Contents lists available at ScienceDirect Ecological Modelling journal homepage: www.elsevier.com/locate/ecolmodel Comparison of aquatic ecosystem functioning between eutrophic and hypereutrophic cold-region river-lake systems Eric Akomeah a, ⁎ , Karl-Erich Lindenschmidt a , Steven C. Chapra b a Global Institute for Water Security, University of Saskatchewan, 11 Innovation Boulevard, Saskatoon, Saskatchewan, S7N 3H5, Canada b Department of Civil and Environmental Engineering, Tufts University, United States ARTICLE INFO Keywords: Global sensitivity analysis Identifiability Cold region River-lake systems Under-ice processes Surface water quality modelling ABSTRACT Located on the same river, the degree of eutrophication in the upper and middle reaches of the Qu’Appelle River in Saskatchewan are different. While the upper Qu’Appelle is eutrophic, the middle Qu’Appelle River is hy- pereutrophic. To manage the river sustainably, there is a need to understand key processes governing eu- trophication in both systems. In this study, a comprehensive global sensitivity analysis technique, Variogram Analysis of Response Surface (VARS), was applied to gain insights to the functioning of the two systems. Eutrophication in both systems was modelled using the Water quality Analysis Simulation Program (WASP 7.52). The performance of the model to predict key variables of eutrophication was measured using relative root mean square error. The global sensitivity analyses showed that although diffuse loading has significant influence on the systems, prevailing processes governing eutrophic state in the upper Qu’Appelle River include: nutrient and, phytoplankton cycles. Meanwhile, in the middle Qu’Appelle River a number of processes including phy- toplankton cycle, nutrient cycle, diffuse loading and DO balance together sustain its hypereutrophic state. 1. Introduction The use of mechanistic models for social learning, scenario analysis, decision-making and adaptive management of aquatic ecosystems is now extensively applied (Kelly (Letcher) et al., 2013; Döll et al., 2015; Zhou et al., 2015; Landis et al., 2017; Farag et al., 2017). The utility of these models for extrapolation is founded on the ability to complement them with causal hypotheses that are based on prevailing views of how internal limnologic processes of aquatic systems function. Despite their usefulness in providing a rational basis for under- standing freshwater aquatic systems, mechanistic models are often criticized on the issue of parameter identifiability (Beck, 1987; Dietzel and Reichert, 2014; Her and Chaubey, 2015; Han and Zheng, 2016). In the bid to represent internal processes of a system for a foreseeable future scenario, the models end up being complex and poorly identified (Brun et al., 2001). However, addressing identifiability of complex models is well treated in the literature (Brun et al., 2001; Anh et al., 2006; De Pauw and De Baets, 2008; Arhonditsis et al., 2008; Rode et al., 2008; Wagener et al., 2009; Otero-Muras et al., 2010; Cibin et al., 2010; Parslow et al., 2013; Ghasemizade et al., 2017). It has been argued that since field measurements rarely provide enough information to quantify most model parameters, ascertaining practical parameters that re- present the variability in field measurements and conducting sensitivity analysis to unravel most important model parameters and how they interact are the most reasonable means to gaining insights into the aquatic systems’ behaviours (Brun et al., 2001). Regional sensitivity analysis of water quality models with an effi- cient algorithm has been proposed by Spear (1997) to determine rea- sonable parameter values using the Monte Carlo sampling scheme. This Bayesian method, together with other methods such as generalized likelihood uncertainty estimation (GLUE) (Beven and Binley, 1992; Freer et al., 1996; Jiang et al., 2018) still suffers very extensive calcu- lations for the identification of all the parameters within a factor space. In this study, the recently developed Variogram Analysis of Response Surface (VARS) based global sensitivity analysis (GSA) was used to explore the behaviours of reaches of the same river with dif- ferent trophic states. VARS comprehensively assigns the variability of model responses to governing factors at a low computational cost (number of model runs) using directional variogram and covariogram functions (Razavi and Gupta, 2016a). VARS, which has recently been used predominantly as a diagnostic tool in hydrological (Haghnegahdar et al., 2017) and hydrodynamic models (Sheikholeslami et al., 2017) will now be used to understand the aquatic ecosystem functioning of water quality models of the upper and middle reaches of the Qu’Appelle River. Understanding the dominant processes influencing the beha- viours of these two unique systems is key to managing eutrophication. https://doi.org/10.1016/j.ecolmodel.2018.12.004 Received 24 August 2018; Received in revised form 28 November 2018; Accepted 7 December 2018 ⁎ Corresponding author. E-mail address: eric.akomeah@usask.ca (E. Akomeah). Ecological Modelling 393 (2019) 25–36 0304-3800/ © 2018 Elsevier B.V. All rights reserved. T