Advertisement

Hydrobiologia

, Volume 824, Issue 1, pp 215–228 | Cite as

A paradox of warming in a deep peri-Alpine lake (Lake Lugano, Switzerland and Italy)

  • Fabio Lepori
  • James J. Roberts
  • Travis S. Schmidt
LARGE AND DEEP PERIALPINE LAKES

Abstract

We investigated the effects of seasonal air temperature on the north basin of Lake Lugano (Switzerland and Italy), a deep peri-Alpine lake that is recovering from eutrophication. A priori ideas concerning the effects of temperature on key ecosystem responses were formalized in a conceptual model, which was tested against observed responses (from 28 years of monitoring data) using structural equation modeling. The results broadly supported our model, and indicated that air temperature had pervasive effects on the lake’s ecosystem. Warmer-than-usual winters restricted the depth of vertical mixing during turnovers and, as a result, reduced the associated replenishment of phosphorus (P) to surface waters. Moreover, although the reduction of P increased the potential for nutrient limitation, warmer winters and summers led to increases in summer chlorophyll a content (an index of phytoplankton biomass) through complex direct and indirect (food-web) effects. This apparently paradoxical effect on chlorophyll mimics a symptom of eutrophication, and suggests that climate warming might impede the recovery of the lake from eutrophication or increase the management interventions that will be required to meet the restoration targets.

Keywords

Eutrophication Lake ecosystem Structural equation modeling Plankton Warming effects 

Introduction

Climate has been warming during the last decades and is projected to continue to warm for at least another century (IPCC, 2013). Therefore, ecologists have become increasingly concerned about the effects that this change will have on biological communities. In general, the effects can be split into two broad categories. First, warming can influence biological communities through effects on the physiology of organisms (Hughes, 2000). Second, warming can influence biological communities by reinforcing the effects of other global anthropogenic perturbations, such as eutrophication or the spread of invasive species (Dukes & Mooney, 1999; Moss et al., 2011). These interactive effects warrant increased attention because they remain less understood than the isolated effects.

Eutrophication, an enrichment with nutrients, is among the most widespread environmental problems affecting fresh waters worldwide (Smith et al., 1999; Smith, 2003). In Europe and North America, eutrophication generally reached a peak during the middle or second half of the twentieth century, following decades of rapid urbanization. Subsequently, government agencies attempted to restore the lakes affected through management, which usually aimed to reduce the loadings of nutrients entering the lakes (Jeppesen et al., 2005; Schindler et al., 2016). However, because nutrient management is a costly and lengthy process, to date, several lakes have only partly recovered (e.g., Lepori & Roberts, 2017). Moreover, during the last few decades, climate warming appears to have been reinforcing some of the symptoms of eutrophication (Jeppesen et al., 2010; Moss et al., 2011; Winder, 2012; Zhou et al., 2015). In lakes recovering from eutrophication, therefore, warming could hinder recovery and increase the management effort that will be necessary to achieve management targets (Moss et al., 2011).

Although past work on the interactions between warming and eutrophication has emphasized the reinforcing effects, which include increased nutrient concentrations (Nicholls, 1999; Malmaeus et al., 2006), increased dominance of cyanobacteria (Posch et al., 2012), and increased algal biomass (Moss et al., 2011), the outcomes are not necessarily consistent across lakes. For example, in Lake Tanganyika, a deep African lake, warming has been reported to decrease the phosphorus (P) concentration and the algal production of surface waters (Verburg et al., 2003). Therefore, in this lake, warming did not mimic eutrophication, but rather its reverse, i.e., oligotrophication. The varying outcomes imply that the effects of warming on trophic state are not fixed but are moderated by lake-specific characteristics, such as morphometry (e.g., shallow vs. deep) and nutrient-pollution history. The limited, current knowledge base precludes any development of generalizable predictions on the responses of lakes to climate change. Therefore, much can be gained by developing and testing hypotheses about how individual lake ecosystems respond to warming and (or) the interaction between warming and eutrophication.

Here, we examined the effects of air temperature on a deep peri-Alpine lake that is recovering from eutrophication (Lake Lugano, Switzerland and Italy). We focussed on lake-ecosystem responses that are typically used to define the trophic state in restoration schemes, that is, the concentration of P during spring turnovers (PTURN) and the content (mg per square meter) of chlorophyll a during summer (CHLSU; Dillon & Rigler, 1975). In addition, to help elucidate the mechanisms involved, we assessed the effects on the maximum depth of mixing during turnovers (MIXDEPTH), an important determinant of surface-water P concentrations (Salmaso et al., 2013), and the summer biomass of large herbivorous zooplankton (hereafter, large grazers, GRAZERSU), an important regulator of phytoplankton (Lepori & Roberts, 2017).

Our approach was threefold. First, we summarized existing literature into a conceptual model of the causal effects of seasonal air temperature on the lake-ecosystem responses of interest (PTURN, CHLSU, MIXDEPTH, and GRAZERSU). Second, we assessed the fit between the model and the actual responses observed in the lake (based on 28 years of monitoring data, from 1989 to 2016) using structural equation modeling (SEM). Third, we revised our conceptual model by removing hypothesized paths that emerged to be unimportant and (conversely) adding new links that were identified as important. Finally, we discuss the effects of warming on the trophic state of the lake (as defined by PTURN and CHLSU) supported by this revised model and examined the implications for lake recovery from eutrophication.

