The study was completed at Sleeping Bear Dunes National Lakeshore in Loon Lake (44° 42′ 30.24″ N., 86° 07′ 48″ W.), Benzie County, Mich. (fig. 1). Loon Lake is 38.4 hectares in surface area with a maximum depth of 20.1 meters (m) and mean depth of 9.0 m (Stockwell and Gannon, 1975). The Platte River enters Loon Lake on the eastern shoreline and outlets at the northernmost extension of the lake (Stockwell and Gannon, 1975). The lake has a well-established population of zebra mussels and few extant native mussels (Nichols and others, 2003; D. Woolnough, Central Michigan University, unpub. data, March 2024).

In June 2024, divers completed reconnaissance surveys throughout Loon Lake to assess zebra mussel density and to locate suitable study sites. The substrate in the littoral zone is sand and few cobbles and an overlay of marl. The lakebed drops rapidly to greater than (>) 6 m in depth within ~30 m from the shore. The substrate at depths >3 m is primarily silt and fine organic material. Most of the available substrate for zebra mussel colonization is woody debris. Divers reported that zebra mussel populations were sparsely distributed in clumps in association with the available hard substrate. We avoided the northern half of the lake between the inflow and outflow of the Platte River because it is heavily used by recreational boaters. Sites were selected south of the Platte River inflow along the eastern shoreline, between ~1 and 3 m in depth, and containing some mussel-colonized substrate. Three sites were selected (sample size = 3 benthic mat, 3 carbon dioxide mat, 3 reference) for replication of reference, benthic mat, and carbon dioxide mat treatments (fig. 1). Our experimental unit was the site. Within a site, three 4-m x 4-m plots were delineated for each treatment and a reference. Plots within each site were separated by ~8–16 m; water depth in the plots ranged from 0.5 to 2.0 m. Small floats marked the corners of each plot, and the midpoint of each carbon dioxide plot was recorded from a portable Global Positioning System (table 1).

The material used for the benthic and carbon dioxide infusion mats was woven polyethylene with a vinyl coating that was 14 mil thick, acid resistant, and nonreactive. Each mat was 4.25 m x 4.25 m, which included a 0.25-m apron for placement of anchoring weights. The benthic mat was rolled onto a 6-m (length) x 15-centimeter (cm; diameter) polyvinyl chloride (PVC) pipe for deployment. The mat was carried from shore and unrolled over the plot by divers. The perimeter of the mat was anchored by sandbags (~10 bags per mat) and cinder blocks to prevent influx of freshwater under the mat.

The carbon dioxide infusion mat was supported by a fiberglass and PVC frame (fig. 2) consisting of a center hub and four arms extending to each corner of the mat. The top of the hub had a connecting port for inflowing carbon dioxide-infused water, and the lower part of the hub had a base (strainer grate, 0.5-m height x 15-cm diameter) to hold the center of the mat above the substrate. Four arms (3 m in length) made of fiberglass rods surrounded by 19-millimeter diameter PVC pipe extended from the hub to each corner of the mat, creating a frame for attachment of the mat. The arms were connected to the hub by 180-degree hinges that allowed unfolding the mat for deployment and collapsing the arms for retrieval. Vinyl loops (~5 cm x 15 cm) were glued to the top of the mat to secure the arms of the frame. When fully deployed, the mat covered 16 square meters (m²), not including the apron (0.25-m width) around the perimeter, which was used for placement of sandbags to anchor and seal the mat edge. Additionally, a chain (15.8 mm width) was placed on the apron around the perimeter of the mat to reduce loss of carbon dioxide from the mat.

Carbon dioxide was applied as Carbon Dioxide–Carp (U.S. Environmental Protection Agency registration no. 6704–95) under the regulatory authority of Michigan Department of Environmental Quality Aquatic Nuisance Control (Certificate of Coverage no. MIG032005) and U.S. Environmental Protection Agency Experimental Use Permit exemption for application outside the scope of the current label. Hereafter, we refer to Carbon Dioxide–Carp as “carbon dioxide.” The carbon dioxide infusion and delivery system consisted of two diaphragm pumps (Seaflo model SVDP1, 20-liter per minute flow rate) to deliver lake water, a 30-pound cylinder of carbon dioxide (food-grade compressed carbon dioxide gas), gas regulator and flow meters, two downward-flow gas infusion chambers, and one outflow hose. The infusion and delivery system was mounted to brackets on a 5.5-m (length) aluminum boat (fig. 3), and the proximate end of the delivery hose was secured to either a navigation buoy or a cylinder block adjacent to each carbon dioxide plot. Water was pumped from the lake into the infusion chambers and a regulator and flow meter were used to adjust delivery rate of carbon dioxide to a large air stone within the infusion chambers. The outflowing water from the two chambers was connected to a single hose for delivery of carbon dioxide infused water to the center of the mat.

