Abstract
A constructed in-lake water quality mitigation system has proven itself to be effective at reducing Machado Lake phosphorus (P) levels, but ineffective at reducing nitrogen (N) levels. A combination of lake sediment dredging and capping, oxygenation, and a recirculating wetland have reduced lake water column P levels by nearly 50%, as compared to pre-project levels. Key to this result has been the dampening of seasonal P recycling in the sediments. A new lake water quality numerical model is presented, with applications to both pre- and post-project conditions. Model auditing has revealed very good results with respect to predicting mitigation impacts on P but poor results with respect to predicting the performance, or lack thereof, of the N mitigation system. Model sensitivity analyses indicate that the P reductions are primarily attributable to the sediment dredging and capping. Conversely, seasonal data, supported by modeling, suggest that the poor performance of the N mitigation system may be attributable to incomplete removal, or sequestration, of sediment N mass during dredging and/or a lack of impact from the oxygenation system. Future mitigation efforts for the lake should focus on reducing the substantial watershed nutrient loads to the lake and further in-lake P inactivation.
HIGHLIGHTS
A new lake water quality model is presented.
A major urban lake restoration project is reviewed.
Nitrogen and phosphorus mitigation actions are evaluated.
Lake mitigation actions include dredging, sediment capping, oxygenation, and recirculating wetlands treatment.
This study provides an auditing of both the constructed mitigation system and numerical model predictive power.
INTRODUCTION
Urban lakes often suffer from eutrophication (e.g. Waajen et al. 2014; Teurlincx et al. 2019), characterized by high nutrient concentrations and high phytoplankton algae levels. This condition negatively impacts the lake with respect to aesthetics, ecosystem health, aquatic life, and primary (e.g. swimming) and secondary (e.g. paddling) contact recreation. Nutrient enrichment in these lakes, primarily phosphorus (P) and nitrogen (N), typically occurs due to loads originating in the urban watershed and delivered via stormwater to the lakes. Sources of P and N in an urban watershed include fertilizers, domestic and industrial sewage and grey water, vegetation clippings, pet waste, atmospheric deposition, and detergents. All lakes, particularly shallow ones, also tend to ‘recycle’ aquatic nutrients, with large releases of bioavailable nutrients from the bottom sediments to the overlying water column during certain times of the year (typically in the summer) (de Medina et al. 2003; Heinen & McManus 2004; Lim et al. 2011). ‘Legacy’ nutrient sources to lakes can also be significant. These sources are derived from historical land use practices (such as agriculture) and retained in soils, and lake bottom sediments, for long periods of time.
The mitigation and restoration of urban lakes is a priority for many cities, motivated by the public visibility of the lakes and a desire to restore ecosystems and/or provide contact recreation opportunities in the lakes. Urban lake restoration, however, is complex and challenging, with no known ‘silver bullet’ (Teurlincx et al. 2019; Lürling & Mucci 2020; Abell et al. 2022). Lake nutrient mitigation options include reducing catchment contaminant loads through source controls, constructed wetlands to intercept and filter stormwater, recirculating wetlands to remove nutrient mass from the lake water column directly, dredging to remove nutrient-enriched sediments, aeration, and oxygenation to inhibit sediment nutrient releases, the addition of P inactivation chemicals such as alum, and algae management practices such as sonication or the use of algaecides.
