Managing Cyanobacteria Blooms in Lake Hume: Abundance Dynamics Across Varying Water Levels

Abstract

Lake Hume, a critical reservoir within the Murray River system, Australia, has been identified as a potential source of cyanobacteria in downstream rivers during past mega-blooms. This study aims to evaluate the impact of lake-level fluctuations on cyanobacterial abundance at the dam outlets, with the goal of mitigating the risk of cyanobacteria intake from hydropower and irrigation outlets during periods of low dam levels. Utilising a one-dimensional vertical hydrodynamic model (LAKEoneD), the study simulated time series data on water temperature and stratification within Lake Hume. These outputs were then incorporated into a cyanobacteria growth model driven by water temperature, mixing dynamics and light. Despite inherent uncertainties in the models, the simulated cell counts effectively mirrored bloom occurrences. Consequently, a series of simulations across varying water levels in the lake revealed a consistent risk of significant cyanobacteria intake through both the hydropower and irrigation outlets when water levels dropped below specific thresholds. Notably, water levels below 20 m and 10 m posed heightened risks of releases of seed populations of cyanobacteria from the hydropower and irrigation outlets, respectively.

1. Introduction

Cyanobacteria, also known as blue-green algae (BGA), are ancient prokaryotes capable of forming dense, sometimes toxic blooms in still waters. These blooms pose significant threats to aquatic ecosystems and human health [1]. Cyanobacterial blooms can obstruct sunlight penetration, and their decomposition depletes oxygen, which is essential for the survival of other aquatic organisms [2]. Furthermore, cyanobacteria frequently produce cyanotoxins, such as microcystins, saxitoxins, anatoxins and cylindrospermopsins, being of primary concern [3,4]. These toxins pose significant threats to water safety for humans, aquatic organisms, and livestock, and their treatment is often expensive [5,6]. A global survey indicates that cyanobacterial blooms are increasing in range, magnitude, frequency, and duration in freshwater lakes worldwide [7,8,9,10]. These increases are likely to continue given the predicted impact of climate change and global warming [11,12,13], as cyanobacteria growth depends strongly on climatic drivers such as temperature as well as non-climatic drivers such as riverine nutrient run-off [14].

In Australia, surveys of cyanobacteria blooms identified the same increasing trend in lakes and reservoirs [3,15]. In contrast, large-scale cyanobacterial blooms in the Murray River, the most economically important waterway in Australia, were uncommon. Following the Millennium Drought (2001–2010), however, their frequency in the downstream River increased significantly, with five mega-blooms occurring in the last 13 years compared to only four in the preceding 67 years [16,17,18,19,20,21]. There is evidence now that these blooms were seeded from Lake Hume water storage, or from billabongs (oxbows) adjacent to the Murray River rather than in flowing sections of the river [22,23]. Lake Hume is the largest storage in the Murray River system, primarily providing water for large-scale irrigated agriculture further downstream during the Austral spring and summer. The mode of operation of the storage aims for maximal capture and storage of early spring run-off from the extensive catchment, followed by large releases during the main growing season (October to February). Consequently, the water level in the reservoir varies seasonally with changes up to about 20 m, with the maximum in early spring and falling to a minimum in late summer. Under these summer low-water-level conditions, the bottom of the seasonal mixed layer [24] sometimes sinks to the level of either of the two reservoir outlets (separate levels for irrigation releases, and hydropower generation). These low water levels in summer are known to be conducive to conditions favourable to the growth of cyanobacteria.

The formation and development of cyanobacterial blooms is a complex process influenced by the combined effects of multiple factors, from biogeochemical to hydrodynamical factors. It is well established that hydrodynamic changes, including water turbulence, water stratification, shear stress, sediment deposition and distribution, etc., could alter the environmental conditions for algal growth [25]. Water-level fluctuation affects the occurrence of cyanobacterial blooms through retention time, turbulent mixing intensity, water column nutrient concentrations, and temperature. Many studies conclude that an increase in algal blooms is associated with a reduction in water level, whether through nutrient concentration [26], precipitation, wind speed and temperature [27], or rainfall patterns [28]. Li et al. [29] stated that as the water-level decline increased, there was a significant increase in algal diversity and a notable decrease in algal cell density. Bloom occurrence in Lake Hume is linked to its water level; decreasing levels means increasing the risk of blooms through increased nutrient exchange across the metalimnion. It is assumed that at 10% reservoir capacity, there is a high likelihood of bloom formation. The water level in Lake Hume at the onset of this project was much higher than 10%, but with the main irrigation releases in the summer, the dam water level was expected to fall.

The concentration of buoyant cyanobacteria is highest at the surface layer under calm conditions. This mixed surface layer is defined as the depth from the surface of waters affected by turbulence generated at the surface by meteorological factors, primarily the surface wind stress and the daily cycle of surface heating and cooling [30,31]. Increasing the mixing depth also releases the nutrients from the hypolimnion into the upper water column, facilitating the subsurface bloom. Moreover, under conditions of strong winds (high turbulence), the cyanobacteria are redistributed downward into the deeper surface layers. Combining with the low-water-level condition, the cyanobacteria are potentially spread to the level of the dam outlets. Water releases under these more highly mixed conditions, when cyanobacterial concentrations are higher at the level of the outlets, have the potential to discharge seed concentrations downstream.

