Surface water samples were collected from both North and South Florida. The surface water in North Florida was collected from rivers (n = 5) and a lake (n = 1). The surface water in South Florida was collected from canals (n = 8). The North Florida samples were collected from each site monthly over 12 consecutive months beginning in November 2011. The South Florida samples were collected 12 times between May 2015 and November 2016. All samples were collected as previously described by McEgan et al. (2013). Briefly, a sterile carboy (Nalgene, Rochester, NY) was fitted with 4 kg lead weights, attached to a rope, and lowered 20 cm below the water surface at least 3 m from the shore. Carboys were filled with at least 1 L of water, transported to the lab at 4°C, and processed within 24 h of collection. Sampling always occurred before solar noon.

At each sampling physicochemical water quality, including turbidity, water temperature, pH, conductivity, colorimetric reading, and oxidation-reduction potential (ORP), was assessed. Each physicochemical parameter was measured in triplicate, and the value averaged. Turbidity was measured using a portable colorimeter (DR/850; Hach Company, Loveland, CO, USA) according to the manufacturer's instructions. Water and air temperatures were measured with a portable temperature probe (SH66A; Cooper Instrument Corporation, Middlefield, CT, USA). The ORP and pH were measured with a portable ORP/pH meter (pH 6 Acorn series: Oakton, Vernon Hills, IL, USA). Conductivity was measured with a portable conductivity tester (HI98304 DIST1 4 EC, HANNA Instruments, Woonsocket, RI, USA).

The weather data, including air temperature, relative humidity, and rainfall, were obtained for each sampling from the Florida Automated Weather Network (https://fawn.ifas.ufl.edu/) using the closest weather station to the sampling site. To characterize the land use around each sampling site, land cover data were downloaded from the National Land Cover dataset (https://www.mrlc.gov/national-land-cover-database-nlcd-2016). The proportion of land within 122,366 and 1,098 m of each site under pasture-hay, cropland, forest-wetland, and developed (>20% impervious) cover was then calculated using the code developed by D. Weller (https://github.com/wellerd2/Calculating-land-use-land-cover-and-landscape-structure-parameters) as previously described (Liao et al., 2021). Buffer distances (122,366 and 1,098 m) were selected based on the recommendations from the Leafy Green Marketing Agreement on how far pre-harvest agricultural water sources should be from land uses that may contaminate the water with human pathogens (Table 7 Crop Land and Water Sources Adjacent Land Uses in California Leafy Greens Marketing Agreement, 2020). For example, the recommended distance from a crop land to a concentrated animal feeding operation (CAFO) with >1,000 animals is 1,200 feet (~366 m); this buffer (366 m), and buffers 1/3 smaller (122 m) and 1/3 larger (1,098 m) were used here. The elevation was obtained from the United States Geological Survey (https://apps.nationalmap.gov/elevation/#/%23bottom) for each water collection site.

The aerobic plate count (APC) (CFU/100 mL), total coliform (MPN/100 mL), and E. coli (MPN/100 mL) levels were enumerated in each sample as previously described (McEgan et al., 2013). Briefly, for APC, water samples were serially diluted in 0.1% peptone water (Difco, Sparks, MD), and 100 μL aliquots were spread plated in duplicate on a tryptic soy agar (TSA) (Difco, Sparks, MD, USA). The TSA plates were incubated at 35 ± 2°C for 24 h. Colonies were enumerated by hand and CFU/100 mL calculated. Colisure presence/absence snap packs (IDEXX Laboratories, Inc., Westbrook, ME, USA) were used to determine the coliform and E. coli most probable numbers (MPN) in a five-by-three MPN configuration (10-, 1-, and 0.1 mL dilutions). The tubes were incubated at 35 ± 2°C for 24 h. The yellow color indicated coliforms, and E. coli was identified by observing fluorescence using a 6-watt fluorescent, 365 nm long-wave UV lamp with bulb from IDEXX Laboratories, Inc., Westbrook, ME. The MPN/100 mL was determined from the table in Standard Methods for the Examination of Water and Wastewater, 18th ed (American Public Health Association, 1992).

