Next Article in Journal
Response Surface Methodology–Artificial Neural Network (RSM-ANN) Approach to Optimise Photocatalytic Degradation of Levofloxacin Using Graphene Oxide-Doped Titanium Dioxide (GO-TiO2)
Previous Article in Journal
Evaluation of Water Resource Carrying Capacity and Analysis of Driving Factors in the Dadu River Basin Based on the Entropy Weight Method and CRITIC Comprehensive Evaluation Method
Previous Article in Special Issue
BloomSense: Integrating Automated Buoy Systems and AI to Monitor and Predict Harmful Algal Blooms
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Water
Volume 17
Issue 16
10.3390/w17162361
water-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Investigating the Relationship Between Microcystin Concentrations and Water Quality Parameters in Three Agricultural Irrigation Ponds Using Random Forest

by
Jaclyn E. Smith
1,2,3,*,
James A. Widmer
4,
Matthew D. Stocker
1,
Jennifer L. Wolny
5,
Robert L. Hill
2 and
Yakov Pachepsky
1
1
Environmental Microbial Food Safety Laboratory, United States Department of Agriculture—Agricultural Research Service, Beltsville, MD 20705, USA
2
Environmental Science and Technology Department, University of Maryland, College Park, MD 20742, USA
3
Oak Ridge Institute for Science and Education, Oak Ridge, TN 37830, USA
4
Departments of Food Science and Technology, University of Georgia, Athens, GA 30602, USA
5
Office of Applied Microbiology and Technology, Human Foods Program, US Food and Drug Administration, College Park, MD 20740, USA
*
Author to whom correspondence should be addressed.
Water 2025, 17(16), 2361; https://doi.org/10.3390/w17162361
Submission received: 27 June 2025 / Revised: 1 August 2025 / Accepted: 2 August 2025 / Published: 8 August 2025
(This article belongs to the Special Issue New World: Advancing Water Applications Through Machine Learning and Artificial Intelligence)
Browse Figures
Versions Notes

Abstract

Cyanotoxins in agricultural waters pose a human and animal health risk. These toxins can be transported to nearby crops and soil during irrigation practices; they can remain in the soil for extended periods and be adsorbed by root systems. Additionally, in livestock watering ponds, cyanotoxins pose a direct ingestion risk. This work evaluated the performance of the random forest algorithm in estimating microcystin concentrations using eight in situ water quality measurements at one active livestock water pond and two working irrigation ponds in Georgia and Maryland, USA. Measurements of microcystin along with eight in situ-sensed water quality parameters were used to train and test the machine learning model. The models performed better at the Georgia ponds compared to the Maryland pond, and interior models performed better than nearshore or whole-pond models. The most important variables for microcystin prediction were water temperature and phytoplankton pigments. Overall, the random forest algorithm(RF), augmented with a ‘trainControl’ function to perform repeated cross validations, was able to explain 40% to 70% of the microcystin concentration variation in the three agricultural ponds. Water quality measurements showed potential to aid water monitoring/sampling design by predicting the microcystin concentrations in the studied ponds by using readily available and easy to collect in situ data.