Published results of mitigation projects that target in-lake nutrient control are mixed. P inactivation through chemical dosing has been widely proven as a means of controlling both water column P and, potentially, phytoplankton levels in urban lakes. For example, Kozak et al. (2017) reduced P levels in their shallow urban lake by approximately 50% primarily through alum dosing. Similar results have been reported by Lopata et al. (2013) and Huser et al. (2016) (alum dosing), Grochowska et al. (2019) (P inactivation and aeration), Costadone et al. (2021) (alum dosing and aeration), and Rosińska et al. (2019) (P inactivation in combination with biomanipulation and aeration). The impacts of aeration in the Kozak et al. (2017) study were less conclusive, for both P and N. Aeration does appear to have played a significant role in more modest reductions of lake N levels in the urban lake case studies presented by Rosińska et al. (2019), Grochowska et al. (2019) and Costadone et al. (2021). Zębek & Napiórkouska-Krzebietke (2016) identified a significant decrease in both phytoplankton and phosphate levels in a shallow urban lake due to aeration and stormwater controls. Urban lake dredging was shown to be highly effective at reducing both P and N in the short term in the case study presented by Ruley & Rusch (2002). However, these authors report a gradual recovery of pre-dredge P levels after approximately 15 years. Similar transient impacts of dredging a shallow urban lake were reported by Norris & Laws (2017). Conversely, studies have shown that dredging may cause short term increases in bioavailable ammonia from lake bottoms by exposing deeper sediments (Jing et al. 2013; Zhong et al. 2018). In terms of wetlands, Saviolo Osti et al. (2019) quantify an approximately 40% decrease in water column P concentrations due to the implementation of a floating wetland in their urban lake.
Numerical modeling can be used in support of lake restoration efforts to gain a greater depth of understanding of current lake contaminant dynamics and sources and to make predictions about lake response to mitigation options (Muhammetoğlu et al. 2002; Ruley & Rusch 2004; Xing et al. 2014; Vinçon-Leite & Casenave 2019). For example, modeling can be used to quantify the role of lake internal recycling of nutrients in the overall lake nutrient budget. It can also be used to predict the impact of reducing such recycling rates through in-lake water quality controls. Understanding model limitations and areas of uncertainty are key to appropriately applying models for such predictive purposes.
In 2015, the City initiated restoration efforts to reduce nutrient levels and improve trophic conditions in the lake. The constructed mitigation system includes a recirculating 2.5-acre lakeside wetland and a speece cone oxygenation system (Horne et al. 2019). Additionally, approximately 2-feet of enriched surface sediments were dredged from the lake bottom during the construction of the system, and a low-permeability cap (bentonite clay) was applied to the remaining sediments to sequester residual pollutants. Collectively, these restoration efforts are herein referred to here as ‘the project’. The oxygenation system is designed to dampen sediment nutrient releases by eliminating, or reducing, lake bottom anoxic conditions. Sediment nutrient release rates are generally known to be higher under anoxic, as compared to oxic, conditions (de Medina et al. 2003; Haggard et al. 2005). The constructed wetland is designed to remove dissolved and particulate nutrients directly from the water column, via a range of wetland biophysical processes, as lake water is moved through the wetland. Dredging aimed to remove a legacy nutrient pool in the shallow sediments thereby reducing internal sediment nutrient releases and recycling. Similarly, the bentonite cap is intended to create a physical barrier to inhibit exchanges of solute between sediment porewater and the overlying water column. The full project has been in operation since late 2016. Routine water quality monitoring has occurred since that time.
The objectives of the study presented here were to:
assess the effectiveness of the Machado Lake mitigation project with respect to targeted nutrients, P and N;
perform an ‘auditing’ of a predictive numerical model previously used to support project design; and
gain and disseminate insight on urban lake mitigation options generally (‘lessons learned’) to be of value to urban lake mitigation and restoration efforts globally.
METHODS
The City performed routine monitoring of lake water quality, including nutrients, for a period of approximately seven years prior to the implementation of the mitigation project (2008–2014). Stormwater inputs to the lake, at three major drain outfalls, were also sampled routinely from 2008 to 2010. The sampling frequency for both in-lake and stormwater outfall sampling was primarily bi-weekly. Post-project in-lake sampling commenced in December 2016 with a variable frequency of approximately once or twice per month. Post-project stormwater outfall sampling commenced in July 2016, with an approximately bi-monthly frequency. Stormwater outfall sampling includes both wet weather events and dry weather baseflow sampling. All data used in this study were provided by the City.
