<<

1 Stormwater Research Summary Banner

2 Objective 3: Phosphorus Release from Stormwater

3 Authors: Vinicius Taguchi, Tyler Olsen, Ben Janke, Heinz G. Stefan, Jacques Finlay, John S. 4 Gulliver 5 6 Background 7 There is growing concern that aging stormwater ponds may become net sources of phosphorus (P) to 8 receiving waters during high flow events. Decades of research in lentic ecosystems have demonstrated 9 potential for sustained release of P previously deposited in sediments (i.e. internal loading) in and 10 wetlands. This body of work demonstrates that P is a major contributor to that is difficult 11 to effectively manage (Gächter and Wehrli 1998, Søndergaard et al. 2007, Song and Burgin 2017). 12 Stormwater ponds often have high external P loading, and other characteristics that may increase the 13 likelihood of internal loading, such as low availability of metals that immobilize P. However, such ponds 14 have received comparatively little research attention. Objective 3 research examined controls on P 15 cycling in stormwater ponds toward predictive understanding of P retention behavior. Our 16 investigations focused on the examination of P using a data synthesis combined with detailed 17 characterization of sedimentary P release in five ponds (Obj. 3a), and annual P mass balance 18 measurements in three ponds coupled with analyses of factors associated with P retention or release 19 within these ponds (Obj. 3b). 20 21 Data Synthesis 22 To investigate the prevalence of excess P in stormwater ponds, we examined data previously collected 23 by the Riley Purgatory Bluff Creek Watershed District (RPBCWD) in Minnesota. Between 2010 and 2013, 24 RPBCWD surveyed 98 ponds and observed high total phosphorus (TP) concentrations (i.e. >0.5mg/L) in a 25 number of them (RPBCWD 2014). Surface grab samples were collected from the downstream third of 26 each pond between July and September. Some ponds were added during the study, so that ponds were 27 samples between 3 to 18 times (median of 5 samples), except for two that had single samples. We 28 compared RPCWD pond TP to characteristic stormwater runoff from the Twin Cities Metro Area (TCMA) 29 and the U.S. The expectation was that TP concentration in stormwater ponds would be less than the 30 typical inflow TP concentration due to loss of P from the water column from settling of particulates and 31 assuming other non-runoff sources of P loading (e.g., from local inputs of leaf litter, or grass clippings,) is 32 negligible by comparison in most ponds; however, the presence of internal loading from recycling of 33 sediment P within the ponds could cause elevated pond TP. TCMA stormwater runoff was characterized 34 using stormwater monitoring data (19 sites, 2362 storm events, n=2505 TP samples) from several TCMA 35 watershed management agencies (Janke et al. 2017). These data were fit to a log-normal distribution 36 (Van Buren et al. 1997), and the upper 95% confidence interval of expected values (centered about the 37 mean rather than the median as in the case of a percentile), which represents the maximum value for 38 approximately 97.5% of the distribution, was then calculated to be 0.38 mg/L. Thus, 97.5% of the inflow 39 data are estimated to be below 0.38 mg/L. For comparison, the process was repeated with a national 40 stormwater runoff characterization dataset (104 municipalities, 8602 storm events, n=7407 TP samples) 41 from the US EPA NPDES Stormwater Quality Database (Pitt et al. 2008), yielding a 95% CI of 0.42 mg/L 42 TP. We compared sampling data from RPBCWD against the Twin Cities Metro 95% CI threshold value of 43 0.38 mg/L in Figure 1, which shows box plots for each of the 39 ponds with median TP concentration 44 above the 95% CI and condenses the remaining 59 ponds into a single box plot.

1

45 46 Our analysis shows that 39% of the 98 ponds sampled had median TP concentrations greater than the 47 95% CI (Figure 2). For each dataset, measurements below reported limits of detection were 48 approximated as equal to half the detection limit (Kayhanian et al. 2002). We hypothesized that these -3 49 ponds have high internal P loading from the sediments, released as orthophosphorus (ortho-P or PO4 ), 50 and are therefore more likely to experience high P concentrations. We address this hypothesis through 51 the laboratory column studies and field measurements described below.

52 53 Figure 1. Box plot of RPBCWD TP grab sample data. The 59 ponds with a median below 0.38 mg/L (95% CI for TCMA 54 runoff data compiled in Janke et al. 2017) are combined into the far-right box plot. 55

56 57 Figure 2. Distribution of total phosphorus concentrations from the RPBCWD pond (grab sample) data presented in 58 Figure 1, with median grab sample values from the five intensively studied ponds (A-E) from this study 59 also shown. 60 61 Sediment Core Methods

2

62 To address our hypothesis that P release from the sediments drives high P concentrations in stormwater 63 ponds, we sought to identify and better understand the mechanisms controlling P release from 64 stormwater pond sediments. We collected intact sediment cores and overlying water, using core tubes 65 with an approximately 7 cm outer diameter and approximately 1.5 m in height, from each of 5 66 stormwater ponds (Figure 3, Table 1). These ponds covered a range of characteristics (Table 1), including 67 water column P concentrations (Fig. 2). Six cores were collected from pond A, and five from ponds B-E. 68 69 Table 1. Characteristics of the five studied ponds. All ponds were included in the core study; ponds C, D, and E were 70 studied in the monitoring task. TIA = Total Impervious Area; n/a = not available. Grab Sample TP (n = 10 - 20 71 samples; hypolimnion and water column samples excluded) and influent event mean concentration (EMC) of 72 stormwater TP (n = 20 – 22 samples; snowmelt excluded) come from the monitoring task of the study for ponds C, 73 D, and E (see SI). Grab sample values for ponds A and B come from previous studies (Pond A: Noah Czech, City of St. 74 Cloud; Pond B: RPBCWD 2014).