1. Introduction

Cyanotoxins are toxic compounds which are produced by certain species of freshwater cyanobacteria, also referred to as blue green algae. When harmful cyanobacteria bloom, these naturally occurring toxins can contaminate water bodies such as lakes, ponds, rivers, reservoirs, and wetlands, posing significant risks to aquatic ecosystems, animal health, and human health [1]. Microcystin is the most prevalent cyanotoxin worldwide and is among the most hazardous of the hepatotoxic cyanotoxins [2,3]. In an agricultural setting, microcystins can pose a direct ingestion risk to livestock through drinking water sources [4,5] and feed [6,7]. Moreover, microcystins can be transported to fields during irrigation where they can remain in the soils for extended periods of time [8,9,10] and can be taken up by the root systems of crops [11,12].
Microcystins are one of the most widespread freshwater toxins across the globe with over 300 different congeners identified [13,14]. Microcystins are commonly monitored for in lakes [15,16], reservoirs [17,18], and rivers [19,20] as part of research and resource management programs. Several countries have adopted the World Health Organization’s provisional guideline concentration of 1 ppb microcystin-LR in drinking waters [21,22], and several US states have adopted the United States Environmental Protection Agency’s health advisory limit of 1.6 ppb microcystin in drinking waters [23,24]. However, at the federal level, there are currently no regulations or guidelines for cyanobacteria/cyanotoxins in drinking waters, recreation waters, or agricultural waters, though microbial water quality standards exist for water “that is intended to, or likely to, contact the harvestable portion of covered produce or food-contact surfaces” through the Food Safety Modernization Act [25], studies monitoring the impacts and remediation of microcystin, as well as other cyanotoxins and water quality parameters, are underway in waterbodies >4 ha to provide data that can be used to support Section 305B of the Clean Water Act [26,27], but due to their small size, national assessments in agricultural irrigation ponds and livestock watering ponds are limited.
Concentrations of microcystin exhibit significant spatial and temporal variability in waterbodies, which can complicate effective monitoring design and management. Spatially, microcystin concentrations can vary due to wave action [28,29], weather conditions such as wind [30,31], nutrient inputs [32,33], water column stratification [34], cyanobacteria colony morphology [35], and phytoplankton community structure [36], all of which can lead to zones, or hot spots, of microcystin within a waterbody. Earlier work on Ponds 1 and 2 [31], and Pond 3 [37] determined substantial spatiotemporal variability of microcystin concentrations, with concentrations being greatest along the shoreline. Temporally, microcystin concentrations can vary due to factors such as weather events [38,39], seasonal changes [40], and cyanobacteria bloom dynamics [41], with peak concentrations of cyanobacteria and cyanotoxins often occurring during warmer months and eutrophic periods [42,43].
It has been well-documented that harmful cyanobacteria blooms (cyanoHABs) [1,44], and Microcystis blooms specifically [45], are closely associated with water quality parameters, and numerous models have been developed based on these associations [17,46]. There is growing evidence that microcystin concentrations in freshwater systems are also closely associated with various water quality variables, such as nutrient levels, especially nitrogen and phosphorus [47,48], temperature [49,50], turbidity [51], and pH [16,52]. However, the development of models that can predict cyanotoxins has lagged behind those predicting blooms, likely owing to the constraints of datasets needed to elucidate the complex interactions that precipitate toxin formation. Understanding the relationships between water quality parameters and microcystin may allow for the development of predictive models that can be used to mitigate the risk of microcystin in water sources.
Recently, spatially dense water quality assessments have become more readily available with advancements in satellite imagery, in situ sensing, and aerial and underwater drone technology [53,54,55,56,57]. Traditional water sampling methods, which often rely on manual water collection at discrete locations, have been supplemented by autonomous or remotely operated sensors that provide real-time, high-resolution spatial and temporal data [56,58,59]. Satellite imagery and drones equipped with multispectral and hyperspectral imaging can detect cyanoHABsacross a water body, and coupled with ecological knowledge of the waterways, may indicate times and locations where cyanotoxin concentrations could be elevated [53,60,61,62]. Meanwhile, in situ sensors deployed on buoys or autonomous underwater vehicles can continuously monitor key water quality parameters such as temperature, turbidity, chlorophyll, and dissolved oxygen, providing in-depth temporal water quality data allowing for better predictions of cyanobacteria and microcystin concentrations [57,63,64]. These technologies enhance monitoring capabilities, allowing for rapid detection of potential toxin hot spots and more effective resource management.
Despite the advances in monitoring, understanding the relationship between measurable water quality parameters and cyanoHABs is challenging. The complex nonlinear relationships between easily measured water quality and microcystin concentrations are not well described by traditional statistical methods. The use of machine learning allows for these multifaceted relationships to be modelled where traditional statistics would fail to adequately predict cyanotoxin concentrations at levels of concern [65]. When assessing cyanobacterial water quality, the random forest (RF) algorithm often outperforms other machine learning models, including Extreme Gradient Boosting (XGB), K-Nearest Neighbors (KNN), Stochastic Gradient Boosting (SGB), and Support Vector Machines (SVM), because its ensemble structure of individual decision trees can effectively predict phytoplankton, including cyanobacteria, biomass and relationships with environmental factors [66,67,68,69], though Hwang et al. [70] cautioned that in large bodies of water the best performing model was sampling location dependent due to confounding hydrologic factors. Additionally, RF algorithms have a built-in mechanism for preventing overfitting and combatting multicollinearity in a dataset [71,72]. RFs are often used in the scientific community due to their ease of use, robustness against multicollinearity, minimal tuning requirements, and predictive power [73,74].
While there have been numerous studies documenting the correlations of microcystin concentrations and water quality measurements such as phycocyanin [75], chlorophyll [76,77], turbidity [78], temperature [79], nutrients [80], and pH [52], only a few studies have compared the spatial distributions of microcystin to the spatial distributions of water quality [31,81,82]. Additionally, there are numerous studies using machine learning to predict cyanobacteria in various types of large waterbodies [68,83], but studies on the prediction of toxin concentrations specifically in agricultural ponds less than 4 ha are lacking, primarily because these small waterbodies fall outside of federal oversight [84] and thus lack the spatially- and temporally-robust datasets necessary to develop models using machine learning [85].
Effective monitoring strategies in agricultural waters must account for the variability and natural fluctuations in microcystin to minimize exposure risk to livestock and crops but also should be easy enough to be implemented by resource managers who may not have routine access to conduct laboratory-based assessments. Recently, Mu et al. [86] and Singh et al. [82] called for increased efforts in monitoring cyanotoxins in agricultural waters to improve safeguards to public health. With this in mind, we have focused on assessing the relationship between microcystin concentrations and water quality data that can be collected by hand-held sondes using the RF algorithm in agricultural irrigation and livestock watering ponds of Georgia and Maryland, USA.