The methods for the MPN estimation of Salmonella in each water sample were based on the US Food and Drug Administration (FDA) Bacteriological Analytical Manual (BAM) (Andrews et al., 2011). Briefly, a three-by-three MPN method using dilutions (i) 10 mL of water sample in 10 mL double-strength lactose broth, (ii) 1 mL of water sample in 9 mL single-strength lactose broth, and (iii) 0.1 mL of water sample in 9 mL single-strength lactose broth were done. The MPN tubes were incubated at 35 ± 2°C for 24 h. Selective enrichment was performed by transferring 1.0- and 0.1 mL aliquots of each tube to tetrathionate (TT) broth (Difco) and Rappaport-Vassiliadis (RV) broth (Difco), respectively. The TT and RV broths were incubated at 35 ± 2°C and 41 ± 2°C for 24 and 48 h, respectively. A 10-μL aliquot from each TT and RV broth were streaked onto xylose lysine Tergitol 4 (XLT-4) (Difco) and CHROMagar Salmonella Plus (DRG International, Inc., Springfield, NJ, USA), and incubated at 35 ± 2°C for 24 h. Presumptive Salmonella colonies were streaked on lysine iron agar slants (LIA) (Difco) and triple sugar iron agar slants (TSI) (Difco). The slants were incubated at 35 ± 2°C for 24 h.

Further confirmation was performed using PCR for the invA and oriC genes, as previously described (Malorny et al., 2003). PCR confirmed Salmonella isolates were preserved at −80°C in 15% glycerol. One Salmonella isolate per sample enrichment scheme (e.g., TT-XLT-4, RV-XLT-4) was sent to the National Veterinary Services Laboratory (Ames, Iowa, US) for serotyping.

Samples that were below the limit of detection (LOD) for Salmonella (LOD; <0.48 log₁₀ MPN/100 mL) were considered Salmonella-negative. Samples yielding an MPN value, above the LOD (with an upper limit of detection of 3.3 log₁₀ MPN/100 mL), were considered positive for Salmonella (volume of water tested was 33.3 mL).

All analyses were performed in R version 3.3.5 (R Foundation for Statistical Computing, Vienna, Austria). Initial descriptive analysis was performed, and summary statistics were calculated separately for all continuous factors (e.g., microbial levels, weather conditions) in North and South Florida. Using the vegan package, Simpson's Index of Diversity was calculated to quantify and compare the Salmonella serotype diversity in the North and South Florida water samples. Multiple samples had total coliform levels above the upper LOD, and the upper LOD times 1.5 were used for the total coliform value for these samples in the regression and tree analyses.

Bayesian mixed models were implemented to characterize the differences in microbial concentration, and the presence-absence of Salmonella between regions, seasons, and water types. The outcomes considered are listed in Tables 1, 2. Due to the number of samples below the detection limit for Salmonella and above the limit for total coliforms, hurdle models were implemented. Briefly, logistic regression was used to characterize the associations if a sample was Salmonella positive or negative, or if the total coliform levels were above or below the limit of detection. Then for those samples where Salmonella or total coliforms were enumerable, a separate log-linear model was fit. For models where the outcome was binary, a Bernoulli distribution with a logit link function was used. All models included a random effect of site and fixed effect of the season to account for pseudo-replication and temporal autocorrelation. Separate models were fit for each outcome with either water type (canal, lake, or river) or region (North or South Florida) as the covariates. Separate models were also used to characterize the relationship between Salmonella contamination and log₁₀ E. coli levels (both as a continuous concentration variable, and as two binary variables indicating if E. coli levels in the sample were above or below the geometric mean and STV cut-offs prescribed in the PSR). Models were fit using the brms package for Bayesian Regression Models using 'Stan', uninformative priors, 3 chains, and thinning set to 10 (Bürkner, 2017a,b). While the number of iterations per chain was set to 5,000 (burn-in of 2,500) for most models. The maximum a posteriori (MAP) and 95% credibility interval (CI) for the effect estimates were calculated using the bayestestR package (Makowski et al., 2019a,b). The method of interpreting the MAP and 95% CI estimates is described in the footnotes for Tables 1, 2 as the interpretation of the probability of direction (PD), practical significance (PS), and regional of practical equivalence (ROPE) indices, which were quantified and used to determine if the (i) given factor had a substantial effect on the outcome, and (ii) if a positive or negative effect exists regardless of if that effect is negligible or non-negligible.

