Salinization, warming, and loss of water clarity inhibit vertical mixing of small urban ponds

Urbanization drives multiple environmental changes that influence critical ecosystem processes. Factors such as salinization by deicing road salts, reduced water clarity (and greater light attenuation) from eutrophication and sediment loading, and warming constrain not only the biodiversity of ponds, but also their physical mixing (with consequences for oxygen availability and the provision of ecosystem services). Leveraging an extensive urban gradient in the Greater Toronto Area, we collected summertime depth profiles from 50 stormwater retention ponds to investigate their vertical stratification. We found that water columns were generally stratified but contrary to expectations, we found relatively minor roles of basin area and depth. Instead, we discovered an overwhelming effect of salinity along with significant impacts of temperature and water clarity on water density gradients. Findings extend our fundamental understanding of mixing regimes in small, shallow waterbodies and indicate increasing risks to pond functioning in a warmer and saltier future.

of surface runoff to dampen hydrological responses, precipitate suspended solids, and mitigate the transport of contaminants to receiving waters, there is growing appreciation of their functions for carbon sequestration and as wildlife habitat and public spaces for recreational use (e.g., dog walking, bird watching, and angling; Moore and Hunt 2012).However, by acting as water quality filters for their broader catchments, urban ponds also contend with heightened anthropogenic pressures.For example, in temperate regions that experience winter freezing, aquatic ecosystems face salinization by deicing road salts (especially sodium chloride; Szklarek et al. 2022).At the same time, ponds are threatened by thermal stress from rising atmospheric temperatures as well as summer inflows of heated runoff from roadways and parking lots.Water clarity also plays an important role in temperature, as suspended sediment and excessive primary production (caused by nutrient loading) both increase light attenuation and relative warming of near-surface waters (Fee et al. 1996).While each of these stressors exert direct effects on the organisms and communities residing in ponds, they also influence the physical process of how water columns mix, with consequences for their broader functioning.
Mixing regimes dictate provisions of several critical ecosystem services.If vertical water density gradients become too strong, waterbodies may stratify, resulting in upper (epilimnion) and lower (hypolimnion) layers that mix poorly or not at all.If sustained over time, stratification leads to oxygen depletion (anoxia) in the hypolimnion, especially in highly productive systems impacted by cultural eutrophication (elevated nutrient availability from agriculture and urbanization; Nürnberg 1995).Anoxic waters are detrimental to aquatic life, including invertebrates and fish, and alter the redox potential of bottom sediments.Under cool, anoxic conditions, aquatic sediments may experience reduced aerobic respiration and store more carbon (Bartosiewicz et al. 2019) but also emit potent greenhouse gases like methane (Herrero Ortega et al. 2019) and cycle contaminants, such as phosphorus, back into the water column where they are more biologically available and at greater risk of downstream transport (Sibert et al. 2015;Taguchi et al. 2020).
Although the mixing regimes of lakes are appreciated for their central role in numerous ecological and biogeochemical processes, pond dynamics have largely been ignored as their shallow depths were assumed to prevent the development of meaningful density gradients (Lewis Jr. 1983).However, this long-standing assumption (that ponds are polymictic, or continuously mixed) has been challenged in recent years and it is now recognized that smaller waterbodies do stratify under certain conditions.For instance, Holgerson et al. (2022) revealed a range of thermal profiles in ponds and proposed a novel framework for classifying the summer mixing regimes of shallow waterbodies as a function of surface area and depth.This framework generated several useful insights but was limited in its ability to distinguish between certain categories, possibly because their predictive model overlooked variables such as water transparency and salinity.While loss of water clarity influences thermal stratification (by warming surface waters), several studies have demonstrated the importance of salinity to the establishment of chemical stratification in lakes and ponds during both winter and summer months (e.g., Judd 1970;Marsalek et al. 2000;Novotny and Stefan 2012;McEnroe et al. 2013;Dupuis et al. 2019;Ahmed et al. 2023).If kinetic forces (from wind) exceed the stabilizing effects of temperature-and salinity-driven density gradients then a waterbody will mix, and it is possible to model these physical dynamics to make predictions about when stratification will occur given local conditions (e.g., Ladwig et al. 2023).But despite recent gains in characterizing mixing regimes, smaller urban ponds contend with unique weather and drainage patterns (including impacts of buildings on wind speed and storm runoff) and the practical influence of temperature, salinity, and morphometry is poorly resolved under field-based conditions.
In light of increasing observations of ponds stratifying during the summer and the clear gaps in our understanding of the roles that salinization and other anthropogenic stressors play, we conducted an extensive survey of urban ponds to characterize mixing regimes and test the relative importance of underlying factors.First, we sought to compare vertical profiles across a broad set of ponds to investigate variations in summertime water density gradients and the extent of stratification leading to subsurface anoxia.Second, we placed our findings in the context of recent research on mixing regimes (Holgerson et al. 2022) to extend the predictive framework and identify mixing thresholds for salinity.Finally, we tested relationships between key morphometric (maximum depth and surface area) and anthropogenic factors (maximum specific conductance, surface temperatures, and water clarity) and vertical density gradients to advance understanding of what drives the mixing regimes of ponds and assess their broader sensitivity to increasing urban stressors (salinization, loss of water clarity, and thermal loading).As we hypothesized a central role of salinity, we also tested for interactions whereby salinization moderates (either amplifying or dampening) the effects of other factors.

