- Top of page
- Materials and methods
- Supporting Information
As a result of growing human populations and an associated increase in competition for space, loss of habitat is now one of the major threats to biodiversity and the services provided by ecosystems (Sih, Jonsson & Luikart 2000). Pressures on habitats are particularly severe in coastal areas, as a consequence of the high concentration of human activities (Airoldi & Beck 2007). In order to better manage conflicts of interest, minimize negative impacts and promote sustainable use of coastal environments, tools and information systems for marine spatial planning are currently the focus of extensive efforts. One area of particular interest for efficient planning is the development of methods for coupling human pressures and ecological effects, that is, for predicting the ecological consequences of alternative policy and management scenarios.
The complexity of ecosystems, however, makes prediction of ecological responses to environmental change extremely challenging, and scenario-based assessments are still in their infancy (Coreau et al. 2009; Elith & Leathwick 2009). Scenarios are sets of expectations about plausible futures, which aim to explore the range of potential outcomes starting from a known situation, to aid in management planning (Bennett et al. 2003). Scenario analysis is often perceived only as a qualitative description of possible futures, given a set of assumptions of driving forces or extrinsic stressors. However, using scenarios in combination with predictive ecological models may be a way of producing quantitative estimates of the span of potential future outcomes, encompassing different sources of uncertainty (Coreau et al. 2009).
For ecology to grow into a fully integrated part of decision-making in society, taking these difficult steps towards a quantitative and predictive discipline is crucial (Pereira et al. 2010). One branch that has emerged within predictive ecology is species distribution modelling, in which species–environment relationships are described statistically and used to make spatial predictions of species distributions across space and time (Elith & Leathwick 2009). Species distribution models (SDMs) have the potential to be used as tools for exploring management scenarios relating to the conservation of species and habitats (Robinson et al. 2011). So far, SDMs have primarily been used in large-scale studies of climate change effects (e.g. Keith et al. 2008; Araújo et al. 2011), while their potential for exploring management options relating to human pressures that can be managed at a regional scale is largely unknown. The basis of the method is that by including a measure of the human pressure(s) of interest as predictor variable(s) in the model, the effects of changes in the pressure may be explored. Adding the spatial component of the SDM into a scenario modelling approach greatly enhances the information that can be gained from relating a change in a pressure to changes in the species of interest, when compared to a purely correlative approach. The ability to define spatially explicit changes allows for a more locally relevant and quantitative assessment of expected change, providing answers to the questions of how much change and where. For this approach to be successful, the pressure variable should have a direct, mechanistic effect on the modelled response variable. Further, the future level and distribution of the pressure variable expected under the scenario being evaluated is needed in order to make spatial predictions of the modelled species.
In the current study, we set out to test the utility of SDMs for studying scenarios relating to eutrophication mitigation measures in the Baltic Sea. Eutrophication, due to excessive phosphorus and nitrogen loads, is one of the most serious threats to coastal ecosystems worldwide (Cloern 2001), including the semi-enclosed Baltic Sea (Korpinen et al. 2012). The high nutrient concentrations give rise to excessive growth of planktonic and filamentous algae, which leads to shading, smothering and reduced recruitment success of perennial macrophytes in the shallow sublittoral (Berger et al. 2004; Krause-Jensen et al. 2008), as well as oxygen deficiency and habitat degradation with severe negative effects on ecosystem functioning and the goods and services provided by the sea (Rönnberg & Bonsdorff 2004; Diaz & Rosenberg 2008). These problems, in combination with the requirements of the EU Marine Strategy Framework Directive and the Water Framework Directive, have led the member states of Helsinki Commission (HELCOM, the executive body of the Convention on the Protection of the Marine Environment of the Baltic Sea Area) to take action against eutrophication in the Baltic Sea Action Plan (BSAP). This action plan was adopted in 2007 by all countries surrounding the Baltic Sea (HELCOM 2007a). It has set up a number of environmental objectives for 2021, relating to eutrophication, hazardous substances, maritime activities and biodiversity conservation (Backer et al. 2010). The main indicator for eutrophication is water transparency measured as mean summer Secchi disc depth (HELCOM 2007a). The Secchi disc method (Preisendorfer 1986) is a simple measure of the amount of phytoplankton in the water column, and long time-series are available. While phytoplankton only accounts for 17–40% of light attenuation in the study area, there is a strong relationship between light attenuation and the chlorophyll a concentration in the water column, indicating that Secchi depth is a good indicator of eutrophication status (Fleming-Lehtinen & Laamanen 2012). Specific target and reference levels for water transparency have been set for the different basins of the Baltic Sea. The reference levels are based on historical data, while the target levels are set to 25% lower transparency than the reference level.
