Stratification dynamics in a shallow reservoir under different hydro-meteorological scenarios and operational strategies


  • Onur Kerimoglu,

    Corresponding author
    1. Department of Lake Research, Helmholtz Centre for Environmental Research—UFZ, Magdeburg, Germany
    2. INRA UMR CARRTEL, Thonon-les-Bains, France
    3. Helmholtz-Zentrum Geesthacht, Institute of Coastal Research, Geesthacht, Germany
    • Corresponding author: O. Kerimoglu, Helmholtz-Zentrum Geesthacht, Institute of Coastal Research,Max-Planck-Straße 1, 21502 Geesthacht, Germany. (

    Search for more papers by this author
  • Karsten Rinke

    1. Department of Lake Research, Helmholtz Centre for Environmental Research—UFZ, Magdeburg, Germany
    Search for more papers by this author


[1] Vertical mixing plays a major role in functioning of seasonally stratified aquatic systems. In this study, we employ a 1-D stratification model and a 9 year forcing data set to simulate the thermal dynamics in a large, but shallow reservoir that regularly displays a polymictic character with complete mixing events during summer. Such mixing dynamics is typical for many water bodies in the temperate zone having an intermediate depth. In many cases summer-mixing events were documented to induce severe water quality deteriorations (e.g., cyanobacterial blooms). We examined and quantified the response of summer-mixing behavior to combinations of hydrological regimes, i.e., water level fluctuations and withdrawal depth, and changes in meteorological variables, i.e., air temperature and wind speed. According to our findings: (i) increasing summer air temperatures considerably increase the resistance of the water column against mixing; (ii) while the combination of maintenance of a high and constant water depth and implementation of epilimnetic discharge results in almost complete resistance to mixing, their individual effects are also substantial, being roughly comparable to the effects of 4–6 K increase in air temperatures; (iii) wind is a critical variable, 30% increase of which can compensate up to 5.5 K increase in air temperatures; and (iv) effects of changes in air temperature, wind speed, and water depth are inter-dependent, as indicated by enhanced importance of wind and temperature in response to decreasing water depth, as well as reduced importance of depth in response to decreasing wind speed and increasing temperature.

1. Introduction

[2] A thermally stratified water column is generally conceptualized as a two-layer system, an upper strongly mixed, warm epilimnetic layer and a weakly mixed, cold hypolimnetic layer, separated by a metalimnetic layer characterized by strong temperature and density gradients and hence very limited mixing [Imboden and Wüest, 1995]. Thermal stratification is a key process responsible for the regulation of many physical, chemical, and ecological processes critically related to the functioning of aquatic ecosystems. The critical depth hypothesis, which defines the establishment of thermal stratification as the primary cause of onset of phytoplankton blooms [Sverdrup, 1953] keeps attracting attention and debate to date [e.g., Behrenfeld, 2010; Chiswell, 2011]. Not only the timing, but also the magnitude and composition of phytoplankton blooms are strongly influenced by the temporal [e.g., Kerimoglu et al., 2012; Peeters et al., 2012] and spatial [e.g., Huisman et al., 2004; Kerimoglu et al., 2012] features of stratification. Moreover, as the primary mechanism of oxygen renewal in hypolimnion is the influx from upper layers, hypolimnetic oxygen may become depleted depending on the duration and intensity of stratification [Elci, 2008; Matzinger et al., 2007]. A poor oxygenation state of the hypolimnion, in turn, may result in release of nutrients and heavy metals from the sediments [Belzile et al., 1996; Carignan and Lean, 1991; Middelburg and Levin, 2009], thereby triggering a cascade of processes that have system-wide effects.

