SWAT Literature Database for Peer-Reviewed Journal Articles

Title:Regionalisation of the parameters of a hydrological model: Comparison of linear regression models with artificial neural nets 
Authors:Heuvelmans, G., B. Muys and J. Feyen 
Journal:Journal of Hydrology 
Volume (Issue):319(1-4) 
Article ID: 
URL (non-DOI journals): 
Broad Application Category:hydrologic only 
Primary Application Category:regionalization of input parameters 
Secondary Application Category:calibration, sensitivity, and/or uncertainty analysis 
Watershed Description:25 subwatersheds ranging in size from 2.24 to 209.93 km^2, which are tributaries located in the Flemish part of the Scheldt River in Belgium. 
Calibration Summary:Daily (1990-95): E varied between .70 & .95 (across the 25 watersheds; manual calibration) 
Validation Summary:Daily (1996-2001) E varied between .67 & .92 (across the 25 watersheds) 
General Comments:Calibration of 7 SWAT input parameters, selected as a part of an initial sensitivity analysis, was performed using both linear regression and forced?-forward Artificial Neural Network (ANN). A non-parametric bootstrap method was used to assess uncertainty for both approaches. The ANN delivered more accurate parameter estimates if the non-linearities simulated by the ANNs had physical meaning and watershed characteristics, for extrapolated areas, were similar to characteristics of watersheds that the ANNs were created for. Otherwise, linear regression resulted in better results. Uncertainty was generall higher for the ANNs. 
Keywords:Parameter regionalisation; Parameter uncertainty; Artificial neural networks; Bootstrap; Land cover