Optimizing turbine withdrawal from a tropical reservoir for improved water quality in downstream wetlands

Authors

  • Manuel J. Kunz,

    Corresponding author
    1. Surface Waters–Research and Management, Eawag: Swiss Federal Institute of Aquatic Science and Technology, Kastanienbaum, Switzerland
    2. Institute of Biogeochemistry and Pollutant Dynamics, ETH Zurich, Zurich, Switzerland
    • Corresponding author: M. J. Kunz, Eawag: Swiss Federal Institute of Aquatic Science and Technology, Surface Waters–Research and Management, Seestrasse 79, CH-6047 Kastanienbaum, Switzerland. (manuel.kunz@alumni.ethz.ch)

    Search for more papers by this author
  • David B. Senn,

    1. Surface Waters–Research and Management, Eawag: Swiss Federal Institute of Aquatic Science and Technology, Kastanienbaum, Switzerland
    2. Institute of Biogeochemistry and Pollutant Dynamics, ETH Zurich, Zurich, Switzerland
    Search for more papers by this author
  • Bernhard Wehrli,

    1. Surface Waters–Research and Management, Eawag: Swiss Federal Institute of Aquatic Science and Technology, Kastanienbaum, Switzerland
    2. Institute of Biogeochemistry and Pollutant Dynamics, ETH Zurich, Zurich, Switzerland
    Search for more papers by this author
  • Elenestina M. Mwelwa,

    1. Hydrology Department, ZESCO LIMITED, Lusaka, Zambia
    Search for more papers by this author
  • Alfred Wüest

    1. Surface Waters–Research and Management, Eawag: Swiss Federal Institute of Aquatic Science and Technology, Kastanienbaum, Switzerland
    2. Institute of Biogeochemistry and Pollutant Dynamics, ETH Zurich, Zurich, Switzerland
    3. Physics of Aquatic Systems Laboratory–Margaretha Kamprad Chair, ENAC, EPFL, Switzerland
    Search for more papers by this author

Abstract

[1] Large reservoirs in the tropics act as efficient nutrient traps and often develop hypoxic conditions in the hypolimnion. Both effects may have severe implications for aquatic ecosystems, such as limited primary production in downstream riparian agriculture and in natural wetlands due to reduced nutrient loads, and, if hypolimnetic waters are withdrawn, hypoxic conditions that pose toxic risks in downstream rivers. This study using Itezhi-Tezhi Reservoir (Zambia) as a model system aims at defining optimized turbine withdrawal to prevent hypoxia and to relieve low-nutrient conditions in the downstream Kafue Flats floodplain. A biogeochemical 1-D model simulating reservoir-internal processes and water quality in the outflow was used for estimating dissolved oxygen (DO) concentrations and inorganic nitrogen and phosphorus loads in the outflow. The water depth of turbine withdrawals was varied in a set of simulations to optimize outflow water quality. Releasing hypolimnetic water was shown to result in lower average outflow DO concentrations of 4.1–6.8 mg l−1 compared to the current 7.6 mg l−1. More importantly, the outflow will remain hypoxic during up to 189 days. Meanwhile, withdrawing nutrient-rich hypolimnetic water compensated effectively for nutrient losses to the reservoir sediment. Both outflow DO concentrations and nutrient output could be optimized in the scenario with 50% epilimnetic turbine discharge originating from ∼13 m depth. In this optimal scenario, hypoxia was prevented permanently, and average DO concentrations decreased moderately to 5.2 mg l−1. Additionally, five-times higher dissolved inorganic N and dissolved inorganic P loads resulted in comparison to the current dam operation.

1. Introduction

[2] Over the past 100 years of intensive river damming the environmental and social impacts of large reservoirs triggered intensive debates [World Commission on Dams, 2000]. In addition to hydrologic alterations, dams retain particle and nutrient loads and change the oxygen regime, and thereby affect downstream water quality [Bosch, 2008; Harrison et al., 2009; Teodoru and Wehrli, 2005; Walling, 2006; Wollheim et al., 2008]. The transformation of a river into a lake facilitates primary production and water column stratification resulting in anoxic conditions and the accumulation of reduced substances in the deep water [Friedl and Wüest, 2002]. Such changes in water quality related to a stratified water column are pronounced in warm, tropical reservoirs [Kunz et al., 2011a, 2011b]. In such systems, the internal biogeochemical processes directly affect sensitive downstream ecosystems, such as wetlands and river deltas [McCartney, 2009; World Commission on Dams, 2000] and their ecosystem services, such as habitats for biodiversity and fisheries [Stockner et al., 2000].

[3] Mitigation strategies are a key element of environmental impact assessments for dam projects and the operation rules of existing dams can be improved for minimizing negative downstream effects. The environmental flows approach aims at mimicking natural flow dynamics [Acreman and Dunbar, 2004; Acreman et al., 2000; Poff et al., 2010] but the concept should be complemented with qualitative perspectives. The “green hydro” assessment concept included some qualitative aspects such as river morphology and biodiversity to promote ecosystem restoration [Bratrich et al., 2004] but ecological criteria such as temperature, particle load, and the concentration of organic matter, oxygen, and nutrients were not considered, although they play critical roles in constraining ecosystem functions [Arthington et al., 2010]. So far, only few dam optimization studies took water quality aspects explicitly into account, and if included, they were not defined as requirements for operation rules [Jager and Smith, 2008] or were focused on pollution or eutrophication in heavily modified river systems [Nilsson and Renöfält, 2008]. Dam optimization in relatively unpolluted rivers, e.g., in developing regions, such as sub-Saharan Africa has so far received little attention. Pristine rivers and ecosystems in these regions often receive only limited amounts of nutrients from natural and anthropogenic sources.

