Correcting On-the-Go Field Measurement-Coordinate Mismatch by Minimizing Nearest Neighbor Difference.

Many current precision agriculture applications involve on-the-go field measurements of soil and plant properties that require accurate georeferencing. Specific equipment configuration characteristics or data transmission, reception, or logging delays may cause a mismatch between the logged data and the GPS coordinates because of time and position lags that occur during data acquisition. We propose a simple coordinate translation along the measurement tracks to correct for such positional inaccuracies, based on the local travel speed and time lag, which is estimated by minimizing the average ln-transformed absolute difference with the nearest neighbors. The correction method is evaluated using electromagnetic induction soil-sensor data for different spatial measurement layouts and densities and by comparing variograms for raw and modified coordinates. Time lags of 1 s are shown to propagate into the spatial correlation structure up to lag distances of 10 m. The correction method performs best when repeated measurements in opposite driving directions are used and worst when measurements along parallel driving tracks are only repeated at the headland turns. In the latter case, the performance of the method is further improved by limiting the search neighborhood to adjacent measurement tracks. The proposed coordinate correction method is useful for improving the positional accuracy in a wide range of soil- and plant-sensing applications, without the need to grid the data first.

Assessment of a green roof practice using the coupled SWMM and HYDRUS models.

Green roof can mitigate urban stormwater and improve environmental, economic, and social conditions. Various modeling approaches have been effectively employed to implement a green roof, but previous models employed simplifications to simulate water movement in green roof systems. To address this issue, we developed a new modeling tool (SWMM-H) by coupling the stormwater management and HYDRUS-1D models to improve simulations of hydrological processes. We selected green roof systems to evaluate the coupled model. Rainfall-runoff experiments were conducted for a pilot-scale green roof and urban subbasin. Soil moisture in the green roof and runoff volume in the subbasin were simulated more accurately by using SWMM-H instead of SWMM. The scenario analysis showed that SWMM-H selected sandy loam for controlling runoff whereas SWMM recommended sand. In conclusion, SWMM-H could be a useful tool for accurately understanding hydrological processes in green roofs.

On the Information Content of Coarse Data with Respect to the Particle Size Distribution of Complex Granular Media: Rationale Approach and Testing.

The particle size distribution (PSD) of complex granular media is seen as a mathematical measure supported in the interval of grain sizes. A physical property characterizing granular products used in the Andreasen and Andersen model of 1930 is re-interpreted in Information Entropy terms leading to a differential information equation as a conceptual approach for the PSD. Under this approach, measured data which give a coarse description of the distribution may be seen as initial conditions for the proposed equation. A solution of the equation agrees with a selfsimilar measure directly postulated as a PSD model by Martín and Taguas almost 80 years later, thus both models appear to be linked. A variant of this last model, together with detailed soil PSD data of 70 soils are used to study the information content of limited experimental data formed by triplets and its ability in the PSD reconstruction. Results indicate that the information contained in certain soil triplets is sufficient to rebuild the whole PSD: for each soil sample tested there is always at least a triplet that contains enough information to simulate the whole distribution.

Temporal Stability of Escherichia coli Concentrations in Waters of Two Irrigation Ponds in Maryland.

Fecal contamination of water sources is an important water quality issue for agricultural irrigation ponds. Escherichia coli concentrations are commonly used to evaluate recreational and irrigation water quality. We hypothesized that there may exist temporally stable spatial patterns of E. coli concentrations across ponds, meaning that some areas mostly have higher and other areas mostly lower than average concentrations of E. coli To test this hypothesis, we sampled two irrigation ponds in Maryland at nodes of spatial grids biweekly during the summer of 2016. Environmental covariates-temperature, turbidity, conductivity, pH, dissolved oxygen, chlorophyll a, and nutrients-were measured in conjunction with E. coli concentrations. Temporal stability was assessed using mean relative differences between measurements in each location and averaged measurements across ponds. Temporally stable spatial patterns of E. coli concentrations and the majority of environmental covariates were expressed for both ponds. In the pond interior, larger relative mean differences in chlorophyll a corresponded to smaller mean relative differences in E. coli concentrations, with a Spearman's rank correlation coefficient of 0.819. Turbidity and ammonium concentrations were the two other environmental covariates with the largest positive correlations between their location ranks and the E. coli concentration location ranks. Tenfold differences were found between geometric mean E. coli concentrations in locations that were consistently high or consistently low. The existence of temporally stable patterns of E. coli concentrations can affect the results of microbial water quality assessment in ponds and should be accounted for in microbial water quality monitoring design.IMPORTANCE The microbial quality of water in irrigation water sources must be assessed to prevent the spread of microbes that can cause disease in humans because of produce consumption. The microbial quality of irrigation water is evaluated based on concentrations of Escherichia coli as the indicator organism. Given the high spatial and temporal variability of E. coli concentrations in irrigation water sources, recommendations are needed on where and when samples of water have to be taken for microbial analysis. This work demonstrates the presence of a temporally stable spatial pattern in the distributions of E. coli concentrations across irrigation ponds. The ponds studied had zones where E. coli concentrations were mostly higher than average and zones where the concentrations were mostly lower than average over the entire observation period, covering the season when water was used for irrigation. Accounting for the existence of such zones will improve the design and implementation of microbial water quality monitoring.