Materials and methods

Study site

Lake Lugano (45o 59′0″N, 8°58′ 0″E, 271 meters above sea level) is a relatively large (surface area: 49 km2) and deep (maximum depth: 288 m) natural lake located at the foothills of the Central Alps, at the border between Switzerland and Italy (Fig. 1). The lake is divided in two main basins, the north basin and the south basin. Because we were interested in the response of deep lakes to climate, in this study we focused on the north basin, which is the deepest and has the longest water renewal time (12.3 years; Barbieri & Polli, 1992). Owing to these characteristics, this basin is nearly meromictic; that is, the water column is nearly permanently stratified into a top layer that mixes during turnovers (mixolimnion, extending from the surface to c. 100 m) and a more isolated deep-water layer that mixes only occasionally (monimolimnion, extending from c. 100 m to the bottom of the lake). However, the maximum depth of mixing varies considerably from year to year (see ‘Results’) and can exceptionally reach the bottom (Holzner et al., 2009). The regional climate is characterized by mild winters (average January temperature: 3.3°C), warm summers (average July temperature: 22.1°C), and high annual precipitation (annual total: 159 mm). In keeping with the mild climate, Lake Lugano does not freeze during winter; and the mixolimnion turns over once a year, usually in February or March (we will refer to this event as the late-winter turnover).
Fig. 1

Geographic position of Lake Lugano, Switzerland and Italy, and the sampling station in the north basin (Gandria). Descriptive point depth values (m) shown for both basins along with contours at 100, 171, 271 (m) intervals for the north basin

The north basin of Lake Lugano has a long history of nutrient pollution, which started around the 1930s (Barbieri & Mosello, 1992). Eutrophication peaked in the 1970s–1980s when the basin reached eutrophic to hyper-eutrophic status. Since the early 1980s, the lake has undergone a nutrient management program, which has achieved a substantial reduction in external total phosphorus (TP) loadings (Lepori & Roberts, 2017). In addition to nutrient pollution, during the last decades, the lake and its tributaries have been affected by climate warming (Lepori et al., 2015; Lepori & Roberts, 2015). For example, between 1972 and 2013, the surface-water summer temperatures of the north basin increased at a rate of 0.6°C per decade. This warming pattern was attributed to the combined effects of global warming and decadal variation in climatic oscillations (Lepori & Roberts, 2015).

Data for this study were sourced mainly from a long-term monitoring program of the lake, which is described further below (‘Data compilation and model parameterization’).

Structural equation modeling

We used SEM (Grace et al., 2010, 2015) to test for the causal effects of air temperature on the lake-ecosystem responses of interest because (i) a wealth of a priori knowledge existed about the lake ecosystem, and (ii) SEM allowed us to test for effects that had a hierarchical (i.e., spatial relationships) and/or temporal (i.e., order of events) structure and were either direct or indirect. Structural equation modeling is usually considered a large-dataset technique, whereas our database had a limited sample size (n = 28, see below). However, we considered that the application of SEM to our data was justified by the simplicity of our model and the low number of parameters estimated.

Our approach to SEM began by reviewing literature on lake ecosystems to develop a preliminary conceptual model that summarized the major causal pathways that might link air temperature to the lake-ecosystem responses of interest (Fig. 2). Second, we compiled the data available and parameterized all model variables for which sufficient information was found. Third, we simplified the preliminary model balancing data availability, model complexity, and information loss to develop a primary model (Fig. 3a). Fourth, we used SEM to assess the strength of each causal path included in the primary model and its global fit, i.e., the overall fit between the model and the data. Finally, we modified the primary model by removing or adding pathways based on path coefficient significance and modification indexes, which resulted in the development of a revised model of our study system (Fig. 3b). What follows is a detailed description of this process.
Fig. 2

Hypothesized causal pathways through which seasonal air temperature influences lake-ecosystem responses, including phosphorus concentration during turnovers and chlorophyll a content during summer. WI = winter, VE = vernal period, SU = summer, SW = surface water, MIXDEPTH= maximum depth of vertical mixing during late-winter turnovers, PTURN = phosphorus concentration in surface waters during late-winter turnovers, PSU = phosphorus concentration in surface waters during summer, GRAZERSU = biomass of large grazers during summer, CHLSU = content of chlorophyll a during summer. Individual paths are identified using the numbers in brackets [1–10]

Fig. 3

Structural equation models explaining the direct and indirect effects of seasonal air temperature (TWI and TSU) on lake-ecosystem responses. a Primary model, b Revised model. The thickness of the arrows reflects the size of the effects (see ‘Materials and Methods’ section). NSP > 0.1, P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001; NS = not significant, indicated with dotted arrows. Numbers are standardized path coefficients. Other conventions as in Fig. 2

Preliminary model

The preliminary conceptual model derived from literature is illustrated as a path diagram in Fig. 2. Individual paths are identified using the symbols [1–10]. Seasonal temperatures, denoting the exogenous variables, are represented by average temperatures during winter (TWI, with winter defined as the period encompassing December, January, and February), the vernal period (TVE, January–April), and summer (TSU, June–August). May temperature was not included because we lacked specific hypotheses linking it to summer trophic state (see below). All lake-ecosystem responses are endogenous variables. According to our model, seasonal temperatures control two key physical features of the lake, the surface-water temperature of the lake (SW-T) and the maximum depth of vertical mixing during late-winter turnovers (MIXDEPTH). Responses in P concentration in surface waters (PTURN), biomass of large grazer during summer (GRAZERSU), and content of chlorophyll a during summer (CHLSU) are hypothesized to be directly or indirectly caused by these physical effects.

Effects on MIXDEPTH

Hypotheses concerning the physical effects of seasonal air temperature on the lake were inferred from previous work. First, an analysis of the effect of climate on the temperature of Lake Lugano during 1972–2013 (Lepori & Roberts, 2015) indicated that SW-T was positively correlated with air temperature during most seasons (fall [September–November] was an exception). Therefore, in Lake Lugano, warmer air temperature was predicted to result in higher SW-T during most seasons (Fig. 2, paths [1a–1c]). Second, work in Lake Lugano and neighboring peri-Alpine deep lakes has shown that winter weather influences MIXDEPTH (Salmaso et al., 2013). Cold winters tend to lead to deep vertical mixing during turnovers, and mild winters to shallow mixing. This effect is expected because, during mild winters, the surface waters remain warmer than the deeper waters, and the resulting temperature gradient impedes mixing (Verburg et al., 2003). Therefore, we expected that atmospheric temperature, through the effect on SW-T (Fig. 2, path [1a]), indirectly influences the depth of mixing during late-winter turnovers (Fig. 2, path [2]).

Effects on PTURN

The P concentration considered in the model is the TP concentration of the surface waters, which reflects P availability to plankton. We hypothesized that a reduction in MIXDEPTH leads to a reduction in PTURN and summer TP concentration, PSU (Fig. 2, paths [3–4]). This effect was expected because in deep lakes (e.g., deeper than 100 m) the replenishment of P in the surface layer is determined mainly by vertical mixing, which transfers P from deep to surface waters during turnovers (Verburg et al., 2003; Salmaso et al., 2013). Moreover, we expected that PSU in Lake Lugano is determined mainly by PTURN (Fig. 2, path [4]) because the stratification of the water column and the reduced discharge of the tributaries during summer should decrease fluxes from other sources of P, including deep waters, sediments, and the watershed.