[4] Here, we examine how negative effects of the Itezhi-Tezhi Reservoir (ITT; Kafue River Basin, Zambia) on the water quality in the downstream Kafue Flats floodplain could be mitigated. The vulnerable Kafue Flats experienced a decrease in their biodiversity potentially caused by dam operations at ITT and at downstream Kafue Gorge (Figure 1). Primarily, dam operations resulted in an altered hydrology characterized by a reduction in the natural discharge variability [Mumba and Thompson, 2005]. A secondary dam-induced impact is related to water quality changes [Mumba and Thompson, 2005]. The extent of such changes induced by ITT is illustrated by the removal of approximately 50% of total nitrogen (N) and 60% of total phosphorus (P) inputs via reservoir net sedimentation and denitrification [Kunz et al., 2011a]. Moreover, the importance of ITT in affecting biogeochemical cycles at local and regional scales is illustrated by the efficient nutrient recycling and relatively high N and P removal efficiencies [Kunz et al., 2011a]. To quantify these reservoir-internal processes, the carbon (C), N, and P cycles in ITT were studied by applying the biogeochemical reservoir model RES1 [Kunz et al., 2011a]. In a next step, an adopted RES1 version was then used to simulate outflow water quality incorporating the upcoming change in the withdrawal regime via the exploitation of hydroelectricity [Kunz et al., 2011a; Piésold, 2003]. The planned turbines are designed to release nutrient-rich and oxygen-poor bottom water from ITT. Such releases may partly compensate for nutrient losses in ITT, but impose the risk of temporal hypoxia in downstream river sections [Kunz et al., 2011a].

Figure 1.

Location of Itezhi-Tezhi Reservoir (ITT) in southern Africa upstream of the Kafue Flats floodplain and the Kafue Gorge Dam (KGD). While hydropower is produced at KGD since 1971, new turbines are proposed at ITT to increase production. Since its completion in 1977, ITT served as a control dam for KGD.

[5] The goal of this paper is to develop a generally applicable optimization procedure for stratified reservoirs in nutrient-poor environments. By assessing how turbine withdrawal from different depths influences downstream water quality, we aim at defining an optimized withdrawal management for ITT with which hypoxic conditions in the outflow are prevented and dissolved oxygen (DO) and nutrient outputs are maximized. To reach this goal, we simulated the biogeochemical cycles of ITT as well as outflow water quality by extending the model RES1 previously applied to ITT [Kunz et al., 2011a]. RES1 was combined with a simple optimization procedure in order to simulate outflow water quality. The study proposes clear operation criteria for the optimized withdrawal depths and shows that biogeochemical modeling offers effective assistance in the design of ecologically optimized hydropower operation schemes.

2. Study Site and Modeling Approach

[6] The reservoir ITT has a surface area of 364 km2, a volume of 5.4 km3, and maximal depth of 50 m [McCartney and Houghton-Carr, 1998; Obrdlik et al., 1989]. Surface water from ITT is currently released to the Kafue Flats at 250 m3 s−1 on average (Figure 2; ZESCO, unpublished data, 2009). The Kafue Flats expand over 6500 km2 and are flooded between November and April (Figure 1). ITT stratifies seasonally during southern hemisphere spring to autumn (thickness of the mixed surface layer ∼20 m), resulting in the development of an anoxic hypolimnion and the accumulation of nutrients and reduced substances, followed by deep seasonal convective mixing from the end of June to early August (i.e., in southern hemisphere winter [Kunz et al., 2011a]).

Figure 2.

Hydrologic scenarios defined by maximum measured discharge at ITT between August 2008 and July 2009 (Qmax), average daily discharge between 1978 and 2009 (Qavg), and minimum discharge between August 1996 and July 1997 (Qmin; ZESCO, unpublished data, 2009). The shaded areas below the maximum turbine capacity (306 m3 s−1, dotted line) depict the turbinated portion, whereas the white areas above the dotted line indicate the spilled portion of the total discharge (solid line).

[7] On the one hand, releasing ITT hypolimnion water through the planned turbines is problematic in terms of downstream DO concentrations. In some river reaches of the Kafue Flats, low DO concentrations of <2 mg l−1 already persist over 150 km along the river due to the mineralization of organic matter originating from the Kafue Flats consuming DO [Zurbrügg et al., 2012]. A further decrease in DO concentrations caused by an altered outflow from ITT has to be avoided with respect to the health of fishes. Fish species unadapted to hypoxic conditions cannot survive if DO <2 mg l−1 [McNeil and Closs, 2007]. Moreover, spawning and the survival of fish larvae were found to be hampered severely under such conditions [Fontenot et al., 2001]. We thus included the threshold of DO <2 mg l−1 as a water quality criterion for evaluating scenario modeling outcomes (see below).

[8] On the other hand, an increase in nutrient output from ITT is beneficial for the particular nutrient-poor Kafue Flats as it compensates for N and P sediment removal in upstream ITT. Reducing sediment removal would move the system toward a more natural, predam-like state. In addition to this fundamental argument, dam-induced ecological changes in the Kafue Flats may have been aggravated by decreased nutrient inputs to the Kafue Flats. Among these changes are the invasion of low-nutrient tolerant, N-fixing plants [Mumba and Thompson, 2005], the reduction in the abundance of Kafue Lechwe, an endemic antelope [Chabwela et al., 1990; Schelle and Pittock, 2005], as well as negative effects on ecosystem services, such as subsistence farming and small-scale fisheries. The latter effect is however aggravated by overfishing, which was found to be the main reason for the recent decline in fish abundance [Deines et al., 2013].