We hypothesise that surface and subsurface blooms might extend to deeper layers of the water column when the water level decreases, hence increasing the risk of cyanobacteria intake through the dam outlets. If releases occur at these times, high cyanobacteria concentrations are advected downstream, providing seeding populations when suitable environments are encountered. Confirmation (or otherwise) of this proposition is a necessary pre-condition for devising mechanisms to mitigate the downstream cyanobacterial blooms ultimately sourced from upstream storage such as Lake Hume. Observational examination of the role of deepening of the mixed layer in increasing outlet cyanobacterial levels is logistically complicated and very resource-intensive, as well as contingent on having the right meteorological conditions at the time of observations. The costs are prohibitive, and we have chosen to examine this question using a well-validated modelling approach. In this paper, we use a well-established 1D hydrodynamic model, coupled with a light- and temperature-driven cyanobacterial growth model, to explore if such enhancement of cyanobacterial concentrations at deeper layers is feasible and whether our hypothesis is supported by the modelling. In situ observations of temperature profiles, cyanobacterial abundance, and local meteorological conditions are used in the calibration/validation of the model, greatly strengthening the reliability of the model results.

Understanding the link between the hydrodynamics of Lake Hume and BGA bloom is essential to explore strategies aimed at mitigating the risk of the release of BGA from Lake Hume downstream. The utilisation of hydrodynamic models holds promise in facilitating the assessment of the dynamics surrounding thermal stratification buildup and dissipation, as well as evaluating the associated risk of cyanobacteria proliferation based on their temperature-dependent growth characteristics. These hydrodynamic models can also be used to run specific scenarios on the effect of inflows from the upstream Dartmouth Dam to change Lake Hume’s thermal and mixing regimes and, thus, possibly suppress cyanobacteria blooms. The model can also run scenarios on the effect of changing lake water levels on cyanobacteria concentration. Various simulation packages are accessible for depicting coupled hydrodynamic and biogeochemical processes [32,33]. However, there is a substantial trade-off between model complexity by including a multitude of processes and in-lake data availability to validate such models in specific cases. The growth dynamics of phytoplankton are intricately linked to the prevailing physical conditions of their environment. While temperature, light availability, and mixing patterns within the water column play pivotal roles in facilitating growth, it is important to note that these factors do not individually dictate the onset and decline of blooms. The primary determinants governing cyanobacteria growth, along with the parameterisation process to formulate a straightforward yet universally applicable model, are extensively detailed by Jöhnk et al. [34].

Lake Hume is a major reservoir of the River Murray. A map of Lake Hume and the Upper Murray catchment can be found in Figure 1. Lake Hume was a source of BGA in the Murray River from the dam wall to Lake Mulwala blooms in 2003, 2005, 2007, 2009 and 2010 [22,23,35,36], and most likely in 2016 [37]. These blooms are often conspicuous due to their occurrence at the surface layer. However, subsurface blooms, which can also harbour dense cyanobacterial populations, present a different scenario. These subsurface blooms are generally less visible and often overlooked, despite their potential toxicity and frequent occurrence in sensitive water bodies, such as irrigation reservoirs like Lake Hume [38].

Blooms observed in Lake Hume were found to coincide with the intermittent disruption of the lake’s seasonal thermal stratification during summer and autumn, facilitating the release of nutrients from the hypolimnion into the upper water column [39]. This phenomenon typically occurs when the water level in Lake Hume declined below 10% of its total capacity, resulting in shallower depths and a diminished stratification compared to the robust and enduring stratification observed during periods of near full capacity [40]. However, the bloom event documented in 2016 represented a departure from this established pattern and may portend a future scenario for Lake Hume. Notably, in 2016, the lake’s water volume was approximately 40% of its total capacity at the onset of the blooms. This cyanobacteria bloom exhibited marked deviations from typical occurrences in several respects, for instance, the occurrence of an atypical species, Chrysosporum ovalisporum (formerly classified as Aphanizomenon ovalisporum). On some occasions, this ecologically plastic species can bloom at 22 $°$C [41]. However, it is found that in this circumstance of the Australian system, Ch. ovalisporum typically has high optimal growth and was documented to dominantly bloom at a warm temperature of above 26 $°$C [42,43]. While this species had been previously identified within the Murray–Darling Basin (MDB), it had not hitherto formed blooms within the river system. Historically rare in Australian aquatic environments, recent observations indicate its proliferation in various lake and river systems [44]. Moreover, the magnitude of the observed bloom surpassed prior occurrences, potentially attributable to an anomalous, enduring warm period prevalent across the region. Notably, the bloom endured well into winter, defying conventional expectations regarding seasonal dynamics. The appearance, and now apparent dominance, in Lake Hume of previously unknown species may be a manifestation of the ecosystem’s response to the warming climate [14]. According to climate statistics published by the Australian Bureau of Meteorology, at the Hume Reservoir station, the mean maximum and minimum annual temperature increased linearly for the periods of 1981–1990, 1991–2000, 2001–2010 and 2011–2020, while the mean annual rainfall (mm) and the mean number of days of rain more than a millimetre decreased in the same periods. The Murray–Darling Basin Authority has published the Hume Dam’s water information since 1969 (https://riverdata.mdba.gov.au/list-view/hume-dam, accessed on 25 January 2025). In 2024, for example, the Hume Dam water volume (full storage capacity of 3005 GL) was in the range of 2000 GL at the start of the water year (July), reaching 2400 GL in spring (September), and down to 1000 GL in summer (February).