Water Drainage Max Median Grab Influent TP Site ID Site Name Area Area Depth Age Land Use TIA Sample TP EMC ha ha m yr % mg/L mg/L Residential/ A St. Cloud 52 0.21 22.1 1.2 18 60 0.57 n/a School B Minnetonka 849_w 0.66 2.8 1.7 >50 Residential 30 2.70 n/a

C William Street Pond 0.23 15.4 1.5 >30 Residential 19 0.36 0.38 Parking Lot, D Church Pond 0.04 1.8 0.8 ~20 >75 0.24 0.12 Residential E Alameda Pond 1.05 115.3 1.8 >30 Residential 20 0.22 0.30 75 76

77 78 Figure 3. Stormwater pond field sites. Pond A is located in St. Cloud, MN, Pond B in Minnetonka, MN, and Ponds C, 79 D, and E in Roseville, MN. 80 81

3

82 Each core was incubated at the St. Anthony Falls Laboratory with 30-40 cm of sediment with 83 approximately one meter of overlying pond water in clear PVC tubes (Figure 4). Cores were incubated 84 under oxygenated and anoxic conditions, during which ortho-P and total phosphorus (TP), dissolved 85 oxygen (DO), temperature, and pH were monitored. Initially, the cores were drained and the overlying 86 water column was filtered through a 1.2 μm glass fiber filter to remove suspended particulates. The 87 filtered water was then added back to the respective core with care taken to avoid sediment 88 resuspension. The water columns were subsequently bubbled with air to simulate the mixing and 89 aeration that takes place in spring. Afterward, air bubbling was halted and cores were left unmixed for 90 one month, the first week of which was used to calculate an overall sediment P flux rate from the 91 measured water column P concentrations. This was done by fitting a linear regression through the P 92 concentration time series data and dividing the slope (P mass per time) by sediment surface area of the 93 cores. Discrete sampling events occurred every one to seven days, depending on the observed rate of 94 change, and are identified by points in Figure 5. Samples were taken from the center of the water 95 column during the mixed oxic phase and from both the center of the water column and 8 cm above the 96 sediment surface during the unmixed anoxic phase. DO, temperature, and pH readings were taken from 97 the center of the water column during the mixed oxic phase and in 15-cm increments, beginning 3 cm 98 above the sediment surface during the unmixed anoxic phase. In order to minimize variables and isolate 99 the sediment P flux for measurement, core incubations were not provided with particulate P additions 100 as would occur in-situ via atmospheric deposition and stormwater runoff. At the conclusion of the 101 month-long unmixed monitoring period for each core, a sequential chemical extraction of the sediments 102 was completed to illustrate differences in P speciation and identify how P was bound in the sediments 103 (Engstrom 2010). This information on P speciation and related P sediment binding characteristics in 104 important to understand mechanisms that control P release and internal loading across ponds with 105 varied P load and morphology. 106 107

108 109 Figure 4. Intact sediment and water cores taken from Pond B. 110

4

111 Sediment Core Results 112 All of the pond sediments released ortho-P under low DO conditions (< 0.5 mg/L; Table 2 and Figure 5). 113 By contrast, ortho-P release was negligible under high DO conditions. For each pond, we used the 114 calculated P flux from sediments to estimate the potential impact of sediment release on water column 115 TP concentration (mg/L) as a simple function of the ortho-P release rate (mg/m2/day), the number of 116 days over which ortho-P release is believed to be occurring (days in season), and the mean depth of the 117 pond (m) as described below: 118 × 180 119 = ÷ (1000 )

𝑃𝑃 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝑖𝑖𝑖𝑖 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝐿𝐿 120 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼 𝑜𝑜𝑜𝑜 𝑃𝑃 3 121 Ponds C and D had a relatively low release rate 𝑀𝑀that𝑀𝑀𝑀𝑀𝑀𝑀 would𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 haveℎ a fairly minor impact𝑚𝑚 upon pond water 122 column TP concentration. Ponds A, B, and E, however, had higher release rates that could have a 123 substantial impact on TP concentration under low DO conditions. 124 125 Table 2. Calculated potential impact on overlying pond water column phosphorus concentration for 26 cores taken 126 from five ponds (six from pond A, and 5 from ponds B, C, D, and E) (P-Flux ± 67% CI of the mean value). 127 Pond Organic Matter Oxic P-Flux Anoxic P-Flux Mean Depth Potential Impact on P** - Content* (mg/m2/day) (mg/m2/day) (m) (mg/L) A 30% -1.27 ±0.71 7.51 ±2.93 0.7 1.8 B 86% -0.14 ±0.76 5.62 ±1.80 1.0 1.1 C 15% -4.38 ±2.89 1.09 ±0.26 2.0 0.1 D 16% -5.80 ±1.94 2.27 ±0.49 1.2 0.3 E 27% -19.78 ±3.37 3.18 ±2.76 1.2 0.5 128 *Top 11 cm **Based on Anoxic P-Flux; Assuming 180 days in season 129 10 400 8 300 6 P (µg/L) - 200 4 ortho

100 2 Dissolved Oxygen (mg/L) 0 0 0 7 14 0 7 14 Days Days 130 A B C D E A B C D E 131 Figure 5. (a) Orthophosphate concentration in sediment-water core waters under low DO conditions (Error bars = 132 67% CI of the mean). The increase over time is due to release from the sediments. (b) Change in dissolved oxygen 133 concentration 8 cm above core sediment (Error bars = 67% CI of the mean).

5

134 Relatively low P release rates from cores from ponds C and D (Figure 5a) may result from reduced 135 sediment microbial activity. These ponds’ sediments demonstrated relatively lower oxygen demand 136 (suggested by lower rates of DO depletion; Figure 5b) and organic matter content, the former being 137 indicative of opportunistic aerobic respiration by microbes and the latter being the microbial food 138 source. 139 140 We examined phosphorus speciation in the sediments to identify which phosphorus forms were 141 potentially responsible for the observed sediment releases. Sequentially extracting different P fractions 142 in the top 4 cm of sediments (1 sample/cm depth; n=4 samples/core) from each sediment-water core 143 (all cores from ponds A and E and 3 cores from ponds B-D were examined) revealed that organic 144 phosphorus may play an important role in urban stormwater pond systems (Figure 6). Loosely bound, 145 iron bound, and labile organic bound P are all considered to be relatively mobile and therefore play the 146 largest role with respect to P dynamics in the system (Olsen 2017). We found that pond sediments were 147 dominated by labile organic bound P, whereas in sediments loosely bound and iron bound P, 148 sensitive to redox conditions, are typically more important (Natarajan et al. 2017). The smaller labile 149 organic bound P content of Ponds C and D corresponds to their lower release rates, suggesting that 150 labile organic bound P may participate in the release of ortho-P from the sediments. More ponds would 151 need to be sampled, however, before robust relationships between sediment speciation and sediment 152 release rate can be determined. 153 154