[9] In line with the invasion of N-fixing plants, a N budget for the whole Kafue Flats illustrated that N fixation was indeed the largest N source accounting for ∼20,000 t N yr−1, which is five to seven times higher than total N inputs from ITT [Zurbrügg et al., 2013]. In the absence of an analogous pathway for P and low fertilizer usage in the catchment, river-transported P must be an important source for the floodplain, despite the generally low P loads in the Kafue River due to burial in upstream wetlands and low erosion rates owing to the flat topography. Unfortunately, no P budget has been calculated for the Kafue Flats so far, and therefore the importance of riverine P inputs for the entire floodplain remains unquantifiable. However, we assume that regularly flooded and Kafue River-adjacent areas of the floodplain largely depend on the riverine nutrient sources. This is illustrated by the intense hydrologic exchange between river and floodplain in these areas. More than 80% of the river flow passes through the floodplain and export large quantities of organic C and N from the Kafue Flats while reentering the Kafue River [Zurbrügg et al., 2013]. Moreover, despite the dam-induced reduction in discharge compared to the predam state during the flooding period, Zurbrügg et al. [2012] showed that the river-floodplain exchange is a year-round phenomenon and not, as one might expect, limited to the flood season.

[10] An optimized outflow water quality from ITT should thus aim at higher nutrient outputs to mitigate nutrient losses in ITT. Ideally, such a nutrient pulse should take place during the flooding season when the hydraulic exchange between river and floodplain is most efficient allowing the nutrients to enter the Kafue Flats. During low-flow conditions, the river-floodplain exchange is reduced; hence, nutrient outputs from ITT may mainly remain in the main river channel flowing through the Kafue Flats [Wamulume et al., 2011; Zurbrügg et al., 2012].

[11] We developed the 1-D model RES1 [Kunz et al., 2011a] to examine the sensitivity of ITT's biogeochemical cycling and its outflow water quality toward different withdrawal scenarios. RES1 is a derivative of BELAMO, a biogeochemical model that was successfully applied to a wide range of different lakes and reservoirs as well as for scenario modeling [Finger et al., 2007; Matzinger et al., 2007a, 2007b; Mieleitner and Reichert, 2006, 2008; Omlin et al., 2001]. RES1 simulates DO, N, and P cycling as well as the processes of growth, respiration, and death of phytoplankton and zooplankton, N fixation, denitrification, particle settling and sediment removal, P adsorption to sinking particles, and mineralization in the water column and at the sediment surface. The physical and the biogeochemical submodels of RES1 were calibrated with data from our 2008/2009 measurement campaign [Kunz et al., 2011a]. The measurements included monthly depth profiles of temperature, DO and the nutrients NO3 and NH4+ (summing to dissolved inorganic nitrogen (DIN)) as well as soluble reactive phosphorus (SRP) and were collected at the deepest site near the dam. RES1 thus simulates the part of ITT that governs the outflow water quality by integrating horizontal inhomogeneities in the reservoir. A more detailed description of RES1 with the available data is given in Table S1 (available as supporting information) and by Kunz et al. [2011a].

[12] To optimize the robustness of model simulations, RES1 was run continuously for several years using a representative range of constant boundary conditions until the model reached a steady state. First, we accounted for the large interannual variability in the inflow to ITT by performing individual runs for three different hydrologic scenarios defined by maximum (Qmax; years 2008/2009), minimum (Qmin; 1996/1997), and daily averaged (Qavg) discharge documented in the hydrological record from 1978 to 2009, respectively (Figure 2 and Figure S1, provided in supporting information; ZESCO, unpublished data, 2009). This range of discharge curves thus also covers the boundary conditions in which hydrologic optimization for ITT, e.g., with environmental flows, can be realized [Beilfuss, 2012]. Second, we accounted for seasonal surface water level fluctuations by running RES1 with high and low levels (zmax = 50 m and zmax = 46 m depth, respectively, corresponding to maximum and minimum recorded levels during the RES1 calibration period).

[13] The available data set constrained RES1 to a relatively simple model architecture which caused limitations in its representativeness. Because our primary interest was in the downstream water quality and not in reproducing spatial heterogeneities within ITT, limiting the spatial resolution to 1-D representing data collected near the outflow was appropriate.

[14] Certain biogeochemical processes in RES1 were represented in a rudimentary way. Most importantly, slow anaerobic degradation of relatively old and refractory organic matter and the resulting release of reduced substances were found to be crucial for the overall DO depletion in the hypolimnion [Kunz et al., 2011a]. This complex degradation process was only implicitly included in RES1 via the exerted DO demand. We, however, accounted for the accumulation of reduced substances in the hypolimnion by estimating the concentration of DO equivalents as described by

display math(1)

where [Red] is the summed concentrations of reduced substances, mainly CH4 and NH4+, but also minor components such as NO2, S(-II), etc. [Matzinger et al., 2010; Müller et al., 2012]. Since only NH4+ is simulated by RES1, [Red] was approximated with [NH4+]/0.3 as NH4+ was found to account for ∼30% of the total accumulated reduced substances ITT [Kunz et al., 2011a] as well as in natural lakes [Müller et al., 2012].

[15] Detailed data about the dynamics of phytoplankton and zooplankton in ITT were lacking, thus preventing a direct calibration of the biological processes. Nevertheless, since the temporal and vertical development of the goal variables—concentrations of DO and nutrients—were well reproduced by RES1, we could show that the model is well capable of representing the current biogeochemical cycling in ITT [Kunz et al., 2011a].

[16] The modeling goal of an enhanced outflow water quality from ITT may be reached by choosing the optimal withdrawal depth. Hence, the set of tested withdrawal scenarios for ITT included the current as well as various simulated turbine withdrawal levels (Table 1). In the “BASE” (current) withdrawal scenario, surface water spillways are the only way to release water from ITT. The projected turbine installation by the dam operators is characterized by deep water withdrawal located in front of the dam and 5 m above the maximum depth of 50 m (withdrawal scenario “TURB1”). The turbines have a projected maximum discharge capacity of 306 m3 s−1; excess discharge is released through the spillways at the reservoir surface (Figure 2) [Piésold, 2003].

