The ability to apply a hydrologic model to large numbers of basins for forecasting purposes requires a quick and effective calibration strategy. This paper presents a step wise, multiple objective, automated procedure for hydrologic model calibration. This procedure includes the sequential calibration of a model's simulation of solar radiation (SR), potential evapotranspiration (PET), water balance, and daily runoff. The procedure uses the Shuffled Complex Evolution global search algorithm to calibrate the U.S. Geological Survey's Precipitation Runoff Modeling System in the Yampa River basin of Colorado. This process assures that intermediate states of the model (SR and PET on a monthly mean basis), as well as the water balance and components of the daily hydrograph are simulated consistently with measured values.

(KEY TERMS: Precipitation Runoff Modeling System; Shuffled Complex Evolution; Colorado; optimization; solar radiation; potential evapotranspiration; water balance; runoff.)
Advertisement

(ProQuest Information and Learning: ... denotes formulae omitted.)

INTRODUCTION

Runoff from winter snowpack is the main supply of water in the intermountain western United States (U.S.). NOAA National Weather Service (NWS) and USDA National Resources Conservation Service (NRCS) issue runoff forecasts for the western U.S. Both of these agencies are attempting to modernize their runoff forecasting tools by incorporating more spatially distributed hydrologie modeling techniques (e.g. Spatially Distributed Hydrologie Modeling, USDA-NRCS, 1998; Carter, 2005). To this end, the NRCS is currently configuring a version of the U.S. Geological Survey's (USGS) Precipitation Runoff Modelling System (PRMS) for 35 snowmelt dominated basins by 2006 with the possible extension to the entire western U.S. (K. Rojas and F. Geber, NRCS, personnel communication, May 2005). PRMS is a distributed parameter, physically based hydrologie model. The ability to apply a distributed hydrologie model to a large numbers of basins in a timely and efficient manner for runoff forecasting purposes requires a quick and effective calibration strategy.

Traditional approaches to calibration and evaluation of distributed hydrologie models compared observed and simulated runoff at the outlet of the basin. This traditional approach is not sufficient by itself in the evaluation of distributed hydrologie models (Refsgaard, 1997). While incorporation of spatial data into the calibration and evaluation process is ideal, research in this area has occurred mainly in heavily instrumented research basins (Refsgaard, 2000). In general, the data available for calibration and evaluation of distributed hydrologie models are limited for the basins in which NOAA and NRCS are forecasting runoff.

Gupta et al. (1998) proclaimed that hydrologie model calibration must consider the multiple objective nature of the problem. The use of multiple objective functions in the calibration of hydrologie models has become increasingly popular. For example, Hogue et al. (2000) examined recessions and low flows, higher flows, and base flows; Turcotte et al. (2000) examined droughts, annual and monthly flow volumes, high flows, high flow synchronization, and snowmelt runoff; Madsen (2000) examined the water balance, hydrograph shape, peak flows, and low flows; and Boyle et al. (2000; 2003) examined three components of the hydrograph described as driven, nondriven quick, and nondriven slow. While these studies used multiple objectives, the only data used was runoff different portions and time steps of the hydrograph were configured for the multiple objective calibrations. Intermediate variables computed by the hydrologic model (such as solar radiation, potential evapotranspiration, snow water equivalent, snow covered area, and soil moisture) could be characterized by parameter values that do not replicate those hydrological processes in the physical system.

Is this paper, a multiple objective calibration strategy that incorporates additional, easily obtainable, data sets is presented. Four variables simulated by PRMS are used as calibration data sets: (1) solar radiation (SR), (2) potential evapotranspiration (PET), (3) water balance, and (4) daily runoff components. The SR and PET datasets are monthly mean values derived from nationwide data sets, making them readily accessible for application in a large number of basins. The parameters influencing each of the model variables are calibrated in a step wise, multiple objective procedure similar to that presented by Hogue et al. (2000). This process gives the user higher confidence in the model output by assuring that intermediate states of the model (as described by monthly mean values of SR and PET), as well as the water balance and components of the daily hydrograph are simulated consistently with measured values.

STUDY AREA

The Yampa River basin at Steamboat Springs (USGS streamflow gaging Station 09239500) (USGS 2005a) in northwestern Colorado was chosen as the study area (see Figure 1). The Yampa River basin is a mountainous basin where the runoff is strongly dependent on snowmelt, peaking during May. The basin is 1,430 km^sup 2^ in area and ranges in elevation from 2,000-3,800 meters. Figure 2 shows the daily basin mean by month for precipitation and maximum and minimum temperature for two eight-year periods: Water Years (WYs) 1996-2003 and 1988-1995. WYs run from October through September. These two eight-year periods were chosen as the calibration and evaluation periods, respectively. The wettest month for the Yampa River basin is February and the driest month is June. The warmest months are July and August and the coldest are December and January.