All lake samples were collected from just below the water surface at two central locations (ML-1 and ML-2, Figure 1). The lake is well-mixed, both vertically and laterally, with minimal difference in observed water quality between sample sites. In addition to the routine monitoring, lake sediment samples were collected as part of this study and analyzed for nutrient content. The sediment sampling was performed on four separate occasions: July 29 and November 18, 2019; and May 26 and September 8, 2020. A Ponar grab sampler, lowered from a boat, was used to collect shallow sediment samples from existing in-lake sampling sites (ML-1 and ML-2).
A lake water quality model was developed to support the original project design, using the Simplified Lake Analysis Model (SLAM). The SLAM is generalized lake modeling software that calculates lake mass and flow balances on a daily timestep, assuming one or more well-mixed lake zones. Each zone follows the conceptual framework often referred to as a ‘continuously stirred tank reactor’ (CSTR), whereby complete and immediate mixing is assumed for each zone in both the vertical and horizontal directions. This assumption makes the model particularly well suited for lakes that are generally well-mixed and can justifiably be divided into a limited number of small and/or shallow zones.
The model targets the key parameters important for eutrophic lakes: phytoplankton (as chl-a), P, and N. An established empirical equation (Walker 2004) is used to describe the relationship between summer phytoplankton levels and lake nutrient concentrations and hydraulics. Lake watershed hydrology and nutrient loadings can either be explicitly calculated by the model or can be user prescribed. The model allows for a simplified representation of a variety of in-lake mitigation options, including sediment dredging, hypolimnetic oxygenation, supplemental water inputs, pump and treat systems, alum application, and recirculating off-channel wetlands treatment.
Additionally, the SLAM includes a dynamic sediment nutrient flux module. This module calculates internal recycled nutrient loads from the sediments to the water column as a function of shallow sediment nutrient dynamics, settled particulate load, and diffusive exchanges between sediment porewater and the overlying water column. This provides a key mechanistic link between external and internal nutrient loads in the model and improves the model's predictive potential.
The original Machado Lake model was calibrated during the design phase of the project using measured data from the period 2006–2013. The model utilized a single well-mixed zone to represent the lake, with an explicit simulation of sediment nutrient dynamics and exchanges. It was known at the time that the model P calibration was more accurate than the N calibration. In other words, confidence was higher in the P simulations than in the N simulations. This was largely the result of the well-behaved seasonal patterns of variability exhibited in the measured P data. The N data were much less cyclical and, therefore, more challenging to replicate with a numerical model based on monthly-varying kinetic parameters.
The original calibrated model was used to guide mitigation system design, quantifying relative differences in the efficacy of individual in-lake mitigation options toward the goal of reducing nutrient and phytoplankton levels in the lake. For example, the model effectively identified, and quantified, the dominant summer internal nutrient loads in the lake and highlighted the need for lake dredging of enriched sediments and controls on sediment nutrient releases.
For our study, the original model (2006–2013) was extended to simulate the post-project system as parameterized with recent external loadings and climate drivers (2016–2021). This version of the model is referred to herein as the ‘auditing model’. The auditing model was used to: (a) assess the model's predictive power with respect to the accuracy of pre-project forecasts (auditing); and (b) provide insight into current lake mitigation system performance and provide guidance for future lake management. The auditing model includes all known information about the post-project physical and environmental systems. In other words, the original model environmental and physical inputs were refined to better represent actual current post-project conditions. These modified inputs include lake bathymetry (dredged depth and volume), daily precipitation records, and an updated stormwater quality database based on recent observations. Monthly stormwater inflow nutrient concentrations were set based on measured wet and dry weather observations. As in the original model construct, inflows to the lake were prescribed based on reported daily precipitation and simple (Rational Method) calculations of urban runoff volumes. Importantly, all other model inputs, including all calibrated kinetic rates and constants, were maintained at original calibrated values. By retaining the fundamental construct and kinetic rate constant parameterization, results of this exercise provide a useful auditing of the original (pre-project) model predictive ability for project performance.