155 156 Figure 6. Phosphorus speciation in core sediments (Error bars = 67% CI of the mean). Loosely bound P is assumed to 157 be primarily dissolved, labile organic bound P can be degraded over time to become ortho-P, mineral bound 158 (primarily calcium) P is associated with minerals other than iron and aluminum, and residual organic bound P is 159 assumed to be non-degradable. Redox P is the sum of loosely bound and iron bound P, and mobile P is the sum of 160 redox P and labile organic P. 161 162

6

163 Field Measurements and Phosphorus Mass Balance for Ponds 164 A second major objective (3b) of the study was to relate pond P retention (or export) to season, pond 165 water quality, pond characteristics and hydrology, through monitoring in three of the ponds (Ponds C, D, 166 and E). This field data informs the laboratory core study by describing the actual physical and 167 biogeochemical conditions at the study sites. 168 169 Water Quality and Site Conditions of Monitored Ponds 170 The three ponds were selected for simplicity of configuration (one inlet, one outlet), for proximity to the 171 University of Minnesota (for logistical purposes), and to include two old ponds and a recently- 172 maintained pond (Table 1; SI). These ponds were generally small (0.04 – 1.05 ha), older (>20 years old), 173 sheltered by mature trees, had low watershed-to-pond area ratios, and were located in predominantly 174 residential areas. The sites had moderate P levels (median water column and epilimnion TP ranged from 175 0.26-0.46 mg/L across the 3 ponds; Table 1 and SI), corresponding roughly to the middle third of the 176 median TP range among ponds surveyed by RPBCWD (Figure 1). Influent TP in stormwater at the study 177 ponds (TP EMC from 0.12-0.38 mg/L; Table 1 and SI) was lower than observed in previous studies, 178 including the 95% CI from the accumulated Twin Cities Metro dataset (0.38 mg/L; Janke et al. 2017) and 179 from Pitt et al. 2008 (0.42 mg/L across all urban land uses). Therefore, these ponds were likely not 180 subjected to extremely high phosphorus loading over their lifetimes, and the relatively low water 181 column TP concentrations suggest that they may not represent poorly-functioning stormwater ponds. 182 183 Methods: For each pond, water volumes and P concentrations were measured in outflows, inflows, and 184 the water column over the study period (roughly 18 months for Ponds C-D and 11 months for Pond E), 185 including winter and snowmelt. Inflows and outflows were monitored with ISCO automatic water 186 samplers (flow-paced samples composited into 1-liter samples) and depth-velocity flow probes installed 187 in inlet and outlet storm pipes for as many rainfall and snowmelt events as possible during the 188 monitoring season (Table 3). Water column (integrated from the water surface to just above the 189 sediment surface), epilimnion (from within 10 cm of the water surface), and hypolimnion (from within 190 25 cm of the sediment surface) water samples were collected as 1-L grab samples from a canoe or 191 through the ice. Vertical profiles of chemistry (temperature, dissolved oxygen, conductivity) were 192 measured manually at 0.25-m increments, with continuous logging of temperature profiles at stations in 193 the middle of the ponds from July 2017 – May 2018. All manual grab samples and profiles were taken 194 from the deepest point of each pond (approximately near the center in all instances) every 1-4 weeks 195 during the study period, and care was taken not to disturb the sediment surface when hypolimnion or 196 water column samples were being collected. All three monitored ponds were sampled on the same day 197 during field trips, generally within an hour of each other. 198 Water samples were analyzed using standard EPA laboratory methods for major forms of P (ortho-P, 199 TDP, and PP) and chlorophyll-a, with a subset analyzed for total suspended sediment. Loads were 200 determined by the product of observed volume and observed P concentration. For un-sampled events, 201 characteristic concentrations were assigned by averaging the mean monthly concentrations with all 202 concentrations from observations within 3 weeks prior to and after the event. Table 3 below 203 summarizes the number of modeled and observed events per site (see also “Task 3b Additional Data” 204 spreadsheet); all annual or seasonal loads reported throughout the remainder of the report include 205 these modeled events. The SI contains further details of laboratory methods and quality control of both 206 chemistry and hydrologic data, including corrections made to the inflow hydrology at Pond C, which was 207 problematic to monitor due to difficult access and generally low flows. 208

7

209 Table 3. Event summary (total events, sampled events, and modeled events) by monitoring site.

Events with Events with Site Total Events Observed Modeled Concentrations Concentrations Pond C Inlet 131 40 91 Pond C Outlet 145 94 51 Pond D Inlet 88 31 57 Pond D Outlet 28 21 7 Pond E Inlet 77 37 40 Pond E Outlet 44 28 16

210

211 Water Column Stratification Results 212 Stratification during summer with turnover and mixing in fall and spring are normal in large lakes, but 213 small ponds are designed to mix during larger storm events. Instead, regular manual temperature, DO, 214 and conductivity profiles at all three ponds during 2017 revealed evidence of stratification throughout 215 most of the year, especially at the larger Ponds C and E (see SI, Figs. III-D and III-E for Ponds C and D; 216 Pond E shown in Figure 7). Note in particular that DO concentration never exceeded 0.5 mg/L at the 217 bottom of Pond E during the whole year except during site visits on 10/20 and 11/2, following pond 218 turnover (likely late September) and a nearly 76-mm storm event on 10/2/17 (Figure 7a). Anoxia was 219 also persistent throughout the year (and within most of the water column) at Pond C (William Street), 220 while at the shallower Pond D, high DO (>5 mg/L) was likely present during and after spring turnover. 221 222 Strong stratification in the three ponds also prevented mixing at finer time scales, with diurnal variations 223 in temperature affecting only the near-surface water, such that no overnight mixing was evident during 224 the summer period examined, even though stormwater ponds are designed to mix frequently (Figure 8). 225 Even moderate stormwater inflows did not result in complete mixing; for example, sharp temperature 226 shifts on 6/11/17 from a 28-mm storm event show evidence of partial mixing down to roughly 1.25 m 227 below the pond surface, with no mixing in the bottom ~0.75 m of the pond (max depth = 2.0 m).

