Parameter uncertainty drives important incongruities between simulated chlorophyll-a and phytoplankton functional group dynamics in a mechanistic management model
Introduction
Research organizations and government agencies conduct long-term water quality sampling to monitor threats posed by phytoplankton blooms, which are intensifying in magnitude, frequency, and duration, in many cases because of the impacts of human activities such as cultural eutrophication and climate change (Carey et al., 2012; Glibert et al., 2005; O'Neil et al., 2012; Paerl et al., 2011). Chlorophyll-a is classically measured as a proxy of phytoplankton biomass within these monitoring programs and used as a response variable in models designed to explain and predict blooms and their secondary effects (e.g., changes in dissolved oxygen). However, chlorophyll-a is an aggregate measure, and its dynamics may not fully reflect shifts in biomass of key functional groups (e.g., Ní Longphuirt et al., 2019). Additionally, changes in phytoplankton community composition and eco-physiological responses to changing environmental conditions can alter the relationship between chlorophyll-a and biomass (Jakobsen and Markager, 2016).
To capture phytoplankton community dynamics underlying changes in chlorophyll-a, scientists and engineers have developed models that simulate phytoplankton functional groups (PFGs), which are groups of phytoplankton with shared characteristics, such as morphological and eco-physiological traits (Kruk et al., 2002; C. S. Reynolds et al., 2002). In theory, these models quantify phytoplankton growth kinetics and temporal variations in the ecological processes underpinning phytoplankton community structure, thereby representing pelagic productivity dynamics more realistically as compared to models that solely simulate chlorophyll-a. In practice, PFG models can be difficult to parameterize and validate because the processes they simulate are contextualized by significant ecological and biological complexity (Arhonditsis and Brett, 2004; Rigosi et al., 2010; Robson, 2014; Shimoda and Arhonditsis, 2016). Consequently, PFG parameterization often occurs subjectively or through calibration (Frede Thingstad et al., 2010). These complexities result in PFG model algorithms and their parameters being largely uncertain. Many PFG model parameters, such as rates associated with growth kinetics, are not extensively measured or documented in existing literature (Anderson, 2005; Franks, 2009; Shimoda and Arhonditsis, 2016).
Although uncertainties propagate through direct and interactive effects to model outputs, it remains difficult to assess the degree to which parameter uncertainties impact PFG model simulation skill, in part because there are relatively few published PFG modeling studies that include quantified performance metrics. A recent review of PFG modeling analyses by Shimoda and Arhonditsis (2016) revealed that only around 30% of published studies applying PFG models within freshwater systems reported PFG goodness-of-fit measures. The lack of information on PFG performance may partly result from limitations in the availability of detailed and long-term observational PFG data (Shimoda and Arhonditsis, 2016). Among the PFG modeling studies that have presented observations to support simulated outputs, the performances are often poor. For example, Shimoda and Arhonditsis (2016) discovered that published PFG modeling studies in which cyanobacteria was included as a state variable (n = 68) simulatedcyanobacteria dynamics with a median model efficiency of 0.06 (calculated as , where = observations, = predictions, and = the mean of observations), a level of performance not ideal for supporting management and policymaking (Ritter and Muñoz-Carpena, 2013). Increasing demands on water managers to mitigate the effects of harmful algal blooms, such as those composed of toxic cyanobacteria, will be met by a growing need for PFG models that can support managers’ decision-making. Thus, it is important that efforts be made to improve PFG model structure and performance by taking a critical look at the processes and parameters contributing uncertainty to simulated PFG dynamics.
One approach that allows for systematic assessment of PFG model structure and performance is Global Sensitivity Analysis (GSA). As with any type of sensitivity analysis, GSA is used to quantify the degree to which model outputs are sensitive to changes in input factors (e.g., parameter values, boundary conditions, algorithm structures, etc.) (Saltelli et al., 2008). GSA contrasts with other sensitivity analysis approaches due to its capacity to assess nonlinear and interactive effects in high-dimensional models (Saltelli et al., 2004, 2008). Results produced from applications of GSA reveal the relative importance of model input factors. Model input factors that are found to dictate model performance should be addressed in order to ensure they are being assigned the most accurate values. Conversely, factors identified as unimportant could potentially be removed from the model, thereby helping to reduce model complexity and uncertainty (Jakeman et al., 2006; Muñoz-Carpena et al., 2007; Saltelli et al., 2004, 2008). In addition, GSA can help identify processes involving important factors that could be targeted for model improvement. Thus, by methodically informing ways to improve upon the algorithmic structures of PFG models, GSA can markedly increase their utility.