Spatial Patterns of Escherichia coli Concentrations in Sediment before and after High-Flow Events in a First-Order Creek.

Understanding spatial patterns of in freshwater sediments is necessary to characterize sediments as microbial reservoirs and to evaluate the impact of sediment resuspension on microbial water quality in watersheds. Sediment particle size distributions and streambed concentrations were measured along a 500-m-long reach of a first-order creek 1 d before and on Days 1, 3, 6, and 10 after each of two artificial high-flow events, with natural high-flow events also occurring within the sampling periods. Spatial variability of was greater in sediments than in water within any given sampling; however, variation between sampling days was greater for water than for sediment. The mean relative difference analysis revealed temporally stable patterns of concentrations in sediments. rich locations along the reach corresponded to areas with higher organic matter and fine particle contents. Although low ( < 0.5 d) or negative survival rates were observed at most locations along the reach during times where no precipitation was recorded, a small number of locations showed such large concentration increase that on average the survival rate remained positive at the reach scale. The studied creek appears to have hot spots of concentration increase, where conditions for populations to increase are much more favorable than in most other locations across the reach. The effect of this increase can be seen at the reach scale but is difficult to discern without individual sampling that is dense in space and time.

Escherichia coli Export from Manured Fields Depends on the Time between the Start of Rainfall and Runoff Initiation.

After rainfall or irrigation begins, surface-applied chemicals and manure-borne microorganisms typically enter the soil with infiltration until the soil saturates, after which time the chemicals and microbes are exported from the field in the overland flow. This process is viewed as a reason for the dependence of chemical export on the time between rainfall start and runoff initiation that has been documented for agricultural chemicals. The objective of this work was to observe and quantify such dependence for released from solid farmyard dairy manure in field conditions. Experiments were performed for 6 yr and consisted of manure application followed by an immediate simulated rainfall event and a second event 1 wk later. The nonlinearity of the release seen in laboratory and plot studies did not manifest itself in the field. The number of exported cells in runoff was proportional to rainfall depth after runoff initiation in each trial. The proportionality coefficient, termed export rate, demonstrated a strong dependence on the runoff delay time that could be approximated with the exponential decrease. The export rate decreased by one order of magnitude when the rainfall depth at runoff initiation increased from 18 to 42 mm. The same dependence could approximate data from the simulated rainfall event 1 wk after the manure application, assuming that the initial content in manure after 1 wk of weathering was 10% of the initial content. Overall, accounting for the dependence of manure-borne export on the runoff delay time should improve the accuracy of export predictions related to the assessment of agricultural practices on microbial water quality.

Development of a Nowcasting System Using Machine Learning Approaches to Predict Fecal Contamination Levels at Recreational Beaches in Korea.

Microbial contamination in beach water poses a public health threat due to waterborne diseases. To reduce the risk of exposure to fecal contamination, informing beachgoers in advance about the microbial water quality is important. Currently, determining the level of fecal contamination takes 24 h. The objective of this study is to predict the current level of fecal contamination (enterococcus [ENT] and ) using readily available environmental variables. Artificial neural network (ANN) and support vector regression (SVR) models were constructed using data from the Haeundae and Gwangalli Beaches in Busan City. The input variables included the tidal level, air and water temperature, solar radiation, wind direction and velocity, precipitation, discharge from the wastewater treatment plant, and suspended solid concentration in beach water. The dependence of fecal contamination on the input variables was statistically evaluated; precipitation, discharge from the wastewater treatment plant, and wind direction at the two beaches were positively correlated to the changes in the two bacterial concentrations ( < 0.01), whereas solar radiation was negatively correlated ( < 0.01). The performance of the ANN model for predicting ENT and at Gwangalli Beach was significantly higher than that of the SVR model with the training dataset ( < 0.05). Based on the comparison of residual values between the predicted and observed fecal indicator bacteria concentrations in two models, the ANN demonstrated better performance than SVR. This study suggests an effective prediction method to determine whether a beach is safe for recreational use.