Effects on GRAZERSU

We focussed on the functional group of large grazers (i.e., grazers with maximum body length ≥ 1.5 mm), represented in the lake by the cladoceran Daphnia longispina-galeata (including D. longispina (Müller, 1776), D. galeata Sars, 1863, and their hybrids) and the calanoid copepod Eudiaptomus gracilis (Sars, 1863). This group was selected over the whole grazer assemblage because the coupling between grazers and chlorophyll a in lakes is often driven by a few key large grazers, such as Daphnia (Carpenter et al., 1985; McQueen et al., 1986). Moreover, large and small grazers can show different relationships with chlorophyll a (Carpenter et al., 1985); as a result, the effects of grazing may be confounded if these groups are lumped together.

Air temperature can influence the biomass of large grazers in summer through effects on phenology and indirect effects mediated by food-web changes. In Lake Lugano, as in other eutrophic and mesotrophic lakes, Daphnia displays spring peaks in April–May, followed by midsummer declines after a clear-water phase in May–June, which are partly mediated by food limitation (Lampert et al., 1986). Effects on phenology were expected because warm winters and springs can accelerate the growth of Daphnia, resulting in an earlier occurrence of spring peaks and earlier (and more pronounced) declines in summer (Straile, 2000; Benndorf et al., 2001; George & Hewitt, 2006). Conversely, cold winters can slow population growth rates in spring, allowing Daphnia to grow well into summer, and thus reach high biomasses as late as July (Straile, 2000). Therefore, we hypothesized that GRAZERSU (especially Daphnia) is negatively influenced by TVE (Fig. 2, path [5]). Additionally, based on long-term monitoring observations (www.cipais.org), we expected that warmer summers also have negative effects on grazers (Fig. 2, path [6]). Although it is unlikely that in Lake Lugano summer temperatures would directly impair the growth or reproduction of Daphnia and Eudiaptomus (Benndorf et al., 2001; Seebens et al., 2007), effects can arise indirectly, owing to positive effects of high temperatures on predation by juvenile fish (Benndorf et al., 2001) or dominance of inedible cyanobacteria (Paerl & Huisman, 2008; Wagner & Adrian, 2009).

We hypothesized that GRAZERSU is further constrained by PTURN (Fig. 2, path [7]) through effects on food quantity and (or) quality. In Southern Alpine lakes, spring peaks of Daphnia are positively correlated with epilimnetic P concentrations during late-winter turnovers, because P promotes phytoplankton production and Daphnia fertility during the months preceding the peak (Manca et al., 2014; Salmaso et al., 2014). Furthermore, in Lake Lugano, PTURN turnover P concentrations are positively correlated with the biomass of summer grazer assemblages, which include Daphnia and Eudiaptomus (Lepori & Roberts, 2017). Because declines in summer grazers (especially Daphnia) can be caused by food limitation during the clear-water phase (Lampert et al., 1986), and P positively affects food quantity and quality (Sterner & Hessen, 1994; Anderson & Hessen, 2005), this correlation suggests that high PTURN favors grazer biomass in summer by alleviating or delaying any food bottlenecks that may develop after spring. Therefore, we expected that PTURN had positive effects on GRAZERSU (Fig. 2—path [7]).

Effects on CHLSU

Seasonal air temperatures could influence CHLSU (an index of phytoplankton biomass) through three different pathways. First, because phytoplankton is thought to be P-limited in lowland lakes (Schindler, 1977), the reduction of P expected following warm winters (Fig. 2, paths [1–[4]) may cause a corresponding reduction in CHLSU (Fig. 2, path [8]). In contrast, the hypothesized decrease in large grazers (Fig. 2, path [9]) could reduce grazing pressure and lead to an increase of CHLSU. Finally, since the growth of phytoplankton is temperature-dependent (Eppley, 1972; Morin et al., 1999), warm summers may also be expected to increase CHLSU (Fig. 2, path [10]). Based on previous observations (Lepori & Roberts, 2017), we expected that grazers played a predominant role in regulating chlorophyll a content in the lake during summer.

Data compilation and model parameterization

To develop the SEM, we garnered data from the north basin of Lake Lugano for the period 1989 to 2016. We focused on this period because between 1988 and 1989 the lake underwent an apparent regime shift (Lepori & Roberts, 2017), which modified the coupling between phytoplankton and zooplankton. Therefore, the same model could not predict observations from before and after 1989. Data on the exogenous variables TWI, TVE, and TSU were available from the Swiss Federal Office of Meteorology (MeteoSwiss, www.meteoswiss.admin.ch). Data on endogenous (lake-ecosystem) variables were available thanks to a long-term monitoring program of Lake Lugano coordinated by the Administration of Canton Ticino and the University of Applied Sciences and Arts of Southern Switzerland (SUPSI). All lake data were collected at a sampling station located adjacent to the village of Gandria, in proximity of the central and the deepest point of the north basin (46° 00′23.77″N, 9°00′56.35″E; Fig. 1). These data were available either through published reports on the state of the lake (www.cipais.org) or unpublished databases held at SUPSI. More details on the sampling and analytical methods are reported in Lepori & Roberts (2017).

During the study period, air temperature was measured daily at the MeteoSwiss station located in the town of Lugano, adjacent to the lake (46°00′00″N, 8° 57′36″E; 273 m a.s.l.). We parameterized TWI, TSU, and TVE by averaging daily temperatures within winter, summer, and January–April, respectively.

We determined the maximum depth of mixing during late-winter turnovers (MIXDEPTH) based on monthly or biweekly profiles of electrical conductivity (vertical resolution ≤ 1 m), obtained using a conductivity–temperature–depth profiler. For each profile, we identified the depth of the mixed layer as the depth at which conductivity differed by > 3 μS cm−1 from the surface value (median of the values measured between 0 and 2 m). We then parameterized MIXDEPTH as the maximum depth reached by the mixed layer within January–April when turnovers occur (Lepori et al., 2018).

