RESULTS and DISCUSSION

Slides:



Advertisements
Similar presentations
Drought Monitoring and Prediction Systems at the University of Washington and Princeton University Climate Diagnostics and Prediction Workshop Lincoln,
Advertisements

Alan F. Hamlet Andy Wood Dennis P. Lettenmaier JISAO Center for Science in the Earth System Climate Impacts Group and Department of Civil and Environmental.
Experimental Real-time Seasonal Hydrologic Forecasting Andrew Wood Dennis Lettenmaier University of Washington Arun Kumar NCEP/EMC/CMB presented: JISAO.
Optimized Flood Control in the Columbia River Basin for a Global Warming Scenario 1Dept. of Civil and Env. Engineering, UW 2CSES Climate Impacts Group,
Alan F. Hamlet Andy Wood Dennis P. Lettenmaier JISAO Center for Science in the Earth System Climate Impacts Group and Department of Civil and Environmental.
Alan F. Hamlet Andy Wood Seethu Babu Marketa McGuire Dennis P. Lettenmaier JISAO Climate Impacts Group and the Department of Civil Engineering University.
Seasonal outlooks for hydrology and water resources: streamflow, reservoir, and hydropower forecasts for the Pacific Northwest Andy Wood and Alan Hamlet.
Andy Wood, Ted Bohn, George Thomas, Ali Akanda, Dennis P. Lettenmaier University of Washington west-wide experimental hydrologic forecast system OBJECTIVE.
Effect of Model Calibration on Streamflow Forecast Results Ali Akanda, Andrew Wood, and Dennis Lettenmaier Civil and Environmental Engineering University.
Figure 1: Schematic representation of the VIC model. 2. Model description Hydrologic model The VIC macroscale hydrologic model [Liang et al., 1994] solves.
Experimental seasonal hydrologic forecasting for the Western U.S. Dennis P. Lettenmaier Andrew W. Wood, Alan F. Hamlet Climate Impacts Group University.
Global Flood and Drought Prediction GEWEX 2005 Meeting, June Role of Modeling in Predictability and Prediction Studies Nathalie Voisin, Dennis P.
Efficient Methods for Producing Temporally and Topographically Corrected Daily Climatological Data Sets for the Continental US JISAO/SMA Climate Impacts.
A Multi-Model Hydrologic Ensemble for Seasonal Streamflow Forecasting in the Western U.S. Theodore J. Bohn, Andrew W. Wood, Ali Akanda, and Dennis P. Lettenmaier.
Retrospective Evaluation of the Performance of Experimental Long-Lead Columbia River Streamflow Forecasts Climate Forecast and Estimated Initial Soil Moisture.
Drought Prediction (In progress) Besides real-time drought monitoring, it is essential to provide an utlook of what future might look like given the current.
Potential for medium range global flood prediction Nathalie Voisin 1, Andrew W. Wood 1, Dennis P. Lettenmaier 1 1 Department of Civil and Environmental.
Sources of Skill and Error in Long Range Columbia River Streamflow Forecasts: A Comparison of the Role of Hydrologic State Variables and Winter Climate.
Assessing the Influence of Decadal Climate Variability and Climate Change on Snowpacks in the Pacific Northwest JISAO/SMA Climate Impacts Group and the.
Hydrologic Forecasting Alan F. Hamlet Dennis P. Lettenmaier JISAO/CSES Climate Impacts Group Dept. of Civil and Environmental Engineering University of.
Alan F. Hamlet Andy Wood Dennis P. Lettenmaier JISAO Center for Science in the Earth System Climate Impacts Group and the Department.
DOWNSCALING GLOBAL MEDIUM RANGE METEOROLOGICAL PREDICTIONS FOR FLOOD PREDICTION Nathalie Voisin, Andy W. Wood, Dennis P. Lettenmaier University of Washington,
VERIFICATION OF A DOWNSCALING SEQUENCE APPLIED TO MEDIUM RANGE METEOROLOGICAL PREDICTIONS FOR GLOBAL FLOOD PREDICTION Nathalie Voisin, Andy W. Wood and.
EVALUATION OF A GLOBAL PREDICTION SYSTEM: THE MISSISSIPPI RIVER BASIN AS A TEST CASE Nathalie Voisin, Andy W. Wood and Dennis P. Lettenmaier Civil and.
Hydrological Simulations for the pan- Arctic Drainage System Fengge Su 1, Jennifer C. Adam 1, Laura C. Bowling 2, and Dennis P. Lettenmaier 1 1 Department.