8

(a)

228 (b)

229 (c)

230 231 Figure 7. Contour plots of (a) dissolved oxygen, (b) temperature, and (c) specific conductivity profiles, measured at 232 Pond E at 0.25-m intervals. Vertical lines show times when profiles were made on site visits; linear interpolation is 233 used to fill the depths between measurements and time series between site visits (note especially the large gap 234 from Feb 22 to May 20, 2017). At the bottom of the pond, DO concentration was typically less than 1 mg/L 235 throughout most of the year. 236

9

237 238 Figure 8. Temperature profile time series for Pond E from June 3 – July 6, 2017. Temperature gauge elevations are 239 fixed and distances represent the distance below the uppermost probe rather than the distance below the water 240 surface. The probe at 0 m likely protruded above the water at times, recording air temperature. 241 242 Chemostratification was present throughout much of the year at two of the three monitored ponds (C 243 and E), where persistently high conductivity levels at the pond bottoms (Figure 7c; SI) were attributed to 244 chlorides from road salt making water denser and more resistant to vertical mixing (Novotny et al. 245 2009). This stratification peaked in February and slowly weakened over the open-water season (likely by 246 diffusion and some entrainment by inflows) until October, when the ponds (C and E) were flushed out 247 by an approximately 76-mm storm event after turning over in late September (Figure 7c). By December, 248 the ponds were already beginning to re-stratify. This October – December period without 249 chemostratification is the only time when DO concentrations at the sediment surface were above 0.5 250 mg/L at Pond E. 251 252 Temperature stratification was also present throughout much of the year at all three ponds (Figure 7b; 253 SI), though its impact relative to chemostratification is uncertain. However, at Pond E, chemo- 254 stratification was strong despite relatively isothermal conditions in the water column early in the season 255 (Feb through May). This pattern suggests that chemostratification may be facilitating earlier onset of 256 stratification and extending the period of anoxia beyond when it could be effected by thermo- 257 stratification alone, which is thought to be the dominant stratifying force in summer (McEnroe et al. 258 2013). 259 260 To examine the prevalence of springtime chemostratification, we surveyed 37 stormwater ponds in St. 261 Paul and Roseville, MN (within several miles of the study sites) after ice-out in Spring 2018 and found 262 that 29 of the ponds had elevated conductivity (> 1500 µS/cm, or the conductivity of a solution with 263 roughly 460 mg/L chloride, or twice the chronic standard; see SI spreadsheet “Task 3b Spring 2018 Pond 264 Survey”). Further, among a sub-set of 25 ponds profiled, 17 showed strong stratification, with another 2 265 having elevated conductivity but too shallow to stratify. Presence of chloride in ponds C, D, and E was 266 confirmed from direct measurement. Chloride concentration ranged from 191 – 949 mg/L (see SI). 267 Furthermore, two of the three ponds (C and E) were part of a recent study of road salt transport and 268 accumulation (Herb et al. 2017) that showed high levels of chloride in the ponds over three winters. 269 These ponds had high chloride loading (concentrations in excess of 2000 mg/L), high specific

10

270 conductivity (>3000 µS/cm) in hypolimnia post-winter, and elevated chloride (>100 mg/L) in pond 271 effluent, including during the open-water seasons. 272 273 Conceivably, certain shallow ponds may never stratify without the presence of chloride-driven 274 chemostratification, especially where physical characteristics (e.g., lack of wind sheltering by trees) 275 promote a well-mixed water column. However, the impact of chloride-induced stratification is that the 276 resulting anoxic conditions favorable for ortho-P release from pond sediments were present beyond the 277 thermally-stratified period in summer. The implication is that in addition to affecting aquatic life in lakes 278 and streams, chloride may also be affecting the temperature and oxygen dynamics of certain 279 stormwater ponds. 280 281 P Retention and Seasonal Patterns in P Concentrations at Three Ponds 282 283 Inflow-Outflow P: 284 A comparison of observed inflow and outflow concentrations at the three pond sites for 16-27 events 285 (Figure 9) shows several contrasts in P dynamics among the ponds. First, the two larger ponds (Pond C 286 and Pond E) show a reduction in TP concentration from inlet to outlet (i.e. many points are below the 287 1:1 line in Figure 9). For Pond E (Alameda), this TP decrease is driven primarily by a reduction in PP 288 concentration across the pond, suggesting settling of particulates. For Pond C (William Street), a 289 substantial reduction of TDP concentration across the pond was observed, and along with a 290 corresponding drop in ortho-P (not shown), is likely evidence of the effectiveness of its iron-enhanced 291 sand filter. The smaller pond (Pond D; Church Pond) shows opposite trends to the two larger ponds, 292 with much higher outflow concentrations for all forms of P. It is important to note that this pond rarely 293 discharges water. Thus despite high outflow TP concentrations, exported loads were small relative to 294 inflow (Figure 11). For all ponds, algal biomass (as chlorophyll a; chl-a) in the outflow was higher than in 295 the inflow, but lower than in the water column samples, suggesting that algae and/or duckweed are 296 potential vehicles for P transport from ponds (see SI for chl-a data). 297 298 Water Column P: 299 Time series of water column, hypolimnion, and epilimnion TP and TDP at Ponds C and E show some 300 noteworthy seasonal patterns (Figure 10; data from Pond D are not as informative given its shallow 301 depth, and are shown in the SI). First, TP concentrations were variable but tended to be higher in 302 summer (e.g., Pond C), with a decrease in early October following pond turnover and a large storm 303 event on 10/2/17. In Pond C, TP concentrations rebounded quickly after this point, driven by particulate 304 P (perhaps from settling of senesced algae and duckweed) as TDP remained depressed until spring. A 305 lack of large outflow events during this time (Fall 2017) also probably promoted build-up of P in all three 306 ponds. In Pond E, both TP and TDP in the hypolimnion appeared to build up over winter following 307 turnover, likely reflecting breakdown of senesced vegetation and settling of particulates as ice cover 308 formed. After ice-out in 2018, hypolimnion TP increased as epilimnion TP decreased, suggesting flushing 309 of epilimnion water from a strongly-stratified pond, with corresponding build-up of P near the pond 310 bottom. Given the intense stratification in these ponds, the epilimnion is most likely to be entrained and 311 exported from the pond during events. Comparison of mean inlet, outlet, hypolimnion, and epilimnion 312 TP concentrations for each pond also suggested that inflows were flushing epilimnion rather than 313 hypolimnion water, as outflow TP tended to be closer to epilimnion TP than to hypolimnion or inflow TP 314 (especially for Ponds C and E; see SI). The unexpected strong stratification of the ponds complicates 315 further interpretation of vertical and inlet-outlet differences in specific P fractions. A subsequent study 316 could examine the impact of stratification on pond mass balances and P forms.