[3] Man-made dams and reservoirs, with an estimated total number of 16.7 million and a combined surface area of 305,723 km2, currently account for 7.3% of the total inland water surface area [Lehner et al., 2011] and this share can be expected to increase further in the view of not-yet-saturated trends in dam building rates [WCD, 2000]. Many reservoirs, particularly those in flat landscapes, are impoundments with a relatively large surface area and a relatively shallow average depth. These water bodies are characterized by small ratios of hypolimnetic to epilimnetic volume, high hypolimnetic oxygen consumption and nutrient release rates, and their thermal stratification being highly sensitive to meteorological variables. An important difference between natural lakes and man-made reservoirs is the water withdrawal regime, while lakes release water only from the surface layer in rates proportional to inflows, reservoirs can be regulated in terms of both the quantity and the depth of withdrawal [Straskraba, 1998]. Depending on the basin morphology and catchment characteristics, the withdrawal regime can have significant effects on the thermal dynamics [Moreno-Ostos et al., 2008] and hence influence biogeochemical processes and water quality in the reservoirs, making withdrawal regulation an important management strategy. For instance, hypolimnetic withdrawal is an established reservoir remediation technique, principal idea of which is to reduce the retention time of the hypolimnion and hence to avoid occurrence of anoxic conditions and accumulation of nutrients in the hypolimnion [Nürnberg, 1987; Olszewski, 1961]. The potential of hypolimnetic withdrawal as treatment strategy has been extensively explored by numerical modeling studies [Caliskan and Elci, 2009; Casamitjana et al., 2003; Marce et al., 2010]. Positive effects of de-stratification on deep water oxygen can be outweighed, however, by potential negative effects on water quality. Intense mixing during storm events [e.g., Robarts et al., 1998], for example, may cause entrainment of nutrient-rich hypolimnetic waters into the surface layer, development of harmful algal blooms [Nürnberg et al., 2003]. In such systems, a stronger stratification might be desired, and might be achieved by regulation of withdrawals, as well. For instance, in the Eau Galle reservoir, Wisconsin, an experimental withdrawal of water only from an epilimnetic outlet resulted in an increased thermal stability and, hence, a lower flux of dissolved phosphorus to epilimnion and reduced growth of cyanobacteria [Barbiero et al., 1997].

[4] Among the meteorological variables affected by climate change, increase in air temperatures is one of the predictions with a relatively low uncertainty compared to others, such as cloud formation, short-wave radiation, precipitation, and storm frequency [Randall et al., 2007]. Although there has been extensive research on the effects of anticipated warming on water bodies [e.g., Lehman, 2002; Livingstone, 2003; Matzinger et al., 2007; Wilhelm and Adrian, 2008; Straile et al., 2010], there is still a need for investigation of the interplay of these changes with potential changes in other meteorological parameters [MacKay et al., 2009]. A few studies made use of regional climate models and applied climate projections from such models for simulating future thermal structure of lakes [Elliott et al., 2005; Kirillin, 2010]. This approach goes beyond an exclusive increase of air temperature as the climate model accounts for changes in all simulated meteorological variables (e.g., cloudiness, humidity, or wind). The analysis of stratification dynamics of lakes hence has to include more meteorological variables than just air temperature. This appears to be particularly important in water bodies having a moderate maximum depth, i.e., which are in the transition zone between deep lakes with stable stratification and shallow lakes with usually unstable stratification [compare Kirillin, 2010]. In such systems of moderate depth, wind plays a crucial role for the phenology of stratification.

[5] Aim of our study was to systematically analyze the stratification dynamics in moderately deep water bodies susceptible for summer-mixing events. We use the Bautzen Reservoir as a model system as this reservoir switches between polymixis and dimixis depending on meteorological and hydrological characteristics. The central motivation for analyzing stratification dynamics came from the observation that in this category of water bodies, summer-mixing events gave rise to mass developments of cyanobacteria and vast deteriorations in water quality. Employing a numerical one-dimensional hydrodynamic model, we performed scenario analyses to investigate the interplay between the major hydrological and climatological drivers of stratification dynamics: the water level, the withdrawal regime, air temperature, and wind velocity.


2.1. Study Site

[6] Bautzen Reservoir is located in the upper catchment of the River Spree (Germany) directly downstream of the city of Bautzen (51.218°N, 14.466°E). Given its relatively large surface area of 533 ha the maximum depth of 13.5 m and average depth of 7.4 m characterizes Bautzen Reservoir as a shallow impoundment. The trophic state of Bautzen Reservoir is eutrophic. The primary use of the reservoir is, besides recreation, maintaining sufficient summer water discharge in the lower Spree in order to supply water to downstream wetlands and power plants. Due to its exposed location in a windy area, stratification is weak and the otherwise dimictic reservoir can display a polymictic character in some years. The reservoir operation often includes high water withdrawal during summer which usually goes along with major drops in water level. More information about Bautzen Reservoir can be found in the literature [Benndorf et al., 2001; Hülsmann, 2003; Rinke et al., 2007].

2.2. Model Description

[7] We used the one-dimensional, process-based hydrodynamics model DYnamic REservoir Simulation Model (DYRESM, version 4) [Imerito, 2007] for simulating the dynamics in vertical distribution of temperature, density, and salinity. Further information about DYRESM can be found elsewhere [Imberger and Patterson, 1981; Imerito, 2007]. As the version of the model used in this study does not resolve the ice formation, models were initialized in every year at the time of spring overturn (Julian day 60–100) and ended on the Julian day 300. We applied the standard parameters employed in DYRESM version 4.0.0 without any calibration as provided in Table 1. Light extinction coefficient, Kd was chosen as 0.5, given that the Secchi-depth in the reservoir generally occurs to be around 3–4 m and that it is generally assumed that Kd * SD = 1.7 [Koenings and Edmundson, 1991].