Capturing Microbial Sources Distributed in a Mixed-use Watershed within an Integrated Environmental Modeling Workflow.

Many watershed models simulate overland and instream microbial fate and transport, but few provide loading rates on land surfaces and point sources to the waterbody network. This paper describes the underlying equations for microbial loading rates associated with 1) land-applied manure on undeveloped areas from domestic animals; 2) direct shedding (excretion) on undeveloped lands by domestic animals and wildlife; 3) urban or engineered areas; and 4) point sources that directly discharge to streams from septic systems and shedding by domestic animals. A microbial source module, which houses these formulations, is part of a workflow containing multiple models and databases that form a loosely configured modeling infrastructure which supports watershed-scale microbial source-to-receptor modeling by focusing on animal- and human-impacted catchments. A hypothetical application - accessing, retrieving, and using real-world data - demonstrates how the infrastructure can automate many of the manual steps associated with a standard watershed assessment, culminating in calibrated flow and microbial densities at the watershed's pour point.

Release from Streambed to Water Column during Baseflow Periods: A Modeling Study.

Streambed sediments can harbor large populations that are released into the water column during high-flow events. Few studies have been conducted on the rates of transfer from streambed sediment to water column in low-flow conditions in natural streams. The aim of this work was to apply the watershed-scale model SWAT (Soil and Water Assessment Tool) to a natural stream to evaluate the need to account for the release from streambed sediments during baseflow periods and to compare the results of simulating such a release by assuming predominantly passive transport, driven by groundwater influx, against simulations assuming predominantly active transport of random or chemotaxis-driven bacteria movement. concentrations in water during baseflow periods were substantially underestimated when release from the streambed was attributed only to streambed sediment resuspension. When considered in addition to the release due to sediment resuspension at high flows, the active and passive release assumptions provided 42 and 4% improvement, respectively, in the RMSE of logarithms of concentrations. Estimated fluxes to water column during the baseflow periods from June to November ranged from 3.3 × 10 colony-forming units (CFU) m d in the game land area to 1.4 × 10 CFU m d in the mixed pasture and cropland. Results demonstrate that release of from streambed sediments during baseflow periods is substantial and that water column concentrations are dependent on not only land management practices but also on in-stream processes.

Modeling the interannual variability of microbial quality metrics of irrigation water in a Pennsylvania stream.

Knowledge of the microbial quality of irrigation waters is extremely limited. For this reason, the US FDA has promulgated the Produce Rule, mandating the testing of irrigation water sources for many farms. The rule requires the collection and analysis of at least 20 water samples over two to four years to adequately evaluate the quality of water intended for produce irrigation. The objective of this work was to evaluate the effect of interannual weather variability on surface water microbial quality. We used the Soil and Water Assessment Tool model to simulate E. coli concentrations in the Little Cove Creek; this is a perennial creek located in an agricultural watershed in south-eastern Pennsylvania. The model performance was evaluated using the US FDA regulatory microbial water quality metrics of geometric mean (GM) and the statistical threshold value (STV). Using the 90-year time series of weather observations, we simulated and randomly sampled the time series of E. coli concentrations. We found that weather conditions of a specific year may strongly affect the evaluation of microbial quality and that the long-term assessment of microbial water quality may be quite different from the evaluation based on short-term observations. The variations in microbial concentrations and water quality metrics were affected by location, wetness of the hydrological years, and seasonality, with 15.7-70.1% of samples exceeding the regulatory threshold. The results of this work demonstrate the value of using modeling to design and evaluate monitoring protocols to assess the microbial quality of water used for produce irrigation.

Enrichment of stream water with fecal indicator organisms during baseflow periods.

