantify uncertainty in hydrological models, and we will compare these methods with the simpler approach of stochastically perturbing model states. · We will describe methods to quantify uncertainty in weather and climate forecasts, and show how information from diverse sources can be integrated to provide probabilistic forecast products on time scales ranging from hours through to seasons. Finally, we will demonstrate how explicit quantification of all these sources of uncertainty is integrated in an operational hydrological forecasting system.
H52A-03
Using High Resolution Numerical Weather Prediction Models to Reduce and Estimate Uncertainty in Flood Forecasting
Forecast rainfall from Numerical Weather Prediction (NWP) and/or nowcasting systems is a major source of uncertainty for short-term flood forecasting. One approach for reducing and estimating this uncertainty is to use high resolution NWP models that should provide better rainfall predictions. The potential benefit of running the Met Office Unified Model (UM) with a grid spacing of 4 and 1 km compared to the current operational resolution of 12 km is assessed using the January 2005 Carlisle flood in northwest England. These NWP rainfall forecasts, and forecasts from the Nimrod nowcasting system, were fed into the lumped Probability Distributed Model (PDM) and the distributed Grid-to-Grid model to predict river flow at the outlets of two catchments important for flood warning. The results show the benefit of increased resolution in the UM, the benefit of coupling the high- resolution rainfall forecasts to hydrological models and the improvement in timeliness of flood warning that might have been possible. Ongoing work aims to employ these NWP rainfall forecasts in ensemble form as part of a procedure for estimating the uncertainty of flood forecasts.
H52A-04
Forecast uncertainty in semi-arid flash flood modeling using radar rain input
Flash floods are extremely dangerous hazards in the semi-arid southwest US at short temporal scales, posing a significant danger to life and property. Attempts to mitigate this flood risk using model-based forecasting are subject to uncertainties in the model and the data. This study reports on such an attempt using the distributed, semi-arid mechanistic rainfall-runoff model KINEROS2 driven by the NEXRAD WSR-88D DHR-based high resolution radar rainfall input. Sources of operational uncertainty considered in an integrated manner include rainfall estimates, model parameters, and initial conditions. Using a variance-based comprehensive global sensitivity analysis on both real and synthetic data from several events, the high predictive uncertainty in the modeled response was seen to be heavily dominated by operational event-specific biases in the radar rainfall depth estimates. Uncertainties in specific influential model parameters and initial conditions to be preferentially reduced were recognized, which show hillslopes to be more influential than channels on the outlet response in small basins. An inconsistency in behavioral/optimal model parameter set values was seen across events. This indicates the requirement of a computationally intensive Monte-Carlo setup that can incorporate such currently wide source uncertainty ranges with continuous incoming event information updating at local Weather Forecast Offices
H52A-05 INVITED
Progress toward the development and application of ensemble-based short-term hydrologic forecasts.
Water resource managers have benefited from the provision of reliable uncertainty information associated with long-range hydrologic forecasts for many years, particularly in the Western U.S. The development and application of short-term uncertainty information is considerably more complex, yet holds singnificant promise for several user sectors including emergency services, resources and environmental managers, flood control, and hydro power producers. Recognizing the benefits, National Weather Service has been working toward the development of ensemble- based short-term hydrologic forecasts for many years. Recently, efforts have intensified and refocused on a fieldable system for all NWS River Forecast Centers. Challenges, progress, and development plans will be presented.
H52A-06
Toward Integrative Uncertainty Accounting in Operational Hydrologic Ensemble Forecasting
Operational hydrologic forecasts are subject to large meteorological and hydrologic uncertainties, i.e., uncertainties in the hydrologic initial and boundary conditions, future boundary conditions, and observations. To produce reliable and skillful hydrologic ensemble forecasts, it is essential that both meteorological and hydrologic uncertainties are accurately accounted for. Toward that goal, NWS is developing a prototype hydrologic ensemble forecasting capability referred to as the eXperimental Ensemble Forecast System (XEFS) for operation at the NWS River Forecast Centers (RFC). It is envisioned that all or parts of this system may be shared with the research community for collaborative research and development toward improved operational hydrologic forecasting. In this talk, we describe the XEFS framework for integrative uncertainty accounting, identify key issues and share initial results.
H52A-07
ACCOUNTING FOR UNCERTAINTIES IN GENERATING RELIABLE PROBABILISTIC FLOOD FORECASTS FOR BANGLADESH
The country of Bangladesh experiences life-threatening floods in the basins of the Ganges and Brahmaputra rivers flowing through the country with tragic regularity. These floods result in loss of life on a scale that often greatly eclipses the deaths due to natural disasters in developed countries. Flooding in these basins can occur on weekly time scales (as occurred during the severe Brahmaputra floods of 2004 and of this year) to seasonal time scales (as occurred during the disastrous floods of 1998). Beginning in 2003, the Climate Forecasting Applications for Bangladesh (CFAB) project began issuing operational probabilistic flood forecasts to the country of Bangladesh over a wide-range of time scales to provide advanced warning of severe flood-stage discharges in the catchments of the Ganges and Brahmaputra basins. In this paper we discuss the uncertainty estimator module to our 1- to 10-day in-advance automated real-time operational multi-model flood forecast scheme for the upper basins of the Ganges and Brahmaputra rivers. These forecasts are based on an application of the European Centre for Medium-Range Weather Forecasts (ECMWF) 51-member ensemble weather forecasts, near-real-time GPCP and CMORPH satellite and NOAA CPC rain gauge precipitation estimates, and near-real- time discharge estimates from the Bangladesh Flood Forecasting and Warning Centre. The uncertainty estimator module estimates multi-model hydrologic error utilizing daily-updated hindcasts, which are separate from the forecasted weather variable uncertainty. Such a separation of error sources is done to maximize the sharpness of the final forecast probability distribution function (PDF), as well as to enhance the utility of the ensemble spread as an indicator of ensemble skill; for this latter feature of ensemble forecasts, we also present a new measure to test the spread-skill utility. In the final step of the uncertainty module, we merge these two sources of uncertainty together while at the same time providing an additional forecast error correction. This last step utilizes a relatively- unused statistical tool that ensures reliability in the PDF while ensuring skill no worse than a climatological forecast or persistence. http://cfab.eas.gatech.edu/shortterm/home.html