H32E-01
Real-Time Operation Of A Multipurpose Multi-Reservoir System Using A Distributed Hydrological Model And Quantitative Precipitation Forecast
Taking advantage of a distributed hydrological model's capabilities such as capturing spatial heterogeneity, this study couples a physically based hydrological model with embedded dam network operation to a heuristic model for real-time operation. The input rainfall is a meso-scale quantitative precipitation forecast at 0.125 degrees resolution issued every 6 hours. It was analyzed 3 different series and the complete 18 hours lead-time. The system attempts to 1) reduce flood peaks down stream and 2) replenish water level at reservoirs after flood event. The proposed scheme takes advantage of the heuristic algorithm in order to evaluate different release combination sets automatically based on stochastic seeding considering the dam constraints and objective function. Latter is defined to minimize the absolute difference between the forecasted flood volume at protecting point and the total released volume from reservoirs. To estimate the flood volume a desirable discharge is to be set at protecting point. The desirable discharge is defined as the average of observed values exceeding the mean annual discharge; however, this can be modified according to flood warning levels and water resources management. The optimization variables are the release-inflow ratios. In addition, it was introduced the standard deviation of the error forecast as a weight in the objective function. The developed system was applied to upper Tone River in Japan using up to three multipurpose reservoirs. The efficiency of the system's response was evident reducing the flood peaks and volume at protecting point comparing the optimized releases against observed data. This approach has shown feasibility to be used by dam operators as a real-time reference tool for more efficient water resources management.
H32E-02
Artificial Neural Network Models for Long Lead Streamflow Forecasts using Climate Information
Information on season ahead stream flow forecasts is very beneficial for the operation and management of water supply systems. Daily streamflow conditions at any particular reservoir primarily depend on atmospheric and land surface conditions including the soil moisture and snow pack. On the other hand recent studies suggest that developing long lead streamflow forecasts (3 months ahead) typically depends on exogenous climatic conditions particularly Sea Surface Temperature conditions (SST) in the tropical oceans. Examples of some oceanic variables are El Nino Southern Oscillation (ENSO) and Pacific Decadal Oscillation (PDO). Identification of such conditions that influence the moisture transport into a given basin poses many challenges given the nonlinear dependency between the predictors (SST) and predictand (stream flows). In this study, we apply both linear and nonlinear dependency measures to identify the predictors that influence the winter flows into the Neuse basin. The predictor identification approach here adopted uses simple correlation coefficients to spearman rank correlation measures for detecting nonlinear dependency. All these dependency measures are employed with a lag 3 time series of the high flow season (January – February – March) using 75 years (1928-2002) of stream flows recorded in to the Falls Lake, Neuse River Basin. Developing streamflow forecasts contingent on these exogenous predictors will play an important role towards improved water supply planning and management. Recently, the soft computing techniques, such as artificial neural networks (ANNs) have provided an alternative method to solve complex problems efficiently. ANNs are data driven models which trains on the examples given to it. The ANNs functions as universal approximators and are non linear in nature. This paper presents a study aiming towards using climatic predictors for 3 month lead time streamflow forecast. ANN models representing the physical process of the system are developed between the identified predictors and the predictand. Predictors used are the scores of Principal Components Analysis (PCA). The models were tested and validated. The feed- forward multi-layer perceptron (MLP) type neural networks trained using the back-propagation algorithms are employed in the current study. The performance of the ANN-model forecasts are evaluated using various performance evaluation measures such as correlation coefficient, root mean square error (RMSE). The preliminary results shows that ANNs are efficient to forecast long lead time streamflows using climatic predictors.
H32E-03 INVITED
Verification: Connecting Research and Operations
A brief study of river forecasts issued by the National Weather Service raises the question "Have river forecasts improved over the past two decades?" Most hydrologic research is conducted with the assumption that improving the underlying science will improve forecasts. Given the variety of sources of uncertainty in a river forecast, this assumption may or may not be true. A tighter bond between hydrologic research and operations is necessary to ensure research output will be of value to forecast practitioners. For example, there is no well understood baseline of forecast skill against which new ideas can be compared objectively. Nor is there a well documented assessment of uncertainty sources in the forecasts to guide researchers. The National Weather Service has undertaken two important projects to help foster the connection between hydrologic research and operations. First they have initiated development of a Community Hydrologic Prediction System and second they have started to define standardized verification procedures for hydrologic forecasts. A review of the verification study and then descriptions of the two projects will be presented.
