Species distribution models grounded in ecological theory for decision support in river management
Introduction
The health of freshwater ecosystems is deteriorating worldwide (Carpenter et al., 2011, Vörösmarty et al., 2010). They are under many pressures, including for example hydrological alteration, eutrophication, organic pollution, etc. (Allan, 2004, Carpenter et al., 2011, Malaj et al., 2014). River managers are faced with the challenge of restoring these ecosystems with limited budgets and conflicting demands. It is thus important to understand which stressors should be addressed first when planning river restoration. In light of this, several efforts have been made for the development of decision support tools for river basin managers (Holguin-Gonzalez et al., 2014, Holzkämper et al., 2012, Klauer et al., 2012, Shaw et al., 2012, Shuker et al., 2012, Turak et al., 2011, Volk et al., 2010).
In the past, the focus often remained on process-based simulation models that simulate the effect of diffuse and point source pollutions on the physical–chemical water quality of the rivers (Chapra et al., 2004, Cools et al., 2011). Since the introduction of the European Water Framework Directive (WFD), more effort has been put into the development of ecological models (Everaert et al., 2010, Feio and Poquet, 2011). Often these ecological models are no species distribution models, but rather models related to the ecological quality indices (Klauer et al., 2012, Mouton et al., 2009b, Van Der Most et al., 2006). Unfortunately this approach fails to deliver a thorough ecological insight and it is difficult to incorporate any ecosystem dynamics in the stressor analysis (Stephens et al., 2015). Furthermore, hydromorphology is rarely considered in these efforts, since limited data were available up until recently (Vaughan and Ormerod, 2010, Verdonschot, 2009).
Species Distribution Models (SDMs) are a way forward to evaluate the effect of changing abiotic variables on the ecological status of a river more explicitly (Domisch et al., 2013, Funk et al., 2013). SDMs are often used as a tool to assess impacts of environmental stressors on the species community and therefore the health of the ecological system. The aim of SDMs is to predict the distribution of a single species in a study area given a set of factors constraining the presence of the individuals (Guisan and Thuiller, 2005). Typically, these are empirical models which relate field observations to environmental predictor variables, based on statistically or theoretically derived response surfaces (Guisan and Zimmermann, 2000). In recent years SDMs have gained predictive power. Not only novel statistical techniques related to machine learning have improved SDM performance (Elith and Graham, 2009, Elith et al., 2006, Van Echelpoel et al., 2015), there are also several applications for individual species where conceptual approaches have been successful. An important field where advances have been made is the inclusion of species migration in the models (Boets et al., 2014, Dedecker et al., 2007, Dedecker et al., 2006, Pauwels et al., 2013, Radinger et al., 2014).
Still, it seems that the application of SDMs in river basin management is limited (Funk et al., 2013, Marsili-Libelli et al., 2013). The reasons for this are threefold. First, the advances in predictive accuracy for one species cannot always be translated to many species at once. Since ecological goals often are defined as a suit of species, for example for WFD applications, having techniques that are applicable for many species is important. Data availability is another limiting factor. Model complexity can introduce many parameters that are hard to estimate, either because the dataset is too small or because the dataset is not suitable for testing the model structures (Guisan and Thuiller, 2005). This means that, for example, interactions between variables often are not included, since testing these interactions increases the parameter space considerably. Third, the usefulness of a decision support system in environmental management also depends on its deemed validity by stakeholders (Junier and Mostert, 2014, Van der Molen and Boers, 2002, Voinov and Bousquet, 2010, Volk et al., 2010). Flexible modelling systems and the inclusion of expert knowledge can increase the trust of end users in the models, as this allows clear communication of assumptions made in the model development (Guisan and Thuiller, 2005, Krueger et al., 2012, Voinov and Bousquet, 2010). To be acceptable for stakeholders, SDMs should consider appropriate spatial scales (Austin, 2002, Jepsen et al., 2005), have a clear indication of sources of uncertainty (Beale and Lennon, 2012, Van der Lee et al., 2006) and allow for the involvement of stakeholders during model development (Voinov and Bousquet, 2010, Volk et al., 2010).