Data on the concentration of TP in surface waters were available as monthly measures from discrete depths of 0.4, 5, 10, 15, and 20 m. We parameterized PTURN as the average TP concentration within the 0-20 m layer during the late-winter turnover, i.e., when the depth of vertical mixing was greatest (see above).

Data on zooplankton (including Daphnia longispina-galeata and Eudiaptomus gracilis) during summer were available as monthly (July–August) or biweekly (June) estimates of population densities (individuals m−2) for each species (further subdivided in life stages). To parameterize GRAZERSU, we estimated the biomass of Daphnia and Eudiaptomus per unit area (mg m−2) for every sampling occasion (n = 3–4) by multiplying their population density by an estimate of individual biomass (SI Table S1). Next, the biomass values of Daphnia and Eudiaptomus were aggregated into a single biomass and averaged across sampling occasions (within summer).

Data on chlorophyll a during summer (CHLSU) were available as biweekly or monthly concentrations (n = 3–4) within the 0–20 m water layer. These concentrations were obtained by filtering lake-water samples, extracting chlorophyll in 90% ethanol, and measuring its concentration using a spectrophotometer (Lepori et al., 2018). To express CHLSU and GRAZERSU using the same unit (biomass per surface unit, in mg m−2), we multiplied the concentration of chlorophyll a (mg m−3) by 20 m, which assumes that all phytoplankton was confined within this water layer. We then parameterized CHLSU as the average chlorophyll content across sampling occasions (within summer).

Primary model

The primary model (Fig. 3a) was obtained by simplifying the preliminary model (Fig. 2). A simplification was necessary owing to the relatively small dataset available (n = 28; each of the 28 years of data was treated as an observation in the model) and the inability to parameterize some model variables (e.g., surface-water temperature). First, because we had no satisfactory measure of SW-T (lake-water temperature was available only as monthly spot measurements), we assumed (based on Lepori & Roberts, 2015) that air temperatures could be considered as an adequate proxy. Second, in our dataset, vernal temperature (January–April) overlapped and was strongly correlated (r = 0.8, P < 0.001) with winter temperature (December–February), while PSU was strongly correlated (r = 0.9, P < 0.001) with PTURN. Therefore, we used winter temperature to represent both winter and vernal temperatures, and PTURN to represent both PTURN and PSU. By comparison, winter and summer temperatures (TWI and TSU) were uncorrelated (r = − 0.2, P > 0.1) and were kept as distinct predictors in the model.

Model assessment

To assess the primary model, we estimated (i) the path coefficient (unstandardized and standardized) of every path included in the model, (ii) the total effects of the exogenous variables TSU and TWI on each endogenous variable, and (iii) the global fit between our model and the data. Standardized path coefficients (denoted by the letter β) indicate the strength of the causal effect of one variable on another and, unlike unstandardized path coefficients, are constrained to vary between 0 and 1 (Kline, 2004). To interpret their strength, we followed Cohen’s (1988) conventions for correlation coefficients, according to which a coefficient of (±) 0.1 represents a weak effect, a coefficient of 0.3 represents a moderate effect, and a coefficient of 0.5 or larger represents a strong effect. Total effects are the sum of all direct and indirect effects between a predictor and a response variable, where direct effects are measured using path coefficients and indirect effects (i.e., the effects mediated by an additional variable) are calculated as the product of sequential path coefficients. Model fit was assessed using the Chi-squared test (χ2; represents the discrepancy between the covariance matrix predicted by the model and the observed covariance matrix; therefore, a high, non-significant P value indicates good fit), the Comparative Fit Index (CFI; a value > 0.90 indicates good fit), the Root Mean Square Error of Approximation (RMSEA; a value < 0.10 indicates good fit), and the Standardized Root Mean Square Residual (SRMR; a value < 0.10 indicates good fit). In addition, we examined fit statistics with robust errors, including the Maximum Likelihood estimation of Chi square (MLM; with robust standard errors and a Satorra–Bentler scaled test statistic), although the results were not shown, because these robust estimates were not meaningfully different from the default maximum likelihood estimates.

Revised model

The revised model (Fig. 3b) was derived from the primary model using a sequential procedure. First, we eliminated hypothesized links that were non-significant at α = 0.10. Next, we used the Modification Index (MI; a value > 3 suggests the inclusion of additional causal pathways) to examine whether any additional causal pathways should be added to the model. This procedure was run iteratively until no further modifications were suggested.

SEM analysis was performed using R version 3.4.3 (R Development Core Team 2008) in R studio version 0.99.902, with packages lavaan (version 0.5-22), semPlot (version 1.0.1), and dplyr (version 0.5.0; available online at https://cran.r-project.org/).

Results

Primary model

A correlation matrix providing means and standard deviations can be used to recreate the following analysis (SI Table S2). Nearly all the pathways hypothesized in the primary model (Fig. 3a) were supported by significant path coefficients and the excellent model fit (Figs. 3a, 4; Table 1). There was a strong negative effect of TWI on MIXDEPTH (β = − 0.60), while MIXDEPTH had a strong, positive effect on TPTURN (β = 0.82). The total effect of TWI (indirect in nature) on PTURN was moderate-to-strong (− 0.49; i.e., − 0.60 × 0.82).
Fig. 4

Plots of the model variables that were causally linked according to SEM (see Fig. 3 and Tables 1 and 2). Conventions as in Fig. 2

Table 1

Lake Lugano 1989–2016: path coefficients, R2, and fit statistics for the primary model (see Fig. 3), estimated using SEM

Variable

Path coefficient

SE

P-value

St. path coefficient

R 2

L_MIXDEPTH

    

0.37

 TWI

− 0.15

0.04

< 0.001

− 0.60

PTURN

    

0.66

 MIXDEPTH

10.00

1.34

< 0.001

0.82

GRAZERSU

    

0.24

 TSU

− 0.64

0.35

0.065

− 0.31

 PTURN

0.29

0.15

0.047

0.38

 TWI

− 0.13

0.46

0.782

− 0.05

L_CHLSU

    

0.36

 GRAZERSU

− 0.05

0.02

0.002

− 0.54

 TSU

0.02

0.03

0.456

0.12

 PTURN

0.00

0.01

0.718

− 0.06

Significant path coefficients at α = 0.1 are shown in bold

Fit statistics: χ2 = 6.849 (d.f. = 6, P-value [χ2] = 0.335), RMSEA = 0.071 (90% confidence intervals = 0.000 and 0.263); CFI = 0.985, SRMR = 0.069

