Numerical study of hydro-environmental processes of Poyang Lake subject to engineering control

Located in Jiangxi province in southeastern China, Poyang Lake is the largest freshwater lake of the country. It sits on the southern bank of the middle and lower reaches of Yangtze River, and has an area of 16.2 × 10⁴ km², accounting for 9% of Yangtze River basin (Zhang et al. 2015b). Poyang Lake connects with the Yangtze River through a channel in the north and also receives water from a complex river system in the south, consisting of five major rivers namely Ganjiang, Fuhe, Xiushui, Xinjiang and Raohe. The lake region serves as a natural hydrological moderator for the Yangtze River, providing huge volume for water retention and pollutant dilution and an irreplaceable habitat for numerous aquatic organisms along the Yangtze River basin. The lake also supports agriculture, tourism and is the key freshwater source for around 12.4 million people in nearby cities of Nanchang, Jingdezhen and Jiujiang (Zhang et al. 2014).

Since the completion of the Three Gorges Dam (TGD) and its associated reservoir in 2003, the flow characteristics of Yangtze River have recorded significant changes which created knock-on effects on its surrounding water bodies (Lai et al. 2014b). More outflows from Poyang Lake to the Yangtze River reduced the water storage capability of Poyang Lake (Guo et al. 2012a). The lake's water levels have witnessed a declining trend in the 21st century. Monitoring stations around Poyang Lake all logged the lowest historic water levels over the last two decades and the average water level drop ranges from 14.0 to 13.3 m over the last 60 years (Guo et al. 2012b). The main reasons behind the water level drop include decreasing rainfall and influx from upstream rivers, leading to a prolonged dry period. As a result, the self-purification ability of the lake reduces and hence the water quality worsens, resulting in the deterioration of ecology. Within the period of 2000–2007, the river length with the level of water quality no better than Class III (poor quality) increased from 170 to 221 km, while the number of polluted river sections increased from 6 to 22 (Gao et al. 2010). The main pollutants in the region include TN, TP, NH₃-N and COD (Luo et al. 2014), which can be directly tracked to the five major rivers as a result of increased industrial activities, municipal wastes and agricultural fertilizers and poultry farming (Chen et al. 2016).

The decline in water quantity and quality in Poyang Lake has sparked concerns over freshwater security for the humans and wildlife in the region. The Poyang Lake barrage project was therefore proposed to prevent the lake's water level from further declining because of climate change, urbanization and hydraulic control on the Yangtze River (Hu & Ruan 2011). The proposed project involves a series of sluice gates being installed along the narrowest section of the channel at the northern mouth of the lake (Figure 1). The main goal of the project is to manage the water level in Poyang Lake by closing the sluice gates during dry seasons. In the wet season, the sluice gates are fully opened allowing free exchange of water, energy and biology between the lake and Yangtze River (Lai et al. 2016). In support of the proposed barrage to be built, local government offices and hydrological authorities have designed various operational schemes for the project (Zhang et al. 2011). These schemes aim to manage water resources and mitigate potential flooding hazards in the Yangtze River basin. Nonetheless, these operations could alter the natural processes in the hydrological, hydrodynamic, sediment transport, water quality and ecological aspects and weaken the existing interaction and natural connection between the Yangtze River and Poyang Lake. Sudden changes in the water level and flow velocity of the hydro-environment could affect the pollutant transport due to the disturbance on the flow driven material convection. Large difference in water level between the Yangtze River and Poyang Lake could also hinder the daily movement and migration of aquatic animals in the region. Serious degradation of water quality and outburst of algae may result in eutrophication, disappearance of wetlands and damage to wildlife habitat for endangered birds, fish and mammals. Therefore, it is crucial that quantitative analysis is carried out to evaluate the impacts of these schemes.