The overall goal of this study was to address the need for further quantitative scrutiny of state-of-the-art PFG models. Our primary objective was to quantify the global sensitivity of PFG model outputs to parameters related to phytoplankton growth and loss processes. To meet this objective, we applied GSA to the Corps of Engineers Integrated Compartment Water Quality Model (CE-QUAL-ICM), a highly-specified and state-of-the-art 3D mechanistic water quality and PFG model that couples with the Environmental Fluid Dynamics Code (EFDC), a 3D mechanistic hydrodynamic model. CE-QUAL-ICM is supported by the United States Environmental Protection Agency and used to inform the development of water quality targets for regulatory purposes (Cerco and Cole, 1995). Specifically, the GSA was used to assess CE-QUAL-ICM's sensitivity to a large number of input factors when simulating chlorophyll-a, cyanobacteria biomass, and eukaryotic phytoplankton biomass over an 8-year period in a flow-through lake located in Florida, USA. The availability of detailed and long-term phytoplankton monitoring data allowed for model goodness-of-fit to be quantified for all outputs and for application of GSA. Parameters were varied across realistic value ranges supported by a survey of the literature, thus making the findings regarding model sensitivity of broad interest. To our knowledge, this is the first application of GSA to a highly complex PFG model.
Section snippets
CE-QUAL-ICM
An application of CE-QUAL-ICM, a 3D mechanistic water quality and PFG model, was used for this study. CE-QUAL-ICM was developed by the U.S. Army Corps of Engineers (Cerco and Cole, 1995), and is described by the U.S. Environmental Protection Agency (EPA) as a model that can support nutrient target development (EPA, 2001). CE-QUAL-ICM couples with the Environmental Fluid Dynamics Code (EFDC), a 3D process-based hydrodynamic and transport model capable of simulating complex physical processes
Model efficiency
The NSE values produced from the 516 simulations for each of the considered outputs were largely negative, though this is unsurprising given that the range of parameter values considered may not have been well-specified for the studied system despite using realistic ranges of variation from the literature. The median NSE values for chlorophyll-a, cyanobacteria biomass, and eukaryote biomass were −3.0, −0.53, and −8.26 respectively; the maximum NSE values, respectively, were 0.36, 0.34, and
Discussion
To address the need for greater in-depth evaluation of mechanistic phytoplankton functional group (PFG) management models, global sensitivity analysis (GSA) was applied to evaluate the importance of 42 parameters on the outputs of an application of CE-QUAL-ICM, a state-of-the-art process-based PFG and biogeochemical model used to inform the development of water quality targets for the management of algal blooms in fresh and estuarine waters. Notably, this study focused on parameters used to
Conclusions
In the present study, we identified PFG model parameters that drive uncertainty in modeled chlorophyll-a, cyanobacteria biomass, and eukaryotic phytoplankton biomass in a large, shallow, subtropical, flow-through lake, and identified opportunities for PFG model improvement. In particular, the considered model outputs were sensitive to chlorophyll-to-carbon ratios, P uptake parameters, basal phytoplankton metabolism rates, and phytoplankton production under optimal growth conditions. Among these
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements and Data
The authors report no conflicts of interest, and thank Peter Sucsy and John Hendrickson of the St. Johns River Water Management District for providing access to information. NGN is supported by the USDA National Institute of Food and Agriculture, Hatch project 1016068, and RMC by Hatch project FLA-ABE-005556. Additionally, this material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No. DGE-0802270, USDA NIFA Hatch Project 1011481, and
References (71)
- et al.
An effective screening design for sensitivity analysis of large models
Environ. Model. Software
(2007) - et al.
Eco-physiological adaptations that favour freshwater cyanobacteria in a changing climate
Water Res.
(2012) - et al.
A practical application of Droop nutrient kinetics (WR 1883)
Water Res.
(2004) - et al.
Stepwise building of plankton functional type (PFT) models: a feasible route to complex models?
Prog. Oceanogr.
(2010) Optical water quality of a blackwater river estuary: the Lower St. Johns River, Florida, USA
Estuar. Coast Shelf Sci.
(2005)- et al.
Ten iterative steps in development and evaluation of environmental models
Environ. Model. Software
(2006) - et al.
A multi-criteria trajectory-based parameter sampling strategy for the screening method of elementary effects
Environ. Model. Software
(2015) - et al.
Bayesian modelling of algal mass occurrences-using adaptive MCMC methods with a lake water quality model
Environ. Model. Software
(2007) - et al.
River flow forecasting through conceptual models Part I-a discussion of principles*
J. Hydrol.
(1970) - et al.
The rise of harmful cyanobacteria blooms: the potential roles of eutrophication and climate change
Harmful Algae
(2012)
Controlling harmful cyanobacterial blooms in a world experiencing anthropogenic and climatic-induced change
Sci. Total Environ.
Performance evaluation of hydrological models: statistical significance for reducing subjectivity in goodness-of-fit assessments
J. Hydrol.
When do aquatic systems models provide useful predictions, what is changing, and what is next?