Separately from the Bayesian regression, conditional inference trees were implemented using the partykit package and the defaults recommended by the package authors. Trees were used to determine if there were differences between regions in environmental factors associated with log₁₀ E. coli, total coliform, Salmonella levels, and Salmonella presence-absence. For the model where the outcome was log₁₀ Salmonella levels, only samples positive for Salmonella were used. For the total coliform model, coliform values for samples where coliform levels were above the upper LOD were set to 1.5^*LOD. Conditional trees were used since they are robust to collinearity and correlation between explanatory factors, can handle missing data, can handle hierarchical relationships (and account for all possible interactions), and can be easily interpreted (Weller et al., 2020a,c). In interpreting the results, it is important to note that region and water type are collinear, with all South Florida sites being canals, and North Florida sites being either lakes or rivers.

Of the 168 samples, total coliform levels in 112 (40 from North and 72 from South) were above the upper limit of detection (log₁₀ > 3.3 MPN/100 mL); no samples fell below the lower limit of detection. For the 112 counts that fell above the limit of detection, the value of 3.3 log₁₀ MPN/100 mL was used. The mean and median log₁₀ MPN/100 mL of total coliforms was >3.0 (Range = 1.7, >3.3) and >3.2 (IQR = >2.7, >3.3) in North Florida, and >3.3 (Range = 2.4, >3.3) and >3.3 (IQR = >3.3, >3.3) in South Florida (Table 3), respectively. Unlike total coliforms, no sample had E. coli levels below the lower limit or above the upper limit of detection. For North Florida samples the mean was 1.8 log₁₀ MPN/100 mL (Range = 0.6, 3.2) and the median was 1.7 log₁₀ MPN/100 mL (IQR = 1.4, 2.2), while in South Florida samples the mean was 1.3 (Range = 0.0, 2.8) and the median was 1.3 (IQR = 0.9, 1.6; Table 3). While only 9% of the samples collected in South Florida exceed the PSR GM standard for E. coli (126 CFU/100 mL), 32% of samples collected in North Florida exceeded this cut-off. Similarly, 3% of South Florida samples and 11% of North Florida samples had E. coli levels that exceeded the PSR STV standard (410 MPN/100 mL). Multiple sources show that the difference between CFU and MPN is not significant to change the interpretation of the findings or conclusions (Cowburn et al., 1994; Hargett and Goyn, 2004; Gronewold and Wolpert, 2008; Fricker et al., 2010).

Salmonella was detected in 26% (44/168) of water samples (Table 4; 124 samples were below the limit of detection, <0.48 MPN/100 mL). More Salmonella was detected in South Florida (27/96; 28%), than in North Florida (17/72; 24%). For the 124 counts that fell below the limit of detection, the value of.48 log₁₀ MPN/100 mL was used. The mean and median log₁₀ MPN/100 mL of Salmonella was <0.5 (Range ≤ 0.5, 1.4) and <0.5 (IQR ≤ 0.5, <0.5) in North Florida, and <0.6 (Range ≤ 0.5, 3.0) and <0.5 (IQR ≤ 0.5, <0.5) in South Florida (Table 3). Figure 1 describes the distribution of total coliforms, E. coli, and Salmonella by region, season, and water type. The Salmonella prevalence fell within the wide range reported by past Florida studies (McEgan et al., 2013; Luo et al., 2015; Topalcengiz et al., 2017). A Central Florida study reported a Salmonella prevalence of 4.8% (26/540) in 250 mL pond samples (Topalcengiz et al., 2017), while an independent Central Florida study detected Salmonella in all 202 10-L samples collected from multiple surface water types (e.g., ponds, canals) (McEgan et al., 2013). The Salmonella prevalence reported here is also within the range reported by studies conducted in other states, including North Carolina [e.g., 54.7% (47/86) of 25 mL water samples; (Patchanee et al., 2010)], and Georgia [e.g., 11.9% (34/285) of 222 mL water samples (Antaki et al., 2016); 79.2% (57/72) of 111 mL water samples (Haley et al., 2009)]. One reason likely responsible for the wide range in Salmonella prevalence reported by previous studies is the methodological differences between these studies. The study that reported the highest Salmonella prevalence (McEgan et al., 2013) collected 10 L samples, 5 to 40 times larger than the samples collected in the other studies discussed here (Haley et al., 2009; Patchanee et al., 2010; Antaki et al., 2016; Topalcengiz et al., 2017), and were 30 times larger than the samples collected in the present study. The likelihood of detecting Salmonella is higher for larger volumes of water, as increasing the volume tested decreases the lower limit of detection. Methodological differences between studies confound the comparison of results and reduce the ability to determine if the observed prevalence is consistent within and between studies. Development of standardized practices, including standard volumes for studies focused on agricultural water used for produce production may be appropriate. Alternatively, it may be possible to develop a statistical model, that accounts for sample volume, to enable the comparison of results between studies.