The purpose of the pretreatment survey was to estimate the percentage of live and freshly dead mussels in the plots before treatment; however, the reconnaissance survey of the study sites determined that hard substrate was scarce in the plots and zebra mussel density was highly clustered on woody debris. Therefore, we limited the pretreatment survey to the reference plots and completed biased quadrat surveys by actively searching for colonized substrate (such as woody debris, rocks, and plastic debris). Divers placed a 1-m x 1-m quadrat in areas with suitable substrate, collected attached zebra mussels and (or) the substrate from within the quadrat, placed the sample in a mesh dive bag, and returned the bag to the surface for enumeration. Four quadrats were collected from each reference plot (total area surveyed was 4 m² per plot). Zebra mussels were assessed as live or freshly dead. Shell was not included in estimates of mussel density. Freshly dead mussels were defined as valves agape with soft tissue intact or intact valves and shiny shell nacre. Older dead shell was defined as single or intact valves with chalky nacre and darkened or eroded periostracum. After completion of pretreatment surveys, divers collected colonized substrates from the area outside of the plots and placed them into plots to augment zebra mussel density.

The macroinvertebrate community at each site was sampled by collecting triplicate petite Ponar (grab) samples (0.046-m² sampling area) in the reference plots. We did not collect Ponar samples inside the treatment plots to minimize disturbance. Samples were washed through a 500-micron screen in the field; preserved in 70 percent ethanol; and transferred to the laboratory for sorting, identification, and enumeration. Macroinvertebrates were identified to family level, except for oligochaetes and chironomids.

Zebra mussels were collected from the lake in areas outside of the study sites for placement as sentinel mussels and for a standard assessment of mortality in each plot. Mussels were gently scraped with a scalpel or paint scraper from docks, rocks, and woody debris 1 to 2 days before treatment began and transferred to a cooler of lake water. Within several hours of collection, we assessed each mussel for viability by applying gentle pressure to open the valves; mussels that resisted opening were deemed suitable for testing. Mussels were indiscriminately distributed in groups of 28–32, caged within a nylon mesh bag (40-cm length) held open with a PVC ring (7.6-cm diameter x 2.5-cm height; fig. 4A), and transferred to totes filled with lake water for overnight holding. The next day, each cage was indiscriminately removed from the tote, placed onto a plastic tray (15 cm x 25 cm) along with a brick, and secured inside a second mesh bag (60-cm length). The tray prevented the mussel cages from sinking into the substrate and becoming covered with sediment, and the brick prevented the tray from overturning (fig. 4B). Within each plot, 12 cages of sentinel mussels were distributed in a similar pattern: 1 at each corner, 4 near the midpoint, and 4 halfway between the midpoint and each corner (fig. 5). A total of 360 mussels were placed in each plot.

Reference plots were located at a distance (8–16 m) from the carbon dioxide mat to prevent the potential effect of carbon dioxide leakage and minimize disturbance in the plot. On the day of mat deployment, the plot corners were marked by small floats, and caged sentinel zebra mussels were placed into the plot by divers, as described in the previous section. A miniDO₂T (PME; Vista, California) dissolved oxygen (DO) and temperature logger was placed in the center of the plot for continuous measurement of these variables.

Benthic mat plots were between the carbon dioxide plot and reference plot but at a distance of 8–16 m to prevent the potential effect of carbon dioxide leakage and minimize between-plot disturbance. On the day of mat deployment, the plot corners were marked by small floats, and cages of zebra mussels were placed into the plot by divers, as described in the previous section. A miniDO₂T DO and temperature logger was placed in the center of the plot for continuous measurement of these variables. The benthic mat was rolled out and over the plot by divers and anchored in place with sandbags.

We used pH as a proxy for carbon dioxide concentration. Before treatment began, a pH-carbon dioxide concentration curve was developed with lake water. A sample of lake water was infused with carbon dioxide to produce ~50, 100, 150, 200, 250, and 300 mg/L carbon dioxide (fig. 6). The concentration of dissolved carbon dioxide at each pH level was determined by titration in triplicate with 0.363 normal (N) or 3.63 N sodium hydroxide to a pH of 8.3. Based on the curve, the target pH to produce 200 mg/L carbon dioxide was estimated at 6.13 (95-percent CI of 5.90–6.30).