Numerous research efforts have been spent in developing hydrodynamic models for Poyang Lake in the past decade. Lai et al. (2011) built a 2-D numerical simulation of hydrodynamic and pollutant transport for Poyang Lake, although the validation and subsequent analysis of the developed model are limited. Later, Lai et al. (2013) built a new model called the CHAM that dynamically couples a 1-D unsteady flow model and a 2-D hydrodynamic model, which was used to compare the significance of lake inflows and Yangtze River on Poyang Lake level (Lai et al. 2014a) and to investigate the impact of TGD impoundment on Poyang Lake (Lai et al. 2012, 2014b). Li et al. (2014) developed an integrated model that consists of a hydrological model WATLAC and 2-D hydrodynamic model MIKE21 as an attempt to simulate the response and characteristics of the dynamic Poyang Lake catchment system. The same research group later investigated the hysteretic relationship between the lake response and the hydrodynamic change in the Yangtze River (Zhang & Werner 2015). Li et al. (2015) combined a 2-D hydrodynamic model in MIKE with dye tracer simulations to investigate the transport time scale in Poyang Lake for various water level variation periods. Li & Yao (2015) expanded the topic by adapting a similar approach using a hydrodynamic model in conjunction with transport and particle tracking sub-models. Both studies provided an insight on the management of chemical and ecological processes in Poyang Lake and demonstrated early attempts in modelling the transport behaviours in such a large river-lake system. Lai et al. (2016) discussed the possible impacts of the barrage project on the hydrology and hydrodynamics of the lake using a 2-D EFDC model, focusing on the water level, velocity, lake current and water exchange period. Zhang et al. (2015a) developed an inundation model for Poyang Lake with remote sensing data as a means of validation.

Continuous advancement of numerical technology suitable for hydro-environmental simulations under complex topography and bathymetry conditions offers an encouraging outlook. Smoothed Particle Hydrodynamics (SPH) is frequently used in the modelling of turbulent open channel flow over rough boundaries as a mesh-free numerical scheme (Kazemi et al. 2017, 2020). Wu et al. (2019), Yang & Liang (2020) and Yang et al. (2020) adopted the sophisticated random walk methods in environmental analyses. Nonetheless, these mesh-free approaches have yet to be applied to the challenging hydro-environment of Poyang Lake. Studies of the barrage project and evaluation of its impact on hydraulic and environmental issues using these alternative approaches are lacking.

The potential environmental effects of the barrage project in Poyang Lake have rarely been explored via numerical modelling. In this study, we aim to utilise a hybrid multi-dimensional hydrodynamic and water quality model to investigate the impact of the various operation schemes of the proposed lake barrage on the hydro-environment of Poyang Lake. The results can help understand the potential consequences of the barrage and contribute towards the sustainability of Poyang Lake and its neighbouring regions.

The finite volume method is used to discretize the solution domain, involving non-overlapping cell blocks and elements. A layered mesh structure is used, consisting of unstructured mesh in each horizontal layer and a structured connectivity in the vertical direction. The vertical discretization is based on sigma coordinates (DHI 2017a). In the numerical solution, the Roe's scheme, which is an approximate Riemann solver, is used to calculate the convective fluxes at the interface of the cells (Roe 1981). In the Roe's scheme, the dependent variables on both sides of an interface have to be estimated. Second-order spatial accuracy is achieved by employing a linear gradient-reconstruction technique.

The hydrodynamic governing equations are discretized using the Preissmann method and the Gauss-Jordan elimination method is used to solve the resulting algebraic equations. The mass transport governing equation is discretized using the implicit upwind scheme to maintain consistency between the hydrodynamic and mass transport equations. The tridiagonal matrix algorithm (TDMA) is used to solve the 1-D solute transport equations.

In multi-layered 3-D water quality models, the simulation is performed using the ECO Lab module as part of the software package MIKE21/3 Flow Model FM. The ECO Lab module is a numerical unit designed specifically for ecological and water quality modelling. It is used as a platform for customizing models that describe water quality, eutrophication, heavy metals and ecology in aquatic ecosystems using process-oriented formulations. The biochemical processes involved in this research include the dissolved oxygen process, the nitrogen cycle and the phosphorus cycle.