Unlike natural lakes, reservoirs are typically subject to more intensive anthropogenic use and management, particularly those serving multiple purposes such as water supply, hydroelectric power, and flood protection [45]. These reservoirs are generally distinguished by smaller ratios of hypolimnetic to epilimnetic volume, elevated rates of hypolimnetic oxygen consumption and nutrient release, and a thermal stratification that is highly sensitive to meteorological variables [46] (a thermally stratified water column is typically conceptualised as a two-layer system: an upper, well-mixed, warm epilimnion, and a lower, weakly mixed, cold hypolimnion, separated by the metalimnion, which is characterised by steep temperature and density gradients and thus inhibits mixing [47]). The utilisation of large water storage reservoirs often leads to significant environmental issues, including eutrophication, which can result in the development of cyanobacterial blooms during the summer months. Many large reservoirs in Australia are prone to these blooms [23].

Falling water levels are critical in terms of bloom formation. Understanding the effect of water-level fluctuations on the promotion of cyanobacterial blooms is important in lake and reservoir management [48], especially in reservoirs, where water is typically removed for human water demand, and the rate and extent of water-level fluctuation are mostly larger than in natural shallow lakes. In this simplified analysis, we focus only on the effects of fluctuations in water level on the cyanobacterial abundance in a water column, which will offer efficient management strategies by monitoring simple parameters like water levels. From an operational standpoint, real-time monitoring systems that continuously evaluate the likelihood of cyanobacterial blooms during water releases are indispensable for guiding river management strategies and optimising water resource utilisation.

The study presented here is part of a research project whose main objective was to undertake monitoring and modelling efforts targeting blooms of novel and pre-existing BGA species recognised as emerging threats to water quality in Lake Hume and downstream regions of the Murray River. This research is particularly pertinent given the anticipated exacerbation of BGA blooms in the future, attributed to escalating global warming resulting from climate change [34,37,49]. This project has developed and validated a hydrodynamic model specific to Lake Hume. Leveraging pre-existing data collected during the Millennium Drought, as well as the installation of thermistor chains to capture contemporary conditions, the model offers a comprehensive assessment tool. It evaluates the potential for blue-green algae (BGA) blooms in Lake Hume under varying climate conditions and different water transfer scenarios to and from the lake. The outcomes of this assessment hold promise for informing the formulation of novel strategies to mitigate BGA blooms within Lake Hume and its downstream areas. Additionally, the project offers actionable recommendations pertaining to reservoir operation, with the aim of curbing the occurrence of cyanobacterial blooms in freshwater lakes in the future [50,51].

The results will be presented in two publications. In this part I, our objective was to evaluate the influence of lake-level fluctuations on cyanobacterial abundance at the dam outlets using hydrodynamics models. A subsequent part II will be dedicated to investigating the effects of managed flows from upstream reservoirs, aimed at altering the mixing dynamics and stratification within the lake, with the potential to mitigate blooms of blue-green algae (BGA). Part II will be analyzed in a future publication.

2. Methods

2.1. Study Site

Lake Hume is a large reservoir with a full storage capacity of 3005 GL and a surface area of 20 km². It was created by the construction of the Hume Dam (36$°$06$\prime$30$″$ S, 147$°$01$\prime$52$″$ E), located below the confluence of the Murray and Mitta-Mitta River. The dam intercepts Australia’s largest river, the Murray River, at 305 km downstream from its source in the Australian Alps and 2225 km upstream from the Murray Mouth, where the river meets the sea. The dam’s construction was completed in 1936, and it serves multiple purposes, including hydroelectric power generation, irrigation, domestic and livestock water supply, as well as urban water provision for Victoria and New South Wales [52].

The reservoir draws its water from the Upper Murray catchment, which spans 15,280 km² [53]. Although this catchment represents only 2% of the Murray–Darling Basin, it contributes 17% of the basin’s total water, making it one of Australia’s largest freshwater storages. While major river valleys in the catchment have been cleared for agricultural purposes, 80% of the area remains forested and largely unsuitable for cultivation. The lake serves multiple purposes: irrigation water supply, flood mitigation, hydropower generation and aquatic recreation.

2.2. Sampling Sites and Physical Measurements

To achieve our objectives of simulating various water-level scenarios to assess their impact on stratification and cyanobacterial development, particularly in the outlet region of the lake, we implemented a comprehensive monitoring approach. This involved the deployment of long-term thermistor chains and temperature loggers at five strategically selected locations (

Table 1. Monitoring station and instrument deployment at Lake Hume.

Station Location	Station	Australian Height Datum	Longitude	Latitude	Instruments Deployed
Dam Wall	HUME_DAM	157 m	36$°$06$\prime$00$″$	147$°$02′00$″$	thermistor-chain, met-station
North of the Dam	HUME01	157 m	36$°$05$\prime$05$″$	147$°$03′07$″$	four loggers located at 0.5, 5, 15, and 25 m below surface
South of the Dam	HUME09	160 m	36$°$07$\prime$02$″$	147$°$02′03$″$	four loggers located at 0.5, 5, 15, and 25 m below surface
Calder Bay	HUME10	167 m	36$°$00$\prime$07$″$	147$°$05′04$″$	two loggers located at 0.5, 15 m below surface
Huon	HUME06	171 m	36$°$01$\prime$00$″$	147$°$04$\prime$05$″$	two loggers located at 0.5 and 15 m below surface

and Figure 1). The locations are chosen as close as possible to monitoring sites in an earlier study of water column temperature structure and dynamics in Lake Hume [40,50]. These instruments facilitated the collection of high-resolution temperature data, with measurements recorded at 10-min intervals. Additionally, supplementary environmental parameters such as wind speed and direction, air temperature, and humidity, as well as short- and long-wave radiation, were also monitored concurrently. In each of the Murray and Mitta-Mitta arms, there are two designated sites, complemented by an additional station situated in the main expanse of the lake. This station is positioned in close proximity to the instrumented pontoon (Figure 2), which is anchored near the dam wall. The pontoon, which is close to the outlets, hosts a meteorological station that records air temperature, relative humidity, wind speed, and wind direction at a height of 2 m above the water surface. It also measures up- and downwelling irradiance at the water surface. Additionally, the pontoon accommodates a logger unit responsible for monitoring a thermistor chain consisting of 24 temperature sensors distributed vertically from the water surface to a depth of approximately 30 m. Continuous temperature logging was successfully obtained for the summer/autumn period of 2017 and for some stations from winter through to the summer of 2017/2018. A complete data collection coverage of the 2019/2020 period was also captured.