On the day of carbon dioxide mat deployment, a YSI 6920 V2 sonde was placed ~1 m from the plot midpoint for real-time, continuous measurement of pH (as a surrogate for carbon dioxide concentration), temperature, and DO. The proximal end of the sonde cable was secured either to a navigation buoy or a cinder block on shore. Divers moved the carbon dioxide mat from the boat or shore and placed it over the plot. The corners were anchored with sandbags, and the perimeter was secured with additional sandbags (~10 sandbags per mat) and chain. The midpoint of the mat was elevated ~0.5 m above the substrate by the central hub to maintain adequate water volume for carbon dioxide infusion.

Carbon dioxide was intermittently infused under the mats for 120 hours (5 days) during August 20–24, 2024, in plot 1 and during August 21–25, 2024, in plots 2 and 3. The pH under the carbon dioxide mat was monitored continuously during infusion, and when the target pH (6.13 plus or minus 0.22) was reached, water injection under the mat was stopped, and the hose and sonde cable were disconnected and secured to the navigation buoy or cinder block on shore. Every 2–4 hours between dawn and dusk, we reconnected to the sonde and measured pH and DO. Additional “bump” treatments of carbon dioxide were applied for pH >6.20 to maintain the concentration near the target.

Water chemistry (alkalinity, hardness) was measured on grab samples collected at two time points, during and after treatment, and two locations during the study. One sample was collected near the site 1 benthic mat plot and another between sites 2 and 3. Total hardness and alkalinity were analyzed in triplicate using the ethylenediaminetetraacetic acid titrimetric method (method 2340C; American Public Health Association, 2018) and by titrating to an endpoint of pH 4.5 (method 2320B; American Public Health Association, 2018), respectively.

Water quality (DO, pH, temperature, and specific conductance) was measured daily at each plot. We used a Van Dorn sampler to collect water ~0.3 m above the bottom. Measurements were made immediately after collection using a water-quality meter equipped with luminescent optical DO, pH, and specific conductance probes (models HQ40d, LDO 10101, PHC 70501, and CDC40101; Hach Company, Loveland, Colorado); temperature was measured with a digital thermometer (Thermapen model Mk4; ThermoWorks Company, American Fork, Utah).

At 5 days (120 hours), the sandbags were removed from the perimeter of the benthic mat, and divers floated the mat to the surface and moved it away from the plot with minimal disturbance of the substrate. In carbon dioxide treatment plots, the perimeter of the mat was lifted to allow gradual influx of freshwater and dilution of the carbon dioxide concentration. This was done to minimize the discharge of carbon dioxide to the surrounding water when the mat was removed. We monitored dissipation of the carbon dioxide by measuring pH every 5–10 minutes at the edge of the mat, middepth, and 3 m from the edge. We used the sonde pH measurement to determine when carbon dioxide under the mat was equal to ambient levels (outside of the mat). The next day, we lifted the mat to the surface and moved it away from the plot with minimal disturbance of the substrate. In all plots the miniDO₂T sensor was retrieved, logging was stopped, and the data were downloaded. Caged sentinel zebra mussels were retrieved by divers and transferred to the boat for immediate assessment of mortality.

After removal of benthic mats and carbon dioxide-infused mats, we assessed mortality of caged zebra mussels and the percentage of live resident zebra mussels. Because of the limited density of resident zebra mussels in the plots, divers searched the entirety of each 4-m x 4-m plot for zebra mussels after mat removal. Colonized debris and hard substrate were collected and placed into a dive bag, transferred to shore, and enumerated for live and freshly dead zebra mussels. Divers retrieved and transferred caged zebra mussels to shore for assessment. Mussels were assessed as dead or alive based on the criteria that were previously described. Live and freshly dead mussels were retained for shell length measurement. We also collected macroinvertebrate samples and measured water quality in each plot following the methods described previously in “Pretreatment Assessment of Zebra Mussel Density and Macroinvertebrate Communities.”

We measured shell length of a subsample of sentinel mussels from cages (number of measurements [n] =100 mussels per site) and resident mussels from posttreatment surveys, using a digital caliper (nearest 0.1 millimeter). The number of mussels collected in surveys varied among sites, ranging from 16 (site 2) to >150 (site 1); therefore, we measured 100 mussels from site 1 and all mussels collected from sites 2 and 3.

