2017: Post-doc, Ifremer Boulogne-sur-Mer

Contact : Paul Marchal (Paul.Marchal@ifremer.fr)

Development of a formal procedure to optimize the calibration and strengthen the robustness of a complex ecosystem model: Atlantis

The evaluation and implementation of ecosystem-based management requires the support of mature ecosystem models. Atlantis is one such model, for which Ifremer has developed and calibrated an Eastern English Channel  application (Atlantis-EEC). One major limitation of all complex ecosystem models, and of Atlantis-EEC in particular, is the lack of a formal mathematical protocol to optimize/calibrate the huge amount of parameters and also the poor knowledge of how parameters’ uncertainty is actually driving the model’s response. This project will develop methods to formalize the calibration procedure and to assess the sensitivity of the Eastern English Channel Atlantis ecosystem model application to parameter uncertainty. Successfully addressing that dual challenge, in collaboration with mathematicians, will boost Atlantis-EEC to the confines of academia, stepping inside advice-giving territory. The methodological developments brought about by this project will thus benefit the ecosystem modelers’ community, as well as managers in charge of mitigating the effects of mankind on marine ecosystems.

Keywords: ecosystem model, Atlantis, calibration, black box optimization, sensitivity analysis, large dimension

Applicants should have a strong mathematical background and hold a PhD in that discipline. Applicants should have established expertise in applied mathematics (statistics, optimisation, machine learning), computing, engineering, and should demonstrate their capacity in scientific programming (e.g., Java, C++, R, SAS,…). Some basic knowledge in marine ecology would be appreciated.

•    Context
Drawing lessons from past failures and successes (Daw and Gray, 2005; Hilborn, 2004), fisheries science has gradually been moving from traditional single-species considerations (Ludwig, 2002; Rosenberg, 2002) towards a comprehensive, ecosystem-based management (EBM) approach (Botsford et al., 1997; Browman and Stergiou, 2004). Over the last few decades, interest in ecosystem modelling has thus grown, and several “end-to-end” models have been developed to emulate the dynamics of marine ecosystems (Plagányi, 2007). One of the first ecosystem models that included a large set of species groups and trophic connections was Ecopath with Ecosim (EwE) (Christensen and Walters, 2004). Recently, other models have emerged focusing on the spatial dynamics of fish and
fisheries distributions, e.g., ISIS-Fish (Mahévas and Pelletier, 2004), and/or taking into account nutrient cycling, food web dynamics, and environmental variability, e.g., OSMOSE (Shin and Cury, 2004; Travers et al., 2009), APECOSM (Maury et al., 2007), NEMURO.FISH (Kishi et al., 2011), SEAPODYM (Lehodey et al., 2008) and Atlantis (Fulton et al., 2011). Atlantis is currently one of the most comprehensive and up-to-date ecosystem model (Plagányi, 2007), and it has successfully been applied to a number of case studies worldwide (Ainsworth et al., 2012; Kaplan et al., 2012; Savina et al., 2013), including the Eastern English Channel (Atlantis-EEC; Girardin et al., 2016). One major limitation of all complex ecosystem models mentioned above, and of the Atlantis-EEC application in particular, is the lack of a formal mathematical protocol to optimize/calibrate the huge amount of parameters and also the poor knowledge of how parameters’ uncertainty is actually driving the model’s response.
The overarching objective of our postdoctoral project will be to address this double challenge, building on the existing Atlantis-EEC application.

