Environmental factors associated with freshwater recreational water quality in Niagara Region, Ontario, Canada: A path analysis

Escherichia coli concentration levels in recreational water are used by beach managers to evaluate the risk of gastrointestinal illness among beachgoers. We examined the relationship between specific environmental factors and E. coli concentration in recreational beaches in the Niagara Region. We analysed E. coli geometric means collected from eight beaches from two of the Great Lakes in the Niagara Region in Ontario, between 2011 and 2019. We applied path analysis to evaluate the relationship between the environmental factors and E. coli concentrations, including whether effects were direct or indirect via a mediator. Turbidity was found to be an important mediator for the indirect effect of environmental variables overall and in beach-specific models. Rainfall and streamflow had a positive indirect effect on E. coli via turbidity and a direct effect in five out of seven beach models. Streamflow was also a mediator for the indirect effect of previous day air temperature in five out of seven models. In three subset models, outfall E. coli concentration was a mediator for the effect of the environmental factors. Using a novel methodological approach, this study identifies important relationships and pathways that predict beach E. coli concentration in freshwater beaches located on two of the Great Lakes.


Introduction
Poor recreational water quality, as indicated by a high concentration of pathogens, represents a risk of gastrointestinal illness for those engaging in water activities such as swimming [1,2]. In Canada, beach water quality in freshwater bodies is regularly monitored by measuring faecal indicator bacteria (FIB) levels, most commonly Escherichia coli (E. coli), as a surrogate for the presence of enteric pathogens and risk of gastrointestinal illness [3]. The Guidelines for Canadian Recreational Water Quality recommend a geometric mean concentration of less that 200 colony-forming units (CFU)/100 ml of E. coli as a threshold level suggesting an acceptable, low risk of gastrointestinal illness [3]. In the Niagara region, beaches along Lake Ontario and Lake Erie are monitored regularly for E. coli levels; however, the culture-based laboratory methods used in Ontario can require 18-24 h for results to be available [4]. As such, decisions about beach posting status (i.e. if the beach water is unsafe for swimming) are based on data from the previous day. This delay could result in a risk of illness to the swimmers as changing environmental conditions could result in a difference in water quality within hours [5].
Freshwater beach water E. coli concentrations are influenced by multiple environmental factors [6][7][8]. Rainfall is regularly reported as one of the most important predictors of freshwater lake quality, in predictive models, with increased rainfall resulting in higher microbial concentrations [6,9]. Conversely, increased solar radiation has been associated with decreased E. coli counts [7]. This relationship is also affected by lake level, wave height and turbidity [7]. Turbidity, while not regularly collected as part of environmental monitoring at many public health units in general, has been increasingly studied as an important predictor for FIB, including E. coli [8].
The factors affecting E. coli concentrations in beach water can directly impact public health management strategies and policies. Research on freshwater beaches considers the unique characteristics that may influence the factors associated with the presence of FIB in freshwater. For example, E. coli is known to survive longer in freshwater and fresh waters lack the salinity and tidal cycles present in marine water [7]. Inland freshwater beaches require focused research and analysis in order to appropriately inform beach water monitoring programmes in making more timely decisions. Currently, there is a limited body of such research in the Ontario and Canadian context. The environmental drivers of E. coli concentration may not act directly or independently of each other. Understanding the complex pathways of the relationships between the factors and E. coli could help direct water quality improvement efforts. Path analysis, is a powerful method that allows for gaining insight of the cause −effect relationships between variables, allowing for further understanding of complex relationships within interactions webs. Using this method, we are able to estimate the direct, indirect and total effects of factors on outcome variables [10].
The unique geography of Niagara Region, Ontario allows for the opportunity to examine freshwater beaches located along two of the Great Lakes. In addition, the management of beach water quality is of significant public health importance, with an extensive beach monitoring programme in place for the many popular beaches. Using daily water quality sampling data collected by the local health unit linked to publicly available federal and provincial environmental data, this study applies path analysis methodology to examine the key environmental factors of E. coli concentration at eight popular beaches.