Data summary and analysis include the following: (1) water hardness and alkalinity, (2) water quality (DO, pH, temperature, and specific conductance) during and after the treatment period, (3) the density and percentage of live resident zebra mussels within the reference plot before treatment, (4) the density and percentage of live resident zebra mussels in all plots 1 day after treatment, (5) percentage of mortality of caged sentinel zebra mussels in all plots, (6) length of resident zebra mussels (collected in plot surveys) and caged sentinel zebra mussels, (7) pH and estimated carbon dioxide concentration during the exposure period, (8) macroinvertebrate diversity and abundance in reference plots before treatment, and (9) macroinvertebrate diversity and abundance in all plots 1 day after treatment. The experimental unit in all analyses was the individual plots. Significance was defined as α=0.05. We used R (ver. 4.5.1; R Core Team, 2025) for confirmatory data analysis, and key packages included vegan (ver. 2.7–1; Oksanen and others, 2025) for multivariate statistics, lme4 (ver. 1.1.37; Bates and others, 2015) and lmerTest (ver. 3.1.3; Kuznetsova and others, 2017) for mixed-effect models, and ggplot2 (ver. 3.5.2; Wickham, 2016) for plotting. Analysis of variance was used to test for treatment and time (pre- and posttreatment) effects on macroinvertebrate abundance and diversity. We used distance-based redundancy analysis with Manhattan distance to examine overall changes in macroinvertebrates by treatment and time (Legendre and Legendre, 2012). We used a generalized linear model with binomial error family and logit link function (also known as a logistic regression) to estimate the effect of treatments on zebra mussel survival for caged sentinel mussels (Bolker, 2008). A generalized linear mixed-effect model with binomial error family and logit link function was used to estimate the effect of treatments and exposure time (pre- and posttreatment) on resident zebra mussel survival collected in surveys with site treated as a random effect (Bolker, 2008). We used a linear model to estimate the association of mussel shell length versus site, group, and the interaction of site and group (Bolker 2008). We defined the association as α=0.05 for the type I error rate. The software release for data analysis is available at Waller and others (2026).

Measurements of water quality before treatment in the study plots averaged a pH level of 8.36 (standard deviation [SD] 0.12), DO of 9.00 mg/L (SD 0.14), and conductivity of 423.3 (SD 124.5; n=3). Water chemistry from site 1 (n=6) and composite samples from sites 2 and 3 (n=6) had mean alkalinity of 142.8 mg/L as calcium carbonate (SD 1.99) and mean hardness of 143.8 mg/L as calcium carbonate (SD 5.26). Daily water quality was measured in reference plots and adjacent to the treatment plots (table 2) with DO >9.0.0 mg/L and pH >8.00.

During treatment, mean pH and DO values were similar at all sampling points and did not differ from the benthic mat and reference plots (table 3). During additional bump carbon dioxide treatments, we monitored pH at the edge of the carbon dioxide mat and detected some leakage of carbon dioxide on three occasions. All occurrences of pH less than [<] 7.0 were at site 3 near the edge of the mat (6.47 on August 23, 2024; 6.98 on August 25, 2024; and 6.84 to 6.94 on August 26, 2024). We immediately added sandbags to the area of the mat where the leak was detected and monitored pH until we confirmed that the leak had stopped.

The DO data for site 1 are missing for the first 24 hours because of a logging error. Continuous monitoring of DO in the reference plots indicated concentrations remained at >8.0 mg/L (fig. 7). Under the benthic mat and carbon dioxide mat, DO was similar at sites 1 and 2. At site 1, DO fluctuated between ~5.0 and 8.0 mg/L, except for the final day of treatment when the concentration declined <4.0 mg/L under the benthic mat and carbon dioxide mat. At site 2, DO ranged between ~4.0 and 7.0 mg/L under the benthic mat and carbon dioxide mat, and the lowest reading was on day 2 of the treatment. DO was most variable and lowest at site 3. Under the benthic mat at site 3, DO varied on a 24-hour cycle, ranging from about <2.0 to 8.0 mg/L. Under the carbon dioxide mat at site 3, DO trended downward in the initial 48 hours and then stayed mostly at <4.0 mg/L for the remaining 72 hours.