Table 1. DYRESM Parameters
Parameter NameUnitValue
Bulk aerodynamic momentum transport coeff. 1.3 × 10−3
Mean albedo of water 0.08
Emissivity of a water surface 0.96
Critical wind speed(m s−1)3.00
Time of day for output of results(s)25,200
Shear production efficiency 0.06
Potential energy mixing efficiency 0.20
Wind stirring efficiency 0.4
Effective surface area coeff.M21.0 × 107
Vertical mixing coeff. 200
Light extinction coefficient(m−1)0.50
Min. layer thickness(m)0.30
Max. layer thickness(m)3.00
Time step(s)3600
Activate nonneutral atmospheric stability TRUE

2.3. Input Data

[8] The input data used to force the hydrodynamic model represents the meteorological, hydrological, and morphological conditions of the Bautzen Reservoir for the period between 2002 and 2011. The hydrological measurements and bathymetric data (depth-area-volume curves) were obtained from the State Reservoir Administration of Saxony. The inflow temperature was assumed to be equal to the average air temperature over the past 4 days with a lower bound of 0.1°C. As calculated surface temperatures overestimated the measurements during periods of large inflows, the maximum inflow temperature was limited to 18°C. Salinity of the inflows was also unknown, therefore, the average salinity measured during 2001–2011 was assumed to be the constant inflow salinity. Finally, to remove the slight deviations between simulated and observed water levels withdrawals were adjusted accordingly.

[9] The meteorological data were obtained from the Görlitz meteorological station operated by the German weather service (Deutsche Wetterdienst), situated about 35 km east of the reservoir. Existing gaps in the meteorological data at the Görlitz station were filled as follows:

[10] Precipitation: Missing precipitation data from Görlitz were substituted by measurements at the Cottbus meteorological station (60 km north of the reservoir, operated by the German weather service). Note that the precipitation data from Cottbus were originally available only at daily resolution, and conversion to hourly resolution was done assuming an even distribution throughout the day, resulting in a lack of high values found in other years (Figure 1).

Figure 1.

Forcing variables used as input data for numerical simulations, from top to bottom: air temperature, wind speed, precipitation, relative humidity, solar radiation, long wave radiation in hourly averages (sums in case of precipitation), inflow, and outflow in daily averages. Measured and estimated (see the text) data are shown in black and gray, respectively. The bottom plot shows the water depth recorded at Bautzen Reservoir.

[11] Solar radiation (SR): For the periods during which SR data was missing, we developed the following linear model to estimate the SR:

display math(1)

where α is the solar altitude, α' is the minimum α detected for nonzero SR values used in the training set, CSR is clear sky radiation and SD is the hourly sun duration (fraction of minutes with sun shine). The coefficients were found to be a = 47.313, b = 0.250, c = 0.357, d = 0.6545 (R2 =0.77 for α > α'). Above, CSR was estimated according to

display math(2)

as provided by Kasten and Czeplak [1980]. Note that, although the regression formulated in (equation (1)) produced slightly better predictions than the Ångström correlation [Prescott, 1940], it is highly specific to this particular system with its three additional parameters (a, c, α'), and thus not likely to be easily transferable to different systems. There exists a number of other alternative methods in the literature to estimate SR, but some of these methods [e.g., Bird and Riordan, 1986] were not accessible with the available data and the relationship described above produced better results than a two-step procedure of estimating first the hourly cloud cover using the SD [Muneer and Gul, 2000] and using these estimations in turn to estimate the SR [Kasten and Czeplak, 1980].

[12] Incident long wave radiation (iLWR): As neither direct measurements of the iLWR nor measurements of cloud cover in a sufficient temporal resolution were available at Görlitz station we calculated iLWR (Figure 1) as follows. We estimated the daytime cloud cover fraction C, from the solar radiation data according to

display math(3)

as provided by Kasten and Czeplak [1980], using the SR and CSR data estimated as described above and filled the nighttime values by linear interpolation. We then estimated the iLWR (W/m2) from temperature, T (K), vapor pressure, E (mbar), and C according to

display math(4)

[13] The coefficients were found to be a = 31.33, b = 7.8 × 10−9, c = 1.76, and d = 2.56 (R2 = 0.83) from a training data set obtained from a meteorological station located at the Lindenberg, Germany. Vapor pressure, E was estimated from the relative humidity and saturation vapor pressure according to Magnus-Tetens equations as given by Murray [1966].