•    Methods, expected results, possible collaborations
In a first step, an optimization framework will be developed to calibrate Atlantis-EEC. Given the numerous parameters to estimate, the
project will consider methods aiming at reducing the complexity by either identifying the most sensitive ones (Morris method, group
screening), or making explicit the relationships between variables. Different calibration methods will then be explored, including some that have already been used for models of similar complexity, e.g. evolutionary strategies as already used for Osmose (Duboz et al, 2010), Bayesian approaches (Kennedy and O’Hagan, 2001; Andrianakis et al., 2015). The adjustment of metamodels is also envisaged to get around the computational costs constraints. A key task here will be the definition of the objective function, which should account for multiple objectives and for the heterogeneous quality of available data. In a second step, an analysis of uncertainty and/or sensitivity of the Atlantis-EEC model will also be carried out to highlight the key uncertain parameters for which a refined tuning would be necessary. An important step forward would hence be to identify and analyse the main sources of uncertainty in our model. Due to the numerous parameters considered in Atlantis, the application of sensitivity/uncertainty propagation analyses would require using, (i) meta-models (Grace et al., 2010), (ii) experimental plans to reduce the number of simulations based on, e.g., the Morris methods or optimized factorial designs (Lehuta et al., 2013), Latin Hypercube Sampling (Gasche et al., 2013), (iii) sobol indices (Sobol’, 2001), to explore the uncertainty in our application or, (iv) other techniques such as adaptive screening already tested on Atlantis (Pantus, 2007).
The first benefit of this project will be to reduce the computational burden and more generally the time needed to calibrate Atlantis-EEC, but also any Atlantis applications developed around the world.
An additional outcome of the project will then be an identification of the most important parameters that should be best informed as a matter of priority, and the resulting impact on data collection. This is particularly important for Atlantis but also for any complex model where the priority is often given to parameters for which data is abundant, although these are not necessarily those having the largest influence on the model response.
Consolidating the estimation procedure by making it more objective and transparent will also contribute to the general credibility of the
model. The science developed within this project will thus allow advancing Atlantis-EEC beyond academic confines, and eventually stepping inside advice-giving territory. Atlantis-EEC could thus be applied to evaluate the ecosystem effects of scenarios combining climate change, human activities (including fishing and non-fishing sectors), and of their management.
Several contacts have already been made within the Atlantis community, and several colleagues from the NOAA (USA) and CSIRO (Australia) have already expressed interest in collaborating on that particular topic. Importantly also, because the successful achievement of that project will require strong mathematical developments, guidance will be provided by Victor Picheny (INRA, Unité de Mathématiques et Informatique Appliquées de Toulouse), and other colleagues from the MEXICO network (Méthodes pour l’EXploration Informatique des modèles Complexes, http://reseau-mexico.fr/), who will also help to identify potential postdoctoral candidates with the appropriate numerate background.

Andrianakis, I., Vernon, I.R., McCreesh, N., McKinley, T.J., Oakley, J.E., Nsubuga, R.N., Goldstein, M., and White, R.G. (2015). Bayesian history matching of complex infectious disease models using emulation: a tutorial and a case study on HIV in Uganda. PLoS Comput. Biol. 11, e1003968.
Ainsworth, C.H., Morzaria-Luna, H., Kaplan, I.C., Levin, P.S., Fulton, E.A., Cudney-Bueno, R., Turk-Boyer, P., Torre, J., Danemann, G.D., and Pfister, T. (2012). Effective ecosystem-based management must encourage regulatory compliance: A Gulf of California case study. Mar. Policy 36, 1275–1283.
Botsford, L.W., Castilla, J.C., and Peterson, C.H. (1997). The management of fisheries and marine ecosystems. Science 277, 509–515.
Browman, H.I., and Stergiou, K.I. (2004). Perspectives on ecosystem-based approaches to the management of marine resources. Mar.
Ecol. Prog. Ser. 274, 269–303.
Christensen, V., and Walters, C.J. (2004). Ecopath with Ecosim: methods, capabilities and limitations. Ecol. Model. 172, 109–139.
Duboz, R, Versmisse D, Travers, M, Ramat, E, Shin, YJ, 2010. Application of an evolutionary algorithm to the inverse parameter estimation of an individual-based model. Ecol. Mod. 221: 840-849
Daw, T., and Gray, T. (2005). Fisheries science and sustainability in international policy: a study of failure in the European Union’s Common Fisheries Policy. Mar. Policy 29, 189–197.
Fulton, EA, Link, JS, Kaplan, IC, Savina-Rolland, M, Johnson, P, Ainsworth, C, Horne, P, Gorton, R, Gamble, RJ, Smith, ADM, Smith, DC, 2011. Lessons in modelling and management of marine ecosystems: the Atlantis experience. Fish and Fisheries 12 (2): 171-188.
Gasche, L., Mahévas, S., and Marchal, P. (2013). Supporting fisheries management by means of complex models: Can we point out isles of robustness in a sea of uncertainty? PLoS ONE 8, e77566.
Girardin, R. (2015). Ecosystem and fishers’ behaviour modelling: two crucial and interacting approaches to support ecosystem based fisheries management in the Eastern English Channel. Université Lille 1, 2015. English.
Girardin, R., Fulton, E.A., Lehuta, S., Savina-Rolland, M., Thébaud, O., Travers-Trolet, M., Vermard, Y., and Marchal, P. (2016) Identification of the main processes underlying ecosystem functioning in the Eastern English Channel, with a focus on flatfish species, as revealed through the application of the Atlantis end-to-end model. Estuarine and Coastal Shelf Science, In Press.
Grace, J.B., Anderson, T.M., Olff, H., and Scheiner, S.M. (2010). On the specification of structural equation models for ecological systems. Ecol. Monogr. 80, 67–87.
Hilborn, R. (2004). Ecosystem-based fisheries management: the carrot or the stick? Mar. Ecol. Prog. Ser. 274, 275–278.
Kaplan, I.C., Horne, P.J., and Levin, P.S. (2012). Screening California Current fishery management scenarios using the Atlantis end-to-end ecosystem model. Prog. Oceanogr. 102, 5–18.
Kennedy, M.C., and O’Hagan, A. (2001). Bayesian calibration of computer models. Journal of the Royal Statistical Society B, 63, 425-464.

