Climatic Change

, Volume 141, Issue 2, pp 347–361 | Cite as

Is the future of large shallow lakes blue-green? Comparing the response of a catchment-lake model chain to climate predictions

  • Fabien Cremona
  • Sirje Vilbaste
  • Raoul-Marie Couture
  • Peeter Nõges
  • Tiina Nõges
Article

Abstract

We constructed a model chain into which regional climate-related variables (air temperature, precipitation) and a lake’s main tributary hydrological indicators (river flow, dissolved inorganic carbon) were employed for predicting the evolution of planktonic blue-green algae (cyanobacteria) and zooplankton (rotifer) biomass in that lake for the mid-21st century. Simulations were based on the future climate predicted under both the Representative Concentration Pathways 4.5 and 8.5 scenarios which, combined with three realistic policy-making and basin land-use evolution lead to six scenarios for future water quality. Model outputs revealed that mean annual river flow is expected to decline between 3 and 20%, depending on the scenario. Concentration of river dissolved inorganic carbon is predicted to follow the opposite trend and might soar up to twice the 2005–2014 average concentration. Lake planktonic primary producers will display quantitative changes in the future decades whereas zooplankters will not. A 2 to 10% increase in mean cyanobacteria biomass is accompanied by a stagnation (−3 to +2%) of rotifer biomass. Changes in cyanobacteria and rotifer phenology are expected: a surge of cyanobacteria biomass in winter and a shortening of the rotifer biomass spring peak. The expected quantitative changes on the biota were magnified in those scenarios where forested area conversions to cropland and water abstraction were the greatest.

1 Introduction

Anthropogenically driven global climate change is expected to alter dramatically community structure and ecosystem functioning of fresh waters worldwide (Jeppesen et al. 2014). Indeed, several climate models predict major shifts in precipitation patterns and air temperature with important feedbacks on lacustrine systems in particular (IPCC 2014, O’Reilly et al. 2015). Shallow lakes, which are the most widespread inland water bodies, are particularly sensitive to hydrological alterations and nutrient concentration due to their large surface to volume ratios (Nõges et al. 2003). The effect of these alterations will be magnified in shallow lakes situated at higher latitudes where ice regime and phenology drive the length of the growing season (Adrian et al. 1999; Weyhenmeyer et al. 2004). In those lakes, climate alteration alone is expected to cause the following (non-exhaustive) changes: increase in sediment resuspension and phytoplankton abundance, decrease in submerged macrophyte biomass and zooplankton size, and a general shift toward an enhancement of pelagic primary production at the expense of benthic production (Jeppesen et al. 2014). It is also likely that climate change will be complemented by additional anthropogenic stressors on lake basins as the impact of global population keeps encroaching into natural ecosystems (Nõges et al. 2016a). The synergistic effect of multiple stressors together is expected to be far more detrimental to ecosystem condition than that of a simple sum of separate stressors. Furthermore, under multiple stress lake basin restoration actions may trigger poorly understood cause-effect chains that instead of recovery it might worsen the ecological condition. However, although the majority of European water bodies are affected by more than one stressor, little is known about their combined effects (Hering et al. 2015).

The mechanistic understanding of how stressors interact and impact upon water resources and status at the river basin and water body scale is of great scientific interest at the present (e.g. Hering et al. 2015). The MARS project (managing aquatic ecosystems and water resources under multiple stress) is an example of such initiative at the European scale. According to Hering et al. (2015), the most common combination of stressors on European lakes is diffuse pollution (e.g. agriculture) and hydromorphological degradation (e.g. water abstraction). The combination of ongoing cultural eutrophication, rising temperature, and hydrological alteration, for example, is suspected to increasingly shift phytoplankton community composition toward cyanobacteria dominance in shallow lakes (Kosten et al. 2012). Planktonic cyanobacteria are considered undesirable as their potentially toxic blooms pose a health risk for humans and domestic animals (Cheung et al. 2013), and they cause important changes in lake ecosystem functioning such as an evolution of lake metabolism toward net heterotrophy (Cremona et al. 2014a), the reduction of piscivorous fish size and abundance, and modifications of zooplankton community composition (Zingel and Haberman 2008). Indeed, as colonial cyanobacteria are mostly unpalatable, large zooplankters are eliminated due to the scarcity of food, leaving small zooplankters like rotifers which feed on detritus and bacteria free from competition (Zingel and Haberman 2008). The mean size of rotifers generally decreases and their abundance increases when the biomass of cyanobacteria is on the rise, resulting in a biomass decrease in eutrophic conditions (Haberman and Virro 2004). Cyanobacteria and rotifers appear thus to constitute valuable indicators of lake ecosystem degradation under multiple stressor conditions.

The objective of this study was to forecast the combined influence of three stressors (climate change, land use, water abstraction) on the phytoplankton (cyanobacteria) and zooplankton (rotifer) biomass of Lake Võrtsjärv, a large lake situated in Estonia (north-eastern Europe), for the mid-21st century (2030–2060). Our main working hypothesis was that the magnitude of the stressors would be positively correlated with the rise of cyanobacteria and decline of rotifers. To test this hypothesis, we used a chain modelling approach consisting in a succession of climate, process-based, and empirical models (Kaste et al. 2006; Couture et al. 2014; Moe et al. 2016).