Study area
Eight beaches monitored by Niagara Region Public Health were selected for the study based on the availability of regular water sampling data (Fig. 1). Two beaches are located along Lake Ontario and six along Lake Erie. The region has a total area of 1852 km 2 and a population of 427 421 [11]. Two major water systems traverse the region in a northward direction, connecting Lake Erie to Lake Ontario: the Niagara River, which contains Niagara Falls and the Welland Canal.

Water quality data
Samples for E. coli collected during the recreational season (May to September) from 2011 to 2019 were obtained by Niagara Region Public Health. Beach water samples were collected six times a week at seven of the participating beach study sites and four times a week at the remaining beach (Queen's Royal). Sample collection took place between 7 and 10 AM each day at knee to waist depth, 15-30 cm below the surface of the water, from five pre-specified sampling locations at each beach, following recommended provincial guidelines [12]. Water samples were centrally processed at a Public Health Ontario laboratory within one calendar day of collection using an accredited modified membrane filtration method [13]. The daily E. coli geometric mean for each beach was calculated by the public health unit based on the laboratory results of the five samples collected. In addition, water outfall samples were collected from stormwater runoff at four participating sites on a weekly basis and processed in the same manner. To address the skew of both the E. coli geometric mean and outfall E. coli values, log transformations were used to satisfy linear assumptions, prior to data analysis.

Environmental data
Total daily precipitation (mm) and maximum, minimum and mean air temperature (°C) data were obtained from the  Canadian Government's Environment and Natural Resources weather station historical data repository [14]. Three weather stations in the Niagara Region were selected based on completeness of data during the study period 2011-2019: Grimsby Mountain, Port Colborne and Fort Erie. Beaches were linked to a weather station based on lake location and proximity to the station (Supplementary Table S1). Wave height (m) and wind speed (knots) were collected from Environment Canada buoy station historical data [14]. Lake Ontario sites were linked to buoy 45 159, while sites located on Lake Erie were linked with buoy 45 142 (Supplementary Table S1, Fig. 1). Stream discharge data were collected from sensors located mid-way through the Niagara River and Welland Canal and publicly available on Environment Canada's streamflow historical data repository [14]. Sites were linked to sensor data based on proximity to the river or canal. Ultraviolet (UV) radiation data were collected from the closest station collecting these data, the U.S National Oceanic and Atmospheric Administration Weather Service station located in Buffalo, New York, United States [15]. Turbidity, measured in nephelometric turbidity units (NTU), was collected by Niagara Region Public Health as part of their water sampling collection process, at the third (middle-most) sampling site at each beach. Testing of the water samples for turbidity took place on-site using a Hach 2100P Turbimeter. To adjust the scale of the variables and to address skew of the data, turbidity and stream discharge were log transformed prior to analysis.