Water quality under the carbon dioxide mat was continuously logged by the YSI sonde (fig. 8; table 4). At the completion of treatment on day 5 at site 1, we determined that the sonde had detached from the cinder block and was near an edge of the mat that had become slack, which resulted in wide fluctuations in measurements (pH range from 6.0 to >8.0) and a higher mean pH (6.80) at site 1 (table 4; fig. 8). The pH measurements were more consistent and remained closer to the target at sites 2 and 3 (fig. 8; table 4). Overall, the mean (and estimated) pH was greater than the target pH of 6.13, and the estimated mean carbon dioxide concentration was <200 mg/L at all three sites (table 4). The total mass of carbon dioxide used to treat the three 16-m² plots for 5 days was 11.5 kilograms (25.3 pounds).

The effects of the benthic mat and carbon dioxide mat on resident mussels in the plots were confounded by low mussel density (table 5), limited hard substrate, and the effects of carbon dioxide on mussel attachment. Substrates that were colonized with zebra mussels and placed into the carbon dioxide plots before treatment had few to no mussels attached at the posttreatment survey, likely because of carbon dioxide effects on byssal thread detachment (Waller and others, 2020). The entirety of each 4-m x 4-m plot was surveyed, and all colonized substrate was collected in the posttreatment assessment. The estimated survival of mussels in the reference plot did not differ from zero for treatment time (fig. 9A; 0.111, z=0.393, 95-percent CI of −0.442–0.664, p=0.694). Therefore, pre- and posttreatment survey data for the reference plots were combined. The percentage of live zebra mussels in the benthic mat (difference of −2.13, 95-percent CI of −2.97–1.29, z=4.98, p<0.001) and carbon dioxide mat treatments (difference of −1.83, 95-percent CI of −2.48–−1.17, z=5.44, p<0.001) was lower compared to that of the reference site (1.70, 95-percent CI of 1.17–2.23; fig. 9B; table 5).

Sentinel zebra mussels in the benthic mat (difference of −0.728, z=−6.21, p<0.01, 95-percent CI of −0.956–−0.50) and the carbon dioxide mat (difference of −4.49, z=−30.5, p<0.01, 95-percent CI of −4.79 to −4.21) treatments had lower survival rates than those in the reference plots (1.96, 95-percent CI of 1.78–2.14) (table 6; fig. 10). In carbon dioxide mat treatments, zebra mussels experienced 100-percent mortality at sites 1 and 3 and >80-percent mortality at site 2. In comparison, mortality in benthic mat treatments ranged from 13.5 to 27 percent (table 6; fig. 10).

Survival of caged mussels varied by location within the plot, but it differed by treatment and site. Zebra mussel survival in the carbon dioxide treatment was higher in cages located midway and at mat corners (only at site 2; difference of 2.48, 95-percent CI of 1.85–3.11, z=3.89, p<0.001), whereas survival under the benthic mats was lower at the corners (difference of −13.81, 95-percent CI of −13.53–−14.09, z=−4.895, p<0.001; fig. 11).

The mean shell length of resident mussels collected in surveys was less than that of caged sentinel mussels (difference of −4.0 95-percent CI of 3.2–5.0, z=8.95, p<0.001; table 7; fig. 12). However, there was no difference in the shell length of mussels within groups at site 1 (mean of 15.1, 95-percent CI of 14.3–15.8) compared to site 2 (difference of −0.2, 95-percent CI of −1.3–0.9, z=−0.380, p=0.704) or site 3 (difference of 0.9, 95-percent CI of −1.3–0.9, z=1.79, p=0.743).

We compared benthic macroinvertebrate abundance and diversity in the reference plots before and after treatment as a measure of temporal variability in the community during the study period. Within a reference plot, no difference was detected between the pre- and posttreatment sampling periods in macroinvertebrate abundance or diversity (degrees of freedom [df]=1, f-statistic [f]=0.2879, probability [p]=0.777; fig. 13; table 8). Macroinvertebrate abundance and diversity 1 day after treatment were compared between the benthic mat and carbon dioxide mat plots and reference plots. The abundance of benthic macroinvertebrates was less in the benthic and carbon dioxide mat treatments compared to the reference plots (df=2, f=5.76, p=0.002; fig. 13). Diversity did not vary between the reference plots and the mat treatments (df=3, f=2.46, p=0.081; table 8). Distance-based redundancy analysis explained 97 percent of predictive variability with one axis, mostly treatment (variance=3389.7, f=5.76, df=2, p<0.001) and a large amount of residual variance (9416.7, df=32; fig. 14).