St. = standardized, TWI = winter air temperature, TSU = summer air temperature, MIXDEPTH = maximum depth of mixing during late-winter turnovers, PTURN = total phosphorus concentration during late-winter turnovers, CHLSU = average chlorophyll a content in summer, GRAZERSU = average biomass of large grazers in summer. L_ = log10 transformation, RMSEA = root mean square error of approximation, CFI = comparative fit index, SRMR = standardized root mean square residual, SE = standard error

The expected direct positive effect of PTURN on GRAZERSU was also moderate-to-strong (β = 0.38). In addition, GRAZERSU was directly and negatively influenced by TSU (β = − 0.31, a moderate effect), whereas the direct effect of TWI was weak (β = − 0.05). The total effect of TWI on GRAZERSU, which included this weak direct component and an indirect one (through the effect of TWI on PTURN), was − 0.24 (a weak-to-moderate effect). Through these direct and indirect effects, seasonal temperatures (TSU and TWI) had a strong combined (negative) effect of − 0.55 on GRAZERSU.

The expected direct negative effect of GRAZERSU on CHLSU was strong (β = − 0.54). In contrast, direct effects of PTURN and TSU on CHLSU were weak and non-significant (β = − 0.06 and 0.12, respectively). There was a positive indirect effect of TWI on CHLSU of 0.10 (a weak effect) through the pathway TWI → MIXDEPTH → PTURN → GRAZERSU → CHLSU and a positive indirect effect of 0.03 through the pathway TWI → MIXDEPTH → PTURN → CHLSU, adding to a total effect of 0.13 (i.e., 0.10 + 0.03). TSU had a slightly stronger and positive total effect of 0.29 (indirect effect = 0.17, direct effect = 0.12, Table 1). The joint effect of TWI and TSU on CHLSU was thus 0.42 (a moderate-to-strong effect), of which the effect mediated by grazers (0.27 = (total effect of both TWI + TSU on CHLSU)–(direct effect of TWI on CHLSU)) represented a major component. All measures of global fit (χ2 = 6.85, degrees of freedom (d.f.) = 6, P-value [χ2] test = 0.34, CFI = 0.985, RMSEA = 0.071, and SRMR = 0.07) indicated excellent agreement between our model and the data (Table 1).

Revised model

We started the revision of the primary model by removing the paths TSU → CHLSU, PTURN → CHLSU, and TWI → GRAZERSU, which were non-significant at α = 0.10. After removing these paths and refitting the model, a MI of 4.1 supported the inclusion of the additional path TWI → CHLSU, which was not hypothesized a priori. We added the suggested path and developed the revised model (Table 2; Fig. 3b). According to this model, the direct effect of TWI on CHLSU (β = 0.30) was positive and significant, increasing the total effect of TWI on CHLSU from 0.13 to 0.41 (indirect + direct effect; Fig. 3b). Conversely, the total effect of TSU on CHLSU was reduced from 0.29 to 0.16 (indirect effect only). Global fit statistics supported these changes with improvements in all fit statistics (χ2 = 3.66, df = 8, P value [χ2] = 0.89, CFI = 1.00, RMSEA = 0.000, and SRMR = 0.06, Table 2). Moreover, the final model explained 6% more of the year-to-year variation in CHLSU than the primary model (Table 2).
Table 2

Lake Lugano 1989–2016: path coefficients, R2, and fit statistics for the revised model, estimated using SEM. Conventions as in Table 1

Variables

Path coefficient

SE

P-value

St. path coefficient

R 2

L_MIXDEPTH

    

0.37

 TWI

− 0.15

0.04

< 0.001

− 0.60

PTURN

    

0.66

 MIXDEPTH

10.00

1.34

< 0.001

0.82

GRAZERSU

    

0.23

 TSU

− 0.63

0.35

0.068

− 0.30

 PTURN

0.31

0.13

0.016

0.40

L_CHLSU

    

0.42

 GRAZERSU

− 0.05

0.01

< 0.001

− 0.54

 TWI

0.06

0.03

0.042

0.30

Fit statistics: χ2 = 3.660 (d.f. = 8, P-value [χ2] = 0.886), RMSEA = 0.000 (90% confidence intervals = 0.000 and 0.105), CFI = 1.000, SRMR = 0.059

Discussion

Air temperature in winter and summer had pervasive effects on key ecosystem characteristics of the north basin of Lake Lugano, including the indicators of trophic state PTURN and CHLSU. A main result of the study was that air temperature had opposite, seemingly paradoxical, effects on these indicators. That is, high air temperature in winter or summer reduced P replenishment during turnovers, thus reducing levels of a limiting nutrient; but at the same time, it increased chlorophyll a content during summer, thus mimicking one of the main symptoms of eutrophication. This dual and previously undocumented effect emphasizes the complexity of the pathways shaping pelagic food webs and has implications for the management of the lake.

Analysis by SEM was able to identify the main pathways underlying these responses, which appeared to be broadly consistent with our predictions. The effect of winter temperature (TWI) on the TP concentration in surface waters during turnovers (PTURN) supported the indirect effect of TWI on the depth of vertical mixing (MIXDEPTH). This effect was expected because, in deep meromictic and oligomictic lakes (e.g., deeper than 100 m), replenishment of P to surface waters is often determined by a transfer from deep (infrequently mixed), P-rich waters during turnovers (Verburg et al., 2003; Salmaso, 2012; Salmaso et al., 2013). Because mild winters tend to reduce the depth of mixing during turnovers (Salmaso et al., 2013), it follows that mild winters should also reduce P replenishment.

According to our model, GRAZERSU was controlled by PTURN and controlled CHLSU, and therefore played a pivotal role in linking nutrient and plankton responses across the food web of the lake. Summer grazer populations can be negatively affected by vernal temperatures through effects on phenology (Straile, 2000; Seebens et al., 2007). However, the positive effect of PTURN, together with the lack of direct temperature effects, suggests that in Lake Lugano GRAZERSU was primarily influenced by food limitation, because P can influence both food quantity and quality (Sterner & Hessen, 1994; Anderson & Hessen, 2005). In other words, high PTURN benefited large grazers in summer by relieving them from food bottlenecks, which were probably most acute during the clear-water phase (Lampert et al., 1986). The negative effect of TSU has a less evident interpretation but is consistent with the notion that higher summer temperatures can promote dominance of inedible phytoplankton over more edible forms and predation by fish (Benndorf et al., 2001; Paerl & Huisman, 2008; Wagner & Adrian, 2009). However, exploratory correlation analyses (SI Table S2) indicated that, of the two large grazers occurring in the lake, Daphnia was associated with PTURN, whereas Eudiaptomus was associated with TSU. Therefore, explanations for these relationships appear to apply to individual taxa, rather than the whole functional group of large grazers.