Fecal indicator organisms (FIOs) are generally believed to be present in surface waters due solely to direct deposition of feces or through transport in runoff. However, emerging evidence points toward hyporheic exchange between sediment pore water and the overlying water column during baseflow periods as a source of FIOs is surface waters. The objective of this work was to (a) propose a mass balance-based technique for estimating changes of FIO concentrations in the same volume of water (or "slug") from the inlet to outlet of stream reaches in baseflow conditions and (b) to use such enumeration to estimate rate of the FIO release to stream water column. Concentrations of Escherichia coli (E. coli) and enterococci were measured in the slug while simultaneously monitoring the movement of a conservative tracer, Br that labeled the slug. Concentrations of E. coli in the slug were significantly larger (P = 0.035, P = 0.001, and P = 0.001, respectively) at the outlet reach in all three replications, while enterococci concentrations were significantly larger in two of three replications (P = 0.001, P < 0.001, and P = 0.602). When estimated without accounting for die-off in water column, FIO net release rates across replications ranged from 36 to 57 cells m-2 s-1 and 43 to 87 cells m-2 s-1 for E. coli and enterococci, respectively. These release rates were 5 to 20% higher when the die-off in water column was taken into account. No diurnal trends were observed in indicator concentrations. No FIO sources other than bottom sediment have been observed during the baseflow period. FIOs are released into stream water column through hyporheic exchange during baseflow periods.

Can E. coli or thermotolerant coliform concentrations predict pathogen presence or prevalence in irrigation waters?

An increase in food-borne illnesses in the United States has been associated with fresh produce consumption. Irrigation water presents recognized risks for microbial contamination of produce. Water quality criteria rely on indicator bacteria. The objective of this review was to collate and summarize experimental data on the relationships between pathogens and thermotolerant coliform (THT) and/or generic E. coli, specifically focusing on surface fresh waters used in or potentially suitable for irrigation agriculture. We analyzed peer-reviewed publications in which concentrations of E. coli or THT coliforms in surface fresh waters were measured along with concentrations of one or more of waterborne and food-borne pathogenic organisms. The proposed relationships were significant in 35% of all instances and not significant in 65% of instances. Coliform indicators alone cannot provide conclusive, non-site-specific and non-pathogen-specific information about the presence and/or concentrations of most important pathogens in surface waters suitable for irrigation. Standards of microbial water quality for irrigation can rely not only on concentrations of indicators and/or pathogens, but must include references to crop management. Critical information on microbial composition of actual irrigation waters to support criteria of microbiological quality of irrigation waters appears to be lacking and needs to be collected.

Rainfall-induced release of microbes from manure: model development, parameter estimation, and uncertainty evaluation on small plots.

A series of simulated rainfall-runoff experiments with applications of different manure types (cattle solid pats, poultry dry litter, swine slurry) was conducted across four seasons on a field containing 36 plots (0.75 × 2 m each), resulting in 144 rainfall-runoff events. Simulating time-varying release of Escherichia coli, enterococci, and fecal coliforms from manures applied at typical agronomic rates evaluated the efficacy of the Bradford-Schijven model modified by adding terms for release efficiency and transportation loss. Two complementary, parallel approaches were used to calibrate the model and estimate microbial release parameters. The first was a four-step sequential procedure using the inverse model PEST, which provides appropriate initial parameter values. The second utilized a PEST/bootstrap procedure to estimate average parameters across plots, manure age, and microbe, and to provide parameter distributions. The experiment determined that manure age, microbe, and season had no clear relationship to the release curve. Cattle solid pats released microbes at a different, slower rate than did poultry dry litter or swine slurry, which had very similar release patterns. These findings were consistent with other published results for both bench- and field-scale, suggesting the modified Bradford-Schijven model can be applied to microbial release from manure.

Survival of Manure-borne and Fecal Coliforms in Soil: Temperature Dependence as Affected by Site-Specific Factors.

Understanding pathogenic and indicator bacteria survival in soils is essential for assessing the potential of microbial contamination of water and produce. The objective of this work was to evaluate the effects of soil properties, animal source, experimental conditions, and the application method on temperature dependencies of manure-borne generic , O157:H7, and fecal coliforms survival in soils. A literature search yielded 151 survival datasets from 70 publications. Either one-stage or two-stage kinetics was observed in the survival datasets. We used duration and rate of the logarithm of concentration change as parameters of the first stage in the two-stage kinetics data. The second stage of the two-stage kinetics and the one-stage kinetics were simulated with the model to find the dependence of the inactivation rate on temperature. Classification and regression trees and linear regressions were applied to parameterize the kinetics. Presence or absence of two-stage kinetics was controlled by temperature, soil texture, soil water content, and for fine-textured soils by setting experiments in the field or in the laboratory. The duration of the first stage was predominantly affected by soil water content and temperature. In the model dependencies of inactivation rates on temperature, parameter estimates were significantly affected by the laboratory versus field conditions and by the application method, whereas inactivation rates at 20°C were significantly affected by all survival and management factors. Results of this work can provide estimates of coliform survival parameters for models of microbial water quality.