The stations’ names and detailed locations are given in Table 1. The names follow previous field studies [40].

Water sampling and profiling, including bio-optical measurements (photosynthetic pigment concentration) and phytoplankton counts, were conducted using various instruments at these five stations during the following field campaigns for this project:

• 28 February–2 March 2017.
• 28–30 March 2017.
• 26–28 April 2017.
• 14 August 2017.
• 28 February 2018.
• 24 January 2020.
• 13 February 2020.
• 3 March 2020.
• 13 December 2020.
• 7 January 2021.

Regular sampling of cell counts and derived biovolume was conducted by WaterNSW and the Australian Water Quality Centre at multiple distinct locations within the lake. Additionally, CSIRO conducted additional cell counts at the pontoon location (station HUME_DAM). Flow cytometry was utilised for these assessments, with each sample analysed in triplicate. Moreover, cross-validation was performed using two sample days of microscopic counting.

2.3. Hydrodynamic and Algal Growth Model

2.3.1. Hydrodynamic Model—Background

We employ a one-dimensional vertical process model to characterise dynamics within the water column, under the assumption of comparability across a sizeable expanse of the lake. This approach disregards larger-scale 2D/3D processes or encapsulates them through parameterisation. As the purpose of this study is to access the abundance of the cyanobacteria at the dam outlet upon release, modelling cell abundance in a single column at the outlet is adequate. Hence, the model developed for this study utilises the one-dimensional, vertical hydrodynamic model LAKEoneD [54,55,56], with a simplified competition model of up to three phytoplankton species. The driving factors in this model are light and temperature, with nutrients assumed to be non-limiting for Lake Hume. This streamlined approach facilitates straightforward deployment, particularly in systems characterised by predominantly unknown biogeochemical attributes and intricate food webs, offering wider simulating applicability given the expansion in species ranges due to climate change impacts.

2.3.2. One-Dimensional Hydrodynamic Model

The 1D-vertical hydrodynamic lake model LAKEoneD used includes a $k-ϵ$ turbulence model [54,55,56] to describe vertical mixing processes. The model solves a set of partial differential equations for momentum and heat balance supplemented with balance equations for turbulent kinetic energy and turbulent dissipation rate [56,57]. The performance of LAKEoneD has been extensively tested in different lake environments [58,59,60,61]. LAKEoneD is driven by hourly meteorological data for short-wave radiation, air temperature, relative humidity, wind speed and direction, and cloudiness. The output of LAKEoneD consists of water temperature ($°$C), $T(z,t)$, and turbulent diffusivity (m²s⁻¹), $D(z,t)$ with depth (m), z, and time (s), t, on regular vertical and time grids. For Lake Hume, a 0.5 m depth grid and an internal time resolution of 5 min were chosen to resolve major processes.

2.3.3. Cyanobacteria Growth Model

The competition model of Jöhnk et al. [34] is used to simulate cyanobacteria growth. The different functional groups implemented in the original model differ in their parameterisation of light and temperature limitation and buoyancy/sinking characteristics. Competition for light between phytoplankton functional groups can then be described by a reaction–advection–diffusion model [62,63] as in Equation (1):

$${\displaystyle \frac{\partial C_i}{\partial t}}=p_i(I,T(z,t))C_i-L_i\left(T(z,t)\right)C_i+v_i{\displaystyle \frac{\partial C_i}{\partial z}}+{\displaystyle \frac{\partial }{\partial z}}\left(D(z,t){\displaystyle \frac{\partial C_i}{\partial z}}\right)$$

(1)

where the population density (cells per unit volume) of species i, $C_i$, is assumed to change via: light intensity ($\mathsf{\mu }$molm²s⁻¹), I; temperature, T; dependent growth, $p_i(I,T)$, and a species-specific loss rate, $L_i$, which is a function of temperature. Cells can either rise and sink (e.g., buoyant cyanobacteria) or sink (e.g., diatoms, green algae) with a constant velocity (m/s), $v_i$. The last term describes the mixing of the cells via turbulent diffusivity, $D(z,t)$. A full description of the model and a glossary with a detailed explanation and physical meaning of the parameters can be found in [34].

Zero flux boundary conditions are presumed at both surface and bottom boundaries. Temperature and turbulent diffusivity parameters are incorporated from the hydrodynamic model. Meanwhile, the underwater light distribution is computed based on incident light (derived from meteorological drivers) and phytoplankton concentration with depth, thus accounting for shading effects by different species. Physiological growth parameters were used according to Mehnert et al. [64].

The specific formulations for light- and temperature-dependent growth and loss rate are provided in Jöhnk et al. [34] for light dependence and temperature dependence using an adapted version of Robson and Hamilton [65]. The growth model does not include a feedback mechanism to the hydrodynamic model, meaning there is no adjustment to the heat balance resulting from changes in light absorption caused by growing algae populations. Moreover, the model targets the growth of a particular cyanobacteria species, reducing the complexity associated with incorporating multiple species’ growth characteristics and the intricate dynamics of phytoplankton.