11

(a) Pond C 1000

TDP

PP

100 TP Outflow Concentration, ug/L Concentration, Outflow

10 10 100 1000 317 Inflow Concentration, ug/L 200 600 (b) Pond D (c) Pond E 500 150 400

100 300

200 50 100 Outflow Concentration, ug/L Concentration, Outflow Outflow Concentration, ug/L Concentration, Outflow 0 0 0 50 100 150 200 0 100 200 300 400 500 600 318 Inflow Concentration, ug/L Inflow Concentration, ug/L 319 Figure 9. Inflow vs. outflow P concentrations (µg/L) for all sampled storm and snowmelt events at the three 320 monitored ponds: (a) Pond C [William St Pond] (n = 27 events), (b) Pond D [Church Pond] (n = 16), and (c) Pond E 321 [Alameda Pond] (n = 25). The dashed line is 1:1, which would indicate that inflow and outflow concentrations were 322 equal. Note the log-log scale in plot (a).

12

1400 (a) Pond C, TP TP, wc TP, epi TP, hypo 1200 1000 800 600 400 200 Water Column P, P, ug/L Column Water 0

323 400 (b) Pond C, TDP TDP, wc TDP, epi TDP, hypo 300

200

100 Water Column P, P, ug/L Column Water 0

324 800 (c) Pond E, TP TP, wc TP, epi TP, hypo

600

400

200 Water Column P, P, ug/L Column Water 0

325 800 (d) Pond E, TDP TDP, wc TDP, epi TDP, hypo

600

400

200

Water Column P, P, ug/L Column Water 0

326 327 Figure 10. Time series of P concentrations in water column samples collected at two of the monitored ponds, Pond 328 C (plot a: TP, plot b: TDP) and Pond E (plot c: TP, plot d: TDP). Samples collected prior to Aug 2017 were integrated 329 water column (‘wc’) samples; those collected after were epilimnion (‘epi’) and hypolimnion (‘hypo’) samples.

13

330 Mass Balance: 331 Annual P loading and retention, defined as the reduction of incoming P loading relative to the Inlet load 332 (i.e. [Inlet Load – Outlet Load]/Inlet Load), is summarized for each pond in Figure 11. Note that while 333 load calculations were completed for the entire data record at each site, results are shown here only for 334 the most recent annual period (June 1, 2017 – May 31, 2018), with the exception of Pond E (Alameda), 335 for which the data record is 11 months (July 1, 2017 – May 31, 2018). All ponds showed retention of TP 336 overall, with slightly better than 50% retention at Pond C (WSP) and Pond E (Alameda). Pond D (Church 337 Pond) very rarely discharged, as evidenced by its extremely high volume retention (~95%; Figure 11d), 338 and combined with low concentrations in influent (TP EMC ~0.12 mg/L; Table 1), provided the greatest 339 reduction of the three ponds for all forms of P, including 78% retention for TP. TDP reduction was high 340 for Pond C (69%), as expected due to the IESF (iron-enhanced sand filter) installed at the site. An 341 explanation for the higher particulate reduction observed for Pond E than for Pond C is unclear, 342 especially due to the IESF at Pond C, though a skimmer wall at Pond E’s outlet may have helped prevent 343 floating particulates such as duckweed and algae from leaving the pond.

(a) Pond C, P Loading (b) Pond D, P Loading 0.8 5 4 0.6 3 56% 0.4 2 46% P Load, kg P Load, kg 78% 69% 0.2 73% 1 86% 5.29 2.33 3.22 1.74 1.94 0.60 0.74 0.16 0.45 0.12 0.28 0.04 0 0.0 TP PP TDP TP PP TDP William St Inlet William St Outlet RCCP Inlet RCCP Outlet 344 (c) Pond E, P Loading (d) Runoff Volumes 20 60

15 27% 40 53% 10

P Load, kg 62% 38% 20 7% 5 95% 16.4 7.68 10.2 3.85 6.16 3.84 ML Volume, Runoff 13.7 12.8 1.12 0.06 62.3 45.4 0 0 TP PP TDP Pond C Pond D Pond E Alameda Inlet Alameda Outlet Inlet Outlet 345 346 Figure 11. Annual inlet and outlet loads (kg) of TP, PP, and TDP and water volumes (megaliters) for the three pond 347 sites. Annual period is June 1, 2017 – May 31, 2018 for Ponds C and D; for Pond E, it is an 11-month period (July 1, 348 2017 – May 31, 2018) due to a later start to monitoring at the site. Percentages shown are retention ([Outlet – 349 Inlet] normalized by Inlet), i.e. the portion of incoming loads retained by the pond. Loads reflect modeled 350 concentrations and hydrology (see SI, Section III). 351

