Selection and validation of a complex fishery model using an uncertainty hierarchy

Abstract

Assessing the validity of a model is essential for its credibility especially when the model is used as decision making tool. Complex dynamic fishery models are recommended to investigate the functioning of fisheries and to assess the impact of management strategies, particularly spatial fishing regulations. However, their use is limited due to the difficulty and computational cost of parameterizing and gaining confidence, particularly for parameter rich models. These difficulties are compounded by uncertainty regarding parameter values, many of which are often taken from literature or estimated indirectly. Here we propose a methodology to improve confidence and understanding in the model, easily transferable to any complex model. The approach combines sensitivity analysis, scalability of parameters, optimization procedures, and model skill assessment in order to parameterize, validate and achieve the most plausible formulation of a model given the available knowledge while reducing the computational load. The methodology relies on five steps: (1) sensitivity analysis, (2) classification of parameters into a hierarchy according to their sensitivity and the nature of their uncertainty, (3) building of alternative formulations, (4) calibration and (5) skill evaluation. The approach is illustrated here by reviewing the parameterization of the ISIS-Fish model of the anchovy fishery in the Bay of Biscay. By using this approach, it is possible to make a thorough assessment of lacking information (e.g. accessibility to fishing and adult mortality) and to identify the strengths and weaknesses of the model in the context of different hypotheses. When applied to the ISIS-Fish model, the results suggest higher egg and adult mortality than formerly estimated, as well as new estimates for the migration towards spawning areas. They show the reliability of the model in terms of correlations with observations and the need for further efforts to model purse seiner catches. The methodology proved to be a cost-efficient tool for objectively assessing applied model validity in cases where parameter values are a mix of literature, expert opinion and calibration. Highlights ► A five-step methodology is proposed to parameterize and validate a complex fishery model. ► It combines scalability of parameters, optimization procedure and model skill assessment. ► It is used to review the ISIS-Fish model of the anchovy fishery in the Bay of Biscay. ► It modifies our understanding of anchovy dynamics and evidences model strengths and weaknesses.

How larval traits of six flatfish species impact connectivity

“…Ideally, sensitivity analyses should consider all parameters together, which is not feasible in complex models requiring important computing resources. An alternative approach proposed by Lehuta et al (), which consists in first assessing the most critical parameters, then estimating uncertainties and processing to calibration, would be worth considering in larval transport models.…”

Integrated ecological–economic fisheries models—Evaluation, review and challenges for implementation

“…12 ISIS-FISH Mahevas & Pelletier, 2004;Pelletier et al, 2009;Drouineau, Mahévas, Pelletier, & Beliaeff, 2006;Drouineau, Mahévas, Bertignac, & Duplisea, 2010;Duplisea, 2010;Lehuta, Mahévas, Petitgas, & Pelletier, 2010;Rocklin, Pelletier, Mouillot, Tomasini, & Culioli, 2010;Lehuta, Mahévas, & Le Floc'h, 2013;Lehuta, Petitgas, et al, 2013;Lehuta, Holland, & Pershing, 2014;Lehuta, Vermard, & Marchal, 2015;Rochet & Rice, 2010;Marchal, Little, & Thebaud, 2011;Marchal, De Oliveira, Lorance, Baulier, & Pawlowski, 2013;Hussein et al, 2011a,b;Vermard et al, 2012;Gasche, Mahevas, & Marchal, 2013;Reecht et al, 2015. 13 BALTIC FLR-SMS Bastardie et al, 2009;Bastardie, Nielsen, & Kraus, 2010;Bastardie, Vinther, Nielsen, Ulrich, & Storr-Paulsen, 2010;Bastardie, Vinther, & Nielsen, 2012;Bastardie, Nielsen, & Vinther, 2015;Nielsen et al, 2011;Feekings et al, (submitted). Macher, Guyader, Talidec, & Bertignac, 2008;Macher et al, 2013;Merzéréaud, Biais, Lissardy, Bertignac, & Biseau, 2013;Merzéréaud et al, 2011;Simmonds et al, 2011;Raveau et al, 2012;Guillén et al, 2013;Guillén, Macher, Merzéréaud, Fifas, & Guyader, 2014;Guillén, Macher, Merzéréaud, Boncoeur, & Guyader, 2015;EU STECF, 2015a…”

A sequential approach to calibrate ecosystem models with multiple time series data

“…We focused on model dependency and time variability to categorize the parameters since these are two criteria which are usually considered to reduce the number of parameters to estimate. This preliminary parameters' classification can lead to fixing parameter values from other models or species/ecosystems or to ignoring time variability in the parameters (Lehuta et al 2013). However, in the present study, we have not considered other useful criteria such as sensitivity analysis, which has been used to reduce the number of parameters to be estimated (Megrey et al 2007, Dueri et al 2012, Lehuta et al 2013.…”

“…This preliminary parameters' classification can lead to fixing parameter values from other models or species/ecosystems or to ignoring time variability in the parameters (Lehuta et al 2013). However, in the present study, we have not considered other useful criteria such as sensitivity analysis, which has been used to reduce the number of parameters to be estimated (Megrey et al 2007, Dueri et al 2012, Lehuta et al 2013. Additionally, a successful calibration does not mean that a model is reliable (Gaume et al 1998), and a proper validation is always required, eventually providing information to improve the model and to revise the calibration (Walter andPronzato 1997, Jorgensen andBendoricchio 2001).…”