Table 1. Definitions of Withdrawal Scenarios
ScenarioSpillway Withdrawal Depth (m)Turbine Withdrawal Depthfepic
zhypoa (m)zepib (m)
  1. a

    Hypolimnetic turbine withdrawal depth.

  2. b

    Epilimnetic turbine withdrawal depth.

  3. c

    Proportion of total turbinated discharge originating from the epilimnetic withdrawal.

BASE2.7
TURB12.745.8
TURB22.720.01
TURB32.745.87.5, 10, 12.5, 15, 17.5, 201/10, 1/5, 1/4, 1/3, 1/2, 2/3, 3/4, 4/5

[17] Withdrawal of deep hypolimnetic water via turbines causes drawdown of relatively warm epilimnetic water. This drawdown heats the hypolimnion and accelerates the convergence of epilimnion and hypolimnion water temperatures during the period of cooling. Consequently, deep seasonal convective mixing sets in earlier and lasts longer compared to the BASE withdrawal scenario. The beginning of the mixing period was set in RES1 to the date when the temperature difference between epilimnion and hypolimnion was <0.05°C. The mixing period was simulated by applying a high vertical diffusion coefficient to the entire water column.

[18] A straightforward solution to increase DO concentrations in the downstream flow is to replace the withdrawal of deep water by water from 20 m depth allowing for temporal turbination of DO-rich epilimnetic water (withdrawal scenario “TURB2”). On the downside, we anticipated that TURB2 would yield lower nutrient release to the downstream wetlands compared to TURB1 due to lower nutrient concentrations in the epilimnion. Hence, a technically more elaborate scenario was simulated with both epilimnetic and hypolimnetic turbine withdrawals (“TURB3”). To evaluate the optimal setup of TURB3, the upper turbine withdrawal level (zepi) and the relative distribution of turbine withdrawal from the two levels were varied. The fraction of epilimnetic turbine withdrawal (Qepi) compared to total turbine outflow (summing Qepi and hypolimnetic withdrawal, Qhypo) was expressed with the parameter

display math(2)

[19] Combining six levels for zepi (7.5, 10, 12.5, 15, 17.5, and 20 m) with eight levels for fepi (1/10, 1/5, 1/4, 1/3, 1/2, 2/3, 3/4, and 4/5) yielded 48 different simulations. Based on the outcome from these 48 TURB3 subscenarios, the optimal withdrawal scheme for ITT will be identified. For simplicity, the TURB3 subscenarios were only based on the Qavg hydrologic scenario. However, once found, the TURB3 subscenario describing the optimal withdrawal scheme will be run for all three hydrologic scenarios.

[20] The TURB3 subscenarios were evaluated based on four constraints rating the water quality released from ITT. DO conditions in the outflow were evaluated based on the duration of hypoxic conditions (i.e., number of days with DO < 2 mg l−1; thypox) and based on the annual mean DO concentration (DOmean, i.e., mean outflow DO concentrations in the mixture of spilled and turbinated water). DOmean allows judging the scenarios in more detail concerning the decrease in outflow DO concentration, especially if tanox = 0. To evaluate nutrient release from ITT, annual output loads of DIN (Nout) and of SRP (Pout) were simulated. Annual loads are expected to be dominated by the seasonal contributions during the flooding season when both discharge (Figure 2) as well as N and P concentrations (see below) are high. Hence, they are a suitable proxy for the ecologically relevant nutrient inputs to the Kafue Flats. An integrated rating for each scenario x was quantified by aggregating the constraints thypox, Nout, and Pout depending on the epilimnetic fraction fepi and depth zepi yielding a first objective water quality function, WQ1, for minimizing hypoxic periods and maximizing nutrient output

display math(3)

and by aggregating the constraints DOmean, Nout, and Pout resulting in a second objective function, WQ2, maximizing both DO concentrations and nutrient output

display math(4)

where thypox,max is the maximum duration of hypoxic outflow, DOmean,min is the minimum annual average DO concentration, and Nout,min and Pout,min are the minimum annual outflow loads of N and P, respectively, for all 48 TURB3 subscenarios. The ratios including thypox and DOmean were squared in equations (3) and (4) to equally weigh DO and nutrient constraints. As both the DO and the nutrient constraints have to be fulfilled, the quotients in equations (3) and (4) were multiplied. Scenarios with lowest WQ1 and lowest WQ2 are considered most beneficial regarding the quality of released water to the Kafue Flats.

3. Results and Discussion

3.1. Temperature in the Water Column and in the Outflow

[21] The thermal regime of ITT was substantially altered in the withdrawal scenarios TURB1, TURB2, and TURB3 (Figure 3). In comparison to the withdrawal scenario BASE, the hypolimnion exhibited higher temperatures due to the drawdown of the warmer epilimnetic water. This effect increased with increasing turbine withdrawal depth and with increasing discharge. Hence, heating of the hypolimnion was largest in the TURB1/Qmax scenario.

Figure 3.

Contour plots illustrating the water column temperature (T, °C) of ITT for the four different withdrawal scenarios (“BASE” to “TURB3”) and for the three different hydrologic scenarios (Qmax, Qavg, and Qmin; Figure 2) with maximum reservoir water level. Withdrawal of hypolimnetic water caused drawdown of warmer epilimnetic water. Consequently, warming of the hypolimnion resulted in earlier deep seasonal convective mixing relative to the BASE scenarios and was most pronounced for TURB1/Qmax. For TURB3 only the three scenarios defined by fepi = 0.5 and zepi = 12.5 m are shown. Red arrows at the right y axis mark turbine withdrawal levels; the green arrow indicates mean spillway outlet levels.