As a second exercise, to gain further insight into current lake system dynamics, the lake model was recalibrated using the post-project measured data set (‘recalibrated model’). The model calibration process involved adjustments to internal model parameters to achieve adequate agreement between measured and modeled lake water quality data. Both water column and surface sediment measured nutrient data were included in this assessment. The recalibrated model was then applied, in simple sensitivity exercises, to gain insight into current system nutrient dynamics and to quantify load reductions required to achieve numeric water quality targets.
RESULTS: ANALYSIS OF MEASURED DATA
Additionally, lake bottom shallow sediment nutrient data provide insight into sediment nutrient dynamics and are useful for lake water quality model calibration. Limited sediment data are available for the pre-project lake (October 2008 and September 2009, only). Four rounds of post-project sediment sampling were performed to-date as part of this study, in 2019 and 2020. A significant reduction in sediment P concentrations in the post-project lake bottom sediment, as compared to pre-project lake bottom sediments, was observed (Table 1). The mean and range of sediment P concentrations measured during the post-project period are many times lower than those measured during the pre-project period. Summary statistics for sediment N concentrations, however, are almost identical in the pre- and post-project data sets. Sediment N has seemingly not been significantly reduced in the post-project system. Note that, given the paucity of data, it is unclear whether sediment nutrient levels in the lake have changed significantly from the initial dredged system in 2016.
Date . | TP (mg/g), mean (range) . | TN (mg/g), mean (range) . |
---|---|---|
September 2009 (pre) | 3.5 (1.7–6.7) | 0.7 (0.4–2.0) |
2019–2020 (post) | 0.6 (0.2–1.5) | 0.7 (0.3–1.1) |
Date . | TP (mg/g), mean (range) . | TN (mg/g), mean (range) . |
---|---|---|
September 2009 (pre) | 3.5 (1.7–6.7) | 0.7 (0.4–2.0) |
2019–2020 (post) | 0.6 (0.2–1.5) | 0.7 (0.3–1.1) |
RESULTS: MODELING
N simulation results for this application were unsatisfactory. Although the overall annual range of variability is reasonably well-captured by the model, TN in the water column is consistently underpredicted by the model (Figure 9(b)). This is particularly true during the critical summer period. The cyclical pattern of seasonal variability, noted previously for the measured data, is not evident in the model output (Figure 7(b)). This suggests that the newly formed (via post-project deposition) sediment N releases have not been accurately predicted by the model.
The primary difference between the post-project and pre-project model parameter sets can be found in the sediment N module. To achieve the observed seasonal fluctuations in lake N levels in the post-project system, model sediment N mineralization and desorption rate constants (kd2) were increased in the post-project calibration, as compared to the original model. Additionally, a greater portion of the sediment N pool was made bioavailable, for release to the water column, by increasing the N burial fraction. The net effect of these parameter changes is an increased sediment N summer recycling rate, increased seasonal sediment N releases to the water column, and reduced retention of N in lake bottom sediment pool. This provides a reasonable replication of the seasonality observed in the measured data (described above). In other words, the model re-calibration reinforces the conceptual model of significantly higher N recycling in the lake than was either predicted for the post-project system or was originally quantified for the pre-project system. Note that the original model parameterization closely, and optimally, replicated observed pre-project conditions (2006–2013). The need for re-calibration, therefore, implies a significant change in lake, and lake sediment, N dynamics in the post-project system compared to pre-project.
Application of the recalibrated model indicates that large reductions in watershed loads to the lake will be required to achieve stated in-lake numeric water quality targets (0.1 TP, 1.0 TN). According to the model, watershed load reductions of approximately 65 and 45%, for P and N respectively, are required to achieve these previously established targets, as annual averages. Note that a previous pre-project TMDL study for the lake quantified watershed load reduction requirements of 91 and 47%, for P and N, respectively.
