A Priori Parameter Estimation for Distributed Models and for Predictions in Ungauged Basins (PUB) I
Presiding: Q Duan, Lawrence Livermore National Laboratory; T Wagener, Pennsylvania State University
H44B-01 INVITED 15:30h
A priori parameters, uncertainties, and calibration in watershed modeling
The availability of new sources of comprehensive, high resolution physiographical information shifted watershed modeling toward more physically-based approaches. As a result, model parameterization can be more closely linked to basin physical properties. However, because of simulation and measurement scale differences as well as model and input data uncertainties, different transformation functions between measured properties and actual model parameters/constants (so called a priori parameters) are required. Derivation of a priori parameters is critical in distributed hydrological modeling when parameters have to be specified for a large number of computational elements. Two approaches are most commonly used: establishing direct physically-based relationships, and statistical regionalization. Both approaches have disadvantages because of poorly defined physical relationships or non-uniqueness of calibration results. This presentation evaluates these approaches for the Hydrology Laboratory Distributed Modeling System. Parameterization results for distributed water balance and channel routing models suggest that combining these two approaches can improve an estimation procedure.
H44B-02 15:50h
Validation of a Distributed Rainfall Runoff Model Within a Regional Flood Frequency Analysis
The discharge index value estimation, used to make dimensional the growth curve in each site, is the major problem in a regional flood frequency analysis. Because of the great site sensitivity, this topic is a critical issue: in gauged sites the time series average is often sufficient, while in non-gauged sites this index value should be evaluated taking into account as much information as possible about the site concerned.The DRiFt (Discharge River Forecast) model, centred on a proficient description of the drainage system in its essential parts - hillslopes and channel networks - is used to tackle this problem. In DRiFt model the hydrologic response at the basin scale is determined considering actual topography, soil properties, and rainfall fields variability. Calibration and validation of model's parameters - performed through different test cases in different sized basins - led to identify a unique set of parameters for the whole region where the model is applied. This parameters' robustness is the characteristic that allows the use of the model as a link between rainfall and flood frequency analysis. Case of study is the Liguria region, Italy.
H44B-03 16:05h
Using Satellite Retrieved Data to Improve the Calibration of a Hydrological Model for Flash Flood Forecast.
An intriguing prospective in distributed modeling is the possibility to incorporate remotely sensed data, such as evaporative fraction and soil moisture conditions, with the purpose of initializing, driving, updating or calibrating models. In this work a parsimonious distributed model devoted to small catchments continuous discharge simulation is presented. Soil type and land use information are used to calibrate a modified Horton method that represents infiltration process. It is shown that the model is able to mimic the different component of the continuity equation when it is calibrated on reliable hydro-pluviometric data, especially during relatively wet periods. The calibration of the model in a data-scarce catchment is improved by using evaporative fraction maps retrieved by a flexible and simple model of satellite data assimilation.
H44B-04 INVITED 16:20h
Soil Hydraulic Parameters for Heterogeneous Landscapes
A priori parameter estimation for distributed hydrologic models remains to be a challenging issue in hydrology. Spatio-temporal heterogeneity/nonlinearity of hydrologic parameters, between-parameter correlations, basin characteristics, flux gradient/direction, and time-varying boundary conditions are some of the key factors compounding the effect of parameter uncertainty. Although no unique parameter estimation scheme exists to account for all hydrologic scenarios, recent advances including the guidelines for developing "effective" hydrologic parameters will be reviewed in this talk. Examples of transfer functions translating pedologic, topographic, and vegetation attributes to soil hydraulic properties for Southern Great Plains region will be presented. Additionally, new schemes for deriving "effective" soil hydraulic parameter for steady state flow in heterogeneous soils will be presented. "Effective" soil hydraulic parameters of the heterogeneous soil formation are obtained by conceptualizing the soil as an equivalent homogeneous medium. It requires that the "effective" homogeneous soil discharges the same mean surface moisture flux (evaporation or infiltration) as the heterogeneous media. Assuming one-dimensional vertical moisture movement in the unsaturated soils, both scenarios of horizontal (across the land surface) and vertical (across the soil profile) heterogeneities are investigated. The effects of hydraulic parameter statistics, surface boundary conditions, domain scales and fractal dimensions in case of nested soil hydraulic property structure will be addressed.
H44B-05 16:40h
Parameter Estimation of a Water Table Dynamics Model Used in Lands Surface Parameterization Scheme
A gridded version of the land-surface parameterization scheme (LSXGW) with a water table representation previously developed by the author was applied to the East River basin in sounth China. The scheme accounts for the sub-grid variability of water table depth and groundwatee rechage to investigate surface water-groundwater interaction. Four critical parameters were calibrated: 1. water table threshold depth, above which groundwater runoff will be triggered. 2. rate constant, which governs the rate of groundwater runoff generation. 3. specific yield, which controls the fluctuation of water table depth, 4. size of surface retention storage, which controls the partition of surface runoff and infiltration. Results indicate that daily and monthly streamflow at several locations of the basin can be simulated reasonably well by the model after calibration, and evaporation is only marginally affected by the four runoff parameters. The model performance for different combinations of 4 parameters are analyzed in order to understanding the relative control and mutual interaction of each parameter. A parameter estimation scheme based on the hydrograph separation designed for the poorly-gaged areas was developed and tested.
H44B-06 16:55h
A Correction to GLUE: Generalized Likelihood Uncertainty Estimation
The widely accepted GLUE approach to describing the impact of parameter uncertainty in rainfall/runoff models produces confidence intervals for model predictions that are generally incorrect. We document this phenomenon using a case when the rainfall/runoff model is a simple linear regression. Beven (1992) has suggested that the choice of the likelihood function used in GLUE is subjective and may change depending upon the goals of the modeler. We observe why the choice of likelihood functions is not arbitrary, but rather the correct likelihood function depends upon the assumed distribution of the model errors. We show how the GLUE methodology can be adapted to provide the variance of model predictions and confidence intervals for model predictions that will agree with approaches based on classical statistics.