Environ. Model. Software
Towards evidence-based parameter values and priors for aquatic ecosystem modelling
Environ. Model. Software
An improved sampling strategy based on trajectory design for application of the Morris method to systems with many input factors
Environ. Model. Software
Phytoplankton functional type modelling: running before we can walk? A critical evaluation of the current state of knowledge
Ecol. Model.
Optimizing the complexity of phytoplankton functional group modeling: an allometric approach
Ecol. Inf.
Nitrogen chlorosis in blue-green algae
Arch. Mikrobiol.
Plankton functional type modelling: running before we can walk?
J. Plankton Res.
Evaluation of the current state of mechanistic aquatic biogeochemical modeling
Mar. Ecol. Prog. Ser.
Uniqueness of place and process representations in hydrological modelling
Hydrol. Earth Syst. Sci.
C-phycocyanin as a storage protein in the blue-green alga Spirulina platensis
Arch. Microbiol.
User's Guide to the CE-QUAL-ICM Three-Dimensional Eutrophication Model, Release Version 1.0
Twenty-one-year simulation of Chesapeake Bay water quality using the CE-QUAL-ICM eutrophication model
J. Am. Water Resour. Assoc.
An effective parameter screening strategy for high dimensional models
Cyanobacteria as biological drivers of lake nitrogen and phosphorus cycling
Ecosphere
A Database of Marine Phytoplankton Abundance, Biomass and Species Composition in Australian Waters
Aquatic Nitrogen Fixation: Patterns, Rates and Controls in a Shallow, Subtropical Lake
Nutrient Criteria Technical Guidance Manual: Estuarine and Coastal Marine Waters
Recommended human health recreational ambient water quality criteria or swimming advisories for microcystins and cylindrospermopsin
Planktonic ecosystem models: perplexing parameterizations and a failure to fail
J. Plankton Res.
The global, complex phenomena of harmful algal blooms
Oceanography
Revisiting the basis of sensitivity analysis for dynamical earth system models
Water Resour. Res.
Statistical procedure for evaluating hydrologic/water quality models
Trans. ASABE
A Three-Dimensional Environmental Fluid Dynamics Computer Code: Theoretical and Computational Aspects
Cited by (12)
Quantifying the contribution of external loadings and internal hydrodynamic processes to the water quality of Lake Okeechobee
2023, Science of the Total EnvironmentDevelopment of a sensitivity analysis framework for aquatic biogeochemical models using machine learning
2023, Ecological InformaticsClimate change impacts on water quantity and quality of a watershed-lake system using a spatially integrated modeling framework in the Kissimmee River – Lake Okeechobee system
2023, Journal of Hydrology: Regional StudiesSensitivity and identifiability analyses of parameters for water quality modeling of subtropical reservoirs
2021, Ecological ModellingCitation Excerpt :The insights provided by the SA concern both the magnitude of the variations and the direction of change (Borgonovo and Plischke, 2016), which can help to understand the model's intrinsic behavior by identifying the governing processes in the system and enhancing model formulations. For instance, the SA can be conducted for several purposes in ecological modeling, including for elucidating which parameters require additional research or are insignificant (Nelson et al., 2020), the consequences from changing the dominant controls of a system (Gal et al., 2009), and the correlation between input parameters and lake characteristics, such as residence time, meteorological conditions and trophic status (Bruce et al., 2018). Finding the global optimal model and, by implication, the true representation of the system, is a difficult task for complex models in face of the non-uniqueness or equifinality of parameter values, i.e., the same output can be reached by multiple optimal parameter values, which give equally acceptable fits to observational data (Beven, 2006).
Elucidating controls on cyanobacteria bloom timing and intensity via Bayesian mechanistic modeling
2021, Science of the Total EnvironmentCitation Excerpt :Additionally, Ml,T represents cyanobacterial dominance parsimoniously without explicitly parameterizing several phytoplankton groups. While ecologically complex models can be used to explore mechanisms in more detail, they can be challenging to parameterize and often have limited predictive capabilities (Nelson et al., 2020; Shimoda and Arhonditsis, 2016). Our final model (Ml,T) explains about 40% of the chl and DRP variability and over 80% of DIN variability (Table 2), on par with the performance of other mechanistic eutrophication models (Arhonditsis and Brett, 2004).
Exploring nutrient and light limitation of algal production in a shallow turbid reservoir
2021, Environmental PollutionCitation Excerpt :Empirical/statistical models assess the significance of various biophysical variables (e.g., nutrients, temperature, wind, vertical light attenuation) related to algal accumulation (Beaulieu et al., 2013; Havens et al., 2003; Rigosi et al., 2014). In contrast, process-based mechanistic models are less reliant on observational data but can be challenging to parameterize (Glibert et al., 2010; Nelson et al., 2020). In addition, model uncertainty and reliability assessment (e.g., validation) are important but often omitted considerations when assessing a model’s capacity for forecasting and informing management (Arhonditsis and Brett, 2004).