Study design
We conducted a survey of the aftbay area of 50 wet, offline stormwater retention ponds located in Brampton, Ontario during the summer of 2022 (Loewen and Jackson 2023).As one location was deemed an outlier due to unusual chemical and thermal profiles, we excluded this site from analyses to avoid skewing results and proceeded with a sample size of 49.Ponds were small (< 3 ha) and selected to capture a range of depths and catchment conditions (based on percentage impervious cover) across residential and commercial areas of the Credit (n = 26) and Humber (n = 23) River watersheds flowing into Lake Ontario (Fig. 1).All ponds were sampled in July and August when thermal stratification was expected to be at its maximum, and a subset of nine locations were resampled in September.As ponds varied in their design and orientation, we collected vertical profiles at 10 cm intervals at approximately the deepest location of the main basin in each pond during each sampling event.In July and September, we measured temperature and specific conductance from 10 cm below the water surface to just above the sediment surface using a YSI Pro30 Conductivity Meter (Xylem, Washington, USA).In August, we measured dissolved oxygen in addition to temperature and specific conductance profiles starting at about 30 cm below the water surface using a YSI ProDSS Multiparameter Meter.We also obtained water samples at 0.5 m depth for analysis of dissolved chloride, total sodium, and total phosphorus.Chemical analyses were performed by Testmark Laboratories Ltd. (Mississauga, Ontario, Canada) following standard methods (see Supplemental Information Appendix S1 for details).Secchi depths were measured in situ during each sampling event and pond areas were estimated from recent satellite imagery using Google Earth (Version 9.188.0.0,Google, Mountain View, USA).

Physical parameters
Water densities were calculated as a function of fieldmeasured temperature and salinity (computed from fieldmeasured specific conductance) following Millero and Poisson (1981) using the rLakeAnalyzer package (Winslow et al. 2019) in the R statistical computing and graphics environment (R Core Team 2021).While this is a standard approach, we acknowledge that the relationship between specific conductance and salinity can vary both within and among sites, and thus water density estimates may be biased by shifts in ionic composition (Boehrer et al. 2010).Water density gradients were used to assess the strength of stratification and calculated as the mean difference in densities across 10-cm intervals at each site during each sampling event (i.e., average change within each profile).Differences in water densities based only on temperature (assuming salinity = 0) and taking salinity into account were used to evaluate the contribution of salinity to water density gradients.For these calculations, salinity was computed from specific conductance measurements as described in Wagner et al. (2006) and implemented using the CompensateR package (Gold 2023).While there are several metrics used to evaluate the strength or depth of mixed layers (see Gray et al. 2020), we examined changes in water density to align with prior work classifying the vertical profiles of ponds and shallow lakes (Holgerson et al. 2022).Following from Holgerson et al. (2022), we assessed stratification using a mixing threshold of 0.0287 kg m À3 across 10-cm intervals for each profile (modified from 0.287 kg m À3 m À1 between the top and bottom of the water column).Maps were generated using ggmap (Kahle and Wickham 2013), data wrangling were performed using several functions from the tidyverse package (Wickham et al. 2019), and all plots were created using ggplot2 (Wickham 2016).