[22] Depending on the warming of the hypolimnion, water column temperatures leveled out earlier and deep seasonal convective mixing started earlier and lasted longer compared to the withdrawal scenario BASE (Figure 3). Of all withdrawal and hydrological scenarios, the earliest beginning of the mixing period was on 22 March (TURB1/Qmax), the latest on 6 June (TURB2/Qmin) as compared to 20 June (BASE). Restratification started on 6 August for all scenarios.

[23] The effect of the changes in the water column temperatures on the outflow temperatures was minor. While withdrawing hypolimnetic water in the TURB1 to TURB3 withdrawal scenarios resulted in some cooling of the outflow during the hottest months October to January, minimum outflow temperatures of ∼18°C during deep seasonal convective mixing in ITT remained unchanged (data not shown). Hence, as the outflow temperatures of the TURB1 to TURB3 withdrawal scenarios remained within the bandwidth of simulated temperatures of the BASE scenarios, no serious temperature effects, e.g., on fishes, have to be expected from turbine installations.

3.2. Dissolved Oxygen in the Reservoir and in the Outflow

[24] Simulation runs with RES1 were used to evaluate the effect of different withdrawal scenarios on biogeochemical cycling in ITT. Both for the scenarios BASE and the withdrawal scenarios TURB1, TURB2, and TURB3, DO is rapidly depleted due to organic matter mineralization in the hypolimnion after the onset of the stratification in August, leading to anoxic conditions below ∼20 m depth from December on (Figure 4) [Kunz et al., 2011a]. Epilimnetic DO concentrations varied between ∼6 and ∼8 mg l−1. DO concentrations were homogenized throughout the water column during the seasonal deep convection.

Figure 4.

Contour plots illustrating the concentration of water column DO equivalents (DOequiv, equation (1), mg l−1) of ITT for the four withdrawal scenarios and the three hydrologic scenarios (as in Figure 3). Negative values (red colors) indicate the accumulation of reduced substances exerting a DO demand. Red arrows at the right y axis mark turbine withdrawal levels; the green arrow indicates mean spillway outlet levels.

[25] The accumulation of reduced substances as included in the definition of [DOequiv] (equation (1)) depended on the turbine withdrawal levels (Figure 4). TURB1 and TURB3 withdrawal scenarios exhibited only moderate negative [DOequiv] (> −1 mg l−1). Withdrawal of deep hypolimnetic water in these scenarios resulted in the drawdown of oxygen-rich epilimnetic water, which partly oxidized reduced substances. In turn, [DOequiv] was minimal for BASE (Qmax: −2.7 mg l−1) and TURB2 (Qmax: −2.4 mg l−1) owing to the lack of deep hypolimnetic water withdrawal and to the limited drawdown of epilimnetic water to the hypolimnion. At the turbine withdrawal levels [DOequiv] was generally low for all withdrawal scenarios (>−1 mg l−1).

[26] Dissolved methane contributes the largest portion of DOequiv (∼70% [Kunz et al., 2011a]), but may be efficiently released to the atmosphere in the turbulent dam outlet [Abril et al., 2005]. Consequently, most of the oxygen demand described by negative DOequiv values in the water column does not result in a substantial decrease in the outflow DO. Even if the outgassing of methane is neglected and assuming that the reduced substances represented by DOequiv are readily oxidized, the outflow DO would be insignificantly lower: the decrease would be maximum (0.6 mg l−1) for the TURB1/Qmin scenario during 35 d in May and June, but <0.2 mg l−1 during 260 d. DOequiv was therefore omitted in the following discussion about outflow water quality.

[27] Outflow DO concentrations were calculated based on simulated DO water column concentrations. In scenario BASE, DO varied between 6.5 and 8.5 mg l−1 during the simulation period, entirely determined by epilimnetic DO concentrations in ITT (Figure 5). In scenarios TURB1, TURB2, and TURB3, outflow DO concentrations decreased in parallel with decreasing hypolimnetic DO concentrations in the water column (Figure 5). During this period (August to early January), spillways were permanently closed in all hydrologic scenarios since Q < 306 m3 s−1 (Figure 2), thus, no DO-rich epilimnetic water was diverted to the outflow. For TURB1 and TURB2, the decrease in outflow DO concentrations culminated in hypoxic DO levels of <2 mg l−1 starting in December. In the most extreme case (TURB1/Qmin), the hypoxic period lasted 189 d and was only relieved by mixing of epilimnetic water during the seasonal deep convection starting in June. For the hydrologic scenarios Qavg and Qmax of TURB1, the hypoxic period lasted only 108 and 84 d, respectively. Surprisingly, it was beneficial only in the Qmin scenario to have relatively high turbine withdrawal levels (20 m depth in TURB2 versus 46 m in TURB1). In this case, the downward mixing of DO-rich epilimnetic water to the turbine withdrawal level during the period of cooling in the epilimnion (January to July) shortened the hypoxic period from 189 to 112 d (Figure 5).

Figure 5.

DO concentration (blue bands, mg l−1) in the outflow of ITT for the four withdrawal scenarios and the three hydrologic scenarios (as in Figure 3). Widths of the blue bands reflect the variability introduced by RES1 runs with maximum and minimum surface water levels. For the most extreme combination of TURB1/Qmin the hypoxic period (DO < 2 mg l−1, dashed line) lasts for 189 d.