2.4. Characterization of Nonwinter Mixing Dynamics

[14] In this study, the focus is on complete circulation (homogenization) events experienced during the vegetation, i.e., nonwinter period. Therefore, we designed a metric that isolates and quantifies those “warm homogenization” events by counting the number of days during which the difference between surface and bottom temperature, ΔT is less than 0.5 K, and the surface temperature, Ts is higher than 10°C. We termed this metric as “count of mixed days.” We considered a second metric, which counts the number of distinct mixing events, i.e., number of instances during a year, where homogenization (ΔT < 0.5) occurred following a stratified state (ΔT > 0.5 K) in daily resolution, and termed this metric as “count of destratification events” (see supporting information). We use the simulation outputs at 07:00 to calculate both mixing metrics. Our central findings are qualitatively insensitive to the exact value of these threshold values (tested ranges: ΔT = 0.5 − 2 and Ts = 8 − 12°C).

2.5. Model Validation

[15] For quantitatively evaluating the performance of DYRESM in reproducing the stratification dynamics, we compared the simulated and measured temperatures at the surface (at z = 0.5 m, n = 2182) and bottom (z = zmax, n = 2180), as well as the simulated and measured temperature differences between the surface and bottom, all recorded at 07:00. Comparison of these modeled and measured temperatures was achieved by the coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE), calculated as follows:

display math(5)
display math(6)
display math(7)

where Si and Oi stand for the simulated and observed temperature on the date i, inline image is the mean of the observed data, and n is the sample size.

2.6. Scenarios

[16] We defined four different hydrological settings in the Bautzen Reservoir by a 2 × 2 matrix along the factors: presence/absence of summer drawdowns and surface/deep water withdrawal (see Figure 2). Within each hydrological setting, we analyzed the effects of higher air temperatures by increasing the air temperature measurements used to force the hydrodynamical model up to +6 (K). Simulations were performed over 9 years between 2002 and 2010. The standard scenario (STD) represents the real, i.e., unchanged, hydrological conditions in Bautzen Reservoir from 2002 to 2010 characterized by strong water level fluctuations (except in 2009 and 2010) and deep water withdrawal (2 m above ground). Changing the water withdrawal toward the surface defines the second setting (SUR). Here, the withdrawal depth (WD) varied between 1.5 m and 0.5 m below surface (initially, WD was at 1.5 m below surface and whenever the water level became shallower than 0.5 m above the WD, WD was readjusted to 1.5 m below surface). As the third scenario (H10), we considered a reservoir with a relatively stable water level as observed in the year 2010 over the whole simulation period, by employing the hydrology (inflow and outflow) and initial water level (∼13 m) of the year 2010, and standard atmospheric forcing over the simulation period. Finally, in the fourth scenario (H10SUR) the 2010 hydrology and initial water level was used in combination with surface water withdrawal. Note that the STD-scenario corresponds to a classical situation in a reservoir while the H10SUR-scenario simulates conditions, as if Bautzen would be an unregulated riverine lake with surface inflow and outflow.

Figure 2.

Description of the four different hydrological settings applied in the first scenario analysis for which the air temperatures were shifted over the range of 0 to +6 (K).

[17] In a second analysis, the combined effects of temperature, wind and water level on the number of mixing days were investigated. We used the H10-scenario for this simulation study as in this setting the intra-annual water level fluctuations remained relatively low, such that the sensitivity to water levels could be tested just by varying the initial water depth. Same would hold also for the H10SUR scenario but the analysis in the first scenario analysis suggested that the H10 scenario, mixing of which is more sensitive to forcing variables, would make a more interesting case. We ran simulations with water depths varying over the range of 8–13.5 (m), observed temperatures shifted between 0 and +6 (K) and the observed wind speed shifted between −30% and +30% of the original measurements. In order to avoid analyzing a single year with some meteorological peculiarities we simulated the 9 years (2002–2009) as a whole and calculated the average number of mixed days over all years.

3. Results

3.1. Phytoplankton Composition and Relevance of Mixing Indices

[18] In Bautzen Reservoir, summer-mixing events occurred frequently, particularly after major drops in water level, and were followed by mass developments of cyanobacteria. During the relatively dry summer of 2008, for example, a major water level drop took place and several wind episodes (wind speed >5 m/s) caused homogenization of the water column as indicated by a close to zero temperature difference between surface and bottom (Figure 3). The cyanobacteria Microcystis became the dominant genus following the first of these mixing events and remained dominant until the end of the year. A contrasting picture emerges in exceptionally wet years when water level not dropped significantly over summer as is exemplified by the year 2010 (Figure 3). Due to high amounts of inflow, no summer drawdown took place. Coincidentally, the wind speed was very low until the end of August (around day 240) and no summer mixing took place. The Microcystis remained in small numbers until the end of the year while green algae and diatoms were the dominant forms. In 2008, the mass development of Microcystis was associated with elevated epilimnetic total phosphorus concentrations, which increased from about 50 to more than 200 µg L−1. In contrast to this, in 2010 maximum values of epilimnetic total phosphorus remained at values around 80 µg L−1.