Statistical analysis
We applied path analysis methodology, a powerful statistical method for examining causal patterns among variables, which is useful in understanding the influence of variables on one another [16]. A hypothesised conceptual path model (Supplementary Fig. S1) was developed based on available literature addressing the environmental factors influencing water quality, as measured by E. coli. This model featured the potential pathways between the environmental factors and the outcome variable, E. coli concentration, as well the intervariable relationships. Multicollinearity diagnostics ensured variables met the assumption of independence [17]. Variables were not included in the model if the variance inflation factor (VIF) exceeded 0.80. A multilevel approach was not required for this analysis as exploratory linear mixed effects modelling, including beach as a random effect, did not identify significant clustering of the E. coli values at the beach level.
To examine the temporal relationship between environmental conditions and the E. coli concentration, we examined values of the previous day for mean stream discharge, mean wave height and mean UV index. Air temperature was also captured as a previous day mean instead of a max and min, as it was expected that these values would be reflected in changing daily mean. We included same-day values of turbidity and streamflow. As we expect E. coli concentration on the previous day to be associated with the current day concentration, this previous day value was also included in the model. A cumulative rainfall variable was generated as the 2-day sum of precipitation (mm), preceding the day of collection of the water sample. All variables in the model were continuous. The pathways from these variables to our outcome variable, log 10 E. coli, were hypothesised to operate via a direct relationship or indirectly via a mediator, such as turbidity, stream discharge and outfall (at applicable beaches).
Specific indirect effects for each antecedent-mediator-outcome path were calculated.
Goodness of fit for each model was assessed by chi-square statistics, comparative fit index (CFI), Tucker-Lewis index (TLI) and root mean square error approximation (RMSEA). The models were considered to have a very good fit if the CFI and TLI were above 0.95 and RMSEA was below 0.05 [18]. Akaike information criterion (AIC) and Bayesian information criterion (BIC) were used to compare and select between models in the model-building phase. Analyses were carried out using Stata version 14.0 using the sem command.

Descriptive data
A total of 5149 observations from eight beach sites in the Niagara Region were included in the analyses and linked with daily environmental factor values (Supplementary Table S2). Overall geometric mean E. coli levels by beach are presented in Figure 2 and Supplementary Table S3 Table S2). The highest outfall E. coli values were seen at Long Beach.
Beach postings are presented in Supplementary Table S4, overall and by beach site. Queen's Royal Beach had the highest proportion of beach days posted as unsafe for swimming overall and the highest annual proportion of days posted in 2013 (54%), which was also the year Queen's Royal Beach had the highest geometric mean. Lakeside Beach was posted as unsafe for swimming for most of 2019 due to high water levels in Lake Ontario that resulted in submersion of most of the beach in the lake. Unsurprisingly, most other beach postings throughout the study period were associated with high E. coli levels, with few resulting from algae, rainfall and other reasons (Supplementary  Table S5). Other reasons were due mainly to visible debris.
Annual environmental factor values are summarised in Supplementary Table S6. Rainfall was presented as an annual cumulative value, while all other variables were presented as a mean or median of daily values. The highest rainfall for the study period was reported in 2018 with 513 mm of rain, followed by 2013 with 471 mm of rain. The highest mean air temperature was in 2018 (24.3°C), followed by 2013 (22.8°C). The highest seasonal turbidity was in 2019 at 6.5 Nt\TU (Supplementary Table  S6). Of note is also the high stream discharge value in the Niagara River, with the highest daily mean reported in 2019 with a value of 7850 m 3 /s.