Grounding SDMs in ecological concepts, such as the Spatially Explicit Species Assemblage Modelling (SESAM) concept as presented by Guisan and Rahbek (2011) could overcome some of these hurdles. Not only could such approaches advance scientific understanding (Austin, 2007), they also provide a pragmatic method for the development of SDMs for many species at once and allow the inclusion of expert knowledge.
We have developed and implemented a model approach to predict aquatic macroinvertebrate assemblages based on the SESAM concept using both data and expert knowledge. The approach is grounded in two ecological theories: the niche theory (Hirzel and Le Lay, 2008), specifically the niche characteristics, and the landscape filter theory (Poff, 1997). In this paper, we present the model approach, the model development process and its implementation for aquatic macroinvertebrates in Flanders. We have implemented the method for many species simultaneously and have selected the most optimal model structure based on the overall highest performance for all species. The method is validated in terms of its accuracy by using statistical criteria and its ecological validity by using expert knowledge. Advantages and drawbacks of the used methodology for freshwater ecosystem management are assessed in the discussion.
Section snippets
Overview of model development process
The research presented in this paper existed of 6 steps (Fig. 1). In a first step, the general model concept was developed based on general ecological theory and stakeholder consultation (Section 2.2). Second, all Flemish monitoring datasets of macroinvertebrate species and abiotic variables were combined and explored (Section 2.3). Stakeholder needs were taken into account for the variable selection and the data preparation. In the third step, the method for the construction of the individual
Construction of habitat suitability indices
Habitat suitability indices (HSI) were constructed for 92 taxa, considering the restriction of 40 data points. The integration of the limnodata-dataset with the Flemish response curves for the continuous variables resulted in a preliminary confidence level for each model (Fig. 4). For 13 species, little to no matching knowledge was available in the limnodata-dataset. For four variables (COD, T, TSS and NO2) the limnodata-dataset could not offer transferable knowledge for the Flemish rivers. In
Discussion
Generally all models show a high agreement with existing knowledge on species preferences. The species responses as described by the habitat suitability indices (HSI) are similar to those in the limnodata-dataset and the general ecological preferences as provided by Verberk and Nijboer (2012). Furthermore, the selection of limiting variables concur with existing ecological hypotheses. For example, hydromorphology is only limiting when considering the interaction between variables, which
Acknowledgments
Elina Bennetsen is supported by a Bijzonder Onderzoeksfonds (BOF) project (B/09924/02). Sacha Gobeyn is supported by a BOF project related to the Ecuador Biodiversity Network of the Vlaamse Interuniversitaire Raad—Universitaire Ontwikkelingssamenwerking (VLIR-UOS). The authors would like to acknowledge the Flemish Environment Agency (VMM) who commissioned the study on which this research was based.
References (70)
Spatial prediction of species distribution: an interface between ecological theory and statistical modelling
Ecol. Model.
(2002)Species distribution models and ecological theory: a critical assessment and some possible new approaches
Ecol. Model.
(2007)- et al.
Combining data-driven methods and lab studies to analyse the ecology of Dikerogammarus villosus
Ecol. Inf.
(2010) - et al.
Coupling a hydrological water quality model and an economic optimization model to set up a cost-effective emission reduction scenario for nitrogen
Environ. Model. Softw.
(2011) - et al.
Development of migration models for macroinvertebrates in the Zwalm river basin (Flanders, Belgium) as tools for restoration management
Ecol. Model.
(2007) - et al.
Choice of study area and predictors affect habitat suitability projections, but not the performance of species distribution models of stream biota
Ecol. Model.
(2013) - et al.
An applicability index for reliable and applicable decision trees in water quality modelling
Ecological Informatics
(2016) - et al.
Ecological niche models for the evaluation of management options in an urban floodplain—conservation vs. restoration purposes
Environ. Sci. Policy
(2013) - et al.
Multimetric Macroinvertebrate Index Flanders (MMIF) for biological assessment of rivers and lakes in Flanders (Belgium)
Limnologica
(2010) - et al.
Predictive habitat distribution models in ecology
Ecol. Model.
(2000)
Bringing diverse knowledge sources together—a meta-model for supporting integrated catchment management
J. Environ. Manage.
Evaluating the effect of corridors and landscape heterogeneity on dispersal probability: a comparison of three spatially explicit model approaches
Ecol. Model.