Remote Sensing Applications to Improve Seasonal Forecasting of Streamflow and Reservoir Storage in the Upper Snake River Basin Marketa McGuire, Andy W.
Ongoing Work As part of a project intended to evaluate the potential for improving water resources management in Mexico through use of climate forecasts,
Long-lead streamflow forecasts: 2. An approach based on ensemble climate forecasts Andrew W. Wood, Dennis P. Lettenmaier, Alan.F. Hamlet University of.
Evaluation of Skill and Error Characteristics of Alternative Seasonal Streamflow Forecast Methods Climate Forecast and Estimated Initial Soil Moisture.
LSM Hind Cast for the Terrestrial Arctic Drainage System Theodore J. Bohn 1, Dennis P. Lettenmaier 1, Mark C. Serreze 2, and Andrew G. Slater 2 1 Department.
Long-Range Streamflow Forecasting Products and Water Resources Management Applications in the Columbia River Basin Alan F. Hamlet, Andy Wood, Dennis P.
Nathalie Voisin1 , Andrew W. Wood1 , Dennis P. Lettenmaier1 and Eric F
Upper Rio Grande R Basin
Andrew Wood, Ali Akanda, Dennis Lettenmaier
Precipitation-Runoff Modeling System (PRMS)
Hydrologic Considerations in Global Precipitation Mission Planning
Use of Extended Daily Hydroclimatalogical Records to Assess Hydrologic Variability in the Pacific Northwest Department of Civil and Environmental Engineering.
(April, 2001-September, 2002) JISAO Climate Impacts Group and the
Challenges in western water management: What can science offer?
Hydrologic implications of 20th century warming in the western U.S.
1Civil and Environmental Engineering, University of Washington
Dennis P. Lettenmaier, Andrew W. Wood, Ted Bohn, George Thomas
Nathalie Voisin, Andy W. Wood and Dennis P. Lettenmaier
Hydrologic ensemble prediction - applications to streamflow and drought Dennis P. Lettenmaier Department of Civil and Environmental Engineering And University.
2006 Water Resources Outlook for Idaho and the Western U.S.
Multimodel Ensemble Reconstruction of Drought over the Continental U.S
150 years of land cover and climate change impacts on streamflow in the Puget Sound Basin, Washington Dennis P. Lettenmaier Lan Cuo Nathalie Voisin University.
THE POTENTIAL FOR FLOOD AND DROUGHT PREDICTION
Kostas M. Andreadis1, Dennis P. Lettenmaier1
Hydrologic Forecasting
Hydrology and Water Management Applications of GCIP Research
Andy Wood and Dennis Lettenmaier
Hydrologic response of Pacific Northwest Rivers to climate change
Long-Lead Streamflow Forecast for the Columbia River Basin for
Effects of Temperature and Precipitation Variability on Snowpack Trends in the Western U.S. JISAO/SMA Climate Impacts Group and the Department of Civil.
Andrew Wood, Alan Hamlet, Dennis Lettenmaier University of Washington
Land surface modeling for real-time hydrologic prediction and drought forecasting Dennis P. Lettenmaier Department of Civil and Environmental Engineering.
A. Wood, A.F. Hamlet, M. McGuire, S. Babu and Dennis P. Lettenmaier
Hydrologic issues in the measurement of snowfall
N. Voisin, J.C. Schaake and D.P. Lettenmaier
Andy Wood and Dennis P. Lettenmaier
Towards a global drought prediction capability
Results for Basin Averages of Hydrologic Variables
Andrew W. Wood Dennis P. Lettenmaier
A Multimodel Drought Nowcast and Forecast Approach for the Continental U.S.  Dennis P. Lettenmaier Department of Civil and Environmental Engineering University.
Evaluation of the TRMM Multi-satellite Precipitation Analysis (TMPA) and its utility in hydrologic prediction in La Plata Basin Dennis P. Lettenmaier and.