Path analysis
A total of 3598 observations were linked to previous day E. coli concentration and environmental variables, for inclusion in the path analysis (Supplementary Table S2). Final models of the significant overall and beach-specific pathways are presented in path diagrams representing the relationship between the factor or exogenous variable and the mediating variable or with log 10 E. coli directly. In the overall model, model fit indices present a very good model fit with RMSEA = 0.016, chi-square = 0.069, Epidemiology and Infection 3 CFI = 0.997 and TLI = 0.992 (Table 1). Model fit indices for the final beach-specific models are presented in Table 2 and suggest very good fit. Estimates of mediation for the indirect effects in the models are presented in Table 2. The overall regional path model presents the results of available observations in all eight participating study beaches, excluding the outfall data (Fig. 3).
The data for seven beaches were used to develop beach-specific models and presented in Figure 4. Due to the low sample size at Queen's Royal beach following the linking of observations to antecedent variables, a beach-specific model was not included in the final analysis. Turbidity had a significant and positive direct effect on E. coli and was an important mediator, both overall and in all beach-specific models. In the overall model, all environmental factors had a significant indirect effect on E. coli via turbidity ( Table 2). Increased log 10 streamflow, which did not have a direct effect on E. coli, did have a positive effect on turbidity and therefore a total effect on E. coli of 0.318. Turbidity was the mediator for significant indirect effects (Fig. 4) in all beach models. Overall and at Sherkston Elco Beach and Wainfleet Beach, temperature was not directly associated with log 10 turbidity; however, an increase in previous day mean temperature had a significant negative effect on streamflow and therefore an indirect effect on log 10 turbidity. A similar pathway was observed at Long Beach and Nickel Beach; however, the indirect effects were not significant ( Table 2).
Cumulative rainfall in the preceding 48 h had a positive direct effect on E. coli, overall and at all beaches except Sherkston Elco and Sherkston Wyldewood. In the overall model, in addition to its direct effect, rainfall had a statistically significant indirect effect via its positive effect on turbidity (b = 0.001, P = 0.017), for a total effect of (b = 0.002, P ≤ 0.001). A similar direct and significant indirect pathway was identified at Bay Beach, while at Lakeside Beach, Long Beach, Nickel Beach and Wainfleet Beach, rainfall had a direct relationship with E. coli but no significant indirect pathway via turbidity.
In all models, average wave height in the previous 24 h was positively associated with turbidity and therefore had a positive and statistically significant indirect effect on E. coli. Interestingly, however, it had a negative direct effect on E. coli. This was consistently reported across beach-specific models as well. Despite the positive indirect effect through turbidity, the     There was consistency in how previous day E. coli presented across all models. An increase in value was associated with E. coli via a significant positive direct effect and a positive indirect effect via turbidity. Previous day UV had a negative and indirect effect on E. coli via turbidity (b = −0.012, P = 0.010) in the overall model. A similar pathway is observed at Wainfleet Beach, with only negative indirect effects on E. coli via log 10 turbidity. At Lakeside and Long Beach, the effects on E. coli were both direct and indirect via turbidity.
As previously described, streamflow was a mediator for previous day air temperature via turbidity, overall and in four out of seven beach models. Direct effects on E. coli, in addition to indirect effects, were only identified at Sherkston Wyldewood Beach and Wainfleet Beach. At Wainfleet Beach, the path model describes a significant and positive indirect effect via turbidity (b = 1.400, P ≤ 0.001); however, the total effect of streamflow was not significant (0.511, P = 0.249), suggesting only partial mediation by turbidity.

Outfall models
A total of 738 observations were included in the outfall path analysis (Supplementary Table S2). The overall model is presented in Figure 5 and includes data from all four beaches that collected weekly outfall data. We remind here that due to low sample size, a beach-specific outfall path analysis model was not developed for Queen's Royal Beach. The three beach-specific models are presented in Figure 5. Model fit indices suggest an excellent fit for the overall model and Long Beach model; however, poor fit is observed at Bay Beach and Nickel Beach (Table 3).
Overall, the addition of the outfall variables resulted in an additional significant mediated pathway for previous day air temperature (b = 0.002, P = 0.004) and rainfall (b = 0.002, P ≤ 0.001) ( Table 4). In addition to the direct effect of outfall on E. coli, there was a positive indirect effect via turbidity (b = 0.114, P ≤ 0.001) ( Table 4). At Long Beach and Nickel Beach, outfall did not mediate the effect of other factors but instead had a positive direct effect, or indirect effect via turbidity (Nickel Beach). At Bay Beach, mean air temperature had a direct effect on E. coli and several indirect pathways via streamflow, which then also had pathways through outfall and turbidity. The total effect of mean   temperature in the Bay Beach model was 0.103 (P ≤ 0.001) ( Table 4). Rainfall only had indirect pathways via outfall and then via turbidity for a total positive effect of 0.022 (P ≤ 0.001) ( Table 4). Similar to Bay Beach, mean temperature at Nickel Beach had a direct effect on E. coli and an indirect effect via streamflow, which then had paths to E. coli or via turbidity. Total effects were 0.048 (P = 0.029).