While the temperature and light dependencies of these species are typically measured in laboratory settings (e.g., [64]), they may adapt to their natural environments and display growth characteristics that differ from laboratory observations [44]. Generally, phytoplankton growth rates follow a hump-shaped curve: they increase gradually until they reach an optimal temperature, where growth is maximised, before sharply declining at higher temperatures. Variations in the optimum temperature for growth can cause substantial shifts in species dominance ([64]). In this study, we used the base growth function of the cyanobacterium Ch. ovalisporum, which has been abundant in the system since 2016, as the foundation for the model. This species has an optimum temperature of about T_opt = 32 $°$C [66]. To explore differences in cyanobacterial development, we also tested Microcystis, another prevalent cyanobacterium species in temperate lakes with an optimum temperature of about 28 $°$C [67], to observe the differences in cyanobacteria development.

2.3.4. Model Simulation with Varying Water Level in the Dam

The key hypothesis of this work posits that BGA blooms in the River Murray downstream of Hume Dam are partially attributed to the dissemination of BGA from Lake Hume by means of releases of BGA-rich water via the hydropower or irrigation outlets. Given that these outlets are comparatively deep in relation to the maximum lake level (as indicated in

Table 2. Depth range of hydropower and irrigation intakes at Hume Dam.

Intake	Depth [m AHD]
hydropower	162.535–168.679
Irrigation	157.710–161.367

), it is anticipated that the potential issue arises primarily under low lake levels. To explore this hypothesis, simulations were conducted with varying water levels to examine the correlation between stratification, cyanobacteria distribution with depth, and temporal dynamics in proximity to these outlets. This investigation aims to identify the lake levels at which seeding risks from the outlets are likely to manifest.

The investigation into lake levels and their impact on the changing depth distribution of temperature and cyanobacteria cell counts was conducted by testing a range of levels from 190 m (Z190) down to 160 m (Z160), with all values referenced to the Australian Height Datum (AHD), in 5 m intervals. This approach allowed for the examination of water levels ranging from 40 m down to 10 m.

2.4. Model Setup and Input Data

The hydrodynamic models are driven by several input datasets, including hourly meteorological data, inflow rates, temperatures of main tributaries, outflow from the lake, and bathymetric information. The main drivers for the change in water temperature in a water body are meteorological forcing, usually given as air temperature, short-wave irradiance, long-wave irradiance from the water body itself, long-wave irradiance into the water body often described via cloudiness, and wind speed and direction. The internal distribution of heat is determined by light absorption, in its most simple form described via a single absorption coefficient determined from chlorophyll biomass or measured Secchi depth. While wind direction is typically used to estimate fetch length in 1D models, its implementation in this lake’s configuration, with two major arms, proved challenging. We cannot estimate the direction due to differential shielding by surrounding hills, and the wind direction was thus omitted. A hypsometric curve was employed to represent the lake’s area–depth relationship. The chosen simulation period spans 2014 to 2021, ensuring the model reaches the necessary steady-state conditions and adequate data coverage.

The growth model relies on short-wave irradiance converted into photosynthetically active radiation (PAR), alongside environmental factors such as water temperature and turbulent diffusivity provided by the hydrodynamic model component. These parameters are outputs of the hydrodynamic model with hourly resolution and a depth discretisation of 0.5 m. Nutrient limitation, specifically phosphorus and nitrogen, is not incorporated as the system is considered eutrophic, and the model does not account for other potential limiting factors.

The experiments were conducted using distinct batch files for each sub-experiment. These files sequentially executed the hydrodynamic model component first, gathering its output, which was then directed into the growth model. Model parameters were passed through external files, facilitating the seamless integration of the two components and ensuring efficient parameter management.

A hypsometric curve describing changes in the lake area with depth was derived from bathymetric information.

Meteorological Data

The hydrodynamic models were driven by meteorological time series for air temperature, relative humidity, wind speed and direction, short-wave irradiance, and cloudiness with a resolution of one hour.

Obtaining meteorological data that accurately represent lake conditions is a major challenge. Long-term data in the region are only available for Albury Airport (7.5 km from the Hume Dam) from the Bureau of Meteorology (BoM). Not all necessary parameters have been measured (e.g., no short-wave radiation, large gaps in cloudiness, and thus incoming long-wave radiation). To overcome this issue, we used simulated data from https://www.meteoblue.com/ (accessed on 10 September 2024) for Lake Hume with a position (grid point) near the dam wall. This commercially available dataset contains hourly gap-free time series since 1985 allowing for immediate input into the hydrodynamic model.

Based on our experience, we acknowledge that temperature, and particularly wind conditions, can vary significantly between the Murray and Mitta-Mitta arms of the lake. But we also acknowledge that our available datasets for driving the models are limited, and in situ data from the pontoon are sparse. Figure 3 presents comparisons among three meteorological datasets for Lake Hume: Meteoblue data, Bureau of Meteorology measurements at Albury Airport, and in situ data from the pontoon at the dam wall. While all three datasets exhibit similar overall patterns, there are discrepancies in detail. Although in situ data and local weather station data generally align across all three parameters, notable deviations occur on specific days, indicating distinct local weather patterns within the lake environment, nestled between hills, compared to the weather station located in flatter terrain. The Meteoblue data tend to underestimate daily minimum temperatures and wind speeds, and the Albury airport BoM data tend to overestimate. However, a comparison of measured and simulated stratification using the aforementioned meteorological sources revealed a good congruence (Figure 4). Thermal stratification is a critical factor driving the cyanobacteria growth model. Hence, this comparison gives us good grounds to choose any of the three data sources. Meteoblue data stand out as the sole dataset covering the entire period (2014 to 2021) to calibrate the model and provide all relevant input data for the hydrodynamic and growth models in a consistent manner.