Figure 3.

Time series of various parameters measured at the Bautzen Reservoir for the years 2008 (left column) and 2010 (right column). (top) water level (black, abbreviated WL), wind speed (blue, W), and temperature difference between the surface and bottom layers (red, ΔT). (bottom) biovolumes of diatoms (brown, D), green algae (green, G), and cyanobacteria (cyan, C).

3.2. Validation of the Hydrodynamic Model

[19] Surface and bottom temperatures of the Bautzen Reservoir were reproduced well by DYRESM as suggested by the time series and scatter plots as well as the calculated RMSE, MAE, and R2 values (Figure 4). Simulated temperatures agreed very well with measurements and observations. Summer surface temperatures are sometimes, however, slightly overestimated. Bottom temperatures were reproduced also reasonably well although in about half of the years the spring warming phase were underestimated. Temperature difference between surface and bottom accumulated these smaller deviations, as reflected by a wider scatter and lower skill scores obtained for surface or bottom temperatures separately. Stratification phenology was fully covered by the model. Particularly in the second half of the individual years, i.e., in the time of the year most sensitive for summer-mixing events, agreement between simulations and observations can considered to be satisfactorily good. Note that, although the model verification was based on the period 2002–2011, the year 2011 was not included in the following scenario analyses, as the available forcing data were not covering the entire period of interest, i.e., surface water was still warm at the end of the simulated period.

Figure 4.

Simulated versus measured temperatures (respectively, the black and gray lines in the time series plots) at the surface, bottom and temperature differences between the surface and the bottom over the entire simulation period (2002–2011). Skill scores (see section 'Model Validation') are provided on the scatter-plot.

3.3. Effects of Changes in Air Temperature in Hydrologically Different Systems

[20] Irrespective of the hydrological setting, the number of summer-mixed days decreased with increasing air temperatures (Figure 5). Reduction in the warm-mixed days with increasing air temperature is due to the enhanced difference between surface and bottom temperatures, which increases the wind energy required to overcome the resulting density gradient. Employing the 2010 hydrology of the Bautzen Reservoir in the H10 scenario, i.e., stable water levels without summer drawdowns increased the resistance to mixing relative to the standard scenario (Figures 5 and 6, also in most years in Figure S1 of the supporting information). The number of summer-mixed days within the H10 scenario varied considerably between years (Figure 5b), pointing to the importance of inter-annual variation in meteorological and hydrological conditions under this scenario. With increasing temperatures, susceptibility to summer mixing decreases gradually, although never completely vanishes in the H10 setting (Figure 5b). Implementation of surface withdrawal (SUR scenario) resulted also in an increased resistance to mixing (Figures 5 and 6, also Figure S1) in comparison to the standard scenario. Export of water at the surface in the SUR scenario protects the cold hypolimnion from being withdrawn, which effectively results in maintenance of the temperature gradient between surface and bottom layers, contrasting with the STD case. Surface withdrawal as represented by the SUR scenario was slightly more effective in prevention of summer-mixing events in comparison to maintenance of the water depth, as represented by the H10 scenario in a majority of years and temperature settings (Figures 6 and Figure S1). Moreover, inter-annual variability was higher in the SUR scenario than in the H10 scenario (Figure 5b versus Figure 5c). Roughly, impact of surface discharge and constant water level on the summer-mixing behavior was comparable to that of a 4–6 K increase in air temperature (Figures 5 and 6). Finally, the combined effects of high water levels and surface discharge (H10SUR scenario) result in a dramatic enhancement of resistance to mixing (Figures 5 and 6, also Figure S1). With temperature increases as small as 2 K, the summer-mixing events disappeared completely in almost all years, i.e., under a wide range of meteorological conditions.

Figure 5.

Response of annual mixing behavior, quantified as the count of summer-mixed days (see the methods) to change in air temperatures up to 6 K in 0.5 K steps in different hydrological scenarios (see section 2) denoted in the top left corner of each plot. Box-whisker plots indicate the median, 25th and 75th percentiles and extremes (plus marks indicate outliers) from the population of 9 years between 2002 and 2010.