The effects of TWI and TSU on CHLSU involved three pathways. First, CHLSU was mainly controlled by GRAZERSU. This effect strongly suggests that although the summer biomass of large grazers depended on spring productivity, in summer, in an apparent reversal of roles, grazers as a functional group regulated phytoplankton ‘from the top down’ (Hrbácêk et al., 1961; Carpenter et al., 1985; McQueen et al., 1986). Because GRAZERSU was causally linked to PTURN, and PTURN was causally linked to TWI, part of the positive effect of TWI cascaded through the pathway TWI → MIXDEPTH → PTURN → GRAZERSU → CHLSU. Based on the stronger coupling between Daphnia and PTURN, it is likely that the effect through GRAZERSU was conveyed mainly through this species (rather than through Eudiaptomus). Second, TSU influenced CHLSU indirectly through the negative effect on GRAZERSU, which in this case was probably conveyed mainly through Eudiaptomus. Third, our revised model indicated that TWI influenced CHLSU not just indirectly through effects on P and zooplankton, but also through a second, apparently more direct (TWI → CHLSU) pathway. This pathway is perplexing because the known effects of winter climate on phytoplankton are usually mediated by other factors, which include alterations in mixing regime (Posch et al., 2012) or disruptions of the coupling between phytoplankton and zooplankton (Winder and Schindler, 2004). Although the direct pathway identified by SEM might partly reflect these factors, it also suggests that other mechanisms play a role, highlighting the limits of our current understanding of the complex relationships between winter climate and lake ecosystems, as well as the need for further research on the topic.

A difficulty in interpreting unanticipated or complex pathways like this one lies in the relative simplicity of our model, which entailed some simplifying assumptions (see, e.g. Fig. 2 in Carpenter et al., 1985, for a more complete model of a lake food web). The complexity was constrained by data availability and the statistical necessity to maintain a reasonable ratio between parameters and observations in the model. These constraints did not prevent the identification of the major pathways at work in the food web (indicated by the excellent fit between model and the data), which was the primary aim of this study. However, developing a more detailed understanding of each individual pathway will require a larger dataset, which would enable more nuanced SEM models. For example, while our model depicts that CHLSU was controlled mainly by GRAZERSU, bottom-up control of phytoplankton biomass in lakes is common (Jeppesen et al., 2005) and probably occurred also in Lake Lugano at times (e.g. in spring, or before 1989, Lepori & Roberts, 2015). Therefore, owing to our focus on summer conditions and data limitations, one should not conclude that bottom-up effects are not important to the food web of Lake Lugano.

The effects of air temperature on the ecosystem observed in this study probably hinged on several contextual factors, including the depth of the lake, the predominance of top-down regulation of phytoplankton, and, perhaps, the species composition of large herbivorous zooplankton. Depth is an important factor because the effects of warming on the replenishment of P are typical of deep lakes. In smaller and shallower lakes, warming may have different effects, e.g. it might lead to greater nutrient concentrations through greater internal recycling or greater export from the watershed (Nicholls, 1999; Malmaeus et al., 2006). The predominance of top-down regulation in summer is probably related to the trophic state of the lake because it is more likely in lakes of moderate trophic state than in either severely eutrophic or oligotrophic lakes (Persson et al., 1988). These considerations suggest that the effects of warming might be difficult to generalize, although they may be expected to conform to our results at least in deep meromictic or oligomictic lakes of moderate trophic state, including lakes that are currently recovering from past eutrophication.

In recovering lakes, the positive effect of warm winters and summers on summer chlorophyll a contents presents a management challenge because it reinforces a symptom of eutrophication. Lake Lugano is undergoing a program of nutrient management, which aims to re-establish mesotrophic conditions through a reduction of the external loadings of P. So far, the program has achieved a substantial reduction of the loadings (Lepori & Roberts, 2017). Moreover, in the north basin of the lake, the loadings have nearly reached the program target, and further reductions would be technically difficult to attain (Barbieri & Mosello, 1992). Despite these achievements, the summer concentrations of chlorophyll a observed during the last decade of the study period (2007–2016) oscillated from moderate to high (4.3 to 19.4 mg m−3), indicating corresponding alternations between mesotrophic and eutrophic states (e.g., Nürnberg, 1996).

Owing to the effect of air temperature on CHLSU demonstrated in this study, during the last decade (after TP loadings were substantially reduced), chlorophyll a concentrations compliant with the restoration targets (< 9–10 mg m−3, indicating mesotrophic conditions; Nürnberg, 1996) occurred during cold or average years, whereas high chlorophyll a concentrations (19 mg m−3) occurred during the warmest years (i.e., in 2015 and 2016, the two hottest years of the study period). Currently, therefore, warmer-than-usual years are preventing the achievement of the restoration targets. Because the climate of Lake Lugano is projected to warm during the next century (IPCC, 2013; Lepori and Roberts, 2015), warm years and high chlorophyll a concentrations should become more frequent, assuming all other conditions remain the same. Therefore, the management program might not achieve its targets unless the loadings of TP are reduced further, to compensate for the interactive effects of the projected warming. These implications support the idea that warming has an important effect on biological communities through interactions with existing environmental pressures, in this case eutrophication, and highlight the importance of considering these interactions when establishing management goals and strategies.

Notes

Acknowledgments

Fabio Lepori worked on this paper during a visit at the Colorado Water Science Center and Fort Collins Science Center in Fort Collins (U. S. Geological Survey, USGS) during October–November 2016, which was possible thanks to a grant from the Swiss National Science Foundation and a sabbatical leave from University of Applied Sciences and Arts of Southern Switzerland. All three institutions are gratefully acknowledged for their support. Sibel Satiroglu and Jill Baron provided welcome comments on an earlier draft of the manuscript. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. Data are proprietary to Swiss organizations, contact Fabio Lepori for more information.