2. Materials and Methods

2.1. Monitoring of Agricultural Ponds

2.1.1. Georgia Field Sites

Water sampling was performed at an agricultural pond on each of two farms in southeastern Georgia, USA. These agricultural ponds are referred to as Pond 1 (Figure 1A) and Pond 2 (Figure 1B) for anonymity. Sampling occurred monthly from June 2022 to October 2023 for a total of 17 sampling dates for each pond. Surface water samples (500 mL) were collected along a fixed sampling grid consisting of interior and nearshore sampling locations, which remained the same throughout the 17-month study.
Pond 1 is approximately 1.6 ha with an average depth of 1.14 m and is currently utilized as a livestock watering pond for a herd of approximately 50 beef cattle and occasionally for irrigation of surrounding fields. Pond 1 contained 18 sampling locations, from which 306 samples were collected over the course of the study. Pond 2 has an area of approximately 3.2 ha with an average depth of 1.17 m and is used for irrigating corn and cotton crops. There were 16 sampling locations in Pond 2, from which 272 total samples were collected over the course of the study.

2.1.2. Maryland Field Sites

Water sampling was conducted at Pond 3 during the growing seasons of 2022 through 2024 at a commercial farm located in central Maryland (location and farm name withheld at the request of the owners). Pond 3 measures approximately 4 ha and has an average depth of 1.22 m. Pond 3 is used to irrigate leafy greens, apple trees, blackberries, blueberries, pumpkins, and various other pick-your-own produce. Sampling of the pond was conducted in a grid-like fashion with 10 sampling locations in 2022 (Figure 2A) and 18 sampling locations in 2023 and 2024 (Figure 2B). The pond is creek-fed from the north, and water that enters the pond passes through two settling ponds before entering the larger pond. Areas in the northern and northwestern portions of the pond were not sampled due to the shallowness of the water.

2.2. In Situ Measurements

Sampling locations remained the same for every sampling date. Water samples were taken at a depth of 0–15 cm. Bank samples were taken with a 500 mL hand grab sampler at approximately 1.5 m from the shoreline. Interior samples were taken from a small boat, GPS was utilized to provide consistency of sampling locations between different sampling dates. Water quality parameters were measured in situ at the same time and place using a YSI EXO-2 sonde (Yellow Springs Instruments, Yellow Springs, OH, USA). The sonde was used to measure a total of eight parameters: temperature (°C, TEMP), pH, specific conductivity (μS cm−1, SPC), fluorescent dissolved organic matter (relative fluorescent units [RFU], FDOM), dissolved oxygen (mg L−1, DO), chlorophyll a (RFU, CHL), phycocyanin (RFU, BGA), and turbidity (nephelometric turbidity units [NTU], NTU). After collection, samples were immediately placed into a cooler with ice packs to keep samples as close to the original water temperature as possible.

2.3. Microcystin Measurements

Samples allocated for microcystin (ppb, MC) analysis went through three freeze–thaw cycles prior to being analyzed using ELISA-ADDA microcystin test kits (PN#520011, Gold Standard Diagnostics, Warminster, PA, USA). Sample dilutions were made with the laboratory reagent blank solution provided in the kit, either prior to analysis or were rerun with proper dilutions to fit within the ELISA kit’s standard curve per the manufacturer’s recommendations. Each ELISA measurement was performed in duplicate and analyzed on a microplate photometer (Model #4300, Gold Standard Diagnostics, Warminster, PA, USA) with absorbance recorded at 450 nm according to kit instructions.