In the 1-D model, the study area covers the majority of the middle and lower reach of Yangtze River from Yichang to Datong, together with Dongting Lake and Poyang Lake, as shown by the grey region in Figure 2. The river networks are simplified into 125 sub-channels, 2,831 cross-sections and 91 junctions. Regions with small water flows and low interaction with the main lakes are excluded from the computation. Topographic survey was conducted around 2008 in the Yangtze main channel, Dongting Lake, Poyang Lake and a series of cross sections in river branches. The surveyed results provide key information required for the cross-sectional approximation. All the major rivers and sub-lakes in the Poyang Lake region are included in the model and a close-up of the cross-section layout around the Poyang Lake is shown in Figure 3. The sampling points at a cross-section are spaced at 20–50 m intervals to ensure sufficient sampling resolution in both narrow trenches and open waters in the broad lake areas. The hydrodynamic simulation was performed for the period 2000–2008, while the major water-quality analysis was done for the period between 2004 and 2008. The computational time step was set at 300 seconds. A roughness coefficient was assigned to each point in a cross-section, ranging from 0.02 to 0.04 s/m^1/3. In order to account for the operation of the barrage, additional calculation points were placed at the barrage position. Further details of the 1-D hydrodynamic modelling in the middle reach of the Yangtze River can be found in Huang et al. (2016). Recently, the 1-D model domain has been extended from Zhutuo, which is a key control hydrological station for the TGP, to the Yangtze Delta near Shanghai.

Multi-layered 3-D model computation is conducted in the main Poyang Lake, extending from the river mouths of the five main branches in the south to the barrage in the north (Figure 4). Given the enormous fluctuation of the water-surface area, between 50 km² in the dry season and 5,000 km² in the wet season, high grid resolution (50–60 m) is required. However, it is computationally unaffordable to adopt uniform grid resolution. The triangular mesh is used to reduce calculations in the regions with gradual variations whilst providing an accurate representation of the narrow watercourses where the flow and pollutant distribution vary rapidly. Irregular inner and outer parts of the lake have been discretized into triangular elements of variable grid sizes. The grid size varies from 60 to 150 m in main channels, and from 600 to 800 m in the homogeneous lake central regions. The final mesh consists of 56,658 vertices, 109,494 elements and 15,556 edges (Figure 5). The 2-D mesh layers are positioned at three depth levels of Poyang Lake to give a 3-D representation of the flow. In total, the simulation consists of 56,658 × 3 nodes with a calculation time step of 300 seconds. This grid system supports an efficient and accurate representation of the topographic connectivity, flow exchange and mass transport between various sub-regions of Poyang Lake. The input files of the multi-layered 3-D model are based on the outputs of the 1-D river network simulation, together with the field-measured data provided by the local water authorities.

In hydraulic model validation, the simulated surface elevations are compared with the field-measured data at several gauge stations in the main Poyang Lake region. The resulting RMSEs are within the range of 0.330–0.631 m and NSEs are in the range of 0.933–0.977 for 1-D hydrodynamic model (Figure 6), while for the multi-layered hydrodynamic model they are 0.518–0.972 m and 0.683–0.976 respectively (Figure 7). The figures illustrate good matches of two sets of data about the hydrodynamics of the lake. In general, the 1-D and multi-layered hydrodynamic models are able to consistently produce the correct trend and magnitude of the water level fluctuation across the Poyang Lake over a long period of time.

Validations of the 1-D and multi-layered 3-D water quality models were performed in similar procedures. Water quality indices; ammoniacal nitrogen (NH₃-N), total phosphorus (TP), dissolved oxygen (DO) and water temperature (TEMP) were computed and compared with the measured data.