[28] The extent of the drawdown of DO-rich epilimnetic water depending on the hydrologic scenarios Qmax, Qavg, and Qmin also influenced outflow DO concentrations (Figure 5). The most efficient warming of the hypolimnion in Qmax (Figure 3) resulted in an earlier deep convective mixing and an earlier reoxygenation of the hypolimnion, and thus outflow concentrations of ∼7 mg l−1 resulted as early as in March. At the other end of the spectrum, this positive effect was nearly absent in TURB3/Qmin due to minimum drawdown of epilimnetic water.

[29] For the TURB3 subscenarios with high fepi and/or shallow zepi, hypoxic conditions in the outflow could be avoided completely (e.g., TURB3 with fepi = 0.5 and zepi = 12.5 m; Figure 5). Nevertheless, DO concentrations decreased substantially between August and early December to a minimum level of 2.1 mg l−1. DO remained the longest at this minimum level for the Qmin hydrologic scenario, for which spillways are not opened at all. In turn, for Qmax and Qavg, the opening of the spillways between February and May resulted in increased outflow DO concentrations. The entire set of TURB3 subscenarios are discussed below.

[30] We accounted for the variability of outflow DO concentrations introduced by the reservoir surface fluctuations by using simulation runs with maximum and minimum surface water levels. We found relatively small variations (<1 mg l−1) in most scenarios (Figure 5).

3.3. Concentrations of Dissolved Nutrients in the Reservoir and in the Outflow

[31] During the stratified period, DIN and SRP accumulated in the anoxic hypolimnion to reach maximum concentrations in May (Figures 6 and 7). The simulated maximum nutrient concentrations ranged from 0.14 to 0.45 mg N l−1 (TURB1/Qmin and BASE/Qmax scenarios, respectively), and from 0.02 to 0.14 mg P l−1 (TURB1/Qavg and BASE/Qmax scenarios, respectively). Owing to a relatively low thickness of the hypolimnion, mixing the seasonally accumulated nutrient stocks with the low-nutrient surface water did not result in a detectable increase in the vertically homogenized concentrations during seasonal deep mixing. Moreover, the epilimnion remains virtually nutrient-depleted (DIN < 0.06 mg l−1 and SRP < 0.02 mg l−1) throughout the year and in all scenarios due to rapid DIN and SRP uptake by organisms [Kunz et al., 2011a].

Figure 6.

Contour plots illustrating DIN concentrations (mg l−1) in the water column of ITT for the four withdrawal scenarios and the three hydrologic scenarios (as in Figure 3). Red arrows at the right y axis mark turbine withdrawal levels; the green arrow indicates mean spillway outlet levels.

Figure 7.

Contour plots illustrating SRP concentrations (mg l−1) in the water column of ITT for the four withdrawal scenarios and the three hydrologic scenarios (as in Figure 3). Red arrows at the right y axis mark turbine withdrawal levels; the green arrow indicates mean spillway outlet levels.

[32] The temporal development of DIN and SRP concentrations in the water column of ITT differed substantially among the withdrawal scenarios (Figures 6 and 7). Hypolimnion nutrient accumulation was lowest in the scenarios TURB1 and TURB3, where turbine withdrawal levels were close to the maximum depth, thus drawdown of DIN- and SRP-poor epilimnetic water diluted the concentrations in the hypolimnion. In turn, drawdown of epilimnetic water to the hypolimnion was smaller in the scenarios BASE and TURB2 allowing DIN and SRP to accumulate to a greater extent.

[33] These trends in the nutrient dynamics in the water column determined DIN and SRP outflow concentrations (Figure 8). Generally, releasing hypolimnetic water in the withdrawal scenarios TURB1 to TURB3 caused higher DIN and SRP outflow concentrations due to the seasonal accumulation of the dissolved nutrients in the hypolimnion. In TURB3, we found ∼fivefold increases in peak DIN and SRP concentrations compared to BASE (0.1 mg N l−1 and 0.02 mg P l−1, respectively; Figure 8). Largest concentration increases were observed in TURB1 due to the highest fraction of deep hypolimnetic water released through the turbines (DIN peak concentrations ∼eightfold, SRP peak concentrations ∼sixfold; data not shown). Among the hydrologic scenarios, Qmin exhibited highest concentrations followed by Qavg and Qmax, as a result of the respective dilution with low-nutrient epilimnetic water in the outflow (Figures 2 and 8).

Figure 8.

(a) DIN and (b) SRP concentration (green and red bands, respectively; mg l−1) in the outflow of ITT for the withdrawal scenarios BASE and TURB3 and the three hydrologic scenarios (as in Figure 3). Widths of the bands reflect the variability introduced by RES1 runs with maximum and minimum surface water levels.

[34] The seasonality of DIN and SRP concentration profiles in the water column was reflected in the outflow concentrations (Figure 8). Outflow DIN reached a first peak in December at the onset of the flooding season (Figure 2) owing to the accumulation of NO3 (Figure 6). For TURB1 and TURB3, DIN peaked a second time prior to deep seasonal mixing caused by the accumulation of NH4+ in the deep hypolimnion. Simultaneously, SRP concentrations exhibited a single peak at the end of the stratified period (Figure 8).

[35] These periods with elevated DIN and SRP outflow concentrations coincided with high-flow rates in the Qmax and Qavg hydrologic scenarios (Figure 2). Consequently, DIN and SRP loads were also highest during these periods, and the largest fraction of the annual loads entered the Kafue Flats synchronous to the predam situation. Moreover, decreased DIN and SRP concentrations during the low-flow period in the TURB1 to TURB3 withdrawal scenarios compared to BASE translate to lower loads (Figures 2 and 8). Lower loads during this period mitigate the increased loads in the BASE scenario caused by ITT relative to the natural conditions.