Dennis P. Lettenmaier Andrew W. Wood, and Kostas Andreadis
UW Hydrologic Forecasting: Yakima R. Discussion
Multimodel Ensemble Reconstruction of Drought over the Continental U.S
Results for Basin Averages of Hydrologic Variables
Presentation transcript:

RESULTS and DISCUSSION The Effect of Hydrologic Model Calibration on Seasonal Streamflow Forecasts Xiaogang Shi, Andrew W. Wood, and Dennis P. Lettenmaier Department of Civil and Environmental Engineering, University of Washington, Seattle, WA, 98195 ABSTRACT 2) Retrospective evaluation of forecast errors 2. Model calibration Cp is shown in figures below for calibrated, uncalibrated, and uncalibrated bias corrected scenarios, for forecasts made from Dec. 1 through June 1, and for forecast periods of 1-6 months length with zero month lead time (forecast period starts on the forecast date). As is expected for snowmelt dominated rivers in the western U.S., Cp generally improves as the forecast date advances from December 1 through the winter and into the spring until about the time of maximum snow accumulation, typically April to May depending on the specific basin, and then begins to decline. For forecasts made on December 1, Cp is often negative, which reflects the fact that much of the snow that will contribute to forecast period runoff has not yet accumulated. Depending on the forecast date, forecasts for longer periods tend to have higher Cps than for shorter forecasts, primarily because errors in forecasting of runoff timing are reduced for longer forecast periods. Model calibration was performed using the Multi Objective COMplex evolution (MOCOM-UA) algorithm of Yapo et al. (1998), which employs a population evolution strategy to find the optimal parameter set. Two objective functions were incorporated in our implementation of the algorithm: the Nash-Sutcliffe efficiency measure of Nash and Sutcliffe (1970) and the absolute value of annual mean volume error between observed and simulated streamflow. The MOCOM-UA method searches the parameter space by sampling the feasible parameter space at a number of points. At each point, the multi-objective vector is computed and then the population is ranked and sorted using the Pareto-rank procedure of Goldberg (1989). The iterative process shown at right causes the parameter set to converge towards the Pareto optimum. A priori estimates of these six VIC model parameters obtained from the N-LDAS experiment (Mitchell et al. 2004; Maurer et al. 2002) were used as the uncalibrated parameter sets. They also were used as the initial values for the automatic calibration. Hydrologic model calibration is usually viewed as central element of the Ensemble Streamflow Prediction (ESP) method of seasonal streamflow forecasting. Metrics for evaluation of forecast errors such as root mean squared error (RMSE) are heavily influenced by bias, which in turn is readily reduced by calibration. On the other hand, bias can also be reduced by post-processing -- e.g., “training” bias correction schemes based on retrospective simulation error statistics. This observation invites the question: how much is forecast error reduced by calibration, beyond what can be accomplished by post-processing to remove bias? We address this question through retrospective evaluation of forecast errors at eight streamflow forecast locations distributed across the western U.S.. Lead times ranging from zero to six months are investigated, for forecasts initiated from December 1 through June 1, which span the period when most runoff occurs from snowmelt-dominated western U.S. rivers. ESP forecast errors are evaluated both for uncalibrated forecasts to which a percentile mapping bias correction approach is applied, and for forecasts from an objectively-calibrated model without explicit bias correction. Using the coefficient of prediction (Cp), which essentially is a measure of the fraction of variance explained by the forecast, we find that the reduction in forecast error as measured by Cp that is achieved by bias correction alone is nearly as great as that resulting from hydrologic model calibration. 3. Ensemble Streamflow Prediction The ESP method estimates the hydrologic state of current physical conditions (especially for the soil moisture and snowpack) at the time of forecasts using hydrologic simulations driven by observed surface forcings up to the time of forecast, followed by resampling of hydrologic ensembles from past observed forcings. For example, January ESP ensembles were initialized with a January 1 nowcast and each hindcast had 30 ensemble members (1970 -1999). The schematic is shown at right. NCDC met. station obs. 1/8 degree weather inputs soil moisture snowpack VIC model spin up streamflow The first day of Dec., Jan.,Feb.,Mar.,Apr.,May,Jun. 10 years back Hydrologic hindcasts 1 - 6 Months INITIAL STATE ESP ensemble traces (30) STUDY BASINS Yellowstone (YELLO) Stehekin (STEHE) North Fork Flathead (NOFOR) Salmon (SALMO) Feather (OROVI) Bruneau (BRUNE) Animas (ANIMA) Gunnison (BLMSA) Eight river basins shown at right were selected across the western U.S. which have minimal effects of diversions and regulation upstream as noted in U.S. Geological Survey metadata, or have records from which upstream diversion and storage effects had been removed. All forecast points have lengthy streamflow records (spanning at least the period 1971-2001) of high quality. The seasonal hydrologic