Figure 8 illustrates the comparison between 1-D computational results and the field-measured data at several gauge stations. The simulated NH₃-N concentrations and the TP concentrations are in the same order of magnitude as the field-measured data, but they generally do not match the fluctuating trends well, with the predicted local maximums and minimums often missing the measurements. The resulting RMSEs and NSEs range from 0.184 to 0.372 mg/L and from −0.569 to 0.151, respectively. These represent a large discrepancy and a pure prediction performance. The comparisons between simulated predictions and field-measured data are generally good in the case of DO and TEMP, with the simulated results satisfactorily matching the trends of the field measurements. Across all of the gauge stations with valid field measurement data, validations of DO and TEMP are categorically more satisfactory than those of NH₃-N and TP in the 1-D water quality model validation. The DO validation records RMSEs and NSEs to be 1.253–1.862 mg/L and −0.706–0.499, respectively. The TEMP validation records RMSEs and NSEs to be 2.115–3.238 °C and 0.851–0.932, respectively.

The multi-layered 3-D water quality modelling results are compared with the field-measured data in 2008 in the validation. The multi-layered 3-D model enables the study to focus on the regions of broader water surface near the central part of the lake. Figure 9 presents the typical validation results in the multi-layered 3-D simulations.

As opposed to the 1-D validation results, the multi-layered 3-D simulated NH₃-N and TP concentrations have demonstrated satisfactory correlations with the field measurements. The NH₃-N prediction achieves the RMSE and NSE values of 0.322 mg/L and 0.928, respectively. The TP prediction achieves the RMSE and NSE values of 0.007 mg/L and 0.795, respectively. Temperature and algal growth are among the major factors that affect the DO variation. The rise in temperature during the summer triggers DO depletion, as it encourages biological activities and oxygen uptake for algal growth during the wet season, resulting in the decline of DO concentrations. The simulation demonstrates the correct trend of DO variation, with DO prediction achieving the RMSE and NSE values of 1.174 mg/L and 0.213 respectively. The predicted temperature records a very good agreement with the field-measured data with RMSE and NSE values of 0.693 °C and 0.993, respectively. Overall, the model is able to generally reproduce the field-measured results of all the water quality indicators.

Deviation of the computed results from the observed measurements can be caused by many reasons. Pollution associated with certain industrial, social and agricultural activities are difficult to monitor and quantify, leading to the misplacement or negligence of pollutant sources and incorrect discharge loadings. Furthermore, vigorous sand dredging activities in the region constantly change the bathymetry of the lake, which further contribute to the modelling difficulty. In addition, the exclusion of sediment modelling in this model is also a factor that might contribute to the unsatisfactory agreement in validation.

The proposal to construct the Poyang Lake barrage was first put forward in 2008. Its aim is to prevent Poyang Lake's water levels from continuously declining, triggered mainly by the deployment of the TGD. The proposal suggests to build a series of sluice gates along the narrowest section of the channel of 2.8 km wide, in the north of the Poyang Lake. Different barrage schemes are proposed by various local authorities interested in the Poyang Lake's environment (Table 1). Scheme S1 is proposed after a series of hydrological studies conducted by the Jiangxi provincial government. Scheme S2, S3 and S4 are proposals put forward by the Yangtze River Water Conservancy Committee. Scheme S5 is designed by the China Institute of Water Resources and Hydropower Research (IWHR). Scheme S0 is referred to the no barrage condition, corresponding to the natural state. These proposals detail the barrage operations and water level controls in terms of the five periods of the Poyang Lake control over a year: (1) inflow period, (2) storage period, (3) recharging period, (4) supply period, and (5) outflow period. The numerical model developed in this study is applied to examine the impact of different barrage schemes on the Poyang Lake's hydro-environment. The operational patterns of the schemes in Table 1 are translated into the variations of water levels at the location of the barrage over a period from 2005 to 2008. These water levels are specified in the model as the boundary condition at the barrage in the multi-layered 3-D simulation. Controlling the water levels at the barrage essentially changes the flow pattern in the whole lake and its interaction with the Yangtze River, and hence the hydrodynamics and solute transport of the entire lake.