Statistical analysis and visualization
To test the hypothesized roles of anthropogenic factors driving the mixing regimes of small urban ponds, we used a space-for-time substitution approach applying generalized linear models (Gaussian distribution with log-link) to infer potential impacts of temporal change from responses along multiplicative interactions of maximum specific conductance with each of the other predictors.All predictors were centered and scaled (as z-scores) to ease the interpretation of interactions and place variables on a comparable unitless scale.As sodium chloride is commonly used for winter deicing in the study area, we also tested the relationship between specific conductance and dissolved chloride concentrations using simple linear regression.Packages ggeffects (Lüdecke 2018) and interactions (Long 2019) were used to predict values for plotting.

Results
Sampling locations spanned broad environmental gradients across the two watersheds, but these gradients did not show clear spatial patterns (Fig. 1).The ponds ranged from sampling events.Thresholds are the maximum values observed for sites with mean water density gradients of less than 0.0287 kg m À3 across 10-cm depth intervals.For reference, the specific conductance of seawater is $50,000 μS cm À1 .
0.4-4.1 m maximum depth and 0.06-2.41ha surface area and showed general declines in maximum specific conductance (range = 330-14,500 μS cm À1 ), surface temperature (21.1-28.3C), and Secchi depths (0.1-2.9 m) between July and August.Correlations among environmental variables were mostly weak (Pearson r < 0.3) except for relationships between maximum depth and surface area (r = 0.39 in July and 0.47 in August) and surface temperature (r = 0.37 in August; Table S1).
Site productivity ranged from meso-to hypereutrophic (nearsurface phosphorus concentrations ranged from 0.017-0.100mg L À1 ) and 34 developed anoxic conditions in August (dissolved oxygen concentrations below 0.5 mg L À1 ).Dissolved chloride concentrations ranged from 17.7-1560 mg L À1 and exhibited a strong, linear relationship with maximum specific conductance, supporting the use of conductance as a surrogate for the mass of solutes in water density calculations (p < 0.001; Supporting Information Table S2; Fig. S1).
Vertical profiles varied considerably among ponds, but showed general trends of declining temperature, increasing salinity (specific conductance), and declining dissolved oxygen with depth (Supporting Information Figs.S2-S5).These patterns contrasted with the outlier site (C45, which was excluded from analysis), where temperatures increased down to 1.3 m depth in July (1.2 m depth in August) and specific conductance peaked at 33,000 μS cm À1 (Supporting Information Figs.S6-S7).Across the remaining 49 sites, we found similar patterns of increasing water density with depth, but gradients were much more varied when taking salinity into account (Fig. 2 and Supporting Information Figs.S9-S11).Although density profiles were unaffected by changes in salinity in some ponds, others were driven entirely by salinity and taking this into account for density calculations increased gradients by an average of 30.3% across sites.We found that most ponds developed stratification but with the number of unmixed sites decreasing across the summer and into the fall (44/49 in July, 30/49 in August, and September 2/9).Among investigated factors, salinity showed the clearest stratification thresholds, as only those ponds with maximum specific conductance < 2200 and 2550 μS cm À1 mixed during the July and August sampling events, respectively (Fig. 3).
Generalized linear models revealed significant effects of maximum specific conductance, surface temperature, and Secchi depth (July only) on water density gradients (Table 1 Loewen and Jackson Stratification of urban ponds Fig. 4).Maximum depth and surface area were nonsignificant during both sampling events, but the morphometric variables and Secchi depth interacted significantly with maximum specific conductance in July (Supporting Information Figs.S12).
Interactions revealed a small but significant increase in the negative effect of Secchi depth, a reversal of the weakly positive effect of maximum depth (becoming negative), and an amplification of the negligible effect of surface area (becoming positive) under elevated salinity.Although the interaction between specific conductance and temperature was nonsignificant, it indicated a potential synergy whereby the effects of warming are amplified under more saline conditions (and vice versa).