2.5. Model Calibration

As noted by Shimoda and Arhonditsis [68], calibration of process-based water quality models typically involves iterative adjustments of model parameters until the model outputs align with observed data. While manual calibration is labour-intensive and heavily reliant on expert judgment, it was employed in this study. The calibration process was approached as a black-box optimisation problem. This involved three main steps: (1) simulating the original process-based models, (2) quantifying the disparity between simulated and observed data, and (3) employing algorithmic methods to search for parameter sets that minimise the disparity between the simulation and observation—the multidimensional parameter space.

The calibration of LAKEoneD hydrodynamics utilised data from the deployed thermistor chains near the dam wall. For the calibration of the cyanobacteria growth model, all available data on cell counts and/or biomass from various locations in the lake were used. Due to the significant spatial variability in cyanobacteria occurrence, which cannot be fully captured by a one-dimensional model, the calibration process was conducted manually. The objective was to achieve a close concordance between the simulated and observed bloom abundances over a period of time. Moreover, due to limited data availability, with most sampling conducted during the summer period and only a few days each month, the calibration process faced additional constraints.

3. Results

3.1. Temperature and Stratification

The temperature calibration of LAKEoneD for Lake Hume relied on data collected from temperature loggers positioned near the dam wall, strategically placed in the deepest sections of the lake. Figure 5 illustrates a comparison between the measured and simulated stratification in the lake, spanning from the onset of stratification just after the pontoon deployment (end of September 2019) until complete mixing in March 2020, followed by the subsequent cooling of the water column until May. While the simulation describes the thermistor chain measurements well during the stratified period, the model inadequately represented this cooling phase (from March to May 2020), primarily driven by large inflows of cooler water into the lake. As the one-dimensional model lacks the capability to account for such processes, its representation of this phase was limited. As a result, the simulated temperature deviated from the measurement for this period, as seen in Figure 5’s lower panel.

However, Figure 6 depicts the surface temperature and mixing layer structure before this cooling/inflow phase from Sunday, 28 September 2019, to Sunday, 7 March 2020, which were well simulated by the model. Even the daily changes were accurately resolved, underscoring the model’s ability to capture short-term variability in surface temperature and diurnal mixing behaviour, a crucial aspect of the growth model.

The elevated water temperature observed in mid-February 2020, coinciding with a cyanobacteria bloom alert at Lake Hume, was not accurately captured in the simulations. This discrepancy may be attributed to the meteorological data used in the simulation, which may not have fully reflected the on-lake situation during this warming phase. Both the BoM and Meteoblue data overestimated the wind speed in comparison with the in situ monitoring during February 2020. However, it is noteworthy that the simulated cell counts during this period align well with the bloom (as discussed in the subsequent section). This highlights the significance of not only water temperature for surface cyanobacteria cell counts but also the profound influence of daily mixing layer dynamics.

3.2. Blue-Green Algal Growth

Cyanobacteria growth is significantly affected by the light and temperature constraints defined in the model, which are tailored to specific species. In Lake Hume, the dominant cyanobacteria species can change annually and even within a single year. In 2016, during a massive bloom in Lake Hume and the downstream Murray River, a new species, Ch. ovalisporum, became dominant. Although this species had previously been present in the system, it had never proliferated so extensively. In subsequent years, this species reoccurred, however, in varying degrees of abundance, with other cyanobacteria species such as Aphanocapsa, Microcystis, etc., in high abundance.

Reducing the optimum growth temperature in the model from 32 $°$C to the actual maximum surface water temperature of 28 $°$C in Lake Hume results in a more rapid and substantial increase in cell counts. This adjustment leads to a significant rise in maximum cell counts by orders of magnitude, an extended growing season, and a greater depth of higher cell count concentration, as illustrated in Figure 7.

The one-dimensional growth model is capable of incorporating multiple species with distinct growth characteristics. However, in the case of Lake Hume, the mechanisms driving the shifts in species abundance remain unclear, as previously noted. This uncertainty could result in unrealistic simulations in which all species experience unchecked growth. Moreover, accurate simulations require precise initial biomass or cell counts and the dominant species. For these reasons, the single-species growth model was parameterised using growth data for Ch. ovalisporum for the years 2015/2016, 2018/2019, and 2019/2021, and for Microcystis for the years 2016/2017 and 2017/2018. These parameters were derived from laboratory measurements by Mehnert et al. [64]. While the simulations based on Ch. ovalisporum parameters closely captured the observed blooming behaviour for the corresponding years, they did not align well with the dynamics observed in the two intervening years. This discrepancy is likely attributed to the limitations of using a single-species model and the reliance on parameters that did not adequately reflect the behaviour of the dominant species during those years. In Figure 8, alongside the simulation of cell counts, we have depicted the dominant cyanobacteria species contributing to the bloom composition for each specific year. This composition varies both annually and within a single season, as illustrated in Figure 9. Furthermore, the starting population size for these years needed to be decreased to more closely align with the observed data. This underscores a common drawback of cyanobacteria growth models when simulating multiple years, as they typically lack a life cycle model to account for growth during winter and spring, as well as the possible influence of resting stages (akinetes) on the initial pool for cyanobacteria growth in the spring [69].