cycle of all basins is dominated by spring snowmelt runoff. The eight forecast points are all locations that are considered critical to water management operations, e.g., inflows to reservoirs or flows at key index sites. Comparison of columns (a) and (b) above shows that in all cases, calibration improves the forecast accuracy. However, uncalibrated bias corrected forecasts (column (c)) in many cases have Cp values that are only slightly lower than for calibrated forecasts, and in the case of four of the basins (ANIMA, OROVI, STEHE, and YELLO), bias correction almost uniformly provides higher Cp values than does calibration (gray boxes in column (d)). The results for other lead times ranging from 1 to 6 months are consistent with the results shown above except YELLO, for which calibration provides a higher Cp than does bias correction for 3-6 month lead times. 4. Bias correction approach To correct bias, we use the “percentile mapping” approach (Panofsky and Brier 1968), as adopted by Wood et al. (2002) and applied to streamflow simulations by Snover et al (2003). In the percentile mapping bias correction process, the percentile derived from the cumulative distribution function (CDF) of a simulated flow, is used to extract the corresponding flow at the same percentile from the CDF of the observations. The percentile mapping bias correction approach not only removes the bias in the mean, but also alters the distribution of the simulated streamflows as well. In this case, we used two-parameter log normal CDFs fitted to both the simulated and observed flows as the basis for the bias correction approach. The figures at left show the Cp differences between calibrated bias corrected and uncalibrated bias corrected forecasts. Cp values for calibrated bias corrected mostly are slightly higher than for bias correction alone, but in the cases of the gray boxes, are lower. It should be noted that most runoff occurs in boxes above the upper left – lower right diagonal in the figure, however, and for those boxes calibration with bias correction usually results in an improvement relative to bias correction alone Assessment of forecast skill METHODS We evaluated forecast skill using the coefficient of prediction, Cp, as defined by Lettenmaier (1984) below. Hydrologic model implementation Cp normally ranges between 0.0 and 1.0, where a value of 1.0 corresponds to the perfect forecast; a value of 0.0 is only achieved when the forecast is always equal to the historical long-term mean. For small sample sizes and forecast methods such as ESP, negative values are possible, which imply forecasts that are worse than climatology. The figure to the right illustrates features of the VIC (Variable Infiltration Capacity) model (Liang et al. 1994). It was implemented for each of the eight river basins in water balance mode. The VIC model produced daily runoff from each grid cell in the simulation domain as baseflow and surface runoff. Grid cell runoff was routed through a grid-based stream network using the routing scheme described by Lohmann et al. (1998 a, b) as shown at lower right. The model was implemented at one-eighth degree spatial resolution with five elevation bands specified as described by Nijssen et al. (1997) and Maurer et al. (2002). CONCLUSIONS RESULTS and DISCUSSION We find that the reduction in forecast error as measured by Cp that is achieved by bias correction alone is nearly as great as that resulting from hydrologic model calibration for seasonal streamflow forecasts in western U.S. rivers. Moreover, the discrepancy between calibrated bias corrected and bias correction alone is generally small. Bias correction is very easy to implement and avoids the time consuming calibration process. Although we emphasize that this work is limited to seasonal streamflow forecasts (and is not applicable, for instance, to shorter lead flood forecasting), it does suggest the potential for using “off the shelf” model parameters (e.g., from data sets like the North American Land Data Assimilation System) and/or regional parameter estimation methods, in lieu of site-by-site calibration Calibration results The figures at right show the long-term monthly means for the calibrated, uncalibrated and observed streamflows respectively, which show a good agreement between observed and calibrated streamflow. As expected, the Nash-Sutcliffe Efficiency is much higher (0.83 to 0.97) for the calibrated period across all study basins than for the uncalibrated simulations (0.29 to 0.71). Forcing Daily averaged wind speed from the lowest level in the NCEP-NCAR reanalysis (Kalnay et al. 1996) Gridded forcing data from NOAA Cooperative Observer (Co-Op) station data Other variables indexed to daily temperature and/or the daily temperature range (see Maurer et al. 2002) Internal time step VIC: daily VIC snow: 1 hour Output Daily streamflow REFERENCES Lettenmaier, D. P., 1984: Limitations on seasonal snowmelt forecast accuracy. J. Wat. Res. Plan. Mgmt., 108 , 255-269. Wood, A.W., E. P. Maurer, A. Kumar, and D. P. Lettenmaier, 2002: Long-range experimental hydrologic forecasting for the eastern United States. J. Geophys. Res., 107(D20), 4429, doi:10.1029/2001JD000659. Yapo, P. O., H. V. Gupta, and S. Sorooshian, 1998: Multi-objective global optimization for hydrologic models. J. Hydrol., 203, 83-97. Note: See the author for other references, or www.hydro.washington.edu (“publications”).