Supplementary material

10750_2018_3649_MOESM1_ESM.docx (14 kb)
Supplementary material 1 (DOCX 14 kb)

References

  1. Anderson, T. R. & D. O. Hessen, 2005. Threshold elemental ratios for carbon versus phosphorus limitation in Daphnia. Freshwater Biology 50: 2063–2075.CrossRefGoogle Scholar
  2. Barbieri, A. & R. Mosello, 1992. Chemistry and trophic evolution of Lake Lugano in relation to nutrient budget. Aquatic Sciences 54: 219–237.CrossRefGoogle Scholar
  3. Barbieri, A. & B. Polli, 1992. Description of Lake Lugano. Aquatic Sciences 54: 181–183.CrossRefGoogle Scholar
  4. Benndorf, J., J. Kranich, T. Mehner & A. Wagner, 2001. Temperature impact on the midsummer decline of Daphnia galeata: an analysis of long-term data from the biomanipulated Bautzen Reservoir (Germany). Freshwater Biology 46: 199–211.CrossRefGoogle Scholar
  5. Carpenter, S. R., J. F. Kitchell & J. R. Hodgson, 1985. Cascading trophic interactions and lake productivity. BioScience 35: 634–639.CrossRefGoogle Scholar
  6. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. L. Erlbaum Associates, Hillsdale.Google Scholar
  7. Dillon, P. J. & F. H. Rigler, 1975. A simple method for predicting the capacity of a lake for development based on lake trophic status. Journal of the Fisheries Board of Canada 32: 1519–1531.CrossRefGoogle Scholar
  8. Dukes, J. S. & H. A. Mooney, 1999. Does global change increase the success of biological invaders? Trends in Ecology & Evolution 14: 135–139.CrossRefGoogle Scholar
  9. Eppley, R. W., 1972. Temperature and phytoplankton growth in the sea. Fishery Bulletin 70: 1063–1085.Google Scholar
  10. George, D. G. & D. P. Hewitt, 2006. The impact of year-to-year changes in the weather on the dynamics of Daphnia in a thermally stratified lake. Aquatic Ecology 40: 33–47.CrossRefGoogle Scholar
  11. Grace, J. B., S. M. Sheiner & D. R. Schoolmaster, 2015. Structural equation modeling: building and evaluating causal models. In Fox, G. A., S. Negrete-Yankelevich & V. J. Sosa (eds.), Ecological Statistics: Contemporary Theory and Applications, 1st ed. Oxford University Press, Oxford: 168–199.CrossRefGoogle Scholar
  12. Grace, J. B., T. M. Anderson, H. Olff & S. M. Scheiner, 2010. On the specification of structural equation models for ecological systems. Ecological Monographs 80: 67–87.CrossRefGoogle Scholar
  13. Holzner, C. P., W. Aeschbach-Hertig, M. Simona, M. Veronesi, D. M. Imboden & R. Kipfer, 2009. Exceptional mixing events in meromictic Lake Lugano (Switzerland/Italy), studied using environmental tracers. Limnology & Oceanography 54: 1113–1124.CrossRefGoogle Scholar
  14. Hrbácêk, J., M. Dvorakova, V. Korinek & L. Prochazkova, 1961. Demonstration of the effect of the fish stock on the species composition of zooplankton and the intensity of metabolism of the whole plankton association. Verhandlungen der Internationalen Vereinigung für Theoretische und Angewandte Limnologie 14: 192–195.Google Scholar
  15. Hughes, L., 2000. Biological consequences of global warming: is the signal already apparent? Trends in Ecology & Evolution 15: 56–61.CrossRefGoogle Scholar
  16. IPCC, 2013. Climate change 2013: the physical science basis. In Stocker, T. F. (ed.), Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge.Google Scholar
  17. Jeppesen, E., M. Søndergaard, J. P. Jensen, K. E. Havens, O. Anneville, L. Carvalho, M. F. Coveney, R. Deneke, M. T. Dokulil, B. O. B. Foy & D. Gerdeaux, 2005. Lake responses to reduced nutrient loading: an analysis of contemporary long-term data from 35 case studies. Freshwater Biology 50: 1747–1771.CrossRefGoogle Scholar
  18. Jeppesen, E., B. Moss, H. Bennion, L. Carvalho, L. De Meester, H. Feuchtmayr & L. Liboriussen, 2010. Interaction of climate change and eutrophication. In Kernan, M., R. W. Battarbee & B. Moss (eds.), Climate Change Impacts on Freshwater Ecosystems. Wiley-Blackwell, Oxford: 119–151.CrossRefGoogle Scholar
  19. Kline, R. B., 2004. Principles and Practice of Structural Equation Modeling., Methodology in the Social Sciences The Guilford Press, New York.Google Scholar
  20. Lampert, W., W. Fleckner, H. Rai & B. E. Taylor, 1986. Phytoplankton control by grazing zooplankton: a study on the spring clear-water phase. Limnology & Oceanography 31: 478–490.CrossRefGoogle Scholar
  21. Lepori, F. & J. J. Roberts, 2015. Past and future warming of a deep European lake (Lake Lugano): what are the climatic drivers? Journal of Great Lakes Research 41: 973–981.CrossRefGoogle Scholar
  22. Lepori, F. & J. J. Roberts, 2017. Effects of internal phosphorus loadings and food-web structure on the recovery of a deep lake from eutrophication. Journal of Great Lakes Research 43: 255–264.CrossRefGoogle Scholar
  23. Lepori, F., M. Pozzoni & S. Pera, 2015. What drives warming trends in streams? A case study from the alpine foothills. River Research and Applications 31: 663–675.CrossRefGoogle Scholar
  24. Lepori, F., M. Bartosiewicz, M. Simona & M. Veronesi, 2018. Effects of winter weather and mixing regime on the restoration of a deep perialpine lake (Lake Lugano, Switzerland and Italy). Hydrobiologia.  https://doi.org/10.1007/s10750-018-3575-2 CrossRefGoogle Scholar
  25. Malmaeus, J. M., T. Blenckner, H. Markensten & I. Persson, 2006. Lake phosphorus dynamics and climate warming: a mechanistic model approach. Ecological Modelling 190: 1–14.CrossRefGoogle Scholar