Results of the P model sensitivity analyses indicate that the dredge and cap, the wetlands, and the oxygenation system have all contributed to the observed reductions in TP, with an approximate relative magnitude of contributions in that same order.
DISCUSSION
The Machado Lake constructed mitigation system has been highly effective at reducing P levels in the lake, despite watershed P loads remaining high. Lake TP levels have been reduced by over 50% since implementation of the in-lake mitigation project, without any measured reduction in watershed external loads to the lake. Supported by modeling, the data suggests that lake seasonal sediment P releases have been greatly dampened (90% reduction) by the mitigation system. The large and conspicuous seasonal patterns of variability observed in the pre-project data are no longer present in the post-project system, as evidenced by both the measured data and modeling results. This is likely primarily attributable to the dredging and capping element of the constructed system. However, the fact that the P levels are relatively stable throughout the post-project period also suggests that the oxygenation system is effectively increasing P sequestration in the sediments by eliminating anaerobic conditions.
The original lake water quality model predictions of post-project P appear to be highly accurate, particularly considering the dramatic changes to the lake that have occurred. This lends confidence to the P model for future applications.
N levels remain high in the lake. While watershed N loads also remain high, it appears that the constructed in-lake mitigation system has not been as effective at reducing lake N levels as it has for P levels. More specifically, current seasonal sediment N releases appear to be as high as, or higher than, pre-project values. This same result was reported by Zhong et al. (2018) and Jing et al. (2013), for their laboratory experiments. Legacy sediment N mass at deeper depths may have been exposed by the dredging, and some of this mass may be seeping through the constructed bentonite cap during seasonal release periods. This may have been exacerbated by sediment N released to the water column, likely as soluble ammonia, during the dredging operations and then settling, on top of the bentonite cap, to replenish the active N pool. The latter hypothesis is supported by the work of Wang & Feng (2007), who presented similar results, with a similar explanatory hypothesis, for their shallow urban study lake.
Model auditing of N predictions indicate significant discrepancies between modeled and measured post-project N concentrations. Possible reasons for these discrepancies include a missing, or under-simulated, dry weather catchment N load and/or inaccurate original model parameterization with respect to sediment N recycling and the associated response to the mitigation actions. The latter, as a likely explanation, is supported by a re-calibration exercise which quantified significantly higher rates of sediment N recycling (mineralization), as compared to original calibrated parameters. In other words, this exercise suggests that the original model, and the calibrated parameter set, may have significantly over-estimated the mitigation impact of sediment oxygenation and dredging on N releases.
N:P ratios in the measured data indicate that the constructed restoration elements have lowered P levels to the point where the lake is now, at times, P-limited. This implies that, going forward, phytoplankton growth will be equally sensitive to either N or P reductions, which provides more options for mitigation. P inactivation through chemical dosing (e.g. alum) now appears to be a viable option for P and phytoplankton control in the lake. This option has substantial support in the urban lake literature with respect to effectively reducing TP.
Sensitivity modeling indicates that, of the mitigation actions taken, the removal of sediment P during the dredging process has provided the single largest positive impact on lake P levels. The low-permeability bentonite cap may also play a role in eliminating upward vertical migration of any residual ‘legacy’ P in the remaining sediments. The model assumes both a completely ‘clean’ lake bottom starting point, with respect to sediment nutrients, and a lack of any mobile underlying deeper legacy nutrients in its simulation of the dredged system. This appears to be an accurate representation of the post-project system, where minimal sediment P influences are apparent in the measured water column data. In other words, the sediment P fluxes have clearly been greatly dampened by the constructed system, and this is well-captured by the model's representation of the dredged lake bottom.