The total coliform and Salmonella levels reported are higher in South Florida than North Florida; the opposite is reported for E. coli (Figure 1). Total coliform and Salmonella levels reported are highest in canals, with total coliform levels higher in rivers, than lakes, and Salmonella levels were higher in lakes than rivers; for E. coli levels appeared lowest in canals, followed by rivers and lakes (Figure 1).

In the Bayesian mixed models, the log₁₀ E. coli levels (measured as both a continuous factor, and as a binary factor representing if samples were above or below the PSR cut-offs of 126 and 410 CFU/100 mL) and log₁₀ total coliform levels (measured as both a continuous factor, and a binary factor representing if samples were above or below the upper LOD) differed substantially between North and South Florida, and between water types (Table 1). The E. coli levels were ~0.50 log₁₀ MPN/100 mL higher in North Florida than in South Florida (MAP = 0.50; 95% CI = 0.12,0.90). The results of the conditional inference trees were generally consistent with the regression models. However, given collinearity between the region and water type (i.e., that canals were only sampled in South Florida, and lakes and rivers in North Florida), it is difficult to determine if differences are driven by water type or region. To assist in overcoming this difficulty (i.e., collinearity), and probe if differences are driven more by water type or region, future studies should include a variety of water types within each region. The different surface water sampling locations reflect the different, and complex, watersheds in the state, resulting from disjointed drainage systems and atypical elevation gradients. In North Florida a karst topography results in numerous rivers, streams, lakes, and springs; in South Florida, canal systems were developed to divert the water that historically flowed as a sheet of water to allow for agricultural production (Purdum et al., 2002). The findings, in light of these geographical differences, highlight the heterogeneity inherent to freshwater environments, and the need for improved understanding of pathogen ecology for specific, intrastate produce growing regions.

Collinearity between the region and water type is not a factor for Salmonella result interpretation since neither Salmonella levels nor the odds of Salmonella detection differed substantially between regions. The Salmonella levels and the odds of Salmonella detection did differ between water types (e.g., canals, rivers, lakes) (Table 2). Specifically, the odds of Salmonella detection (OR = 0.70; 95% CI = 0.30, 1.59), and Salmonella levels (Effect Estimate = −0.23; 95% CI = −0.62, 0.24) were lower for river samples, compared with canal samples (Table 2) based on the practical significance (PS) index being ≥0.75, indicating the observed effect is both non-negligible and, in the direction indicated by maximum a posteriori estimate (MAP) (Table 2). The Salmonella tree (Figure 2) found evidence of a significant regional difference. However, this difference was dependent on environmental conditions (i.e., when temperatures were high, Salmonella levels were higher, regardless of region, but at lower water temperatures, Salmonella levels were higher in South Florida than in North Florida). These findings indicate that microbial water quality varied both between regions and between water types. This finding was not surprising, as past studies that compared microbial quality between growing regions have also found evidence of regional differences (Strawn et al., 2013a, 2014; Chapin et al., 2014; Weller et al., 2020b). Weller et al. (2020b) sampled Arizona canals and New York streams and reported higher E. coli levels and a higher prevalence of Listeria monocytogenes, pathogenic E. coli markers, and Salmonella in New York streams. Additionally, Strawn et al. (2014) found a higher prevalence of Salmonella positive overall environmental samples and water samples in south Florida (35 and 38%, respectively), compared with New York (5 and 9%, respectively). The findings reported here, coupled with previous studies (Strawn et al., 2013a, 2014; Chapin et al., 2014; Weller et al., 2020b) highlight the differences in microbial water quality between water sources and growing regions, and underscores the challenges of developing a one-size-fits-all approach for managing microbial hazards in agricultural water. The findings reported here, identify the differences in microbial water quality, and subsequent resulting challenges with recommendations for managing microbial hazards, even within a single state. These findings suggest that risk management approaches may need to be tailored to specific water types within localized (e.g., intra-state) regions.