Kishi, M.J., Ito, S., Megrey, B.A., Rose, K.A., and Werner, F.E. (2011). A review of the NEMURO and NEMURO.FISH models and their application to marine ecosystem investigations. J. Oceanogr. 67, 3–16.
Lehodey, P., Senina, I., and Murtugudde, R. (2008). A spatial ecosystem and populations dynamics model (SEAPODYM) – Modeling of tuna and tuna-like populations. Prog. Oceanogr. 78, 304–318.
Lehuta, S., Mahévas, S., Le Floc’h, P., and Petitgas, P. (2013). A simulation-based approach to assess sensitivity and robustness of
fisheries management indicators for the pelagic fishery in the Bay of Biscay. Can. J. Fish. Aquat. Sci. 70, 1741–1756.
Ludwig, D. (2002). A Quantitative Precautionary Approach. Bull. Mar.Sci. 70, 485–497.
Rosenberg, A.A. (2002). The precautionary approach in application from a manager’s perspective. Bull. Mar. Sci. 70, 577–588.
Mahévas, S., and Pelletier, D. (2004). ISIS-Fish, a generic and spatially explicit simulation tool for evaluating the impact of management measures on fisheries dynamics. Ecol. Model. 171, 65–84.
Pantus, F.J. (2007). Sensitivity analysis for complex ecosystem models. PhD Thesis Sch. Phys. Sci. Univ. Qld.
Plagányi, É.E. (2007). Models for an ecosystem approah to fisheries. FAO Fish. Tech. Pap. No. 477. Rome, FAO. 2007, 108 pp.
Maury, O., Faugeras, B., Shin, Y.-J., Poggiale, J.-C., Ari, T.B., and Marsac, F. (2007). Modeling environmental effects on the size-structured energy flow through marine ecosystems. Part 1: The model. Prog. Oceanogr. 74, 479–499.
Savina, M., Forrest, R.E., Fulton, E.A., and Condie, S.A. (2013). Ecological effects of trawling fisheries on the eastern Australian
continental shelf: a modelling study. Mar. Freshw. Res. 64, 1068–1086.
Shin, Y.-J., and Cury, P. (2004). Using an individual-based model of fish assemblages to study the response of size spectra to changes in fishing. Can. J. Fish. Aquat. Sci. 61, 414–431.
Sobol’, I.M. (2001). Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates. Math. Comput.
Simul. 55, 271–280.
Travers, M., Shin, Y.-J., Jennings, S., Machu, E., Huggett, J.A., Field, J.G., and Cury, P.M. (2009). Two-way coupling versus one-way forcing of plankton and fish models to predict ecosystem changes in the Benguela. Ecol. Model. 220, 3089–3099.