H32E-04
A Multi-Site Streamflow Forecast Framework: Application to the Upper Colorado River Basin
The multi-site streamflow forecast framework is a simple and parsimonious method for incorporating large-scale climate information into basin scale streamflow forecasts. The method is parsimonious because predictors need only be developed at one index gage, which is the sum of the seasonal flows at many spatial locations. In an application to the Upper Colorado River Basin (UCRB), multi-model ensemble (MME) forecasts were made of the seasonal (April-July) flows at the index gage. A K nearest-neighbor (KNN) nonparametric disaggregation technique is implemented which provides seasonal forecasts at four spatial locations and in turn monthly forecasts for the peak flow season (April-July). The predictions made in a retroactive forecast mode were comparable to the Colorado Basin River Forecast Center (CBRFC) predictions which are made using the Ensemble Streamflow Prediction (ESP) model. The earliest forecast of the ESP model is January 1 because of its heavy reliance on snowpack information. The multi-site framework provides skillful predictions as early as November 1 by its inclusion of large scale climate information such as geopotential height, zonal winds, meridional winds and sea surface temperature. It is possible that the ESP model and the multi-site framework could be combined in a Bayesian context that could incorporate professional judgment. http://cbracken.info/REU2007/REU.html
H32E-05 INVITED
The U.S. Bureau of Reclamation's use of Climate Information Products to support Reservoir Operations and Water Management
Climate forecast information plays an integral role in Reclamation's operation of surface water systems located throughout the western United States. These systems include over 300 reservoirs, 16000 miles of canals, and 245 million acre-feet of storage capacity. Combined, their operation leads to approximately $9 billion in annual agricultural benefits, enough energy to supply 6 million homes, 308 public recreation areas, and billions of dollars in avoided flood damages. Reclamation's use of climate information varies with decision application, which might be characterized by lead- time, application horizon, and reversibility. This presentation will provide an overview of Reclamation's short- to long-term climate-affected decisions, where and why climate information products are currently used, where products are desired but are absent, and where products are available but are not used for various reasons. The presentation will also highlight lessons learned from recent efforts to introduce new uses of climate forecast information in Reclamation decision processes (e.g., use of short-lead teleconnections to potentially support Spring season flood control management in the Pacific Northwest, use of CPC local 3-month temperature outlooks to support Summer-Autumn stream temperature management in California, and development of downscaled WCRP CMIP3 climate projections to support long-term system evaluations throughout Reclamation's service regions).
H32E-06
DROUGHT MONITORING AND FORECASTING FOR THE U.S. USING CLIMATE MODEL SEASONAL FORECAST
Drought is the most costly natural hazard to the U.S. economy. Drought preparation and mitigation require skillful predictions of drought on-set, development, and recovery. A model-based Drought Monitor and Prediction System (DMAPS) is presented, and it provides a real-time quantitative drought assessment and prediction capability for the U.S. Using the North America Land Data Assimilation System (NLDAS) realtime meteorological forcing and the Variable Infiltration Capacity (VIC) land surface model, DMAPS is capable of capturing the development of the recent severe droughts in the West and Southeast of the U.S. since the beginning of 2007. Using seasonal climate forecasts from NCEP's Climate Forecast System (CFS) as one input, DMAPS also successfully predicted the evolution of the droughts several months in advance. The realtime monitoring and prediction of drought using DMAPS provides invaluable information for drought preparation and drought impact assessment at national and local scales. The prediction element of the DMAPS is also tested and evaluated in a hindcast mode for selected historical U.S. droughts. In these hindcasts, the system uses information from multiple climate model forecasts. In the presentation, an evaluation of the predictive skill of DMAPS is presented that includes quantitative metrics that measure the severity, area, duration of the drought forecasts. http://hydrology.princeton.edu/forecast
H32E-07
Drought Research: Challenges in Meeting the Needs for Drought Information on the Canadian Prairies
Through the Drought Research Initiative (DRI), Canada is undertaking a focused 5-year science program to study techniques for characterizating drought; to better understand drought processes, and to develop methods for improving the prediction of drought. Drought research has generated a great deal of interest among those federal and provincial resource agencies on the Canadian prairies, which are sensitive to water stresses that occur during periods of drought. Consultations with these agencies have indicated that they have a range of information needs to support decisions related to mitigating the effects of on-going droughts or planning for future droughts. Furthermore, these information needs vary by sector and user. Over the past six months, through consultations with responsible agencies and knowledgeable managers, these needs have been more clearly articulated. This presentation is a summary of such requirements by resource sector. It explores how these needs vary with respect to location, scale, season, duration and intensity. The presentation also takes a preliminary look at how DRI could be augmented in order to deal with some of these existing needs and position itself to assist with future research-policy linkages.