Serotyping was performed on one representative Salmonella isolate per isolation scheme (up to 4 isolates per sample) and yielded 45 Salmonella isolates from the 44 positive samples (Supplementary Table 3). One of the 44 positive samples yielded two serovars: S. enterica subspecies enterica Invernes and Muenchen (in North Florida waterways) (Supplementary Table 3). The remaining 43 Salmonella-positive samples represented one serovar including Anatum (1), Florida (3), Hartford (1), Inverness (4), Muenchen (4), Saintpaul (3) and IV 40:z4,z24 (2) in North Florida; and Agona (1), Baildon (1), Braenderup (2), Enteritidis (2), Javiana (1), Litchfield (1), Muenchen (1), Rough (1), Rubislaw (2), Tennessee (4), Typhimurium (1), III 60:r:z (1), and IV 53:z4.z23 (9) in South Florida waterways (Supplementary Table 3). Overall serotype richness was higher in South Florida (13 serotypes) than North Florida (7 serotypes), as was Simpson's Index of Diversity (0.67 in South Florida;0.58 in North Florida). Salmonella serotype IV 53:z4.z23 was isolated most frequently from water samples in South Florida waterways, consisting of approximately 33% (9/27). While no one serotype was predominant among the Salmonella isolates from North Florida waterways, four serotypes (Florida, Inverness, Muenchen, and Saintpaul) represented 78% (14/18). A previous study in central Florida identified 33 Salmonella enterica serotypes from 165 surface water samples with the most frequent serotypes being Muenchen, Rubislaw, Anatum, Gaminara, and IV_50:z4,z23:- (McEgan et al., 2014). When Strawn et al. (2014) looked at the Salmonella diversity between two growing regions (South Florida and New York), it was identified that a high PFGE type diversity (Simpson's diversity index,0.90,0.02) was observed among Salmonella isolates across both regions and only three Pulsed-field gel electrophoresis (PFGE) types were shared between the two regions. Similarly, prior research has shown that specific Salmonella serovars may be associated with certain regions, such as Salmonella Newport repeatedly being isolated from the eastern shore of Virginia (Greene et al., 2008; Truitt et al., 2018) while several Salmonella strains, all with the same PFGE type, have been repeatedly isolated from the surface water in the same region in California (Gorski et al., 2013). These previous findings demonstrate that the diversity of Salmonella varies by space and sub-regions. These findings indicate that not only did Salmonella levels (under specific weather conditions) differ significantly between Florida regions, but that the composition and diversity of the Salmonella populations also differed substantially.

Bayesian regression indicates that microbial targets considered showed evidence of seasonal patterns (Figure 1, Tables 1, 2). The E. coli levels were 0.33 (95% CI = −0.54, −0.09) log₁₀ MPN/100 mL lower in spring than in fall; the differences between summer and fall, and winter and fall were negligible and of indeterminate direction based on PD and ROPE. Odds of total coliform levels being above the upper LOD was higher in summer (OR = 2.27; 95% CI = 0.60, 7.82), and lower in spring (OR = 0.51; 95% CI = 0.17, 1.59) and winter (OR = 0.28; 95% CI = 0.09, 0.85) based on the PS (probability of significance) index being ≥0.75, indicating the observed effect is both non-negligible and in the direction indicated by MAP. Based on the PS from the Bayesian regression, the likelihood of Salmonella detection was higher in spring, summer, and winter compared with fall, but that Salmonella levels (in Salmonella positive samples) were only higher in summer compared with fall (Table 2). Overall, the identification of seasonal patterns in water quality is consistent with the literature (Carter et al., 1987; Haley et al., 2009; Gorski et al., 2011; Cooley et al., 2014). For example, Haley et al. (2009) found that Salmonella concentrations in Georgia surface waters were significantly higher in the summer months compared with other seasons (P < 0.05). While other Florida studies either did not sample in Summer due to fewer crops or found no association with summer, the results from this and previous studies suggest an elevated risk during the summer months and therefore, future work will need to test this hypothesis.