Simulations with each scheme's barrage control produced hydrodynamic results on the year 2007. Three locations, including Xingzi (northern), Wucheng (northwestern) and Kangshan (southern), are chosen to represent the changes in water levels at different parts of Poyang Lake. The surface elevations at these locations under the influence of the schemes are illustrated in Figure 10. The divergence of water levels under different schemes only occurs during the winter months (January to April, October to December), when the barrage is in operation. The convergence of all schemes into the same water levels is observed throughout the wet seasons from May to September. All the lines overlap during the period from May to September, which verifies the intention of the barrage to have no control on the lake's flow during the wet summer season.

Among the five schemes, S4 exerts the least degree of barrage control and thus observes the lowest water levels at the barrage during the dry period, i.e. closest to the natural water levels. The water levels in S2 and S3 follow a similar trend and gradually increase. Both S1 and S5 result in the highest water levels in all three locations, demonstrating that the hydrodynamics of Poyang Lake reacts to the barrage control according to the degree of intervention by the barrage.

The magnitude of the water level differences between the schemes, however, depends on locations. The difference between water levels generated by different schemes in the winter season at Kangshan is smaller than those in Wucheng and Xingzi. Kangshan is at a southern location further away from the barrage and with a higher elevation than the other two monitoring points, so the barrage's influence decreases with the distance. In contrast, locations in the northern and eastern regions will be influenced by the barrage in a greater degree, as at Xingzi and Wucheng.

The rise in water levels due to barrage control will lead to the increase of water volumes and the expansion of water surface areas in Poyang Lake during the dry seasons. Under natural conditions, 2007 recorded a water surface area of 1,831 km² and the water volume in the lake of 66.9 × 10⁸ m³. Results generated from the numerical model under different schemes are summarized in Table 2 below.

As expected, S5 gives the largest increase in both the water surface area and volume among all the schemes. S4 has the smallest impact on the surface areas and volume. In general, the influence of barrage on the water volume is relatively small with the biggest impact coming from S5 at 14.5 × 10⁸ m³, which is still less than 20% of the lake capacity. The increase in volume in S4 is only at 2.1 × 10⁸ m³, which is less than 3% of the lake capacity. Because the Poyang Lake is broad but shallow, the barrage has led to the significant increase in the surface areas through the rise of water levels.

The increase in water levels under the barrage control can bring subsequent environment concerns. Reduction in water level fluctuation in the lake will diminish the flow exchange between the Yangtze River and Poyang Lake, hence decreases the water flow rate and velocities. Lower water turnover rate could result in longer retention time and may cause eutrophication. Moreover, rise in water level will inundate lake regions that potentially include wetland areas and wildlife habitats, in which temporary dry condition and sun radiation are essential. Further investigation is needed on the impact of the barrage scheme on these aspects.

The effect of the barrage schemes on the Poyang Lake water environment was also analysed. Detailed simulation temporally and spatially across the Poyang Lake are shown in the form of subplots for the water quality indicators DO, NH₃-N and PO₄-P. Given the shallow nature of Poyang Lake with water depth generally below 10.0 m, temperature stratification is expected to be very weak even after the construction of the barrage. Temperature is therefore excluded from the detailed discussions due to its small variation. In the submaps, the colour spectrum ranges from purple (low concentration) to red (high concentration) with white region representing dry out region where water level is low and the flow is negligible. Most of the white areas belong to the shallow sand shoals at relatively high elevations.

In Figures 11 and 12, the DO concentrations across the whole of Poyang Lake domain are present. Figure 11 represents DO concentrations in its natural state in 2007. Figure 12 shows the DO distribution under the influence of S1. In its natural state, the DO concentrations in the early months (Figure 11(a) and 11(b)) show great variation along the main channel ranging between 4.0 and 11.0 mg/L. As the water levels rise in the flooding season, the DO concentrations begin to decrease from 5.0 to 8.0 mg/L in April (Figure 11(c)) to below 3.0 mg/L in July (Figure 11(e)). Although there are influxes of high DO concentrations from some of the boundaries in the west side of the lake, the overall DO concentration in the domain remains below 3.0 mg/L (Figure 11(f)). The increase in temperature, reduction in flow rate and growth of algae and microorganisms are potential causes of such concentration decline during the wet season. As the dry season returns, water levels drop and the water flow in streams increases and facilitates oxygen recharge, so the narrow line of higher DO concentration is seen again in Figure 11(h).