The simulations shown in Figure 8 and Figure 9 broadly reflect the observed bloom formation and decline in Lake Hume, as observed through species counts at various locations. The occasional occurrence of very high cell counts is attributed to the presence of exceedingly small cyanobacteria cells, which minimally contribute to the overall biovolume in the lake (e.g., Merismopedia in 2019/2020).

3.3. Stratification and Algal Growth in Relation to Outlets

Figure 10 illustrates the temperature stratification in the lake under different lake levels. The shallower lake exhibits a faster cooling rate and reaches lower temperatures during winter, requiring several months to attain comparable temperatures to the deeper, stratified areas during spring. Additionally, shallow sections tend to experience slightly higher temperatures in summer, while deeper sections distribute heat downward. The temperature differences within the epilimnion fluctuate throughout the year, with differences of up to 5 °C in winter and potential increases of 2 °C in surface temperatures when the lake depth is considerably smaller. These discrepancies diminish with greater lake depths (Z180 vs. Z190 in Figure 10c).

Moreover, as lake levels decrease, cyanobacteria may be mixed downward to areas where water is extracted (see Figure 11), potentially leading to significant seeding effects downstream in the river and, therefore, management implications. Figure 11 shows two examples of water levels, 30 m and 20 m, for the 2019/2020 bloom period. For the 30 m water level, neither the hydropower nor the irrigation outlet will be impacted by blooms. Under lower water levels, here 20 m, the effect of potential drawing in of cyanobacteria is higher but related to deep mixing events.

Based on the simulations of water temperature stratification and the depth distribution of cyanobacteria, it is feasible to assess the likelihood of one of the dam outlets accumulating significant volumes of cyanobacteria. This assessment is inherently reliant on factors such as water level and stratification dynamics. To explore this, we conducted simulations with varying lake levels (maintained constant throughout the entire simulation period) to evaluate the relationship between hydropower and irrigation outlets at Hume Dam and the stratification of water temperature and cyanobacteria.

Appendix A contains simulations conducted over the period 2015 to 2021. The shallower the lake, the higher the frequency of cyanobacteria concentration at the outlets. The deep mixing events are longer in the shallower cases, extending to the greater depths, thus putting more cyanobacteria in the outlets. These simulations indicate a consistent risk of significant cyanobacteria accumulation when the water level drops below 20 m and 10 m for the hydropower and irrigation outlets, respectively. However, it is important to acknowledge that intense discharge events through the outlets may disrupt stratification near the dam wall, potentially drawing in cyanobacteria from shallower layers and exacerbating the risk.

Furthermore, the situation may worsen during periods of high bloom activity, as observed in 2016, when deep mixing events occur in late summer or early spring. Under these conditions, both outlets are susceptible to prolonged periods of elevated cyanobacteria concentrations, even when water levels are high.

4. Discussion

As a pivotal reservoir within the River Murray system, Lake Hume serves as a vital resource for numerous community services, including recreation, tourism, agriculture, and aquatic ecosystems. This study explored the vertical spatial and temporal abundance of blue-green algae and the risk of significant cyanobacteria intake through both the hydropower and irrigation outlets when water levels drop below specific thresholds at Lake Hume. By isolating the hydrodynamic influence of water-level variations on bloom dynamics, the simulation demonstrated a consistent risk of significant cyanobacterial presence in water intakes. The findings of this study align with previous research [27,28,29] indicating that reduced water levels contribute to an increased prevalence of cyanobacterial blooms. Specifically, the results revealed elevated risks for both hydropower and irrigation outlets when water levels dropped below 20 m and 10 m, respectively. Our results underscore the possible scope of using reservoir water-level management during low water levels while having bloom alerts to mitigate the downstream risks of riverine cyanobacterial blooms.

The temperature distribution through the water column, simulated using our 1D hydrodynamic model, was clearly concordant with the observed temperature profile at the pontoon site in the deepest part of the reservoir. This concordance is on three different time scales: a seasonal time scale where the temperature through the water column gradually increases over spring and summer but at different rates depending on the position in the water column. This leads to the formation of a seasonal thermocline. The depth of this thermocline is accurately captured (+/− 0.5 m) on a weekly time scale (Figure 6). The precision of the temperature and stratification model breaks down in the Austral autumn (April) with the entry of cold river inflows. These are not represented in the model which, accordingly, cannot accurately reflect their effects. This occurs at a time when the water is too cold for the cyanobacteria to grow and thus does not impact our conclusion. In contrast, the cyanobacterial population increases at the highest temperatures during the summer stratification. On the shortest time scale (a day), the simulated temperature clearly shows the characteristic small diurnal fluctuation in the surface layers during summer, which diminishes with depth.

The simulation of the cyanobacterial abundance is driven by the output of the temperature/stratification model, and the accurate capture of these parameters noted above gives confidence that the cyanobacterial model output will closely reflect reality. However, this is contingent on the accuracy of the model. The model has been used repeatedly in a variety of different contexts [34,59,60,61,69] with good results, adding to our confidence that the results are an accurate representation of the dynamics of the buoyant cyanobacterial population in response to the local meteorological conditions.