Conditional inference tree analysis indicates that water temperature may drive seasonal trends in total coliform and Salmonella concentrations, and Salmonella detection (Figures 2, 3). Across all three trees, warmer water temperatures were associated with elevated levels or an increased likelihood of detecting the target (Figures 2, 3). This is consistent with past studies reporting seasonal trends in microbial water quality and linked elevated temperatures with an increased likelihood of detecting foodborne pathogens (Polo et al., 1999; Martinez-Urtaza et al., 2004; Haley et al., 2009; Huang et al., 2014; Antaki et al., 2016; Liu et al., 2018). Haley et al. (2009) and Antaki et al. (2016) both note higher Salmonella concentrations when water temperatures are warmer in Georgia, USA. Higher Salmonella detection in water samples collected seasonally from 34 locations along the Puzih River in Taiwan rates in the summer coincided with higher air and water temperatures (Huang et al., 2014). Since conditional trees can visualize hierarchical relationships, it is of interest that in the present study, the primary split for indicator organisms like total coliforms and generic E. coli were based on region and water type (Figures 3, 4), while the primary split for Salmonella was based on weather (Figure 2). Across all three microbial targets, splits for land use are below those for region and water type, which indicates that the land-use relationships identified here were based on region and or water type-specific (Figures 2–4). The tree analysis indicates that E. coli levels were highest in North Florida (lakes and rivers), and lowest in South Florida (canal) sites, with more than ~3% of the land (in a 366 m buffer) under developed cover. Conversely, when water temperatures were higher, Salmonella levels were higher regardless of region, and only at lower water temperatures did Salmonella levels begin to differ between regions (with levels in samples from South Florida being higher, compared with levels from North Florida; Figure 2). According to the tree analysis, the likelihood of Salmonella detection was only dependent on on-air temperature, with no differences between regions (air and water temperature are correlated). The relationships identified in the total coliform tree are more complex, with the highest levels being observed in South Florida (canals) when samples were collected with developed cover (in a 1,098 m buffer) above ~6% and the water temperature was above ~27°C; and in North Florida (river, lake) when samples were collected from water temperature above ~24°C and forest-wetland cover (in a 1,098 m buffer) was above ~63% (Figure 3). Overall, the fact that the first split in the Salmonella tree was based on weather indicates a stronger effect of weather than of region or water type on microbial water quality. The opposite conclusion can be made about land use in the present study, since (i) land use variables were all lower in the trees and thus dependent on specific weather and either region/water type conditions being met. These findings suggest that, for Florida, weather conditions may be useful for monitoring when food safety hazards are more likely to be present in agricultural waterways. Additional research is needed to confirm this finding as previous studies, have found varying pathogen-temperature relationships, and that those relationships are complex based on spatiotemporal factors (e.g., year or site of sample collection) (McEgan et al., 2013; Topalcengiz et al., 2017; Weller et al., 2020b). Weller et al. (2020b) compared pathogen levels between Arizona and New York found a positive relationship between temperature and likelihood of detecting Salmonella in Arizona, but a complex, polynomial relationship in New York.

The PSR proposed microbial standards for pre-harvest agricultural water and are under review at the time of writing this manuscript. Currently, the PSR standards require (i) that agricultural surface water used during production establish a microbial water quality profile (MWQP) using 20 samples collected over 2 to 4 years on a rolling basis, and (ii) that geometric mean (GM) and statistical threshold value (STV) of E. coli in this sample be ≤ 126 and ≤ 410 CFUs/100 mL, respectively (US Food Drug Administration, 2015). An MWQP for each site could not be created in the present study as we did not collect 20 samples over 2–4 years, we were able to compare the likelihood of Salmonella detection and Salmonella concentration in Salmonella-positive samples to E. coli levels in the same samples. E. coli levels in individual samples were more likely to exceed both PSR mean (OR = 5.43; 95% CI = 1.19, 52.52) and STV (OR = 4.22; 95% CI = 0.56, 47.22) cut-offs in North Florida than South Florida (Table 1). Salmonella levels in Salmonella positive samples were not associated with log₁₀ E. coli levels (as a continuous factor) or if the E. coli levels exceeded (or failed to exceed) the PSR cut-offs (Table 1). Odds of Salmonella detection was not associated with log₁₀ E. coli levels or if the levels exceeded the PSR mean cut-off; odds of Salmonella detection was 2.54 higher (95% CI = 0.66, 8.75) in samples that exceeded the PSR STV cut-off than in samples that met the cut-off. The association between E. coli levels and foodborne pathogen presence in agricultural water is consistent with some studies reporting an association and others failing to detect an association (Harwood et al., 2005; McEgan et al., 2013; Pachepsky et al., 2016; Truitt et al., 2018). McEgan et al. (2013) found that the presence and strength of the E. coli and Salmonella relationship differed between sites in the same region of Central Florida. E. coli was an adequate predictor of the presence of Salmonella in 150 mL samples in West Central Florida ponds; when E. coli populations were higher, Salmonella presence was more likely, but the relationship between populations differed between ponds (Havelaar et al., 2017; Topalcengiz et al., 2017). The results presented here support the conclusion from earlier studies that E. coli levels are unreliable as an indicator for the presence and concentration of microbial hazards in agricultural water. As E. coli is an indicator of fecal contamination and not an index for pathogen presence, this aligns with traditional convention.