For the Baltic Sea region, the countrywise nutrient reductions required to reach these goals, as well as the necessary measures, have been defined (HELCOM 2007b; Wulff et al. 2007). Costs associated with eutrophication mitigation are high, and therefore, it is important not only to consider the cost-efficiency of different actions (Elofsson 2010), but also to study more in detail how ecosystem components and functions may respond to a decrease in eutrophication. So far, however, the potential effects of the agreed reduction in nutrient loads on the ecosystems and associated goods and services have only been afforded a cursory evaluation. Qualitative studies indicate that most provisioning, regulating and cultural services of coastal habitats would increase with an improved eutrophication situation (Rönnbäck et al. 2007; BalticSTERN ; Ahtiainen & Vanhatalo 2012). However, detailed studies, entailing quantitative measures of how species distributions and ecosystem services may change with a decrease in eutrophication, are lacking.
In this paper, we show how SDMs can be used for exploring the effects of alternative management measures on the distribution of habitats. We utilize a simple scenario approach, where we account for different sources of uncertainty. Specifically, we use an ensemble of species distribution models for studying the consequences of eutrophication mitigation on the distribution of key coastal macrophyte and fish species in a 40 000-km2 archipelago area in the northern Baltic Sea. The angiosperm eelgrass Zostera marina L. and the brown alga bladderwrack Fucus vesiculosus L. are both considered to be sensitive to eutrophication and changes in water transparency (e.g. Berger et al. 2004; Krause-Jensen et al. 2008), and as such, they are used as important indicators for monitoring environmental status in most Baltic countries (Bäck et al. 2006; Schories, Pehlke & Selig 2009). The fish species perch Perca fluviatilis L. and pikeperch Sander lucioperca L., differ in their preferences for recruitment areas with respect to water transparency, with perch preferring clear water and pikeperch turbid waters (Sandström & Karås 2002; Ljunggren & Sandström 2007; Veneranta et al. 2011). Based on this knowledge of eutrophication responses of the study species, we hypothesized that the distribution of eelgrass and bladderwrack and recruitment areas of perch would increase with a higher water transparency, while recruitment areas of pikeperch would decrease.
- Top of page
- Materials and methods
- Supporting Information
The resulting distribution models for bladderwrack, eelgrass, perch and pikeperch had a moderate to high level of precision and stability (Table 1), implying that they had an adequate discriminatory ability and predictive performance (Maggini et al. 2006). The models for perch and pikeperch performed better than those for eelgrass and bladderwrack. Bladderwrack showed a positive response to Secchi depth, while the partial response of eelgrass differed from negative to positive, depending on modelling method. Perch had a positive response to Secchi depth, while pikeperch responded negatively (Fig. 2). Thus, all species except eelgrass responded as hypothesized. A comparison of the contribution of the three predictor variables to the explained variation of the models indicated that the influence of the three predictor variables varied between species and models (Table 2). For pikeperch, Secchi depth consistently was the most important predictor among methods. For bladderwrack and perch, Secchi depth was on average the second most important variable, while for eelgrass this predictor variable contributed least to the explained variation.
Table 1. Performance of bladderwrack, eelgrass, perch and pikeperch distribution models. For maximum entropy models (Maxent) and generalized additive models (GAM), performance is measured as cross-validated area-under-curve (cvAUC±SE) and for random forest models (RF) as out-of-bag error rate (OOB)
| ||Maxent (cvAUC)||GAM (cvAUC)||RF (OOB; %)|
|Bladderwrack||0·84 ± 0·01||0·86 ± 0·01||25|
|Eelgrass||0·85 ± 0·02||0·74 ± 0·03||12|
|Perch||0·98 ± 0·01||0·85 ± 0·01||2·8|
|Pikeperch||0·96 ± 0·01||0·94 ± 0·01||6·1|
Table 2. Contribution (%) of each predictor variable in bladderwrack, eelgrass, perch and pikeperch distribution models
|Species||Secchi depth||Depth||Wave exposure|
|Eelgrass|| 0|| 1||30|| 7||40||29||93||58||41|
Figure 2. Partial responses to Secchi depth, water depth and wave exposure for the three different modelling methods for bladderwrack, eelgrass, perch and pikeperch.
