Comparison of three regionalization techniques for predicting streamflow in ungaged watersheds in Nebraska, USA using SWAT model

Authors

Michael W. Van Liew, University of Nebraska - LincolnFollow
Aaron R. Mittelstet, University of Nebraska-LincolnFollow

Date of this Version

2018

Citation

Int J Agric & Biol Eng, 2018; 11(3): 110–119

Comments

Open access

DOI: 10.25165/j.ijabe.20181103.3528

Abstract

This study compared three approaches, regional averaging, nearest neighbor, and donor techniques, to regionalize parameters in the Soil and Water Assessment Tool (SWAT) on eleven watersheds located in the Dissected Plains, Plains, and Rolling Hills Landforms in the eastern portion of the State of Nebraska, USA. Within the Rolling Hills Landform, three watersheds were randomly selected as calibration watersheds while two were randomly selected as validation watersheds. Two watersheds were randomly selected as calibration watersheds while one was randomly selected as a validation watershed within each of the Dissected Plains and Plains Landforms. The seven calibration watersheds were used to provide the necessary calibrated parameter sets to execute each of the regional approaches, while the four validation watersheds were used to assess the impact of applying each of these approaches to an uncalibrated watershed. Percent Bias (PBIAS) and the Nash Sutcliffe Coefficient of Efficiency (NSE) were used to assess model performance. Test results of this study show that all three methods performed poorly, since the majority of watersheds among each method tested exhibited PBIAS values greater than ±25% and/or NSE values less than 0.50, which were considered to be unsatisfactory in terms of model performance. The average regionalization, nearest neighbor and donor methods resulted in only four (two calibration and two validation), zero and one satisfactory set of simulated watershed results, respectively. The findings from this study indicate that although each watershed was successfully calibrated with NSE values ranging from 0.51 to 0.84, none of the three regionalization methods provided suitable calibration data sets to define parameter values for performing satisfactory simulations on ungaged watersheds across the eastern Nebraska landscape.