Discussion
While previous studies have examined the factors influencing inland freshwater recreational beach quality, this study used a novel approach in applying path analysis to explore the relationships between environmental factors and their association with E. coli. The Niagara Region provides a unique opportunity to explore these dynamics at beaches located along two of the Great Lakes in one of Canada's top tourist destinations, with approximately 14 million tourist visiting the region annually [19]. We used the water quality data, as captured through the measurement of geometric mean of E. coli, to explore the overall and beach-specific pathways at seven popular beaches in the region. Exploratory models also aimed to account for a potential annual effect by including 'year' as a random effect; however, this did not have a significant impact on the model. The final beach-specific models demonstrate heterogeneity in the significant pathways across the beaches. While some pathways and factors were consistently significant across study sites, there were some notable trends and differences.
An increasing number of studies present a strong relationship between turbidity values and FIB concentration [8,20,21]. Suspended particles in the water may shield pathogens from environmental stressors such as UV radiation penetration [3]. In addition, there is increasing evidence that a significant portion of pathogens are associated with particles [22]. A study by Lawrence [8] near Atlanta, Georgia found that that in 34-42% of surface water samples E. coli were attached to particles. In our study, turbidity was a consistently important variable in the overall and beach-specific models as having both a direct and positive relationship with E. coli concentration and as a mediator for other environmental factors. In the overall model it was a mediator for the effect of all environmental factors. Our results suggest that turbidity is an important factor and mediator in the pathways between environmental factors and E. coli concentrations, which needs to be considered and examined further as part of public health strategies. Based on our results, we recommend that public health authorities in Ontario and elsewhere collect turbidity measurements on-site during routine beach water sampling.
Heavy rainfall has been associated with increased microbial concentrations in beach waters [23]. While this analysis did not specifically examine heavy rainfall (e.g. 95th percentile), which has been associated with increased E. coli levels [24,25], it identified increased rainfall in the preceding 48 h had a direct and indirect effect on E. coli in 5/7 beaches. Antecedent rainfall was associated with increased E. coli concentrations and increased turbidity in the overall and Bay Beach models; however, in the subset outfall data model, rainfall was no longer significant and was instead replaced by outfall E. coli concentration, which was associated with both variables. This suggests that rainfall's effect could at least partially be due to outfall levels. Storm water runoff from impervious surfaces has been suggested to be the most important pollution source causing beach closures [25]. Many urban beaches will be automatically posted as unsafe for swimming following rainfall and increased runoff, without water sampling results [25,26]. Interestingly in 2/7 beaches (Sherkston Elco and Sherkston Wyldewood Beaches), rainfall was not a significant pathway in the models. These findings could suggest that the presence of heavy outfall water flow could be considered to be an indicator of poor water quality conditions.
Overall and at most beaches, stream discharge was indirectly related to E. coli concentration via its positive effect on turbidity. This suggests that streamflow may increase turbidity, which was consistently associated with increased E. coli concentrations. At Bay Beach, we found there was a positive direct effect on E. coli concentration, not mediated by turbidity. Of the beaches examined, this study site was closest in proximity to the start of the Niagara River on the Lake Erie side, which features a strong northward flow through the powerful Niagara Falls. This may be a possible explanation for the important direct effect of streamflow at this location. As presented in the conceptual path diagram (Fig. 3), we expected total rainfall in the previous 48 h to have an effect on streamflow. In an analysis of stormwater samples throughout the duration of storms, it was found that E. coli concentration increased during a rain storm were highest in the early stages of a storm [22]. Interestingly, this pathway was consistently not statistically significant across all models; however, as previously described, this study did not examine heavy rainfall specifically, instead only antecedent total rainfall and perhaps this measure does not capture the effect of heavy rain on streamflow specifically.
Research focusing on the effect of UV irradiation on E. coli has been mostly limited to the laboratory scale [27]. UV irradiance results in damage to microbial pathogens and is a common method for drinking water treatment [27]. In a laboratory study specifically examining the effect on E. coli concentrations, researchers found that exposure to UV light decreased the number of bacterial colonies formed [28]. As previously described, increased turbidity could prevent penetration of UV light and therefore decrease its ability to have an effect on E. coli concentrations in water [3]. We used the previous day UV index as water samples are collected early in the morning and have not yet been subjected to extensive UV irradiance on the day of sampling. Research also suggests that increased UV exposure is important for effective treatment [7,28]. Interestingly, increased UV index had a negative indirect effect on E. coli concentration via its negative relationship with turbidity.
Wave activity has been associated with resuspension of bacteria from sediments and therefore elevated E. coli concentrations in beach water; however, we did not find this relationship in our models [26]. In contrast with our conceptual diagram, increased wave height was interestingly inversely associated with E. coli levels. We did, however, identify that wave height was consistently positively associated with turbidity, which is consistent with the literature, therefore having a positive indirect effect on E. coli via that pathway. Due to the greater negative direct effect, the total effect of wave height was negative in most models. Due to collinearity issues, we did not include wind speed as a variable in the models; however, we hypothesised that increased wave height was a proxy for increased wind activity. Important to note is the far offshore location of the buoys used to measure this variable, which may not reflect wave activity at the shoreline. More importantly, strong offshore energy could result in increased water circulation through stronger currents, therefore resulting in increased flushing of shoreline bacteria [29]. In addition, higher energy waters reduce the opportunity for sediment buildup, which are known to provide a habitat for bacteria survival [29].
Temperature is an important factor influencing E. coli survival and growth [30]. We examined the relationship of E. coli concentrations with air temperature, excluding water temperature due to collinearity issues. Air temperature had a positive direct relationship with E. coli concentrations, overall and in all beach models. Several studies have described an inverse relationship between air temperature and streamflow as it can result in marked declines in streamflow [31]. These findings are consistent with the relationships identified in our models. Increased air temperature had a negative effect on streamflow in the overall model as well as five of the beach models.