2 Methods

2.1 Study site

Lake Võrtsjärv is a large (270 km2) lake located in southern Estonia (north-eastern Europe) which belongs to the southern boreal (or hemiboreal) forest zone. It is a shallow (average depth: 2.8 m) and eutrophic water body: mean total phosphorus (TP) is 48 μg l−1, total nitrogen (TN) 0.91 mg l−1, and chlorophyll a concentration (Chl a) 36 μg l−1 (Cremona et al. 2016). The lake is dominated by planktonic primary producers, especially cyanobacteria which represent 60–95% of the biomass during the ice-free period (Cremona et al. 2014a). The dominant cyanobacterial species are Limnothrix planktonica (Wolosz.) Meffert, L. redekei (Van Goor), and Planktolyngbya limnetica (Lemm.) Kom.-Legn. During the last decade, Lake Võrtsjärv has been ice-covered 135 days per year on average (from mid-November to mid-April) and has been mostly turbid during the ice-free period (Secchi depth < 1 m) owing to frequent sediment resuspension. As the flat relief restricts outflow, the lake experiences large water level changes yearly, in particular during the snow melt at the end of winter (Nõges et al. 2003). The main tributary of Võrtsjärv and contributor of 40% of its total inflow on average is the River Väike Emajõgi. It is a medium-sized (length: 82 km) river with a 1173 km2 large catchment that is roughly equally divided between forested (51%) and agricultural areas (46%) with marginal settlements (3%) and wetlands (1%). Since the mid 1960s, there has been no major change of land use in Väike Emajõgi catchment. The flow regime of the Väike Emajõgi is natural.

2.2 Data collection and analyses

Data for model input variables was gathered on a monthly basis in 10 years (2005–2014) with the exception of dissolved inorganic carbon (DIC) concentrations which were measured for 6 years (2008–2013) and air temperature, precipitation, and river flow which were measured daily.

Air temperature (°C), precipitation (mm), and stream flow values of the River Väike Emajõgi (m3 s−1) were obtained from Estonian Environment Agency. Since precipitation was measured at Valga and Tõlliste stations both located within the Väike Emajõgi catchment area, we averaged these two time-series in order to better represent conditions prevailing in the catchment.

DIC concentrations were determined at the Institute of Agricultural and Environmental Sciences of Estonian University of Life Sciences (Tartu) using the TOC analyser and standard methods (ISO 8245, 1987; EN 1484, 1992). An exhaustive description of the method is available in Pall et al. (2011) and in Cremona et al. 2014b.

Phyto- and zooplankton were collected from the monitoring station near the eastern shore where the water depth corresponded to the mean lake depth. Sampling and quantification of Chl a concentration and plankton biomass were conducted using standardized methods described in Cremona et al. (2014a).

2.3 Storylines

The employed modelling scenarios that we employed were constructed along three narratives called “storylines”, which refer to a near future fictive sequence of events describing economic, environmental, political and climatic developments (Kriegler et al. 2012, MARS 2015). Three storylines (Techno, Consensus, Fragmented) have been outlined in the MARS project, corresponding to different sets of economic, environmental, policy-making, and water management conditions. They are detailed on MARS website (http://www.mars-project.eu/files/download/fact_sheets/MARS_fact_sheet03_storylines.pdf).

The downscaling of the socio-economic factors and anticipated land-use change for Estonia and the Võrtsjärv catchment was performed by stakeholders of the Estonian Environmental Agency. The water abstraction and land use changes concurrent to each scenario were assessed according to storyline environmental impacts described by the stakeholders (Supplementary table 1). The percentages of water abstracted from the Väike Emajõgi and forest terrains replaced by agriculture were thus raised as follows: Consensus<Techno<Fragmented (Table 1).
Table 1

Details of the six simulations and one calibration selected for process-based and empirical modelling. Each run code corresponds to a distinct scenario. Run codes that start with model acronym (GFDL, IPSL) are predictive scenarios set in the future where anthropogenic pressures range from low (Consensus storyline), to moderate (Techno), to strong (Fragmented)

Run code

Storyline

Land use (%)*

Water abstraction

(% of mean annual M0 flow)

Climate data source

RCP

(W m−2)

Simulation timeline

For

Agr

Set

Wet

Hist_M0

51

46

2

1

Estonian Environment Agency

2005–2014

GFDL_M4

Techno

35

60

4

1

15

GFDL

8.5

2030–2060

GFDL_M5

Consensus

45

51

3

1

10

4.5

GFDL_M6

Fragmented

25

70

5

0

20

8.5

IPSL_M4

Techno

35

60

4

1

15

IPSL

8.5

IPSL_M5

Consensus

45

51

3

1

10

4.5

IPSL_M6

Fragmented

25

70

5

0

20

8.5

For, forest; Agr, agricultural; Set, settlement; Wet, wetlands. The first row of land use selected for run code Hist_M0 corresponds to actual land use taking place in the Väike Emajõgi catchment. The other land use numbers are time-extrapolated impact stemming from storylines

2.4 Scenario construction

Supplementary material presents the study workflow and the downscaling of variables at the basin-scale acquired from the outputs of two of the five global climate models of the ISI-MIP assessment (Warszawski et al. 2014) which were used to generate temperature and precipitation data for the period 2006–2099. The models chosen are that of Geophysical Fluid Dynamics Laboratory (GFDL) run by the National Oceanic and Atmospheric Administration (Princeton, New Jersey, USA) driven by the earth system model 2M (ESM2M) and the climate model 5 (CM5) from the Institute Pierre Simon Laplace (IPSL). To each climate change model and radiative forcing scenario—the Representative Concentration Pathways (RCP)—were associated with air temperature and precipitation predictions (IPSL 4.5, IPSL 8.5, GFDL 4.5, GFDL 8.5, van Vuuren et al. 2011, 2014). Bias-corrected time-series of air temperature and precipitation downscaled at a 0.5° resolution (Hempel et al. 2013) were used here. Additional bias-correction was done with the linear scaling method described in Shrestha (2015, summarized in supplementary material) using observed data (2005–2014) from Väike Emajõgi catchment as reference values.