Modeling shows that the wetland treatment system and the oxygenation system will both play important roles in long-term P mitigation. The former provides the only permanent removal of nutrients in the mitigation system, while the latter provides for enhanced sequestration of nutrients in the sediments. The model shows that both systems may have relatively minor roles in the P mitigation observed to-date, yet both are important for the longer-term sustainability of the mitigation system. This will especially be the case if the wetland system is optimized with respect to residence time and pumping rates and is properly maintained. The benefit of dredging is greatest in the period immediately following the removal of enriched sediment. Consequently, modeling indicates that sediment P levels will continue to rise toward pre-project values, as long as external P loads from the draining catchment remain high. As this occurs, the wetland and oxygenation systems are expected to play more prominent roles in lake P mitigation.
There are significant differences in the modeled N dynamics for the post-project system, as compared to the pre-project system. These model differences are supported by a comprehensive measured data set. Specifically, an increased dominance of seasonal sediment N releases is simulated in the post-project calibrated model compared to the pre-project model. This is supported by measured data showing consistently elevated summer and autumn lake concentrations and shallow sediment N concentrations as high as pre-project levels. The dampening of sediment fluxes observed for P is clearly not occurring with N. The ‘recycling’ of lake N appears to be occurring at rates at least as high as those quantified for the pre-project system. This implies that the mitigation mechanisms in place are less effective for N than for P. Possible explanations include that dredging may not have removed enough of the N-enriched sediment, that the bentonite cap may not be as effective for N as it is for P, that soluble N may have been exposed, or physically released, during the dredging process (and consequently remained in the lake mass pool), and/or that the oxygenation system may not be impacting N sequestration in the sediments to the extent expected.
CONCLUSIONS
A sophisticated lake water quality mitigation system has provided mixed results for Machado Lake. Phosphorus is a success story, with sediment P releases greatly dampened in the post-project system and overall lake P levels approximately halved, as compared to pre-project. Important implications for other urban lake nutrient mitigation projections have been highlighted, including the effectiveness of reducing P recycling rates through lake sediment remedial action. Reducing sediment P releases, primarily through dredging and capping, has been shown to be highly effective at reducing overall lake water column P levels, despite continued pressure from watershed loads. A collateral benefit of this result is that the lake is now, at times, P-limited with respect to phytoplankton growth. This implies that further reductions in P could have significant impacts on lake algae levels.
With respect to N, however, further work is required to appreciably reduce lake N levels. In-lake N mitigation has not proved to be effective, particularly with respect to dampening seasonal sediment N releases. While oxidation appears to be generally ineffective with respect to reducing N recycling in this lake, we hypothesize that the primary reason for the observed differences in N and P results lie in the efficacy of the lake dredging with respect to removal of the two nutrients.
Significant lake N reductions may not be achievable until watershed N loads are reduced. Watershed loads to the lake, for both P and N, remain high. Even with the existing in-lake mitigation system, and the significant progress achieved in P levels, modeling indicates that watershed TP and TN loads to the lake need to be reduced by 65 and 45%, respectively, to achieve lake numeric water quality targets. It is well documented that external load reductions are critical to lake restoration efforts (Abell et al. 2022). Future modeling, and monitoring, efforts should focus on integrating watershed processes with lake water quality response (e.g. Wang et al. 2019), to improve model utility and guide future mitigation efforts.
Auditing of a previously constructed lake water quality model indicates strong predictive power with respect to P response to mitigation efforts, but weak predictive power for N. A recalibrated model provides insight into current nutrient dynamics in the lake and provides for improved predictive power in the future. Model sensitivity analyses indicate lake sediment dredging and capping has provided for the largest improvement in water quality, with secondary contributions from the recirculating wetland and oxygenation systems. Based on this analysis, future mitigation efforts for Machado Lake should focus on optimizing existing controls and further P inactivation (e.g. dosing with binding agents such as alum), in conjunction with timely implementation of planned stormwater controls in the upstream drainage areas.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.