Figure 6.

Count of warm-mixed days, in hydrologically different systems indicated in the legend (see section 2.6 and Figure 2) with changes air temperature by +2, +4, and +6 (K) shown respectively in Figures 6b, 6c, and 6d to be compared with the mixing behavior under the standard temperatures shown in Figure 6a. Note that the simulations used for these plots is a subset of simulations used for Figure 5.

3.4. Effects of Changes in Air Temperature, Wind, and Initial Water Level on a Reservoir in the Absence of Summer Drawdowns (the Scenario H10)

[21] As a general observation, the number of summer-mixed days decreases with increasing temperature, increasing initial water column depth and decreasing wind speed (Figure 7), all being intuitive. A closer inspection of the isolines connecting the identical number of mixing days in each of the plots in Figure 7 serves in comparing the magnitude of effects caused by the changes in each factor. For example, in Figures 7a and 7b, stabilizing effect of a 20% reduction in wind speed can be seen to be equivalent to 4–5 K increase in air temperatures. Further, Figure 6 reveals the inter-dependence of water depth, temperature and wind in determining the mixing characteristics of the system. While being an important variable in the base scenario, water depth gradually loses importance with increasing temperatures until around ΔT = +4 K, at which the system becomes almost completely resistant to mixing (Figures 7a and S2a). Similarly, water depth loses importance with decreasing wind speed, and becomes irrelevant at wind speeds 30% lower than the standard values hardly ever mix over the range of depths studied (Figures 7b and S2b). Both air temperature and wind are more important in shallower systems, as indicated by the greater gradient in the count of summer-mixed days across the temperature and wind speeds (Figures 7 and S2). The importance of temperature increases with increasing wind speeds (Figures 7c, 7d, S2c, and S2d, compared along the vertical axis) while that of wind speed slightly diminishes if temperature rises to very high levels (Figures 7c, 7d, S2c, and S2d, compared along the horizontal axis).

Figure 7.

Combined effects of changes in water depth, temperature and initial water depth on the mixing behavior as quantified by the average count of summer-mixed days (n = 9).

4. Discussion

[22] Role of water column stability in development of cyanobacterial blooms is being intensely debated [Carey et al., 2012; Taranu et al., 2012; Wagner and Adrian, 2009] especially as water bodies are predicted to experience more intense and prolonged durations of stratification under climate change. As exemplified in Figure 3, in the Bautzen Reservoir, autumnal cyanobacteria blooms have been observed to occur especially during the years with low water levels, and presumably associated with that, water column being prone to mixing. Cyanobacteria being positively related with the propensity for mixing might seem to be in contradiction with the view that considers water column stability as a factor that favors cyanobacterial forms [Carey et al., 2012; Huisman et al., 2004], however, a recent analysis by Taranu et al. [2012] revealed that this view holds exclusively for dimictic systems. We leave a deeper analysis of the relationship between cyanobacterial growth and mixing dynamics in the Bautzen Reservoir as a potential future topic. Hereafter, we concentrate on how the propensity for mixing, characterized by the count of warm-mixed days respond to changes in hydro-meteorological settings, noting that the patterns suggested by the count of warm-mixed days metric are in concurrence with the count of destratification events metric (Figures S1 and S2).

[23] The employed hydrodynamic model, DYRESM is a process-oriented, deterministic model that has been successfully applied to numerous lakes and reservoirs characterized by various hydro-meteorological conditions and morphological properties, such as Lake Kinneret that displays large water level fluctuations [e.g., Gal et al., 2003], the shallow Prospect reservoir [e.g., Romero et al., 2004] and large and deep Lake Constance [Rinke et al., 2010]. Implementation of the model to the Bautzen Reservoir resulted in a satisfactory model performance (Figure 3). Simulated surface-bottom temperature differences being higher than the observations in most years is an indication of an underestimation of mixing. However, these deviations are likely to be related to the inaccuracies of the forcing data, especially as the meteorological data were not in situ and the inflow temperatures had to be approximated by the average air temperatures. Nevertheless, as our analyses do not require making precise predictions for a particular reservoir, but rather the ability to meaningfully relate the hydro-meteorological forcing to seasonal stratification dynamics in shallow reservoirs in a general sense, we consider the model system to be adequate for the purposes of the current study.