3.4. Water Quality in the Outflow to Downstream Wetland

[36] The biogeochemical model scenarios and the resulting DO, DIN, and SRP profiles in ITT allowed exploring the resulting changes in ITT outflow quality. The optimal epilimnetic turbine withdrawal level in the TURB3 withdrawal and Qavg hydrologic scenarios should avoid hypoxic conditions, minimize DO concentration deficits, and maximize DIN and SRP output loads. The sensitivity of these constraining parameters was evaluated by calculating their values for all 48 TURB3 subscenarios (Figures 9a–9d). The aggregation of the constraints to the objective functions WQ1 and WQ2 according to equations (3) and (4) resulted in an optimized solution (Figures 9e and 9f) as discussed in the following sections.

Figure 9.

Number of days of (a) hypoxic outflow (thypox), (b) mean outflow DO concentration (DOmean), (c) DIN (Nout), (d) SRP output loads (Pout), and objective functions (e) WQ1 (equation (3)) and (f) WQ2 (equation (4)) as a function of the ratio of epilimnetic water in the turbines outflow (fepi, x axis) and the epilimnetic turbine withdrawal levels (zepi, y axis). The TURB3/Qavg subscenarios based on 48 fepi and zepi combinations are depicted by gray crosses. The white shaded areas highlight the most desirable, dark blue areas the most undesirable subscenarios. The objective functions (e) WQ1 and (f) WQ2, which aggregate the Figures 9a, 9c, and 9d and 9b, 9c, and 9d, respectively, illustrate that the scenario with fepi = 0.5 and zepi = 12.5 m had minimal values for WQ1 of 0.0 and for WQ2 of 0.27 and is thus considered most beneficial.

3.5. Maximizing Outflow Dissolved Oxygen

[37] Hypoxia in the outflow from ITT can be avoided (thypox = 0) if at least one third of turbinated water originated from the epilimnion (fepi > 1/3) and if the upper turbine withdrawal level zepi was above 12.5 m depth (Figure 9a). Lower fepi values and a deeper zepi resulted in thypox > 0. The maximum thypox of 105 d was obtained with a minimum fepi of 0.1 and a maximum zepi of 20 m depth.

[38] Although hypoxia can be prevented in several TURB3 subscenarios, outflow DO levels can be further improved by decreasing zepi and increasing fepi resulting in higher DOmean concentrations (Figure 9b). Naturally, highest DOmean (6.8 mg l−1) resulted for the maximum fepi of 0.8 and the minimum zepi of 7.5 m depth. In this scenario, DOmean dropped by only 0.8 mg l−1 relative to BASE (DOmean = 7.6 mg l−1 for Qavg; Figure 5). In the most extreme scenario with minimum fepi = 0.1 and maximum zepi = 20 m, DOmean dropped by 3.5–4.1 mg l−1.

[39] There are technical solutions to avoid hypoxic conditions in the outflow. The hypolimnetic turbine outflow may be oxygenated in stepped spillways [Baylar et al., 2006]. However, the river slope downstream of ITT dam is likely to be insufficient for such structures. Alternatively, hypolimnetic aeration or oxygenation systems [Singleton and Little, 2006] in front of the turbine intakes or after the water outlet may be installed to oxygenate the turbine outflow. High running/maintenance costs are the major disadvantage of these systems. Hence, the installation of an intake tower in front of the dam that allows for withdrawing a mixture of epilimnetic and hypolimnetic water is the most feasible, operationally robust and cost-effective solution to oxygenate ITT's outflow.

3.6. Nitrogen and Phosphorus Output Loads

[40] Dissolved nitrogen output loads (Nout) were maximized in TURB3 subscenarios with high fepi and deep zepi (Figure 9c). These subscenarios withdrew predominantly water from the metalimnion at ∼20 m depth, where DIN was maximum (Figure 6). For fepi = 2/3 and zepi = 20 m, we estimated a maximum Nout of 570 t N yr−1. Subscenarios with shallower zepi withdrew water from the DIN-poor epilimnion. Minimum Nout of 250 t N yr−1 thus resulted for fepi = 0.8 and zepi = 7.5 m. The range of the estimated Nout translated to a sevenfold to threefold increase, respectively, relative to the BASE/Qavg scenario (76 t N yr−1).

[41] Pout was less sensitive toward fepi and zepi for the various TURB3 subscenarios compared to Nout, as illustrated by the narrow range between minimal and maximal Pout. The maximum Pout of 97 t P yr−1 for fepi = 0.25 and zepi = 12.5 m translated to a fourfold increase relative to BASE (23 t P yr−1), while the minimum of 64 t P yr−1 for fepi = 0.8 and zepi = 7.5 m resulted still in a threefold increase (Figure 9d). Additionally, the scenario with the maximum Nout (fepi = 2/3 and zepi = 20 m) also yielded a fourfold Pout increase. This relative insensitivity is consistent with the moderate differences in SRP across the water column (Figure 7). Generally, zepi had a small impact on variations in Pout since SRP concentrations were homogeneous in the zepi range of 7.5–20 m, leading to almost vertical contour lines in Figure 9d.

[42] The increase in Pout of TURB3 relative to BASE (range of Pout,TURB3(min,max) − Pout,BASE = 41–74 t P yr−1) has the potential to compensate 20–37%, respectively, of P burial into the sediments of ITT (200 t P yr−1 [Kunz et al., 2011a]). Hence, by turbinating hypolimnetic water, the compensation of riverine transported P losses to ITT sediments is considerable. For total N, this compensation effect is less pronounced since it was shown that DIN is a minor fraction of total N outputs (<5%), and that total N water column profiles did not substantially vary with depth because refractory dissolved organic N contributes the most significant fraction to total dissolved N (∼95% [Zurbrügg et al., 2013]). Consequently, discharge fractionation between epilimnetic water through spillways and hypolimnetic water through turbines may not substantially influence bulk N output loads.