Discussion
Our results corroborate recent studies (e.g., McEnroe et al. 2013;Holgerson et al. 2022;Ahmed et al. 2023) challenging the long-held idea that shallow waterbodies are generally mixed during summer.However, rather than depth and area, we discovered that water density gradients were primarily driven by salinity, temperature, and water transparency across a broad set of urban ponds.While ponds are influenced by geological setting and may exhibit natural salinity in urban centers (electrical conductivity up to 3000 μS cm À1 ; Oertli and Parris 2019), our survey highlights the legacy of deicing road salts in temperate regions.Even though deicing is limited to the winter, we found that many ponds were brackish throughout the summer, suggesting chronic impairment (Lawson and Jackson 2021).Other studies have highlighted increased sensitivity of smaller, urban systems to road salt runoff, with small lakes ($ 4-5 ha) showing greater increases in chloride concentrations over time (Scott et al. 2019).A survey conducted by Hassall and Anderson (2015) found that conductivity was highest in ponds < 1 ha in area and recorded values up to 3700 μS cm À1 in a smaller stormwater pond (< 0.3 ha) sampled in May (when ponds were receiving salts released from melting snow).While our data are limited to summer, we anticipate dimictic regimes where salinity declines (due to flushing) across the open water season leading to weakened stratification and eventual mixing in the fall (although more study is needed).As prolonged stratification is associated with subsurface anoxia and declining ecological function, our