[24] A fundamental assumption of a one-dimensional model is that gradients in the horizontal plane are much lower than along the vertical axis. This normally holds true for stratified lakes but can be violated by strong wind events inducing upwelling of cold water at the upwind end [e.g., see Rinke et al., 2009]. Given the large surface area of Bautzen Reservoir and the high wind velocities we checked for the occurrence of upwelling events by calculating the Wedderburn number [Imberger, 1985] using the monitoring data from our study period. The Wedderburn number uses the measured stratification characteristics (thermocline depth, temperature gradient) and the prevailing wind intensity and indicates upwelling when its value becomes smaller than one. The calculated Wedderburn number gave only limited evidence of upwelling in Bautzen Reservoir (not shown, only one instance of W < 1). Upwelling, however, can occur over short periods and during these events the model probably underestimates the exchange between epi- and hypolimnion.

[25] For being able to analyze the effects of a large set of hydro-meteorological conditions on the mixing dynamics, we employed two mixing metrics that summarize nonwinter mixing dynamics. These metrics need to be handled with care: first, their absolute values are of limited value because of their sensitivity to the specification of threshold values. On the other hand, in a relative sense, the patterns, e.g., differences in Figure 6 and gradients in Figure 7, are robust to the specification of threshold values. Second, they can be misleading when interpreted locally, e.g., for individual years, as the transformation of continuous variables (e.g., temperature differences) into binary variables (i.e., mixed or not), may either nullify or amplify small differences between two given systems, potentially resulting in very different scores. However, when considered in an average sense, these nullifications and amplifications cancel out, and consistent patterns emerge: for instance, Figure 7 reflects the average of 9 years. In conclusion, we stress that due to the limitations of the model forcing data we used as explained earlier and due to the potential local errors and sensitivity of the absolute values of the metrics we used, our findings should not be perceived as predictions for a specific system, but rather in their heuristic value as an inventory of effects of hydro-meteorological scenarios on the mixing dynamics in shallow reservoirs. A seasonal drawdown of the water level can impose a polymictic character to an otherwise dimictic system and trigger the development of harmful algal blooms [Hambright et al., 1997; Zohary and Ostrovsky, 2011]. With the help of a numerical scenario in which the 2010 flows and initial water depth was employed, we tested whether a constantly high water depth throughout the season would prevent full circulation of the water column. Although the resistance of the water column against mixing was considerably enhanced in this scenario, certain meteorological conditions still caused full summer-overturns (Figures 5 and 6). If mixing is not desired, water-level management alone hence may fail to prevent mixing in the reservoir. Whether the water column mixes or not, predominantly depends on the intensity of storm events—if a wind event is sufficiently strong (in Bautzen Reservoir: wind speeds above 10 m s−1 over several hours) the reservoir mixes irrespective of the prevailing water level.

[26] Hypolimnetic withdrawal is an established reservoir remediation technique [Nürnberg, 1987; Olszewski, 1961], by means of which accumulation of excess nutrients, organic material and metals and occurrence of anoxic conditions, and in turn, their detrimental effects [Belzile et al., 1996; Nürnberg et al., 2003; Müller and Stadelmann, 2004] can be prevented. However, as withdrawing the hypolimnetic water may destratify the water column and subsequently result in entrainment of nutrients to the epilimnion, it may give rise to the development of harmful algae [Barbiero et al., 1997]. The technique is, therefore, potentially unsuitable for reservoirs experiencing harmful algal bloom problems, possibly as in the case of the Bautzen Reservoir (Figure 3). According to our numerical scenarios, implementation of epilimnetic withdrawal would considerably increase the water column stability in the Bautzen Reservoir (Figures 5 and 6). If it would be combined with constantly high water levels the water column would become almost completely resistant to mixing (Figures 5 and 6).

[27] Increasing air temperatures have been observed and are anticipated to further enhance the intensity and duration of stratification in aquatic systems [Lehman, 2002; Schindler et al., 1996], which consequently results in an improved resistance to mixing [DeStasio et al., 1996; Livingstone, 2003; Matzinger et al., 2007; Wilhelm and Adrian, 2008]. Concurrently, our findings suggest that increasing air temperatures will result in a reduction of summer-mixing events. The magnitude of this effect depends, however on the hydro-meteorological settings and the outflow regime of the system under consideration. The Bautzen Reservoir, with its realized hydro-meteorological conditions resulting in summer drawdowns, would principally retain its polymictic character, even under a +6 K increase in the air temperature (Figures 5 and 6). However, if the stability of water column was enhanced by hydrological conditions characterized by a constant water level as in 2010 or epilimnetic withdrawal instead of hypolimnetic withdrawal, or especially by a combination of both, consequences of increasing air temperatures on the resulting system would be more pronounced (Figures 5 and 6). Further investigation of the role of water level indicated that the stabilization effect of a 4 K increase in air temperatures can be compensated by decreasing the water level from 12 to 8 m (Figure 7a). These findings show that water managers have two powerful instruments at hand for influencing the physical structures in standing waters, which will remain effective also under changing climatic conditions: management of water level and withdrawal regime.