Limitations
Given the importance of animal faecal material and microbial loading at beaches, future models could benefit from the inclusion of these data [23]. A microbial source tracking (MST) study conducted in a Toronto beach found that waterfowl was the main source of water contamination and strategies to reduce bird presence were successful in improving water quality [32]. Public health units could consider occasional MST at beaches to improve water quality through targeted approaches [32]. While total rainfall in the previous 48 h was found to have a linear relationship with E. coli either directly or via its relationship with turbidity, further exploration by examining rainfall thresholds or heavy precipitation events (e.g. 95th percentiles) could be beneficial to further understand the relationship between rainfall and E. coli levels in beach water. Bather load information, which was not available for this analysis, could also be beneficial in informing future models given its suggested association with transference of microorganisms from the sand and sediment to swimming waters and through increased turbidity [3]. Finally, other considerations for future analyses include the addition of information about surrounding cattle and agricultural runoff and proximity to combined sewage systems.

Conclusion
With almost half a million residents and a large in-flow of tourists during the summer season, the quality of the recreational waters of the Niagara region is of major public health importance. Poor water quality could result in an increased risk of recreational water illness among the thousands of beachgoers that visit the region's many popular beaches. Given the delay in receiving water sampling results due to the culture-based laboratory methods used in the province, a greater understanding of the environmental factors and dynamics for E. coli concentration could better guide beach managers in the decision-making and risk communication process. To our knowledge, this is the first application of the path analysis methodology to examine factors of freshwater beach quality. This methodology allowed for the exploration of intervariable relationships via mediation. We identified some clear trends and the importance of some key variables, such as turbidity. We also present the heterogeneity that exists across beaches, despite some clear trends and therefore the need for consideration of site-specific factors when evaluating beach quality. The importance of these results extends beyond the Niagara Region and could be applied to other inland freshwater beaches.
Supplementary material. The supplementary material for this article can be found at https://doi.org/10.1017/S0950268821002120