One-dimensional (1D) models used to simulate water temperature stratification effectively captured daily and seasonal variations across different depths. However, inherent uncertainties were observed in simulating the depth and intensity of the thermocline. These uncertainties stem from the limitations of one-dimensional models, which cannot fully represent three-dimensional (3D) transport and mixing processes within a lake. To address this, parameterisation using a single parameter for background turbulent diffusivity was employed. Enhanced accuracy could be achieved using a 3D water quality model such as Delft-3D (https://oss.deltares.nl/web/delft3d, accessed on 25 January 2025). However, one-dimensional hydrodynamic models offer a faster way to simulate dynamic processes in a water column. This allows for a detailed analysis of vertical temperature stratification and mixing processes, both the most important factors in algal growth besides underwater light climate and nutrient availability. Their need for detailed input is less demanding than for three-dimensional hydrodynamic models. As an example, in terms of meteorological drivers, a single station time series is necessary for a 1D model while 3D models need a more detailed picture of the local meteorological field over a lake. Due to their fast computational run times, it is much less effort for a 1D model to estimate the effects of differences in such driving factors. This will lead to more effective cyanobacteria management.

The 1D modelling approach adopted in this study simplifies the system to basic growth relationships dependent solely on physical parameters (such as light, temperature, and mixing), and, if available, nutrient limitations. While this approach inherently limits its applicability range and precludes the examination of more intricate scenarios, it nonetheless yields convincing results for the behaviour during the primary blooming phase, which is predominantly driven by the aforementioned physical parameters. Consequently, it approximates the effects of daily mixing, the short-term impacts of weather events, and, to a lesser extent, the long-term evolution of cyanobacteria abundance. The simulated cell counts effectively mirrored the occurrence of blooms, underscoring the influence of water temperature and daily mixing layers. Challenges in growth modelling arise when prevailing processes surpass the model’s capacity to capture them. For instance, nutrient pulses stemming from wind-driven resuspension could significantly impact model results, particularly in shallow lake systems or bays. Additionally, the dissipation of blooms may not be accurately represented by most simulation models, given the multifaceted nature of factors involved, including decay induced by viral lysis, nutrient limitations beyond the conventional phosphorus and nitrogen, and exposure to UV radiation, among others [70,71]. There is no easy way to predict or model such phenomena. These factors underscore the requirement for further refinements of cyanobacteria growth models to improve their predictive accuracy.

Ingleton et al. [72] have demonstrated that although some cyanobacterial cells may be destroyed during their passage through the outlet works of a reservoir, a significant portion remain viable and capable of further growth in downstream river environments, provided that in-stream conditions are conducive to their proliferation. Steinberg and Hartmann [73] initially posited that rivers were not conducive habitats for planktonic cyanobacteria due to the assumption of their susceptibility to in-stream turbulence. Williamson et al. [74] conducted a study on the dynamics of cyanobacteria concentration following the release of cold water from major headwaters reservoirs into five rivers in New South Wales. The findings indicated that under low-flow conditions, cyanobacteria presence diminished rapidly with increasing distance downstream of the rivers. Subsequent observations have contradicted these notions: cyanobacteria have been found to survive for considerable distances, spanning hundreds of kilometres, following their release from major headwater reservoirs into various rivers. Notably, instances of this phenomenon have been documented in rivers such as the lowland Murray River in Australia [22,23,36,75,76].

Implementing real-time monitoring systems capable of continuously assessing the likelihood of cyanobacterial blooms during water releases is essential for effective river management and optimising water utilisation. Such systems provide informed risk assessments, enabling proactive decision-making to mitigate cyanobacterial bloom issues. An example is the AquaWatch Program (https://research.csiro.au/aquawatch/, accessed on 25 Jannuary 2025), where real-time images of the water surface to detect scum formation and quantify local cyanobacterial abundance are provided through inexpensive in situ hyperspectral cameras. Here, the 1D model can use the realistic starting value of cyanobacteria biomass derived from the hyperspectral reflectance, combined with other essential hydrodynamic measurements, to further improve the accuracy and precision of modelling and forecast cyanobacteria bloom. These assessments can support the application of strategies such as hypolimnetic withdrawal or artificial mixing techniques, which effectively manage and reduce bloom risks in freshwater systems. The technique of hypolimnetic withdrawal or introduction is a recognised method for reservoir remediation. The principal concept involves removing nutrient-rich water from the bottom of the lake, including the removals of dissolved metal species such as ${\mathrm{Fe}}_2^{+}$ and ${\mathrm{Mn}}_2^{+}$, ${\mathrm{NH}}_4^{+}$, phosphate released from the hyperlimnetic sediments, and water depleted in DO. The process reduces the residence time of the hypolimnion, thereby preventing the onset of anoxic conditions and the accumulation of nutrients within this lower layer of the lake.

5. Conclusions

Despite the inherent limitations of the model, this study effectively captures the bloom dynamics of blue-green algae in Lake Hume across varying water levels. The simulations across varying water levels in the lake revealed a consistent risk of significant cyanobacteria intake through both the hydropower and irrigation outlets when water levels drop below certain thresholds. Notably, water levels below 20 m and 10 m posed heightened risks for the hydropower and irrigation outlets, respectively. Moreover, under conditions of high bloom persistence extending into autumn and coupled with the potential for deep mixing events, the associated risk of cyanobacteria intake through the outlets was heightened. This underlines the importance of physical dynamics for transferring materials located in the epilimnion downstream. Moreover, the results highlight the substantial impact of seasonal variations and bloom dynamics on cyanobacteria intake potential.

Overall, by explicitly isolating the hydrodynamic influence of water-level variations on bloom dynamics, this study identifies the mechanism relating water-level fluctuations and the risk of cyanobacteria intake through the lake outlets. Understanding these dynamics is essential for developing effective management strategies to mitigate the adverse impacts of algal blooms on downstream water quality and aquatic ecosystems. The model provides water managers with a convenient tool to explore a strategy for releasing cyanobacteria and other potential pollutants.