2.5 Process-based modelling

2.5.1 Model description

The integrated catchment model for carbon (INCA-C; Futter et al. 2007) was employed for simulating river flow and DIC fluxes in the River Väike Emajõgi. INCA-C involves four components: (1) a semi-distributed module that defines sub-catchment boundaries and land cover uses, (2) an external rainfall-runoff model called PERSiST (precipitation, evapotranspiration and runoff simulator for solute transport; Futter et al. 2014), to calculate hydrologically effective rainfall (HER) and soil moisture deficit (SMD) needed as input for INCA-C, (3) a land-phase hydrochemical model simulating material fluxes through the soil column and transformations between chemical stocks, and (4) an in-stream model simulating the transformations in the aquatic phase. INCA-C is commonly used to model C dynamics in boreal catchments (Futter and de Wit 2008, de Wit et al. 2016); its mathematical model and conceptual diagram are described more thoroughly in Futter et al. 2007. Briefly, the model represents the main terrestrial, soil and in-stream stocks of organic and inorganic carbon, and the transfer of carbon between them. Each land-cover class consists of organic and mineral layer soil boxes. The stream is modelled as a single continuously mixed system.

2.5.2 Model calibration

We selected data from 2005 to 2014 as the baseline of our simulation after which scenarios were constructed. Before starting to calibrate the INCA-C model per se, four input parameters (SMD, HER, air temperature, precipitation) are required. For calculating SMD and HER, we employed the PERSiST model which has the same daily time step as INCA-C. The posterior flow values predicted by PERSiST were considered to match successfully the observed data (r2 = 0.7, n = 3652, N-S=0.29) allowing us to employ predicted HER and SMD as INCA-C inputs parameters (Whitehead et al. 2006).

We used the latest version of INCA-C version (1.1 beta 7) and calibrated it using the observed flow in the Väike Emajõgi, air temperature and precipitation to calculate values of SMD and HER. The calibration process was carried out in two phases as suggested by Ledesma et al. (2012): (1) calibration of the hydrological module in order to achieve a similar performance than the one obtained with PERSiST, (2) manual calibration of the carbon biogeochemistry sub-models using an iterative method. A base flow index of 0.7 was calculated for the Väike Emajõgi based on the 2005–2014 flow values. As a rule of thumb, a gradient of decreasing soil organic carbon (SOC) and dissolved organic carbon (DOC) in the three INCA-C boxes (direct runoff organic layer, mineral layer) was kept, with the following gradient between land uses: wetland>forest>farmland>settlement. For DIC, the gradient was the following: farmland>=settlement>forest>wetland. These assumptions were built on previous research showing larger DIC export to the ocean from agricultural and urbanized catchments compared to forested catchments (Barnes and Raymond 2009). We employed initial conditions described in Futter et al. (2007). The calibration was considered successful when no further iteration could improve a simulation for which at least half (r2 ≥ 0.5) of the variance of both flow and DIC was explained. The main parameter values are described in more details in Supplementary table 2.

2.6 Empirical modelling

INCA-C output values (DIC, river flow) and air temperature were used together as input values for predicting lake-related ecological variables with boosted regression trees model (BRT, Elith et al. 2008). Empirical modelling consisted of two steps. Firstly, we numerically explored the possible dependence between air temperature, river flow, and DIC, on one hand, and three Võrtsjärv-specific ecological variables (cyanobacteria biomass, Chl a, rotifer biomass) on the other hand, using the monitored values in 2005–2014 as a baseline. Secondly, the numerical relationships found were used for predicting the ecological variables based on the outputs of process-based model INCA-C. We ran BRT under the R environment (R Core Team 2013) using “gbm”, “dismo”, and “usdm” statistical packages. Tree complexity was set to 2, learning rate to 0.001, and bag fraction to 0.6.

Two-way analysis of variance (ANOVA) was done on the model outputs to test scenario (random effect) and seasonal (fixed effect) differences. These analyses were performed with JMP (version 10; SAS Institute Inc., Cary, North Carolina).

3 Results

3.1 Calibration

3.1.1 PERSiST

The simulated time-series by the hydrological model PERSiST described relatively well the observed flow in the Väike Emajõgi, with an r2 of 0.7, a RMSE of 89.3 m3 s−1, and a N-S coefficient of 0.29 (n=3652, Fig. 1) over the calibration period (2005–2014). Especially, the occurrence and timing of spring floods was well captured by the model. However, the magnitude of these floods was exaggerated by PERSiST, and in general, the magnitude of low flows between flood events was also overestimated.
Fig. 1

Calibration results for the Väike Emajõgi flow (top panel) and DIC concentrations (bottom panel) using, respectively, PERSiST and INCA models. Time-series of observed (blue solid line or dots) and predicted (red solid line) values comprise also statistical results of the model run

3.1.2 INCA-C

Small adjustments in INCA-C hydrological module improved slightly the predictive power when compared to the PERSiST outputs (r2 increased from 0.70 to 0.72, not shown). For DIC, the parameter set detailed in Supplementary table 2 yielded satisfactory time-series calibration (r2 = 0.65, N-S = −0.77, RMS = 26.27, RE = −4.84, n = 71), although the amplitude of DIC variation was greater in the simulated time series than in the observed ones (Fig. 1).