Irrigation waters and pipe-based biofilms as sources for antibiotic-resistant bacteria.

The presence of antibiotic-resistant bacteria in environmental surface waters has gained recent attention. Wastewater and drinking water distribution systems are known to disseminate antibiotic-resistant bacteria, with the biofilms that form on the inner-surfaces of the pipeline as a hot spot for proliferation and gene exchange. Pipe-based irrigation systems that utilize surface waters may contribute to the dissemination of antibiotic-resistant bacteria in a similar manner. We conducted irrigation events at a perennial stream on a weekly basis for 1 month, and the concentrations of total heterotrophic bacteria, total coliforms, and fecal coliforms, as well as the concentrations of these bacterial groups that were resistant to ampicillin and tetracycline, were monitored at the intake water. Prior to each of the latter three events, residual pipe water was sampled and 6-in. sections of pipeline (coupons) were detached from the system, and biofilm from the inner-wall was removed and analyzed for total protein content and the above bacteria. Isolates of biofilm-associated bacteria were screened for resistance to a panel of seven antibiotics, representing five antibiotic classes. All of the monitored bacteria grew substantially in the residual water between irrigation events, and the biomass of the biofilm steadily increased from week to week. The percentages of biofilm-associated isolates that were resistant to antibiotics on the panel sometimes increased between events. Multiple-drug resistance was observed for all bacterial groups, most often for fecal coliforms, and the distributions of the numbers of antibiotics that the total coliforms and fecal coliforms were resistant to were subject to change from week to week. Results from this study highlight irrigation waters as a potential source for antibiotic-resistant bacteria, which can subsequently become incorporated into and proliferate within irrigation pipe-based biofilms.

Solid Manure As a Source of Fecal Indicator Microorganisms: Release under Simulated Rainfall.

Understanding and quantifying microbial release from manure is a precondition to estimation and management of microbial water quality. The objectives of this work were to determine the effects of rainfall intensity and surface slope on the release of Escherichia coli, enterococci, total coliforms, and dissolved chloride from solid dairy manure and to assess the performance of the one-parametric exponential model and the two-parametric Bradford-Schijven model when simulating the observed release. A controlled-intensity rainfall simulator induced 1 h of release in runoff/leachate partitioning boxes at three rainfall intensities (30, 60, and 90 mm h(-1)) and two surface slopes (5% and 20%). Bacterial concentrations in initial release were more than 1 order of magnitude lower than their starting concentrations in manure. As bacteria were released, they were partitioned into runoff and leachate at similar concentrations, but in different volumes, depending on slope. Bacterial release occurred in two stages that corresponded to mechanisms associated with release of manure liquid- and solid-phases. Parameters of the two models fitted to the bacterial release dependencies on rainfall depth were not significantly affected by rainfall intensity or slope. Based on model performance tests, the Bradford-Schijven model is recommended for simulating bacterial release from solid manure.

Release and Removal of Microorganisms from Land-Deposited Animal Waste and Animal Manures: A Review of Data and Models.

Microbial pathogens present a leading cause of impairment to rivers, bays, and estuaries in the United States, and agriculture is often viewed as the major contributor to such contamination. Microbial indicators and pathogens are released from land-applied animal manure during precipitation and irrigation events and are carried in overland and subsurface flow that can reach and contaminate surface waters and ground water used for human recreation and food production. Simulating the release and removal of manure-borne pathogens and indicator microorganisms is an essential component of microbial fate and transport modeling regarding food safety and water quality. Although microbial release controls the quantities of available pathogens and indicators that move toward human exposure, a literature review on this topic is lacking. This critical review on microbial release and subsequent removal from manure and animal waste application areas includes sections on microbial release processes and release-affecting factors, such as differences in the release of microbial species or groups; bacterial attachment in turbid suspensions; animal source; animal waste composition; waste aging; manure application method; manure treatment effect; rainfall intensity, duration, and energy; rainfall recurrence; dissolved salts and temperature; vegetation and soil; and spatial and temporal scale. Differences in microbial release from liquid and solid manures are illustrated, and the influential processes are discussed. Models used for simulating release and removal and current knowledge gaps are presented, and avenues for future research are suggested.