14

352 Phosphorus Retention: 353 The time series of TP retention at Ponds C and E (Figure 12; see SI for TDP retention time series), which 354 were constructed using the entire data record at each site, showed consistent retention through much 355 of the season at both ponds, including during the generally stormier summer season. Some net export of 356 TP occurred during late snowmelt and ice-out (Apr 2018) at both sites, and during January / February 357 snowmelt in 2017 at Pond C, when the pond froze at a higher water elevation the previous fall than 358 during winter 2017-18. In addition, during some very wet periods (mid-May and June 2017 at Pond C, 359 early October 2018 at Pond E), the ponds were net sources of TP to downstream. While retention time 360 series were not constructed for Pond D due to the rarity of discharge at the site, it tended to also export 361 P during wetter periods or when the pond was full of recent snowmelt (e.g., April 2018). 362 Hydrologic Controls on Pond P Retention: 363 Together, the retention and concentration results (Figures 9, 11) suggest a strong role of hydrology on 364 the performance of ponds for P retention. Pond D, for example, had much higher P concentrations in 365 outflow than in inflow (Figure 9d), yet it had nearly 80% retention of TP over an annual period due to 366 the high loss of water from the pond (likely due to infiltration or evapotranspiration). Pond C had a 367 nearly neutral water balance, with inflows roughly equal to outflows (despite evaporation rates similar 368 to Pond E; see Table 3), suggesting some groundwater inputs to the pond and tiles in the IESF. Yet both 369 this pond and Pond E had high overall retention, exporting P only during wetter periods (snowmelt and 370 high rainfall). Accordingly, antecedent pond water storage capacity (approximated by mean water level 371 over the three days prior to a runoff event) was strongly related to retention percentage for an event 372 (Figure 13). In other words, across all three ponds, net TP export (retention percentage < 0) was far 373 likelier to occur if the pond was approaching the limit of its treatment volume at the onset of a storm or 374 melt event.

15

(a) TP Retention, Pond C 0.25 0.5 Event Retention Cmltv Retention

0.0 0.00

-0.5 -0.25 -1.0

-1.5 -0.50

-2.0

-0.75 (KG) RETENTION EVENT -2.5 CUMULATIVE RETENTION (KG) CUMULATIVE -3.0 -1.00 1/1/17 2/1/17 3/1/17 4/1/17 5/1/17 6/1/17 7/1/17 8/1/17 9/1/17 1/1/18 2/1/18 3/1/18 4/1/18 5/1/18 6/1/18 375 11/1/16 12/1/16 10/1/17 11/1/17 12/1/17 0.2 0.05 (b) TP Retention, Pond D Event Retention Cmltv 0.1 0.0 0.00 -0.1 -0.2 -0.05 -0.3 -0.4 -0.10 -0.5 -0.6 -0.15 EVENT RETENTION (KG) RETENTION EVENT -0.7 CUMULATIVE RETENTION (KG) CUMULATIVE -0.8 -0.20 1/1/17 2/1/17 3/1/17 4/1/17 5/1/17 6/1/17 7/1/17 8/1/17 9/1/17 1/1/18 2/1/18 3/1/18 4/1/18 5/1/18 6/1/18 376 10/1/17 11/1/17 12/1/17 2.0 (c) TP Retention, Pond E Event Retention Cmltv Retention 0.25 1.0 0.0 0.00 -1.0 -2.0 -0.25 -3.0 -4.0 -0.50 -5.0

-6.0 -0.75 (KG) RETENTION EVENT

CUMULATIVE RETENTION (KG) CUMULATIVE -7.0 -8.0 -1.00

377 378 Figure 12. Time series of event and cumulative TP retention at (a) Pond C (William Street Pond) from Nov 2016 379 through May 2018, (b) Pond D (Church Pond) from Jan 2017 – May 2018, and at (c) Pond E (Alameda Pond) from 380 July 2017 – May 2018. Time periods shown are the entire duration of monitoring at the ponds. Loads reflect both 381 sampled and modeled concentrations and hydrology (see SI, Section III), and retention is calculated relative to Inlet 382 loading, such that negative indicates storage of P in pond. See SI for similar plots of TDP retention. 16

100%

50% y = -0.0029x2 - 0.0979x + 0.1824 R² = 0.4426 0%

-50%

TP Retention Retention % TP Pond C

-100% Pond D

Pond E -150% -25 -20 -15 -10 -5 0 5 10 383 Mean 3-Day Antecedent Pond Level, cm 384 Figure 13. TP Retention (%) vs. antecedent pond level (mean over previous 3 days, in cm) for all sampled events at 385 the three ponds. Pond level datum (0 cm) is the level above which discharge occurs for each pond (i.e. negative 386 level indicates storage capacity). Retention is calculated as (Inlet – Outlet)/Inlet, such that positive values indicate 387 retention by the pond for the event. Storm size caused some of the residual variation evident in this plot. 388 Finally, pond water level dynamics can illustrate some differences in pond function and water budgets. 389 In particular, water level time series for Ponds D and E (Figure 14; Pond C water level data were 390 provided by CRWD and are not published here) show the rarity of export events and rapid drawdown of 391 water level in Pond D, especially relative to the more frequent discharge and longer receding limbs of 392 discharge from Pond E. All three ponds were well sheltered by mature trees, but the smallest pond 393 (Pond D) may have been losing substantial water to evapotranspiration from surrounding trees. 394 Approximating drawdown (evaporation) rates from all three ponds during dry periods in 2017 may 395 support this hypothesis (Table 4); Pond D had the highest overall drawdown rates (~1 cm/day during 396 late spring and early summer), while all three ponds showed seasonality of loss rates consistent with 397 evapotranspiration patterns, with generally highest rates in late spring-summer (May, June, July), and 398 lower rates in fall (Sep, Nov). 399 400 Table 4. Rates of decrease in water level (cm/day) in the three monitored ponds during four dry intervals during 401 the 2017 monitoring season.

Period Pond C Pond D Pond E 5/23/17 - 6/11/17 0.75 1.01 0.65 7/1/17 - 7/17/17 0.92 1.02 0.6 9/8/17 - 9/16/17 0.7 0.75 0.64 11/6/17 - 11/20/17 n/a 0.45 0.25 402

17

403

404 Figure 14. Time series of water level in (a) Pond C (William St Pond), (b) Pond D (Church Pond) and in (b) Pond E 405 (Alameda Pond) from March 2017 – May 2018. In both ponds, the reference elevation is the point above which 406 discharge occurs from the ponds. Level data for Pond C were provided by Capitol Region Watershed District. 407 Precipitation (in.) is daily totals from the Minneapolis-St. Paul International Airport.