Download figure to PowerPoint
In the scenario analyses, the predicted responses to changes in eutrophication were quite different among the study species. Based on the ensemble of SDMs, the distribution area of bladderwrack and perch increased with Secchi depth, while eelgrass did not show a marked response to changes in water transparency. The distribution area of pikeperch was predicted to decrease (Fig. 3). For pikeperch, the different modelling methods gave highly consistent results, while for the other species the estimated change in distributions varied between methods.
Figure 3. Predicted effects on the distribution of bladderwrack, eelgrass and recruitment areas of perch and pikeperch as a response to changes in water transparency according to a set of eutrophication scenarios. Curves show percentage change in areal cover with changes in Secchi depth, where numbers on x-axis denote % deviation from current Secchi depth level. Dotted lines show standard errors of predictions from three separate modelling methods. The arrows indicate, from left to right, the Secchi depth changes according to the scenarios business-as-usual (BAU), Baltic Sea Action Plan (BSAP) target and reference levels.
Download figure to PowerPoint
The business-as-usual scenario, involving a 10% decrease in Secchi depth, predicted changes in species distributions ranging from a 16 ± 1·5% (mean ± SE) increase for pikeperch to a 44 ± 13% decrease for perch, with decreases also for bladderwrack and eelgrass. An increase in Secchi depth to the BSAP target level (+11%) again predicted stronger average responses in the fish, with a decrease in pikeperch recruitment areas by 15 ± 1·0% and an increase in perch recruitment areas by 46 ± 20%. The distribution area of bladderwrack increased by 14 ± 6·0% in the BSAP target scenario, while for eelgrass there was a large discrepancy between the methods and no clear trend. The responses in species distributions showed only weak signs of levelling off towards the BSAP reference-level scenario, indicating that larger increases in water transparency than the BSAP target would continue to change species distributions more or less linearly. For the two fish species and bladderwrack, reaching the BSAP reference level of Secchi depth, corresponding to an increase by 48%, is predicted to change distribution areas by 52–215%.
- Top of page
- Materials and methods
- Supporting Information
By utilizing water transparency as a predictor in the models, we show how SDM can be used to predict changes in the distribution of key species in the Baltic Sea as a response to changes in eutrophication status. The analyses suggest that we can expect notable effects on the distribution of species if the Baltic Sea Action Plan targets are reached, and that there will be large differences in response among species. A lowered nutrient load, accompanied by an increasing water transparency, is predicted to cause an increase in the areal cover of bladderwrack, while eelgrass would remain largely unaffected. For fish, recruitment areas of perch are predicted to increase strongly, while recruitment areas of pikeperch would likely decrease. It is important to note that a change in the availability of recruitment areas for fish is not necessarily manifested in a change in stock sizes, if recruitment areas are not limiting population growth (Levin & Stunz 2005). For these two species, however, there is a strong positive relationship between the availability of recruitment areas and the size of the adult stocks (G. Sundblad , U. Bergström, A. Sandström , P. Eklöv, unpublished data), suggesting that the predicted changes would have significant effects on population sizes.
All four of the species studied are vital for the functioning of the coastal ecosystem of the northern Baltic Sea, providing important provisioning, regulating and cultural services. Canopy-forming vegetation provides the basis of diverse habitats important for zoobenthic and fish production, wave energy absorption and aesthetic values (Boström, Baden & Krause-Jensen ; Rönnbäck et al. 2007), and changes in their distribution may widely affect the services provided by the system. Perch and pikeperch generate a suite of goods and services, for example food, recreation and trophic control, and are among the most valued species for both recreational and commercial fisheries in the area (FGFRI 2009; Thörnqvist 2009; ICES catch statistics). Being the most abundant large predatory fish at the coast, they are also important regulators of mesopredatory fish, with cascading effects in both algal and eelgrass habitats (Eriksson et al. 2009; Baden et al. 2010). The scenarios predict extensive changes in the distribution of all these species, except eelgrass, as a response to an improved eutrophication situation, suggesting that there may be large modifications in ecosystem services provided by the coastal system should the politically agreed nutrient reductions in the BSAP come into effect. Notably, the changes in species distributions are both positive and negative depending on species, indicating that the effects on ecosystem services are multifaceted and require detailed analyses to be quantified. To this end, SDMs in combination with scenario analysis as applied here can constitute a good basis for evaluation of the effects of management measures on ecosystem services.