A decision support system for the implementation of the Water Framework Directive in the Netherlands: process, validity and useful information
Environ. Sci. Policy
Limiting factors and thresholds for macroinvertebrate assemblages in European rivers: empirical evidence from three datasets on water quality, catchment urbanization, and river restoration
Ecol. Indic.
The role of expert opinion in environmental modelling
Environ. Model. Softw.
A new instream flow assessment method based on fuzzy habitat suitability and large scale river modelling
Environ. Model. Softw.
Knowledge-based versus data-driven fuzzy habitat suitability models for river management
Environ. Model. Softw.
Modelling a pike (Esox lucius) population in a lowland river using a cellular automaton
Ecol. Inf.
FIDIMO—a free and open source GIS based dispersal model for riverine fish
Ecol. Inf.
Does it help? Testing the usefulness of a tool to aid Integrated Catchment Management
Procedia Environ. Sci.
Species distribution models for sustainable ecosystem management
Modelling with stakeholders
Environ. Model. Softw.
Distinguishing the effects of habitat degradation and pesticide stress on benthic invertebrates using stressor-specific metrics
Sci. Total Environ.
Landscapes and riverscapes: the influence of land use on stream ecosystems
Annu. Rev. Ecol. Evol. Syst.
Incorporating uncertainty in predictive species distribution modelling
Phil. Trans. R. Soc. B.
Modelling everything everywhere: a new approach to decision-making for water management under uncertainty
Freshwater Biol.
Using an integrated model approach for risk assessment of the “killer shrimp” Dikerogammarus villosus
River Res. Appl.
State of the world's freshwater ecosystems: physical, chemical, and biological changes
Annu. Rev. Environ. Resour.
QUAL2K: A Modeling Framework for Simulating River and Stream Water Quality, Version 1.3: Documentation and Users Manual
Effects of physico-chemistry, land use and hydromorphology on three riverine organism groups: a comparative analysis with monitoring data from Germany and Austria
Hydrobiologia
Method for biological quality assessment of watercourses in Belgium
Hydrobiologia
Development of an in-stream migration model for Gammarus pulex L.(Crustacea, Amphipoda) as a tool in river restoration management
Aquat. Ecol.
Do they? How do they? WHY do they differ? On finding reasons for differing performances of species distribution models
Ecography
Novel methods improve prediction of species’ distributions from occurrence data
Ecography
Development of data-driven models for the assessment of macroinvertebrates in rivers in Flanders
Cited by (27)
A large flood resets riverine morphology, improves connectivity and enhances habitats of a regulated river
2024, Science of the Total EnvironmentAllocation of ecological water rights considering ecological networks in arid watersheds: A framework and case study of Tarim River basin
2022, Agricultural Water ManagementCitation Excerpt :Water is one of the most crucial factors maintaining the stability and integrity of dryland ecosystems (Bennetsen et al., 2016; Boithias et al., 2016; Gu et al., 2016).
Towards a food web based control strategy to mitigate an amphibian panzootic in agricultural landscapes
2020, Global Ecology and ConservationA conceptual model for simulating responses of freshwater macroinvertebrate assemblages to multiple stressors
2020, Ecological IndicatorsCitation Excerpt :However, a few studies on habitat models have shown that biotic constraints can also be partly included, in a simplified way. Dispersal constraints can be accounted for in habitat models by adding a general geographical dispersal limitation term to model dispersal from source pools (Bennetsen et al., 2016) and by considering species pools and species-specific (re)colonisation potential in the presence of migration barriers (Kail et al., 2015). Biotic constraints can be included indirectly by accounting for ‘biotic resistance’ to biological invasions of sites with a high macroinvertebrate diversity (Boets et al., 2015).
Multi-objective optimisation of species distribution models for river management
2019, Water ResearchCitation Excerpt :This long history of standardized sampling in the Netherlands increases the change for having more records of species sampled in extreme conditions (e.g. very low oxygen concentration). In addition, the distribution of the input features available in the Limnodata will be quasi-similar compared to other studies, i.e. positively skewed for pollutants (Bennetsen et al., 2016). Even more, sample prevalence of the species are comparable to those reported in other studies (for example low prevalence in case of Baetidae).