3.1.3 BRT

Using boosted regression trees, strong relationships were found between lake-related and river-related variables (Supplementary table 3). Chl a concentration was the best predicted variable as BRT explained nearly two-thirds (63%) of its variance, followed by cyanobacteria biomass (60%) and rotifer biomass (47%). For all three lake-related variables, air temperature was the best predictor of posterior distribution, followed by river flow and DIC concentration. Since all three variables were significant (although DIC displayed much less predictive power than the two others) we decided to use them for the following step—running the six climate change scenarios described previously.

3.2 Predictive scenarios

3.2.1 Flow

Five simulations out of six predicted a net decrease of river flow during the 2030–2060 period compared to the reference period (Fig. 2, Tables 2 and 3). The disparity between reference and predicted periods was mostly visible in late winter and spring. Simulations performed with the IPSL model consistently predicted lower mean flow than simulations which employed the GFDL model because of generally higher temperatures and lower precipitation input values from the former. For each model, annual mean flows were higher for the consensus (M5) than for the Techno (M4) and Fragmented (M6) scenarios. As scenarios using the Techno and Fragmented storylines were constructed with the same climate predictions, their output values were closer to each other than to Consensus outputs. Thus, the lower flow values of Fragmented scenarios compared to Techno ones can only be attributed to greater water abstraction and less water retention by agricultural soils in Fragmented scenarios.
Fig 2

Annually averaged time series of the Väike Emajõgi flow (first row, m3 s−1), DIC (second row, mg L−1) and Lake Võrtsjärv Chl a (third row, μg L−1), Bcyan (fourth row, mg ww L−1), and Broti (fifth row, μg ww L−1) which were modelled by INCA and BRT for predictive scenarios using GFDL and IPSL climate models data as input values. Thicker, double-headed arrow on the y axis corresponds to the range in reference condition

Table 2

Comparison matrix of two river-related variables (flow, DIC) and three lake-related variables (Chl a, cyanobacteria, and rotifer biomass) for the six scenarios (2030–60) and reference period (2004–2015). River-related variables values were obtained with process-based model INCA-C and lake-related variables with empirical model BRT. Tests were performed with monthly averages with a two-way ANOVA

Month

Variablesa

 

Chl a

Bcyan

Broti

Flow

Air T

DIC

Jan

 

 

 

 

*

Higher

*

Higher

Feb

 

*

Higher

 

*

Higher

*

Higher

*

Higher

Mar

 

*

Higher

 

 

*

Higher

*

Higher

Apr

*

Higher

*

Higher

*

Higher

*

Lower

*

Higher

*

Higher

May

 

*

Higher

*

Lower

 

*

Higher

*

Higher

Jun

 

*

Lower

 

 

*

Higher

*

Higher

Jul

*

Lower

*

Lower

 

 

*

Higher

*

Higher

Aug

*

Lower

*

Lower

 

 

*

Higher

*

Higher

Sep

*

Lower

*

Lower

 

 

*

Higher

*

Higher

Oct

*

Lower

*

Lower

 

 

*

Higher

*

Higher

Nov

*

Lower

*

Lower

 

 

*

Higher

*

Higher

Dec

 

 

 

 

*

Higher

*

Higher

aUnder each variable name, the first column indicates (with a “*”) if reference values of that variable are significantly different from at least one group of scenarios (GFDL or IPSL, p < 0.0001). The second column indicates whether variable value in scenarios is significantly lower or higher than in reference conditions

Table 3

Summary of expected changes of mean river-related and lake-related variables according to the six climate change scenarios

 

Flow (m3 s−1)

DIC (mg L−1)

Chl a (μg L−1)

Bcyan (mg ww L−1)

Broti (10−2 mg ww L−1)

Reference

16.8

51

32.6

8.7

9

GFDL_M4

16.3 (−3)

80 (57)

32.4 (−0.6)

8.9 (2.3)

9.2 (2)

GFDL_M5

17.4 (3.7)

72 (41)

32.7 (0.3)

8.9 (2.3)

8.7 (−3)

GFDL_M6

15.5 (−7.7)

80 (57)

32.3 (−0.9)

9.2 (5.7)

9.2 (2)

IPSL_M4

14.3 (−14.9)

93 (82)

33.1 (1.5)

9.5 (9.2)

8.9 (−1)

IPSL_M5

14.4 (−14.3)

81 (59)

32.8 (3.7)

9.6 (10.3)

9.1 (1)

IPSL_M6

13.5 (−19.6)

110 (115)

32.3 (−0.9)

9.6 (10.3)

8.8 (−2)

Number in brackets corresponds to quantitative change (in %) from reference values

3.2.2 DIC

INCA-C predicted a strong increase in the Väike Emajõgi DIC concentrations during the 2030–2060 period compared to references values of 2005–2014 (Fig. 2, Tables 2 and 3), with no scenario exhibiting DIC concentrations that would be within the range of reference. Future DIC concentrations were supposed to increase from 41 (Consensus M5 scenarios with GFDL model) to 115% (Fragmented M6 scenario with IPSL). Similarly to flow, DIC simulations that were ran with IPSL climate data predicted larger deviation from the reference (i.e. higher DIC concentrations) than those employing GFDL data. As there is an inverse relationship between flow and DIC in the Väike Emajõgi, the lower average flow in the future would concentrate the amount of mineral carbon that is in the water column.