The reliability of such scenario analyses depends upon a number of assumptions that have the potential to change not only the quantitative estimates of the predicted changes, but in severe cases also the qualitative predictions. To validate the scenario predictions qualitatively, they may be compared to mechanistic studies and, in cases where management actions are expected to reduce the levels of a human pressure, also to historical data (Elith & Leathwick 2009). As eutrophication in the Baltic Sea has increased steadily during the last decades (HELCOM 2009), there are, to some extent, old data available that correspond to the scenarios of eutrophication mitigation that have been modelled in this study. For the macrophytes, mechanistic studies suggest that both bladderwrack and eelgrass may gain from increases in water transparency (Berger et al. 2004; Krause-Jensen et al. 2008), which agrees with the model predictions for bladderwrack, while for eelgrass the models predict no change in distribution. Historical data from the study area show that bladderwrack declined substantially between the 1940s and the 1980s (Kangas et al. 1982; Kautsky et al. 1986). For eelgrass, available long-term studies from the northern Baltic Sea suggest no declines in the distribution in Finland (Boström et al. 2002) and Estonia (Möller & Martin 2007). In the study region, eelgrass distribution appears to be influenced by a trade-off between light availability and physical exposure. In contrast to marine coastal regions, eelgrass in our study area thrives on exposed sandy sea floor in the outer archipelago areas with relatively good water transparency (Boström et al. 2006). However, with increasing Secchi depth, exposure values peak beyond the suitable range for eelgrass, and the subsequent response in areal distribution remains weak. The lack of response in eelgrass can also be explained by other factors, such as stochastic events, competition with freshwater plants in sheltered areas and availability of suitable sandy substrates, may be more important in the northern Baltic Sea (Krause-Jensen et al. 2008). Thus, for bladderwrack, the results of our scenario analyses agree well with current knowledge, while for eelgrass mechanistic studies and historical data from the study area are somewhat contradictory. These results illustrate the potential utility of SDMs for exploring how the distribution of species suggested as indicators within the Water Framework Directive and the Marine Strategy Framework Directive responds to changes in specific pressures.
The responses of perch and pikeperch are consistent with knowledge about water transparency preferences for early life stages of the two species. Perch recruits are known to gain from a high water transparency, while young pikeperch is adapted to turbid waters (Sandström & Karås 2002; Ljunggren & Sandström 2007). When comparing to historical data, based on commercial catch statistics and coastal fish monitoring, the picture becomes more complex. In the commercial fishery, pikeperch catches increased dramatically from the 1950s to the 1980s, that is, at the onset of eutrophication, while perch catches decreased several-fold during the same period in the northern Baltic Sea. These changes probably reflect the increase in eutrophication status and turbidity (Lehtonen 1985; Hansson & Rudstam 1990; Swedish official catch statistics), in line with our predictions. On the other hand, fish monitoring data, which are available from the 1980s onwards, point at increases in perch populations in the study area during the last decades, while pikeperch populations have been stable or even decreasing. The development in the last decades is probably the result of interacting effects of climate change, eutrophication and fishing (HELCOM 2012; Olsson, Bergström & Gårdmark 2012, N. Mustamäki , K. Ådjers , U. Bergström, J. Mattila, unpublished). These dynamics illustrate the complex nature of population regulation in fish stocks and show that SDMs incorporating only one pressure variable may not be successful in predicting the dynamics of populations. In this study, however, we wanted to test the hypothesized effects of the politically agreed eutrophication targets in isolation from other potential changes in the environment, and as such, the scenario analyses are still valuable.
Species distribution modelling, and predictions of future distributions in particular, may be affected by uncertainty stemming from both the modelling process and the specification of scenarios (Coreau et al. 2009; Elith & Leathwick 2009). Quantifying and communicating these uncertainties is a central part in scenario analysis, as this information is vital for politicians or managers using the scenarios in decision-making. For the Baltic Sea, it is difficult to predict how soon nutrient reduction measures will translate into alleviated eutrophication symptoms of the Baltic Sea, including changes in water transparency, and how much change we can expect, especially as climate-induced changes will likely counteract these measures (Meier, Eilola & Almroth 2011). To accommodate for this source of uncertainty, we made predictions of species distributions across a range of Secchi depths, not only those specified by the scenarios. Thereby, we could graphically illustrate how species respond to Secchi depth. This approach is helpful in the interpretation of results in cases where the magnitude of change in the pressure variable is uncertain. These graphs show that the response to water transparency changes is predicted to be fairly linear, not levelling off until Secchi depth approaches the BSAP reference level. This is true also for a situation with no action taken. If no nutrient reductions are achieved and Secchi depth continues to decrease, then a close to linear negative effect can be anticipated in bladderwrack and perch, and a positive one in pikeperch. The lack of thresholds and other complex responses suggests that the qualitative conclusions from the analyses are robust and that the quantitative effects on habitat distributions will be nearly linearly related to how large improvements in Secchi depth can be reached for these key species of the northern Baltic Sea. Nevertheless, it is important to bear in mind that changes in coastal ecosystems as a result of nutrient reduction have been observed to be nonlinear with respect to nutrient loads, water transparency and time. These complex trajectories, which can probably be explained by concurrent changes in several human pressures (Duarte et al. 2009), are not captured in simplified models taking into account only one of these pressures.
