Towards evidence-based parameter values and priors for aquatic ecosystem modelling

https://doi.org/10.1016/j.envsoft.2017.11.018Get rights and content

Highlights

  • Current practices in parameterising aquatic ecosystem models can be improved.

  • We present a new resource to inform specification of parameter values.

  • An online tool helps modellers find, filter & transform process rate and trait data.

  • Parameter distributions are presented in a probabilistic framework.

  • This will improve connection between observational science and modelling.

Abstract

Mechanistic models rely on specification of parameters representing biophysical traits and process rates such as phytoplankton, zooplankton and seagrass growth and respiration rates, organism sizes, stoichiometry, light, temperature and nutrient responses, nutrient-specific excretion rates and detrital stoichiometry and decay rates. Choosing suitable values for these parameters is difficult. Current practise is problematic. This paper presents a resource designed to facilitate an evidence-based approach to parameterisation of aquatic ecosystem models. An online tool is provided which collates relevant, published biological trait and biogeochemical rate observations from many sources and allows users to explore, filter and convert these data in a consistent, reproducible way, to find parameter values and calculate probability distributions. Using this information within a traditional or Bayesian paradigm should provide improved understanding of the uncertainty and predictive capacity of aquatic ecosystem models and provide insight into current sources of structural error in models.

Section snippets

Software and data availability

Software and data described in this paper are available from the Parameter Library Exploration Interface at http://shiny.csiro.au/CDM/parameterlibrary/latest/

Current state of the art

Parameterisation of aquatic ecosystem models has historically been a difficult and labour-intensive task. Parameter values are typically required for a range of model components, defining traits and responses of phytoplankton, zooplankton and other aquatic biota, characteristics of sediments and detrital material, physical and chemical process rates, and, more recently, bacterial traits and processes. Although mathematical methods have long been available to automate parameter estimation and

Collation of parameter values

There have been several efforts to collate and tabulate parameter values from across the literature. Some of these have been broadly focused on the most commonly used parameters in aquatic ecosystem models (e.g. Bowie et al., 1985) or eco-chemistry (Jorgensen et al., 2000), while others have comprehensively reviewed the literature for a narrower range of parameters relating to a particular component or process of an aquatic ecosystem (e.g. phytoplankton: Edwards et al., 2015, Eppley, 1985;

Online interface

We have implemented a preliminary online user interface to allow further exploration of the parameter library, which can be found at http://shiny.csiro.au/CDM/parameterlibrary/latest/. A screenshot is presented as Fig. 1. In addition to calculating and visualising probability distributions (gamma, normal, log-normal or uniform) fitted to each parameter, the interface allows users to filter datasets where relevant information exists in the database. In most cases, there is an option to restrict

Summary of parameter data

Table 1 summarises the data included in the parameter library as at May 2017. We anticipate that we will add to this resource over time, drawing on community contributions.

Making effective use of parameter prior distributions

The simplest approach to using the information presented via this online resource may be to select the subsample of data most relevant to the application at hand, plot a distribution, and calibrate within the range suggested by the 5th and 95th percentiles. After calibration, the final parameter set can be compared with the expected prior distributions. If more than a few calibrated parameter values fall close to the outer limits of the expected distribution, this may warrant further

Future development

The beta version of the parameter database presented here and currently available online is a static resource, manually updated. It is our intention (subject to funding) to maintain and update this to provide a modern data service. This will include not only expansion of the database itself and improvement of the online user interface, but also improvement of the underlying data structure, consideration of issues of semantics, inter-operability, provenance and data delivery, and provision of an

Conclusion

By presenting physiological trait and biogeochemical process rate information through an online tool that not only synthesises observations from multiple sources but also supports consistent and reproducable processing of these data (for instance, by applying a consistent temperature correction function to observed metabolic rates), we facilitate more evidence-based Bayesian parameterisation of aquatic systems models as well as improved model evaluation and development processes.

Acknowledgements

This work was funded in part through the eReefs project, a public-private collaboration between Australia's leading operational and scientific research agencies, government, and corporate Australia. JB and LG contributed to this work during internships at CSIRO Land and Water, completed as part of their studies. Thanks to Philip Balding for assisting with phytoplankton data entry.

References (44)

  • B.J. Robson et al.

    Three-dimensional modelling of a Microcystis bloom event in the Swan river estuary, western Australia

    Ecol. Model.

    (2004)
  • Y. Shimoda et al.

    Phytoplankton functional type modelling: running before we can walk? A critical evaluation of the current state of knowledge

    Ecol. Model.

    (2016)
  • J. Skerratt et al.

    Use of a high resolution 3D fully coupled hydrodynamic, sediment and biogeochemical model to understand estuarine nutrient dynamics under various water quality scenarios

    Ocean Coast. Manag.

    (2013)
  • W.T. Zhang et al.

    A Bayesian hierarchical framework for calibrating aquatic biogeochemical models

    Ecol. Model.

    (2009)
  • C. Andrieu et al.

    Particle Markov chain Monte Carlo methods

    J. R. Stat. Soc. Ser. B Stat. Methodol.

    (2010)
  • L.M. Berliner

    Physical-statistical modeling in geophysics

    J. Geophys. Res. Atmos.

    (2003)
  • G.L. Bowie et al.

    Rates, Constants, and Kinetics Formulations in Surface Water Quality Modeling

    (1985)
  • J. Bruggeman

    A phylogenetic approach to the estimation of phytoplankton traits

    J. Phycol.

    (2011)
  • N.A.C. Cressie et al.

    Statistics for Spatio-temporal Data

    (2011)
  • M. Dowd et al.

    A statistical overview and perspectives on data assimilation for marine biogeochemical models

    Environmetrics

    (2014)
  • M.R. Droop et al.

    Light and nutrient status of algal cells

    J. Mar. Biol. Assoc. U. K.

    (1982)
  • K.F. Edwards et al.

    Allometric scaling and taxonomic variation in nutrient utilization traits and maximum growth rate of phytoplankton

    Limnol. Oceanogr.

    (2012)
  • Cited by (28)

    • Can a simple water quality model effectively estimate runoff-driven nutrient loads to estuarine systems? A national-scale comparison of STEPLgrid and SPARROW

      2022, Environmental Modelling and Software
      Citation Excerpt :

      A recent review also confirmed the predominance of these water quality models and showed that 44% of articles of 3,282 reviewed used SWAT (Fu et al., 2019). These commonly-used mechanistic and highly-specified models are generally viewed as being state-of-the-art given that they represent many processes known to drive water quality dynamics (Robson, 2014; Robson et al., 2018). Complex process-based models offer insight into the roles of key environmental processes in driving nutrient pollution, and help to evaluate the effectiveness of proposed conservation management strategies and policies (National Research Council, 2007).

    View all citing articles on Scopus
    View full text