3.7. Optimized Withdrawal Scenario

[43] Optimal withdrawal scenarios were identified by aggregating the constraints thypox, DOmean, Nout, and Pout. The objective function WQ1 (equation (3)) was minimal for TURB3 subscenarios defined by an epilimnetic fraction fepi range of 1/3–0.8 and a zepi range of 7.5–12.5 m (Figure 9e). In this relatively broad optimal range of WQ1, hypoxia was permanently avoided resulting in the minimum value of 0 for thypox, the dominant constraint for WQ1 (Figure 9a). Consequently, WQ1 was insensitive toward fepi and zepi in this range. Using WQ2 allowed narrowing the selection of optimal TURB3 subscenarios. WQ2 was minimal for the TURB3 subscenario with fepi = 0.5 and zepi = 12.5 m (Figure 9f). WQ2 had a higher sensitivity toward fepi than WQ1 because DOmean depends on fepi. To estimate the ranges of the DO and nutrient constraints depending on the amount of outflow, the hydrologic scenarios Qmin and Qmax were run in addition to Qavg for this optimal scenario. The simulations are illustrated in Figures 38 and suggested ranges of DOmean of 5.0–5.5 mg l−1 (5.2 mg l−1 for Qavg), thypox = 0, Pout of 35–130 t P yr−1, and Nout of 250–560 t N yr−1. These amounts of Pout are equivalent to 18–66% of P burial into ITT sediments.

3.8. Implications for Dam Management

[44] Our results illustrate the applicability of a combined approach involving biogeochemical reservoir modeling and a simple water quality optimization procedure for ecologic dam management optimization. This approach is a valuable addition to hydrologic optimization strategies and in line with the Brisbane Declaration [2007], which defines environmental flows as both a quantitative and qualitative concept to protect wetlands affected by river damming. Water quality aspects are thus explicitly incorporated as requirements for operation rules in our approach, which was postulated to be most important for ecosystem health [Jager and Smith, 2008].

[45] The ITT model system illustrated that selective withdrawal of hypolimnetic and epilimnetic water provides the required flexibility to effectively optimize downstream water quality. Similarly, others have found that selective withdrawal offers a possibility to improve the reservoir water quality without hampering downstream water quality [Lehman et al., 2009]. Optimized downstream water quality may contribute to the restoration of the Kafue Flats downstream of ITT, particularly to support endemic flora and fauna, such as the Kafue Lechwe [Chabwela et al., 1990; Mumba and Thompson, 2005; Schelle and Pittock, 2005] and to maintain ecosystem services for subsistence farming and small-scale fisheries.

[46] Fortunately, dam optimization does not necessarily result in an economic trade-off. Applying the findings of this study would not result in a lower power generation at ITT, since the amount of discharge available for the turbines is independent of the withdrawal depth. Moreover, Tilmant et al. [2010] showed in their Zambezi-wide study that dam optimization in the Kafue River Basin involving environmental flows would not result in economic losses, as the power output at Kafue Gorge Dam may be maintained under such an optimized flow allocation regime.

4. Conclusions

[47] This study illustrates that optimized turbine withdrawal levels have the potential to substantially enhance outflow water quality. The combination of a biogeochemical reservoir model with scenario analysis of different turbine withdrawal levels and simple ecological goal functions offers an efficient approach to assess options of the dam operation for ecological requirements of downstream aquatic and terrestrial ecosystems.

[48] In this study, we demonstrated that outflow water quality from ITT reservoir to the downstream Kafue Flats may be optimized depending on different withdrawal and hydrologic scenarios at the dam. Simulations of biogeochemical cycles in ITT using the model RES1 were evaluated based on the ecological goal functions that aggregated the constraints “duration of hypoxia,” “average oxygen concentrations,” and “dissolved inorganic nitrogen and phosphorus loads.” From the scenario simulating the projected turbine installation, which are characterized by a deep water withdrawal located in front of the dam, both negative and positive effects for the outflow water quality were found. While hypoxic conditions in the Kafue River are likely to result in fish kills and thus negatively impact fisheries in the Kafue Flats, increased nutrient loads help to compensate for nutrient removal in ITT's sediments. Maximum nutrient loads accounting for up to 66% of phosphorus retention in the sediments may be achieved by withdrawing deep hypolimnetic water. However, this scenario leads to hypoxic conditions in the outflow during up to ∼190 d yr−1. Our scenario simulations suggest that this undesirable situation could be avoided if turbines withdrew a mixture of oxygen-rich epilimnetic and nutrient-rich hypolimnetic water. Turbinated water should thus consist of ∼50% epilimnetic water originating from ∼13 m water depth, and ∼50% deep hypolimnetic water (>30 m depth). In this scenario and for average discharge conditions, outflow DO decreased moderately from 7.6 to 5.2 mg l−1, and dissolved nitrogen and phosphorus output loads could be drastically increased by a factor of 5 compared to the current dam operation.

Acknowledgments

[49] The project was funded by the Competence Center Environment and Sustainability (CCES) for the African Dams Project (ADAPT) and by Eawag. Modeling support was provided by Martin Schmid. We thank Moses Mbuta (ZESCO) for making discharge data available and for his support during fieldwork. We profited greatly from the help of Christian Dinkel, Roland Zurbrügg, Jason Wamulume, Jeremias Fellmann, Ruth Stierli, Adrian Vollenweider, Silvan Thüring, Daniel Ineichen, Leticia Stojkovski, Gijs Nobbe, and Alois Zwyssig (all Eawag) during field and laboratory work. Three anonymous reviewers helped significantly improving an earlier version of the manuscript.

Ancillary