18

408 Discussion for Objective 3b 409 The primary results of the phosphorus mass balance and field-monitoring task are highlighted here: 410 • Pond water column P concentrations were often highest during summer months, particularly in 411 early summer, with hypolimnion concentrations usually higher than epilimnion or inflow 412 concentrations. However, outflow concentrations were generally lower than inflow or 413 hypolimnion concentrations (except at Pond D), suggesting that inflows were not entraining 414 much hypolimnetic water in strongly stratified ponds. 415 • Decreasing water column P concentrations over the summer may indicate biological uptake 416 (algae, duckweed) and particulate settling from inflows, while a rapid increase from early 417 October onward may have been caused by turnover and P release from senescent algae and 418 duckweed. 419 • Despite their age (pond D, E) and near neutral water balance (pond C), the three ponds 420 examined performed as expected with respect to P retention, with annual removals of 53%, 421 78%, and 56% for Ponds C, D, and E, respectively. Wet ponds generally have an expected 422 performance of 52% phosphorus removal (Weiss et al. 2007). 423 • Hydrologic function is crucial for P removal performance of ponds. In the case of Pond D, the 424 high P retention was driven by extremely high hydrologic retention (>90%), while for all three 425 ponds, P export tended to only occur during wetter antecedent conditions (snowmelt, ice-out, 426 or heavy rain), when the ponds had reduced volume storage capacity (approximated by mean 427 water level prior to runoff events). While perhaps an unsurprising result, this suggests that 428 increasing volume treatment capacity of ponds (e.g. through increasing storage or promoting 429 evapotranspiration) may provide a means to improve P removal. Monitoring of pond level could 430 be a helpful and cost effective method to evaluate P retention in urban ponds. 431 • Accounting for pond water balance is essential for understanding pond functioning for P 432 retention or export. Concentrations may be high in ponds, but if water is infiltrating or 433 evaporating, retention may be higher than concentration-based comparisons suggest. The 434 amount of P lost to groundwater via infiltration is at present unknown in the study ponds. 435 • Although the three monitored ponds had anaerobic dissolved oxygen conditions near the 436 sediment bed throughout the growing season (and throughout most of the year at two of the 437 ponds, C and E), the ponds did not become net exporters of phosphorus. This observation also 438 corresponds to the lower release observations of the column studies for ponds C and E. Further 439 study is necessary to understand the role of redox vs. decomposition mediating P release in 440 storm water ponds. 441 • Strong and persistent density stratification of ponds due to salinity from road salt applications 442 can cause oxygen depletion and reduction in boundary layer mixing near the sediment bed. This 443 effect may cause earlier onset of stratification (and resulting anoxia), which persists over the 444 summer and into fall turnover. 445 • This persistent stratification (and anoxia) has potential consequences for P fluxes from 446 sediments and for pond hydraulic function. For example, temperature and conductivity profile 447 data and similarity of epilimnion and outflow TP suggested that stormwater inflows did not mix 448 with the lower water column due to the intense stratification of the ponds, and instead were 449 flushing epilimnetic water. However, this hydraulic disconnection of the hypolimnion may 450 prevent P released from sediments from being exported from the pond. Future work would be 451 needed to test this hypothesis. 19

452 Conclusions 453 There is growing concern that aging stormwater detention ponds, originally constructed for volume and 454 sediment control, may become sources of phosphorus to downstream waters as they accumulate 455 phosphorus and lose storage capacity from accumulated sediment and organic matter. Locally, 456 phosphorus concentration in roughly one-third of 98 urban stormwater ponds sampled by the RPBCWD 457 exceeded the 95% CI of the expected TP in stormwater in a dataset compiled for several TCMA 458 watersheds (Janke et al. 2017), indicating substantially elevated TP in ponds, with potential to export to 459 downstream waters. In this study, we investigated the theory that sediments are releasing ortho-P to 460 the water column, a process known as internal loading. Phosphorus release from the sediments of 461 ponds may be facilitated by low DO concentration. In the three ponds monitored in this study, and 462 potentially in many other detention ponds in the TCMA, stratification is present through most of the 463 summer, a condition that is potentially exacerbated by accumulation of road salt in winter and spring 464 snowmelt, leading to subsequent chemostratification. This results in a low DO concentration that may 465 promote P release and internal loading. Sediment-water phosphorus flux was determined for five ponds 466 in the column study: three ponds with the highest P fluxes also had the highest microbial activity (high 467 sediment oxygen demand) and largest amounts of mobile sediment P (labile organic P + redox P). The 468 other two ponds had lower P release due to lower mobile sediment P combined with lower microbial 469 activity in the sediment, as indicated by lower sediment oxygen demand. These results suggest that low 470 oxygen alone does not lead to P release; presence of bioavailable P and organic matter also play a role. 471 472 Mass balance measurements of phosphorus in three ponds in Roseville, MN, showed annual retention 473 of P of 53% - 78%. Net export of P was observed at the event scale during snowmelt and for some large 474 events, when the ponds had very little volume storage capacity. Anoxic release of P was observed from 475 sediments collected from these ponds (Task 3a). Despite these loss mechanisms, all three ponds 476 exhibited substantial net retention. Reduction in P loading was due to effects of both water loss from 477 ponds which decreased outflow volumes and, in some cases, decreases in water column P 478 concentrations relative to influents. Our results suggest that significant P release rates from sediments, 479 as observed in a laboratory setting, are not alone an indicator of poor pond performance for P removal. 480 481 Further work remains to understand how results of our study compare to the large number of 482 stormwater ponds in the region. Our study ponds had relatively low rates of P loading (i.e. influent EMC 483 at or below the 95% CI of TP, 0.38 mg/L, for TCMA stormwater) and lower water column TP relative to 484 many other ponds in the Twin Cities (e.g., in the analyses of data from RPBCWD ponds), suggesting they 485 had sufficient capacity to detain and remove P. The ponds we studied were stratified for much of the 486 year, and it is not yet know how this unexpectedly strong stratification influences P retention at annual 487 scales. It is possible that more frequent mixing may provide a mechanism to draw sediment-released P 488 into the upper water column, where it could be exported by outflow from storm events; as the pond re- 489 stratified, more P would be released from anoxic sediments (perhaps on the order of days; Fig. 5). 490 491 Ponds are present in the TCMA with far higher water column TP than observed in our monitored ponds 492 (RPBCWD 2014). The cause of these high P levels is not well understood, and similarly, the potential for 493 this P to be retained or exported is not well known. Lower rates of annual retention may be expected for 494 ponds with higher P loads and water column concentrations, such as may be found in watersheds with 495 higher road density, larger watershed-to-pond ratios, and with higher tree canopy cover. P retention is 496 influenced by a wide range of factors such as dissolved oxygen dynamics, duration (and frequency) of 497 stratification and mixing events, wind sheltering, storage capacity, and seasonality. Further 498 investigations are clearly warranted to understand the role of these factors on P retention and release in 499 ponds, towards better prediction of pond performance for P removal.