High water levels induced by the barrage scheme S1 have created a vastly different spatial distribution of DO concentration in the early months. During the dry season, rise in water levels reduces the current-induced aeration in the lake and both Figure 12(a) and 12(b) show a majority of DO concentrations at around 3.0 mg/L with a burst of high DO concentrations in the northwestern part of the lake due to high-DO inflows at the boundaries. As the flooding season approaches, continuous increase on water levels, combined with the rise in temperature, leads to a reduction in the DO concentrations. Throughout the summer period (Figure 12(d)–12(f)), the majority of the lake's DO concentrations are below 3.0 mg/L, which are below the DO standard of 5.0 mg/L for aquatic wildlife. As water levels decline in the winter months, little variation can be seen in DO concentrations (Figure 12(g)–12(h)). High DO influxes from the northwestern boundaries alter the general low DO concentrations in the lake.

The rise in water levels hinders the re-aeration process in the lake and the overall reduction in DO concentrations is evident from the results. The elevated water levels reducing lake-river exchange and water velocities. Sufficient DO concentrations are vital for the survival and growth of aquatic wildlife and water vegetation. Large increase in water levels in the region under the barrage control could damage and endanger the living habitats of many wildlife species in the water environment.

In Figures 13 and 14, the spatial distributions of the NH₃-N concentrations in Poyang Lake are shown. The natural state of the distribution with no barrage influence is shown in Figure 13. During the early dry period in the lake, large quantities of NH₃-N enter the lake from boundaries with a concentration ranging from 0.025 to 1.00 mg/L, resulting in distinct pathways of higher NH₃-N concentrations (Figure 13(a) and 13(b)). The vigorous fluctuations of NH₃-N concentration under different schemes during the first few months of the year could be due to the influence of nearby inflows with rapidly changing NH₃-N concentrations. This could happen when the inflows have multiple upstream pollutant sources such as industrial plants and agricultural farms which discharge chemicals and fertilizers to water bodies. As a result, the NH₃-N concentrations fluctuate greatly. When the flood season begins, increase in water levels dilutes the NH₃-N concentration (Figure 13(c) and 13(d)). Blue regions in the figures with NH₃-N concentrations of 0.050–0.075 mg/L start to fade and disappear into the purple region that has concentrations below 0.010 mg/L. During the wet season, the NH₃-N concentration in the domain is thus diluted, with the concentration below 0.010 mg/L (Figure 13(d) and 13(f)). Inflows with higher NH₃-N concentration from the west boundaries of the lake are quickly neutralized and do not affect the overall concentration of the domain. In winter, drop in water levels results in the significant rise of NH₃-N concentration in narrow and deep streams along the lake's main channel, as shown in Figure 13(h).

Under the influence of the barrage, the NH₃-N concentrations in S1 are reduced in the dry season due to the increase in water levels in this period. As shown in Figure 14, the very high NH₃-N concentrations along the narrow water causeways have disappeared. Boundary inflows with high NH₃-N concentrations can still be seen in the east and northwest regions (Figure 14(b)), however these influxes are soon diluted once they enter the lake area. With high water levels maintained throughout the year, the dilution effect is apparent throughout the flooding period (Figure 14(d) and 14(f)), prior the winter months when the concentration slightly rises again. The increased water level and water volume in Poyang Lake thus enhance the dilution of NH₃-N and hence result in a reduction in the NH₃-N concentration.

