Extended Abstract (252K)J5.8
Distributed model flow sensitivities to input and parametric uncertainty: Case studies for three watersheds in the Central U.S
Theresa M. Carpenter, Hydrologic Research Center, San Diego, CA; and K. P. Georgakakos and J. A. Sperfslage
Distributed hydrologic modeling is a very active area of hydrologic research as investigators develop and enhance methods to incorporate expanding databases of spatial and temporal data. Precipitation from weather radar is a spatio-temporal database used as input to distributed models, while distributed parametric model information is offered by various static spatial databases such as digital terrain elevation, land cover and land use, and soils databases. However, significant uncertainty exists in both the values of radar rainfall estimates, and in the development of model parameter values from the static spatial databases. How does such uncertainty impact streamflow estimates derived from distributed hydrologic models? Given such uncertainties, how do the flow simulations from distributed hydrologic models compare for larger basins to those from simpler spatially-lumped models? This research examines these questions in the context of operational streamflow forecasting.
As a participant of the National Weather Service Office of Hydrology “Distributed Model Intercomparison Project” (DMIP), a case-study analysis was developed for the Illinois River, Blue River, and Elk River watersheds in parts of Arkansas, Oklahoma, and Missouri. The basins include NWS operational river forecast locations at: the Illinois River at Watts, OK (1644 km2); the Illinois River at Tahlequah, OK (2483 km2); the Blue River at Blue, OK (1232 km2); and the Elk River at Tiff City, MO (2258 km2). For each basin, the sensitivity of flow simulations from a given distributed model was examined for various cases of both radar-rainfall uncertainty and hydrologic model parametric uncertainty within a Monte Carlo simulation framework. The soil model components are based on current operational models used in flow forecasting of large basins, and the analyses utilized an archived operational radar precipitation database for the WSR-88D weather radar at Tulsa, OK (hourly resolution, Stage III product). The archived database covered the period 5/1993 through 5/1999. Sensitivity results are summarized by normalized measures of the range of flows computed in the Monte Carlo simulations for selected events and for various locations within each study watershed. These measures are of similar magnitude for the case of rainfall input uncertainty alone and for parametric input uncertainty alone. The sensitivity measures also tend to decrease as drainage area increases. This indicates that the sensitivity of flow simulations to input and parametric uncertainty is scale dependent. The work expands previous results by Carpenter et al (2001), which examined only one watershed for limited input uncertainty cases and for limited events.
Reference: Carpenter, T.M., Georgakakos, K.P., and J.A. Sperfslage, 2001: “On the parametric and NEXRAD-radar sensitivities of a distributed hydrologic model suitable for operational use”, J.Hydrology, 253, 169-193.
Joint Session 5, Flood Hydrology, Management, Information Systems: Longer-Term Planning, Management, Impacts, and Forecasting Issues (Joint with the Symp on Impacts of Water Variability: Benefits and Challenges and the 17th Conference on Hydrology)
Tuesday, 11 February 2003, 1:30 PM-5:30 PM
Previous paper Next paper