[28] Cooling of the surface of a stratified water body, which happens almost every night but also sometimes day time during cold air outbreaks, causes a process called convective mixing [Imboden and Wüest, 1995]. As DYRESM explicitly accounts for convective mixing [Imerito, 2007], all our runs were performed with meteorological forcing at an hourly resolution, and as our analyses are based on simulation results obtained at 07:00 every day (during which the effects of nocturnal convective mixing should be most observable), our results are supposed to reflect the effects of nocturnal convective mixing. In our numerical scenario analysis of increased temperatures, our simplifying assumption of uniform diurnal temperature increase is in contradiction with the findings of Easterling et al. [2000], who detected that warming trends during nighttime can be twice as high as the warming during daytime. Therefore, it can be argued that, in reality, intensity of nocturnal convective mixing will be lower than in our simulations of increased temperatures. In order to determine the influence of such a differential temperature increase on our findings, we conducted a numerical experiment, in which we modified the air temperatures at an hourly resolution assuming as much as twofold increase during nighttime relative to the daytime, assuming a sinusoidal function (Figure S3). Differences between the number of mixing days simulated with diurnally variable increases and those with uniform increases were negligible (Figure S4). A closer inspection of simulated surface and bottom temperatures revealed that the nocturnal cooling would be too weak to overcome the large temperature gradients to emerge under increasing air temperatures in our study system, no matter whether the increases were diurnally uniform or variable (Figure S5). Temporal courses of surface and bottom temperatures under standard and increased air temperatures (Figure S5) illustrate also that the decrease in mixed days and mixing events with increasing temperatures are associated with a reduced efficiency of autumnal convective mixing in general.

[29] In contrast to the existence of numerous studies focusing on the effects of anticipated increases in the air temperature on the functioning of aquatic ecosystems, effects of potential changes in the wind speed were largely ignored see the reviews by Adrian et al. [2009] and MacKay et al. [2009]. Anticipations of changes in the typical and extreme wind speeds have strong spatial components, such as differences between coastal versus inland regions [Griffin et al., 2010] and, e.g., southern versus northern Europe [Beniston et al., 2007]. For Europe, the anticipated changes in the 90th percentile wind speeds vary between −20% and +20%, and the frequency of storm surges over northern Europe, where the study site is located, is anticipated to increase [Beniston et al., 2007]. Further, wind speeds can be drastically modified by the land management practices, such as forestration, which can in turn change the thermal regimes of especially small lakes and reservoirs. Tanentzap et al. [2008] showed that a 35% decline in the average wind speed, explained by forest regrowth and tree planting was partly responsible for a decrease of mixed layer depth from 11 to 7 m in a small lake (surface area 77 ha, max. depth: 21.5 m) in Canada. Our numerical scenario analysis demonstrates that the wind is a crucially important variable in determining the mixing behavior in lakes and reservoirs (Figure 7). For instance, destabilization effect of a 20% increase in wind can be compensated by increasing the water depth considerably, from 8 to 13 m. Similarly, our findings indicate that a 30% increase in wind can compensate between 4 and 5.5 K rise in air temperatures, depending on the depth of the reservoir (Figures 7c and 7d). However, it should be kept in mind that the importance of wind speed crucially depends on the surface area, and that in very small lakes the energy induced by wind might not be sufficient to play a major role in the deepening of the mixed layer [Fee et al., 1996; Robertson and Imberger, 1994]. Nevertheless, our findings point to a need for a systematic assessment of potential changes in the wind speed and storm frequencies and interaction of these potential changes with anticipated changes in air temperatures in characterizing the future mixing behavior of aquatic systems with regard to their morphology.

[30] In this study, we explored the response of the summer-autumn mixing dynamics to a combination of hydrological management strategies and a wide spectrum of changes in air temperatures and wind speed. We reiterate once again, that the inventory of such responses obtained with a model that greatly simplifies a complex system, are of heuristic value, and should not be considered conclusive.


[31] We acknowledge G. Ackermann and V. Neumann from the “State Reservoir Administration of Saxony” (Landestalsperrenverwaltung Sachsen) for stimulating this research and for providing data and financial support. We are grateful to Serghei Bocaniov, Martin Schultze, Bertram Boehrer, and Christoph Jäger for their helpful comments on earlier stages of this work. Finally, we want to express our gratitude to Georgiy Kirillin and two anonymous reviewers for their diligent reviews that significantly improved the quality of this work.