The spatial distributions of the PO₄-P concentration in Poyang Lake are shown in Figures 15 and 16. In Figure 15, no barrage is included and the PO₄-P distribution shown is at its natural state in the Poyang Lake environment. In the early months of the year, the PO₄-P concentration is high along the main channels, as the low water levels restrict the spread of the input pollutants out of the channels. Concentrations in the range of 0.015–0.075 mg/L can be observed (Figure 15(a) and 15(b)). When the water levels rise in April, the PO₄-P concentration spreads out to a larger region as indicated by the increase in blue color in Figure 15(c). The influx of PO₄-P from boundaries remains high in June at 0.060–0.065 mg/L as seen in Figure 15(d). As the flooding season arrives, the increase in water volume dilutes the PO₄-P concentration. Although inflows with high PO₄-P concentration can still be observed in Figure 15(e) and 15(f), the overall concentration in the majority of the lake is below 0.005 mg/L. In the later months of the year, drop in water levels in the lake triggers the increase of PO₄-P concentration along the lake's main channel, as seen in Figure 15(h).

The PO₄-P distributions under the influence of S1 are shown in Figure 16. The increase in water levels has alleviated the PO₄-P concentration in the middle of the lake domain during the dry season (Figure 16(b)). As the water levels rise further due to flooding, the PO₄-P distribution becomes similar to that in S0, including the outburst of PO₄-P concentration in June (Figure 16(d)), which is due to the inflow of a few highly concentrated pollutant streams from Raohe and Xinjiang in the east. The dilution effect on PO₄-P concentration is observed throughout the flooding period until October, by which time the winter season begins with lower water levels and higher PO₄-P concentrations (Figure 16(f)). The spatial distributions of PO₄-P concentration have shown that the implementation of barrage has successfully reduced PO₄-P concentration.

The aim of this study is to develop a numerical model that can simulate the hydro-environmental evolution in the Poyang Lake, which has complicated geometry and experiences large water level fluctuations. The model has been used to assess the hydraulic and water-quality changes as a result of the different operations of a proposed barrage project, which is designed to control the lake water level and water exchange between the Lake and Yangtze River. The combined 1-D and multi-layered 3-D hydro-environmental models have been successfully calibrated and validated.

In terms of hydrodynamic impact of the barrage, the schemes can effectively raise water levels in the lake in winter months, although the water level rise becomes insignificant in the southern part of the lake, which has a large distance from the barrage. A drawback of the water level increase is the inundation of wetlands and wildlife habitats, as well as the reduction of water exchange inside the lake and thus the local accumulation of contaminants in the long term. In terms of the impact on water quality, all five operation schemes are able to reduce the nutrient concentrations in Poyang Lake during the dry season because of the enhanced dilution effect associated with the increased water level and water volume. The concentrations of both NH₃-N and PO₄-P are below the present values with no barrage in place. The different degrees of reduction depend on the designed water level increments in different schemes. The greater the increase in water levels, the lower the nutrient concentrations. This is a welcoming change, as the barrage can mitigate the pollution issues in the Poyang Lake during dry seasons. However, the rise in water levels also triggers a reduction in DO concentration due to a lower level of water exchange and the reduced re-aeration rate as a result of the reduced flow speed. The oxygen depletion may be hazardous to aquatic wildlife and ecosystem. Schemes 1 and 5 considered in the paper give the largest increase in water levels and thus the greatest impact on the lake hydro-environment.

Apart from the positive and negative impacts of the barrage on the Poyang Lake environment, further investigations will be carried out concerning the barrage influence on downstream regions along the Yangtze River. It should also be noted that our present study does not consider sediment transport, which can be justified as the sediment concentration is low in the lake and the simulation time is relatively short. For long-term predictions, the influence of the sediment transport and other broad climate, hydrological and environmental changes shall also be considered. The barrage should incorporate a flexible design so that its operation can be easily altered to adapt to the new situation in the future.

We thank the support by the National Key Research and Development Program of China (2016YFC0402605). G. Huang appreciates the financial support by the National Natural Science Foundation of China (52079130, 51509137), and the Model Development and Application of Water Environment in the Yangtze River Basin (2019-LHYJ-0102) H. Ho thanks the PhD scholarship provided by the Croucher Foundation. D. Liang thanks the financial support by the Royal Academy of Engineering (ISS1516\8\34).

Data cannot be made publicly available; readers should contact the corresponding author for details.