3.2.3 Chl a

Boosted regression trees predictions for Chl a differed between GFDL and IPSL groups of scenarios (Fig. 2). Two of the three GFDL scenarios predicted a decrease of Chl a whereas two of IPSL scenarios forecasted a slight rise of Chl a. In all cases, predicted Chl a remained within or very close to the range of the reference period.

Although statistical analyses did not show large deviations of Chl a concentration in scenarios compared to reference conditions, monthly averages differed significantly between reference conditions and scenarios (Two-way ANOVA, p < 0.0001, Tables 2 and 3). Indeed, significantly greater phytoplankton biomass was forecasted during the spring peak period in April whereas the biomass was expected to decline afterwards during most of the summer season. The planktonic algae biomass would remain stable during the rest of the year.

3.2.4 Cyanobacteria biomass (Bcyano)

Mean cyanobacteria biomass in Lake Võrtsjärv showed modest (2% GFDL_M5) to notable (10% IPSL_M6) rise according to the prospective scenarios (Fig. 2, Table 2). As for Chl a, the increase in Bcyan relative to the annual mean reference value (≈8.7 mg ww L−1) was stronger in the case of IPSL Techno and Fragmented scenarios than in their GFDL counterparts. The phenology of cyanobacteria is predicted to mirror those of the whole phytoplankton community (Table 3). Cyanobacteria biomass would be lower in summer and autumn compared to reference conditions while it would be larger from winter to late spring. These temporal trends were consistent with the predicted decrease of river flow, particularly in spring.

3.2.5 Rotifer biomass (Broti)

Scenarios predicted modest increase (GFDL_M4, M6; IPSL_M5) or decrease (GFDL_M5, IPSL_M4, M6) of Broti from the 0.09 μg ww L−1 average reference value (Fig. 2, Table 2) as the differences were not significant on an annual basis (p > 0.05). Interestingly, there were also no significant differences between monthly averages of reference and scenario values, except for the period spanning on April and May. The modelled monthly average Broti value collapsed in May (0.1–0.13 μg ww L−1 compared to 0.15 μg ww L−1) after having initially exceeded reference biomass in April (0.1–0.13 μg ww L−1 compared to 0.08 μg ww L−1 for reference) showing that, according to the simulations, the spring peak of Broti is predicted to increase in intensity but to last for a shorter time over the next decades.

4 Discussion

4.1 Scenario predictions

According to three (GFDL_M5, IPSL_M4, M5) of the predictive scenario runs, Lake Võrtsjärv will experience a slight net increase in annual mean phytoplankton Chl a in the next decades caused in particular by more favourable conditions (higher temperature, lower tributary flow) in spring, while algal biomass will slightly decrease in summer and autumn compared to reference conditions. However, three other scenarios predict either a small decrease in Chl a or no quantitative change. Paleolimnological, modelling studies, and mesocosm experiments generally describe a positive relationship between anthropogenically driven climate change and phytoplankton biomass in shallow lakes (Jeppesen et al. 2014). Mooij et al. (2005) noted that climate change will improve the carrying capacity of phytoplankton, thus mimicking eutrophication in shallow lakes. Malmaeus et al. (2006) predicted stronger nutrient release and augmented phytoplankton biomass in lakes under warmer conditions, even for lakes with short water residence times like Võrtsjärv. In Võrtsjärv, Nõges et al. (2010) has demonstrated that warmer periods (which are associated with lower inflow and lake water levels) were positively correlated with in-lake total phytoplankton biomass. Together, these support the three scenarios predicting a rise of planktonic algae biomass in Lake Võrtsjärv in the next decades over the three predicting the opposite.

Although the total Chl a will increase only modestly according to our forecast, the cyanobacteria biomass might rise up to 10% compared to reference conditions. Consequently, the relative contribution of cyanobacteria to phytoplankton community biomass will probably increase in the future. These findings are consistent with the literature, as global warming is expected to promote a net increase of cyanobacteria biomass in shallow lakes (Kosten et al. 2012). The decadal, steady growth in cyanobacteria biomass in Võrtsjärv that Jeppesen et al. (2015) observed for the 1978–2012 period is thus expected to continue. Additionally, the model’s overestimation of river flow translates to an underestimation of residence time, and this to an underestimation of the dominance of cyanobacteria. Furthermore, although the models we have employed are not able to trace or forecast cyanobacteria bloom episodes, there is a strong presumption that an increase in the annual cyanobacteria biomass will be linked to the advent of harmful cyanobacteria blooms (Kosten et al. 2012) with dire consequences for the whole lake ecological processes such as dissolved oxygen concentration, metabolic balance, and carbon cycling (Cremona et al. 2014a).