Apart from the specification of scenarios, uncertainty in predictions may also arise from data deficiencies and from the actual modelling process (Elith & Leathwick 2009). Quantifying these methodological errors inherent in the sampling and modelling procedure is crucial for measuring and communicating the precision of the predictions to stakeholders and policy makers. To increase the quantitative predictive ability of the models, it is important to minimize these sources of error, while to test the stability of the predictions, sensitivity analyses need to be performed. Sampling uncertainty, which comes from deficient data sets due to, for example, small sample sizes or unrepresentative sampling, was estimated by cross-validation. The models had a moderate to high level of precision and stability (cvAUC 0·74–0·98, OOB 2·8–25%), indicating that the samples of species occurrence and the predictor variables used for calibrating the models captured the major patterns of species–environment relationships of the complex coastal region. Model uncertainty, that is, method-specific errors in the description of species–environment relationships, was assessed by comparing predictions of three conceptually different modelling techniques. The performance of the methods was comparable, all three producing useful models for the species. For all species except pikeperch, there were, however, relatively large differences in predicted changes in habitat cover between the methods. In the perch, bladderwrack and eelgrass models, the partial response to Secchi depth had a similar shape but varied in importance between models. Consequently, the span in the predictions between the methods was large for these species when applied in geographical space. This can be contrasted to pikeperch, where both response curve shape and variable contribution of Secchi depth were comparable between methods, and the three models produced very consistent estimates of habitat distributions for the different scenarios. Our experience thus suggests that relatively small differences in the shape of the partial response curves, as well as their relative importance (within method), may increase when they are applied in a geographical, predictive, context. In line with previous studies (Araújo & New 2007; Marmion et al. 2009; Mateo et al. 2012), our results highlight the value of applying an ensemble approach for minimizing model-specific errors in predictions of species distributions.
Our results provide a good basis for more in-depth analyses of the potential consequences of eutrophication mitigation measures taken around the Baltic Sea. By coupling the effects on the distribution of key species to ecosystem services, it may be possible to generate spatially explicit cost-benefit estimates (Sanchirico & Mumby 2009), which could add detail to the coarse-scale analyses of the BSAP done so far (BalticSTERN 2013). SDMs coupled with scenario analysis thus have a large potential for use as decision support tools in conservation planning, as they provide a systematic way of comparing the effects of alternative policy options and management measures on species and habitats. This avenue could be highly attractive for management, to help bridging the current gap between economy and ecology in decision-making (Rönnbäck et al. 2007; Carpenter et al. 2009). Another application of SDMs and prediction of future species distributions is in the design of marine protected area networks. By taking into account future changes in species distributions as a response to an altered environment, networks may be designed to be resilient to environmental change (Araújo et al. 2004; Mumby et al. 2011). The MPA network of the northern Baltic Sea is still deficient (Sundblad, Bergström & Sandström 2011), and in the work towards making it ecologically coherent, it would be beneficial to incorporate forecasts of species distributions.
In conclusion, we expect that the effects of eutrophication mitigation on the Baltic Sea coastal ecosystem will be pronounced, with species-specific responses to improvements in water transparency leading to changes in ecosystem functioning. The politically agreed environmental targets of the BSAP will require large-scale and costly actions in the near future, and taxpayers have the right to know how they might be affected by the measures, as well as how the alternative of not taking action would affect them. The most direct experience of changes in the marine ecosystem will relate to the highly populated coastal zone. This study provides a step towards analysing the ecological and economic consequences of the BSAP eutrophication objectives for the coastal ecosystem by providing quantitative estimates of changes in species distributions in relation to water transparency.