Colloid filtration in surface dense vegetation: experimental results and theoretical predictions.

Understanding colloid and colloid-facilitated contaminant transport in overland flow through dense vegetation is important to protect water quality in the environment, especially for water bodies receiving agricultural and urban runoff. In previous studies, a single-stem efficiency theory for rigid and clean stem systems was developed to predict colloid filtration by plant stems of vegetation in laminar overland flow. Hence, in order to improve the accuracy of the single-stem efficiency theory to real dense vegetation system, we incorporated the effect of natural organic matter (NOM) on the filtration of colloids by stems. Laboratory dense vegetation flow chamber experiments and model simulations were used to determine the kinetic deposition (filtration) rate of colloids under various conditions. The results show that, in addition to flow hydrodynamics and solution chemistry, steric repulsion afforded by NOM layer on the plants stem surface also plays a significant role in controlling colloid deposition on vegetation in overland flow. For the first time, a refined single-stem efficiency theory with considerations of the NOM effect is developed that describes the experimental data with good accuracy. This theory can be used to not only help construct and refine mathematical models of colloid transport in real vegetation systems in overland flow, but also inform the development of theories of colloid deposition on NOM-coated surfaces in natural, engineered, and biomedical systems.

Multiscale spatiotemporal variability of fecal indicator bacteria and associated particle size distributions in the sandy bottom sediments of a Pennsylvania creek.

Concentrations of the fecal indicator bacteria (FIB) Escherichia coli and enterococci are used to assess microbial impairment in irrigation and recreation water sources. Although the FIB concentrations' variability at large temporal scales, such as seasons, and large spatial scales encompassing different land use has been studied, the knowledge about smaller scale variability remains sparse. This work aimed to research the small-scale variability of E. coli and enterococci in a montane creek with sandy bottom sediments. Sediment samples were collected weekly for a year in triplicate at sampling sites in a forested headwater, an agricultural area, and a mixed urban-agricultural area. The average weekly change in concentrations was from two times at the forested site to five times at the urban-agricultural site. Mean relative deviations from averages across sampling locations increased from -25% at the forested site to 45% at the urban-agricultural site. This trend was also observed separately over the cold and warm seasons. Over a week without precipitation, E. coli concentrations decreased on average by 20% in warm period and by 45% in cold period; the enterococci concentration declined by 12% in both cold and warm periods. The sediment particle size distributions were significantly different among the three sites and between the cold and warm seasons. Rankings of sediment fine mass fractions and FIB concentrations were positively correlated at two of three sampling sites in more than 70% of observation dates. The results of this work indicate the need to evaluate the uncertainty of sediment FIB concentrations before designing sediment FIB monitoring quality.

Using machine learning models to estimate Escherichia coli concentration in an irrigation pond from water quality and drone-based RGB imagery data.

The rapid and efficient quantification of Escherichia coli concentrations is crucial for monitoring water quality. Remote sensing techniques and machine learning algorithms have been used to detect E. coli in water and estimate its concentrations. The application of these approaches, however, is challenged by limited sample availability and unbalanced water quality datasets. In this study, we estimated the E. coli concentration in an irrigation pond in Maryland, USA, during the summer season using demosaiced natural color (red, green, and blue: RGB) imagery in the visible and infrared spectral ranges, and a set of 14 water quality parameters. We did this by deploying four machine learning models - Random Forest (RF), Gradient Boosting Machine (GBM), Extreme Gradient Boosting (XGB), and K-nearest Neighbor (KNN) - under three data utilization scenarios: water quality parameters only, combined water quality and small unmanned aircraft system (sUAS)-based RGB data, and RGB data only. To select the training and test datasets, we applied two data-splitting methods: ordinary and quantile data splitting. These methods provided a constant splitting ratio in each decile of the E. coli concentration distribution. Quantile data splitting resulted in better model performance metrics and smaller differences between the metrics for both the training and testing datasets. When trained with quantile data splitting after hyperparameter optimization, models RF, GBM, and XGB had R2 values above 0.847 for the training dataset and above 0.689 for the test dataset. The combination of water quality and RGB imagery data resulted in a higher R2 value (>0.896) for the test dataset. Shapley additive explanations (SHAP) of the relative importance of variables revealed that the visible blue spectrum intensity and water temperature were the most influential parameters in the RF model. Demosaiced RGB imagery served as a useful predictor of E. coli concentration in the studied irrigation pond.