The scenarios were contradictory as half of them forecasted a significant rise in Võrtsjärv rotifer biomass in the future whereas the other half predicted a slight quantitative decline. The absence of strong linkages between zooplankton metrics and climate conditions are broadly in agreement with the existing literature. According to Gyllström et al. (2005), although climate was the most important predictor of general zooplankton biomass, there was no conclusive relationship between rotifer share of the biomass and climate-related variables. Adrian et al. (1999) observed a strong relationship between ice phenology and the magnitude of the peak abundance of rotifers in temperate lakes in spring. Indeed, the gradual increase in air temperature will certainly result in a steady reduction in ice cover duration in Võrtsjärv (Jeppesen et al. 2015) which in turn might affect the spring peak of rotifer abundance, as forecasted by our chain models. Although dominating by abundance (up to 99%) rotifers do not constitute the bulk of zooplankton community biomass in Võrtsjärv. This position is occupied by ciliates which make up to 60% of the lake zooplankton biomass (Zingel and Haberman 2008). However, rotifers are a useful indicator of a detrital food web behaviour. A decline in rotifer total biomass and a concomitant rise in cyanobacteria would signify a selective pressure supporting smaller-bodied rotifer species as it is observed during more eutrophic summer conditions (Haberman and Virro 2004). Conversely, a rise in rotifer biomass would correspond to a concomitant decline in ciliate abundance which is rotifers’ main competitor for detritus and bacteria (Nõges et al. 2016b). It is still uncertain whether Võrtsjärv will be pursuing its decadal shift from a grazing to a detrital food web behaviour or if the tendency will reverse.

4.2 Modelling considerations

Using a predictive model such as INCA-C enables forecasting long-term dynamics of the Väike Emajõgi flow and its DIC concentrations with a high temporal resolution under various climate and land-use change scenarios. Furthermore, the simulated river flow proved crucial for assessing ecological variables such as cyanobacteria biomass, which was less dependent on air temperature than the other variables. The PERSiST/INCA-C models have reasonable data requirements and are relatively easy to use thanks to a built-in interface. Although INCA-C hinges on processes found in other widely used models for carbon (Futter et al. 2007), its semi-distributed nature and the complexity of coupled hydro-biogeochemical modelling lead to a large set of parameters to vary during calibration. Hereby limiting ourselves to a time-consuming manual model calibration, we could hardly assess the equability of the parameter set used in calibration, or calibrate multiple reaches simultaneously. Regarding empirical modelling, BRT are considered a versatile tool for ecological modelling (Elith et al. 2008) and provided valuable outputs in this study: the BRT model explained up to two-thirds of the variance in ecological variables, which is substantial considering the large number of observations in our time-series data. Furthermore, the built-in forecast function of BRT enabled reliable, decadal-based predictions of phytoplankton and zooplankton biomass.

Besides model-related uncertainties, the main pitfall of our study design resides in the error magnification which is inherent to chain modelling (Fowler et al. 2007). It is however mitigated by the variety of ecological conditions that are covered under our six scenarios, Consensus and Fragmented storylines representing minimum and maximum environmental perturbations, respectively. Indeed, the results of the study do not represent just possible, but plausible futures based on global economic and political trends that were downscaled to Estonia and Võrtsjärv by national stakeholders. These plausible futures can provide qualitative storylines that are highly relevant to decision-makers.

5 Conclusion

Our main working hypothesis stating that stressor combination would provoke a rise in cyanobacteria biomass and a quantitative decline in rotifers was only partly validated as the modelling approach forecasted an increase in cyanobacteria but was inconclusive regarding rotifer biomass in Lake Võrtsjärv during the 2030–2060 period. The models also predicted a surge of dissolved inorganic carbon concentrations in the main tributary water column. These changes are all clearly indicative of a degradation of the ecological status of the lake. Furthermore, they will trigger other feedback mechanisms on the lake ecosystem and food webs that are beyond the scope of this research but are well described in the existing literature. The anthropogenic pressures occurring in the lake catchment and tributaries, whether they are already existing (land use) or projected (water abstraction), will further increase the degradation of the lake’s ecological status that will be mostly driven by climate-related changes. We recommend to Estonian policy-makers and stakeholders to adopt the “Consensus” storyline described in this article and in MARS project in order to reduce as much as possible the damages done to Võrtsjärv and the ecosystem services this lake provides to the community.

Notes

Acknowledgements

The authors are grateful to Katri Rankinen (Finnish Environment Institute) for providing copies of the INCA-C and PERSiST executable, Alo Laas (Estonian University of Life Sciences) and Ivo Saaremäe (Estonian Environment Agency) for assistance in data collection. Cayetano Gutierrez (Cardiff University) was very helpful with the empirical modelling process. RMC acknowledges funding from the Norwegian Research Council project “Lakes in Transition” (244558). This research was supported by Start-Up Personal Research Grant PUT 777 to FC and IUT 21–2 of the Estonian Ministry of Education and Research, and by MARS project (managing aquatic ecosystems and water resources under multiple Stress) funded under the 7th EU Framework Programme, Theme 6 (Environment including Climate Change), Contract No.: 603378 (http://www.mars-project.eu).

Supplementary material

10584_2016_1894_MOESM1_ESM.docx (118 kb)
ESM 1(DOCX 118 kb)

