The impact of the objective function in multi-site and multi-variable calibration of the SWAT model

Automatic calibration of complex hydro-ecological models is an increasingly important issue which involves making decisions. One of the most relevant is the choice of the objective function, but its effects have been scarcely studied in complex models. We have used the SWAT model to assess the impac...

Full description

Saved in:

Bibliographic Details
Published in:	Environmental modelling & software : with environment data news 2017-07, Vol.93, p.255-267
Main Authors:	Molina-Navarro, Eugenio, Andersen, Hans E., Nielsen, Anders, Thodsen, Hans, Trolle, Dennis
Format:	Article
Language:	eng
Subjects:	Calibration Calibration scheme Catchment model Ecological models Impact analysis Multi-site calibration Nutrients Objective function Performance measurement Studies SWAT Watershed management Watersheds
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Automatic calibration of complex hydro-ecological models is an increasingly important issue which involves making decisions. One of the most relevant is the choice of the objective function, but its effects have been scarcely studied in complex models. We have used the SWAT model to assess the impact of the objective function for a multi-site (4 stations) and multi-variable (OrgP, OrgN, NO3−, PO43−) calibration of the Odense catchment (Denmark). Six calibration schemes were tested, varying the objective function and the nutrient fractions targeted. The best performance metrics (R2, NSE, PBIAS) were obtained when using NSE as objective function and targeting N-fractions and P-fractions separately. The scheme was validated in another SWAT set-up in northern Denmark. Although NSE is often questioned, we found it as an adequate objective function when addressing a multi-site and multi-variable calibration. Our findings may serve as guideline for hydro-ecological modellers, being useful to achieve watershed management goals. •We set-up a multi-site (four stations) and multi-variable (four nutrients) SWAT model.•We assessed the impact of the objective function (its choice and its calculation).•Six calibration schemes for daily nutrient loads were designed and tested.•Best performance was obtained using NSE and targeting N and P fractions separately.•This scheme improved the value of all the performance metrics (R2, NSE, PBIAS).
ISSN:	1364-8152 1873-6726