  26. Manca, M., M. Rogora & N. Salmaso, 2014. Inter-annual climate variability and zooplankton: applying teleconnection indices to two deep subalpine lakes in Italy. Journal of Limnology 74: 123–132.Google Scholar
  27. McQueen, D. J., J. R. Post & E. L. Mills, 1986. Trophic relationships in freshwater pelagic ecosystems. Canadian Journal of Fisheries and Aquatic Sciences 43: 1571–1581.CrossRefGoogle Scholar
  28. Morin, A., W. Lamoureux & J. Busnarda, 1999. Empirical models predicting primary productivity from chlorophyll a and water temperature for stream periphyton and lake and ocean phytoplankton. Journal of the North American Benthological Society 18: 299–307.CrossRefGoogle Scholar
  29. Moss, B., S. Kosten, M. Meerhof, R. Battarbee, E. Jeppesen, N. Mazzeo, K. Havens, G. Lacerot, Z. Liu, L. De Meester & H. Paerl, 2011. Allied attack: climate change and eutrophication. Inland Waters 1: 101–105.CrossRefGoogle Scholar
  30. Nicholls, K. H., 1999. Effects of temperature and other factors on summer phosphorus in the inner Bay of Quinte, Lake Ontario: implications for climate warming. Journal of Great Lakes Research 25: 250–262.CrossRefGoogle Scholar
  31. Nürnberg, G. K, 1996. Trophic state of clear and colored, soft-and hardwater lakes with special consideration of nutrients, anoxia, phytoplankton and fish. Lake and Reservoir Management 12: 432–447.CrossRefGoogle Scholar
  32. Paerl, H. W. & J. Huisman, 2008. Climate. Blooms like it hot. Science 320: 57–58.CrossRefGoogle Scholar
  33. Persson, L., G. Andersson, S. F. Hamrin & L. Johansson, 1988. Predator regulation and primary production along the productivity gradient of temperate lake ecosystems. In Carpenter, S. R. (ed.), Complex Interactions in Lake Communities. Springer, New York: 45–65.CrossRefGoogle Scholar
  34. Posch, T., O. Köster, M. M. Salcher & J. Pernthaler, 2012. Harmful filamentous cyanobacteria favoured by reduced water turnover with lake warming. Nature Climate Change 2: 809–813.CrossRefGoogle Scholar
  35. R Development Core Team, 2008. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, http://www.R-project.org.Google Scholar
  36. Salmaso, N., 2012. Influence of atmospheric modes of variability on a deep lake south of the Alps. Climate Research 51: 125–133.CrossRefGoogle Scholar
  37. Salmaso, N., F. Buzzi, L. Cerasino, L. Garibaldi, B. Leoni, G. Morabito, M. Rogora & M. Simona, 2013. Influence of atmospheric modes of variability on the limnological characteristics of large lakes south of the Alps: a new emerging paradigm. Hydrobiologia 731: 31–48.CrossRefGoogle Scholar
  38. Salmaso, N., F. Buzzi, L. Cerasino, L. Garibaldi, B. Leoni, M. Manca, G. Morabito, M. Rogora & M. Simona, 2014. Influenza delle fluttuazioni climatiche sui grandi laghi a sud delle Alpi: implicazioni nel contesto del riscaldamento globale. Biologia Ambientale 28: 17–32.Google Scholar
  39. Schindler, D. W., 1977. Evolution of phosphorus limitation in lakes. Science 195: 260–262.CrossRefGoogle Scholar
  40. Schindler, D. W., S. R. Carpenter, S. C. Chapra, R. E. Hecky & D. M. Orihel, 2016. Reducing phosphorus to curb lake eutrophication is a success. Environmental Science & Technology 50: 8923–8929.CrossRefGoogle Scholar
  41. Seebens, H., D. Straile, R. Hoegg, H. B. Stich & U. Einsle, 2007. Population dynamics of a freshwater calanoid copepod: complex responses to changes in trophic status and climate variability. Limnology and Oceanography 52: 2364–2372.CrossRefGoogle Scholar
  42. Smith, V. H., 2003. Eutrophication of freshwater and coastal marine ecosystems a global problem. Environmental Science and Pollution Research 10: 126–139.CrossRefGoogle Scholar
  43. Smith, V. H., G. D. Tilman & J. C. Nekola, 1999. Eutrophication: impacts of excess nutrient inputs on freshwater, marine, and terrestrial ecosystems. Environmental Pollution 100: 179–196.CrossRefGoogle Scholar
  44. Sterner, R. W. & D. O. Hessen, 1994. Algal nutrient limitation and the nutrition of aquatic herbivores. Annual Review of Ecology and Systematics 25: 1–29.CrossRefGoogle Scholar
  45. Straile, D., 2000. Meteorological forcing of plankton dynamics in a large and deep continental European lake. Oecologia 122: 44–50.CrossRefGoogle Scholar
  46. Verburg, P., R. E. Hecky & H. Kling, 2003. Ecological consequences of a century of warming in Lake Tanganyika. Science 31: 505–507.CrossRefGoogle Scholar
  47. Wagner, C. & R. Adrian, 2009. Cyanobacteria dominance: quantifying the effects of climate change. Limnology and Oceanography 54: 2460–2468.CrossRefGoogle Scholar
  48. Winder, M., 2012. Limnology: lake warming mimics fertilization. Nature Climate Change 2: 771–772.CrossRefGoogle Scholar
  49. Winder, M. & D. E. Schindler, 2004. Climate change uncouples trophic interactions in an aquatic ecosystem. Ecology 85: 2100–2106.CrossRefGoogle Scholar
  50. Zhou, Y., A. M. Michalak, D. Beletsky, Y. R. Rao & R. P. Richards, 2015. Record-breaking Lake Erie hypoxia during 2012 drought. Environmental Science & Technology 49: 800–807.CrossRefGoogle Scholar

Copyright information

© Springer International Publishing AG, part of Springer Nature 2018

Authors and Affiliations

  1. 1.Institute of Earth SciencesUniversity of Applied, Sciences and Arts of Southern SwitzerlandCanobbioSwitzerland
  2. 2.Colorado Water Science CenterU. S. Geological Survey (USGS)Fort CollinsUSA

Personalised recommendations