References

  1. Adrian R, Walz N, Hintze T, Hoeg S, Rusche R (1999) Effects of ice duration on plankton succession during spring in a shallow polymictic lake. Freshwat Biol 41:621–634CrossRefGoogle Scholar
  2. Barnes RT, Raymond PA (2009) The contribution of agricultural and urban activities to inorganic carbon fluxes within temperate watersheds. Chem Geol 266:318–327CrossRefGoogle Scholar
  3. Cheung MY, Liang S, Lee J (2013) Toxin-producing cyanobacteria in freshwater: a review of the problems, impact on drinking water safety, and efforts for protecting public health. J Microbiol 51:1–10CrossRefGoogle Scholar
  4. Core Team R (2013) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, http://www.R-project.org/. ISBN 3-900051-07-0Google Scholar
  5. Couture R-M, Tominaga K, Starrfelt J, Moe SJ, Kaste O, Wright RF (2014) Modelling phosphorus loading and algal blooms in a Nordic agricultural catchment-lake system under changing land-use and climate. Env Sci Process Impact 16:1588–1599CrossRefGoogle Scholar
  6. Cremona F, Kõiv T, Kisand V, Laas A, Zingel P, Agasild H, Feldmann T, Järvalt A, Nõges P, Nõges T (2014a) From Bacteria to Piscivorous Fish: Estimates of Whole-Lake and Component-Specific Metabolism with an Ecosystem Approach. PLoS One 9, e101845Google Scholar
  7. Cremona F, Kõiv T, Nõges P, Pall P, Rõõm E-I, Feldmann T, Viik M, Nõges T (2014b) Dynamic carbon budget of a large shallow lake assessed by a mass balance approach. Hydrobiologia 731:109–123Google Scholar
  8. Cremona F, Laas A, Arvola L, Pierson D, Nõges P, Nõges T (2016) Numerical exploration of the planktonic to benthic primary production ratios in lakes of the Baltic Sea catchment. Ecosystems. doi:10.1007/s10021-016-0006-y Google Scholar
  9. de Wit HA, Ledesma JLJ, Futter MN (2016) Aquatic DOC export from subarctic Atlantic blanket bog in Norway is controlled by seasalt deposition, temperature and precipitation. Biogeochemistry 127:305–321CrossRefGoogle Scholar
  10. Elith J, Leathwick JR, Hastie T (2008) A working guide to boosted regression trees. J Anim Ecol 77:802–813CrossRefGoogle Scholar
  11. Fowler HJ, Blenkinsop S, Tebaldi C (2007) Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling. Int J Climatol 27:1547–1578CrossRefGoogle Scholar
  12. Futter MN, de Wit HA (2008) Testing seasonal and long-term controls of streamwater DOC using empirical and process-based models. Sci Total Environ 407:698–707CrossRefGoogle Scholar
  13. Futter MN, Butterfield D, Cosby BJ, Dillon PJ, Wade AJ, Whitehead PG (2007) Modeling the mechanisms that control in-stream dissolved organic carbon dynamics in upland and forested catchments. Water Resour Res 43:1–16CrossRefGoogle Scholar
  14. Futter MN, Erlandsson MA, Butterfield D, Whitehead PG, Oni SK, Wade AJ (2014) PERSiST: the precipitation, evapotranspiration and runoff simulator for solute transport. Hydrol Earth Syst Sci 18:855–873CrossRefGoogle Scholar
  15. Gyllström M, Hansson LA, Jeppesen E, Criado FG, Gross E, Irvine K, Kairesalo T, Kornijow R, Miracle MR, Nykänen M, Nõges T, Romo S, Stephen D, Van Donk E, Moss B (2005) The role of climate in shaping zooplankton communities of shallow lakes. Limnol Oceanogr 50:2008–2021CrossRefGoogle Scholar
  16. Haberman J, Virro T (2004) Zooplankton. In: Haberman J, Pihu E, Raukas A (eds) Lake Võrtsjärv. Estonian Encyclopedia Publishers, TallinnGoogle Scholar
  17. Hempel S, Frieler K, Warszawski L, Schewe J, Piontek F (2013) A trend-preserving bias correction - the ISI-MIP approach. Earth Syst Dynam 4:219–236CrossRefGoogle Scholar
  18. Hering D et al (2015) Managing aquatic ecosystems and water resources under multiple stress-an introduction to the MARS project. Sci Total Environ 503–504:10–21CrossRefGoogle Scholar
  19. IPCC (2014) Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate ChangeGoogle Scholar
  20. Jeppesen E, Meerhoff M, Davidson TA, Trolle D, Søndergaard M, Lauridsen TL, Beklioglu M, Brucet S, Volta P, Gonzalez-Bergonzoni I, Nielsen A (2014) Climate change impacts on lakes: an integrated ecological perspective based on a multi-faceted approach, with special focus on shallow lakes. J Limnol 73:88–111CrossRefGoogle Scholar
  21. Jeppesen E, Brucet S, Naselli-Flores L, Papastergiadou E, Stefanidis K, Nõges T, Nõges P, Attayde JL, Zohary T, Coppens J, Bucak T, Rosemberg FM, Sousa Freitas FR, Kernan M, Søndergaard M, Beklioglu M (2015) Ecological impacts of global warming and water abstraction on lakes and reservoirs due to changes in water level and related changes in salinity. Hydrobiologia 750:201–227CrossRefGoogle Scholar
  22. Kaste Ø, Wright RF, Barkved LJ, Bjerkeng B, Engen-Skaugen T, Magnusson J, Sælthun NR (2006) Linked models to assess the impacts of climate change on nitrogen in a Norwegian river basin and fjord system. Sci Total Environ 365:200–222CrossRefGoogle Scholar