FIGURE 4
Relationships between mean water density gradients and environmental variables predicted from generalized linear models (Gaussian distribution with log-link function) for July (a-e) and August (fj).Predictions are conditional effects (holding other terms at their means).
Point color indicates variation in either surface temperature (a and f) or maximum specific conductance (b-e and g-j).Shading represents 95% confidence intervals and * denotes significant main effects following pvalue adjustment by the false discovery rate method.See Supporting Information Fig. S12 for corresponding interaction plots.
results reveal an underappreciated risk posed to aquatic ecosystems by increasing climate change and urbanization.By integrating effects of salinity, our work extends the existing framework for predicting mixing regimes in shallow waterbodies.First, we show the importance of taking salinity into account when calculating water densities, as changes in salinity increased density gradients by an average of 30% across sites.Although Holgerson et al. (2022) were unable to account for this effect (as they lacked conductivity profiles), our maximum depth threshold for July (1.9 m) supported their finding that waterbodies deeper than 2 m rarely mixed.Building on this result, we discovered that sites < 1.9 m deep remained stratified if specific conductance exceeded 2200 μS cm À1 , indicating a potentially critical driver of variation within the "mixing regime transition" (sites with maximum depths between 0.74 and 2 m) where classifications based on purely morphometric variables had limited success (Holgerson et al. 2022).As all our sampling locations were < 4.17 ha (the surface area threshold above which ponds are predicted to mix often; Holgerson et al. 2022), we cannot evaluate the effects of salinity in larger, shallow waterbodies.However, emphasizing salinity's central role in smaller ponds, we found mixing at sites as deep at 3.6 m as density gradients degraded into August, but only in sites where maximum specific conductance was < 2550 μS cm À1 .Although our observations were static (snapshots in time) and thus we were unable to assess weather-related effects or stratification duration (including possible diurnal patterns), we found that maximum specific conductance, surface temperature, and Secchi depth were more important to the mixing regimes of small urban ponds than either maximum depth or surface area.
Though it is well established that water column density gradients depend on salinity (e.g., Judd 1970;Novotny and Stefan 2012;Ladwig et al. 2023;Ahmed et al. 2023), the importance of salinization relative to other factors underlying mixing regimes in small, urban ponds is poorly understood.As empirical studies of water density patterns have typically been restricted to one or a few ponds at a time, McEnroe et al. (2013) performed one of the few broad-scale surveys examining both the thermal and chemical stratification of urban ponds.They found that most of the 45 stormwater retention basins they profiled in southern Ontario, Canada, were stratified, and that water column stability was more correlated with temperature than salinity gradients (measured between the top and bottom of the water column).This key finding is supported by our results indicating that changes in salinity increased density gradients by about 30% on average (whereas the rest of the gradients reflected changes in temperature).However, as stratification necessarily influences subsequent thermal and chemical profiles, these patterns shed little light on which factors contribute to establish the density gradients.Thus, while stratified water columns may reflect greater changes in temperature than salinity along their profile, our results show that density gradients are primarily driven by salinity inputs (with secondary effects of temperature loading and water clarity).
Our tests of salinity moderating the effects of other factors influencing water density gradients were inconclusive.As salinity and temperature affect water density via different but related mechanisms, with salinity increasing mass and warming causing water to expand, we hypothesized that the effects of thermal expansion would differ (possibly increase) in heavier, saline waters (Thom and Ricken 2019).While results indicate that the effect of warming was amplified under more saline conditions (especially in July), this synergistic effect was nonsignificant.Instead, we identified significant interactions with water clarity (represented by Secchi depth), maximum depth, and surface area in July.Here, the negative relationship between Secchi depth and mean density gradient was stronger in saltier ponds, which tended to be less transparent.Interactions with maximum depth and surface area are indicative of the dominant effect of salinity.As predicted, water density gradients declined with increasing surface area and reduced depth when specific conductance was low, but effects were reversed when specific conductance was elevated because salinity overrode the effects of morphometry, and saltier ponds tended to be smaller and shallower.Given these results and potential for feedbacks (e.g., water clarity and salinity as a function of productivity under increasing temperature), further studies with greater sample sizes or more controlled conditions are warranted.
Significant roles of salinity, temperature, and water clarity reveal underappreciated threats to critical ecological and biogeochemical processes posed by urbanization and climate change.As chloride and sodium do not transform biologically, salts used for winter deicing accumulate in soils where they may leach into shallow groundwater and provide a continuous source of salinity to urban ponds and streams throughout the summer (Scott et al. 2019;Lawson and Jackson 2021).Salinization coupled with increased temperatures, whether from atmospheric warming (due to climate change or urban heat island effects), heated runoff, or greater light attenuation in surface waters, pose direct threats to pond biota (Van Meter et al. 2011;Thompson and Shurin 2012;Lawson and Jackson 2021).However, the consequences of reduced mixing may exert broader impacts on ecosystems by depleting oxygen (Dupuis et al. 2019;Ahmed et al. 2023;Jane et al. 2023).We conducted surveys during daylight hours when rates of photosynthesis were at their highest, but conditions are likely to deteriorate further as oxygen is consumed during nighttime respiration (Markager and Sand-Jensen 1989).Given the ubiquity of stormwater retention ponds and their roles in supporting regional biodiversity (e.g., Pinel-Alloul and Mimouni 2013) and maintaining water quality in downstream rivers and lakes (e.g., Taguchi et al. 2020), our work underscores the importance of mitigation strategies (e.g., aeration) and implementation of policies to limit pond thermal loading and salt accumulation.

Fig. 1 .Fig. 2 .
Fig. 1.Sampling locations showing variation in maximum specific conductance (a) and surface temperature at 0.3 m depth (b) during the August sampling event with point size indicating variation in maximum depth (m).Panel (c) shows sampling locations on a regional map.Map tiles by Stamen Design, under CC BY 3.0 (data by OpenStreetMap, under ODbL).

Fig. 3 .
Fig.3.Mixing regime thresholds for maximum specific conductance, surface temperature, and maximum depth during the July (a-c) andAugust (d-f)

Table 1 .
, Results of generalized linear regression models predicting mean water density gradients as a function of environmental variables in July and August.
Generalized linear models are Gaussian models with a log link (n = 49); and * denotes significant coefficient after false discovery rate (FDR) adjustment.