20

500 Acknowledgements 501 We thank many individuals and groups for their assistance with this project. Laboratory analyses were 502 facilitated by Peter Corkery and Rikita Patel (SAFL), and by Shelly Rorer, Tessa Belo, and Claire Jaeger 503 Mountain (EEB). Field sampling was assisted by Krysta Garayt, Claire Jaeger Mountain, and Peter 504 Corkery. Field work was supported by LacCore. V. Taguchi was supported by a National Science 505 Foundation Graduate Fellowship (00039202). 506 507 Data provided by Riley Purgatory Bluff Creek Watershed District (RPBCWD) was collected by David 508 Austin (PI) and Roger Scharf (co-PI) of CH2M (now Jacobs) in a project funded by RPBCWD from 2010 - 509 2012. The Cities of Bloomington, Chanhassen, Eden Prairie and Minnetonka helped monitor the 510 stormwater ponds in RPBCWD. Capitol Region Watershed District provided water level data for Ponds C 511 and E. Research performed using facilities and resources of the Department of Civil, Environmental, and 512 Geo- Engineering, the Department of Ecology, Evolution, and Behavior, and the St. Anthony Falls 513 Laboratory of the University of Minnesota – Twin Cities. 514 515 Project funding was provided through the Clean Water Fund (from the Clean Water, Land and Legacy 516 Amendment), administered by the Minnesota Pollution Control Agency (MPCA). The views expressed 517 within this document do not necessarily reflect the views or policy of the MPCA. 518 519 Links to Other Resources 520 http://stormwater.dl.umn.edu/updates 521 522 References 523 Engstrom, D. 2010. Sediment phosphorus extraction procedure high sample throughput. Modified by 524 Robert Dietz and Michelle Natarajan (2015). 525 Gächter, R., and B. Wehrli. 1998. Ten Years of Artificial Mixing and Oxygenation: No Effect on the 526 Internal Phosphorus Loading of Two Eutrophic Lakes. Environ. Science & Technology 32:3659-3665. 527 Herb, W.R., Janke, B., Stefan, H.G. 2017. Study of De-icing Salt Accumulation and Transport Through a 528 Watershed. William Herb, Minnesota Dept. of Transportation, Research Project Final Report 2017- 529 50. 83 pp., Appendix B. 530 https://www.dot.state.mn.us/research/reports/2017/201750.pdf 531 Janke, B. D., J. C. Finlay, and S. E. Hobbie. 2017. Trees and Streets as Drivers of Urban Stormwater 532 Nutrient Pollution. Environ. Science & Technology 51:9569-9579. 533 Kayhanian, M., Singh, A., and Meyer, S. 2002. Impact of non-detects in water quality data on estimation 534 of constituent mass loading. Water Science and Technology, 45(9), 219-225. 535 McEnroe, N. A., Buttle, J. M., Marsalek, J., Pick, F. R., Xenopoulos, M. A., and Frost, P. C. (2013). Thermal 536 and chemical stratification of urban ponds: Are they ‘completely mixed reactors’? Urban 537 Ecosystems, 16(2): 327-339. 538 Natarajan, P., Gulliver, J. S., and Arnold, W. A. 2017. Internal Phosphorus Load Reduction with Iron 539 Filings. US Environmental Protection Agency (EPA) Section 319 Program and Minnesota Pollution 540 Control Agency (MPCA). 541 Novotny, E. V., Sander, A. R., Mohseni, O., and Stefan, H. G. 2009. Chloride ion transport and mass 542 balance in a metropolitan area using road salt. Water Resources Research, 45(12). 543 Novotny, E.V., Murphy, D. and Stefan, H.G. 2008. Increase of urban lake salinity by road de-icing salt, 544 Science of the Total Environment 406, pp 131-144, doi: 10.1016/j/scitotnev.2008.07.037. 545 Novotny, E.V. and Stefan, H.G. 2012. Road salt impact on and water quality, J. 546 Hydraulic Engineering 138 (12): 169-180. 547 Olsen, T. A. 2017. Phosphorus dynamics in stormwater ponds (Master’s Thesis). University of Minnesota

21

548 – Twin Cities, Minneapolis, Minnesota. 549 Pitt, R., Maestre, A., and Morquecho, R. 2008. National Stormwater Quality Database (NSQD) Version 3 550 Spreadsheet. US EPA National Pollutant Discharge Elimination System (NPDES). 551 Riley Purgatory Bluff Creek Watershed District (RPBCWD). (2014). Stormwater Pond Project 2013 552 Report. 553 Søndergaard, M., E. Jeppesen, T. L. Lauridsen, C. Skov, E. H. Van Nes, R. Roijackers, E. Lammens, and R. 554 O. B. Portielje. 2007. Lake restoration: successes, failures and long-term effects. Journal of Applied 555 Ecology 44:1095-1105. 556 Song, K., and A. J. Burgin. 2017. Perpetual Phosphorus Cycling: Eutrophication Amplifies Biological 557 Control on Internal Phosphorus Loading in Agricultural Reservoirs. Ecosystems 20:1483-1493. 558 Van Buren, M. A., Watt, W. E., and Marsalek, J. 1997. Application of the log-normal and normal 559 distributions to stormwater quality parameters. Water Research, 31(1), 95-104. 560 Weiss, P.T., Erickson, A.J. and Gulliver, J.S. /2007. Cost and pollutant removal of storm-water treatment 561 practices. Journal of Water Resources Planning and Management, 133(3), 218-229.

22