  23. Kosten S, Huszar VL, Bécares E, Costa LS, Donk E, Hansson LA, Jeppesen E, Kruk C, Lacerot G, Mazzeo N, De Meester L, Moss B, Lürling M, Nõges T, Romo S, Scheffer M (2012) Warmer climates boost cyanobacterial dominance in shallow lakes. Glob Change Biol 18:118–126CrossRefGoogle Scholar
  24. Kriegler E, O’Neill BC, Hallegatte S, Kram T, Lempert R, Moss R, Wilbanks T (2012) The need for and use of socio-economic scenarios for climate change analysis: a new approach based on shared socioeconomic pathways. Glob Environ Change 22:807–822CrossRefGoogle Scholar
  25. Ledesma JLJ, Köhler SJ, Futter MN (2012) Long-term dynamics of dissolved organic carbon: implications for drinking water supply. Sci Total Environ 432:1–11CrossRefGoogle Scholar
  26. Malmaeus JM, Blenckner T, Markensten H, Persson I (2006) Lake phosphorus dynamics and climate warming: a mechanistic model approach. Ecol Model 190:1–14CrossRefGoogle Scholar
  27. MARS project (2015) Report task 2.6: definition of future scenarios., p 77Google Scholar
  28. Moe SJ, Haande S, Couture R-M (2016) Climate change, cyanobacteria blooms and ecological status of lakes: a Bayesian network approach. Ecol Model 337:330–347. doi:10.1016/j.ecolmodel.2016.07.004 CrossRefGoogle Scholar
  29. Mooij WM, Hülsmann S, De Senerpont Domis LN, Nolet BA, Bodelier PLE, Boers PCM, Dionisio Pires LM, Gons HJ, Ibelings BW, Noordhuis R, Portielje R, Wolfstein K, Lammens EHRR (2005) The impact of climate change on lakes in the Netherlands: a review. Aquat Ecol 39:381–400CrossRefGoogle Scholar
  30. Nõges T, Nõges P, Laugaste R (2003) Water level as the mediator between climate change and phytoplankton composition in a large shallow temperate lake. Hydrobiologia 506:257–263CrossRefGoogle Scholar
  31. Nõges P, Nõges T, Laas A (2010) Climate-related changes of phytoplankton seasonality in large shallow Lake Võrtsjärv, Estonia. Aquat Ecosyst Health 13:154–163CrossRefGoogle Scholar
  32. Nõges P, Argillier C, Borja Á, Garmendia JM, Hanganu J, Kodeš V, Pletterbauer F, Sagouis A, Birk S (2016a) Quantified biotic and abiotic responses to multiple stress in freshwater, marine and ground waters. Sci Total Environ 540:43–52Google Scholar
  33. Nõges T, Järvalt A, Haberman J, Zingel P, Nõges P (2016b) Is fish able to regulate filamentous blue-green dominated phytoplankton? Hydrobiologia, 1–11. DOI: 10.1007/s10750-016-2849-9Google Scholar
  34. O’Reilly CM et al (2015) Rapid and highly variable warming of lake surface waters around the globe. Geophys Res Lett 42. doi:10.1002/2015GL066235Google Scholar
  35. Pall P, Vilbaste S, Kõiv T, Kõrs A, Käiro K, Laas A, Nõges P, Nõges T, Piirsoo K, Toomsalu L, Viik M (2011) Fluxes of carbon and nutrients through the inflows and outflow of Lake Võrtsjärv, Estonia. Est J Ecol 60:39–53CrossRefGoogle Scholar
  36. Shrestha M (2015) Data analysis relied on Linear Scaling bias correction (V.1.0) Microsoft Excel fileGoogle Scholar
  37. van Vuuren DP, Edmonds J, Kainuma M, Riahi K, Thomson A, Hibbard K, Hurtt GC, Kram T, Krey V, Lamarque JF, Masui T, Meinshausen M, Nakicenovic N, Smith SJ, Rose SK (2011) The representative concentration pathways: an overview. Clim Change 109:5–31CrossRefGoogle Scholar
  38. van Vuuren DP, Kriegler E, O’Neill BC, Ebi KL, Riahi K, Carter TR, Edmonds J, Hallegatte S, Kram T, Mathur M, Winkler (2014) A new scenario framework for climate change research: scenario matrix architecture. Clim Change 122:373–386. doi:10.1007/s10584-013-0906-1 CrossRefGoogle Scholar
  39. Warszawski L, Frieler K, Huber V, Piontek F, Serdeczny O, Schewe J (2014) The Inter-Sectoral Impact Model Intercomparison Project (ISI–MIP): project framework. Proc Natl Acad Sci U S A 111:3228–3232CrossRefGoogle Scholar
  40. Weyhenmeyer GA, Meili M, Livingstone DM (2004) Nonlinear temperature response of lake ice breakup. Geophys Res Lett 31:ᅟ. doi:10.1029/2004GL019530 CrossRefGoogle Scholar
  41. Whitehead PG, Futter MN, Wilby R (2006) Impacts of climate change on hydrology, nitrogen and carbon in upland and lowland streams: assessment of adaptation strategies to meet Water Framework Directive Objectives. BHS 9th National Hydrology Symposium, DurhamGoogle Scholar
  42. Zingel P, Haberman J (2008) A comparison of zooplankton densities and biomass in Lakes Peipsi and Võrtsjärv (Estonia): rotifers and crustaceans versus ciliates. Hydrobiologia 599:153–159CrossRefGoogle Scholar

Copyright information

© Springer Science+Business Media Dordrecht 2017

Authors and Affiliations

  • Fabien Cremona
    • 1
  • Sirje Vilbaste
    • 1
  • Raoul-Marie Couture
    • 2
    • 3
  • Peeter Nõges
    • 1
  • Tiina Nõges
    • 1
  1. 1.Centre for Limnology, Institute of Agricultural and Environmental SciencesEstonian University of Life SciencesTartuEstonia
  2. 2.Norwegian Institute for Water ResearchOsloNorway
  3. 3.Ecohydrology Research Group, Water Institute and Department of Earth and Environmental SciencesUniversity of WaterlooWaterlooCanada

Personalised recommendations