2.4. Machine Learning Algorithms and Performance Metrics

After completing a baseline performance comparison between several ML models (RF, KNN, SGB, SVM, XGB; Supplemental Table S1), the RF algorithm was utilized to predict MC concentrations and the most influential parameters for MC concentrations as it demonstrated the best performance when tuned with repeated cross validations. The ML algorithms were then ranked respect to highest R2 value for each pond and location area (entire pond, interior, nearshore). The RF algorithm was on average ranked the highest amongst the ML models tested for both training and testing R2 values (Supplemental Table S2). The RF algorithm creates predictions by generating multiple decision tress and then averaging their predictions. RF has a built-in mechanism for preventing overfitting of the model due to multicollinearity. The default number of trees was 500. Each dataset was split into two separate sub-datasets: 75% for training, and 25% for testing purposes. All RF analyses were performed in RStudio version 2025.05.0+496 using the ‘caret’ (classification and regression training) package [72]. This package allowed for the use of the ‘trainControl’ function in which a repeated cross validation (10-fold, 5 repeats) was performed during the fitting of the RFs to the dataset. The data showing the change in model performance with the addition of repeated cross validation when compared to the base algorithm can be found in Supplemental Table S3. All results from the repeated cross validations were averaged. By default, the training function automatically tunes the model and chooses the mtry value with the best performance. The metric used to evaluate the RF algorithm performance was the average coefficient of determination ( R 2 ¯ ). The averaging of R 2 was performed across the 50 values obtained through the cross validation and repeats. Various RF models were developed in this study. For each pond, three models were developed: one which included all sampling data, another which only included interior sampling data, and a third which only used nearshore sampling data.

2.5. Data Processing, Software, and Statistics

Prior to analysis, significant MC concentration outliers were removed from the datasets. Outliers were determined using the generalized extreme studentized deviate. Statistical analyses were performed in PAST software v4.15 [87]. Site maps with sampling locations were created using QGIS v3.22 (OSGeo, Basel, Switzerland). All figures and graphs were created using Sigmaplot v13 (Systat Software, San Jose, CA, USA).

3. Results

3.1. Summary of Monitoring Data

Descriptive statistics, consisting of minimums, maximums, medians, means, standard deviations, standard errors, skewness, and kurtosis, for all measured water quality parameters in Ponds 1, 2, and 3 can be found in Table 1. Pond 1 had the highest averages for MC, CHL, BGA, and NTU. The highest averages for DO, SPC, and PH were observed in Pond 2. The highest averages for FDOM and TEMP were observed in Pond 3. The highest maximum values for MC were documented in Pond 1, and the lowest maximum values were in Pond 3. Pond 2 had the highest median value of MC at 5.46 ppb. Only Pond 3 had samples where MC was not detected. The largest standard deviations were seen for SPC and NTU for Pond 1 and 3. At Pond 2 the largest standard deviations were for FDOM, SPC, and NTU; and at Pond 3 the largest standard deviations were for NTU and SPC. In general, skewness measurements were positive (16 out of 27 observations), meaning the data were right skewed with more low values and fewer larger values. MC values for all three ponds had positive skewness measurements. The highest observation of negative skewness values was seen in Pond 1 with FDOM, DO, SPC, pH, and TEMP measurements (5 out of 9 parameters) being negatively skewed or having more values in the higher concentrations than lower concentrations. Kurtosis measures of zero indicate a normal distribution, whereas kurtosis values above zero have a leptokurtic distribution, with more of a peak, and heavier tails and kurtosis values below zero are a platykurtic distribution, which is flatter with lighter tails; the kurtosis of normal distributions should be in a range of −2 to 2 [88]. The majority of the measured parameters fall within this range (17 out of 27). The parameters with leptokurtic kurtosis values are SPC (Ponds 1, 2, and 3), BGA (Pond 2 and 3), CHL (Pond 3), NTU (Pond 2 and 3), and pH (Pond 2). No kurtosis values less than −2 were reported, indicating no parameters were considered to have platykurtic distributions.
Coefficients of variations (CVs) for each pond and sampling date can be found in Supplemental Figure S1. The highest CV values at Ponds 1 and 2 were for MC, CHL, BGA, and NTU, with MC typically having the highest variations within each sampling date. At Pond 1 the highest CVs for MC displayed seasonal trends, with the highest variations of MC being in the months of May through October. This is different from what was seen at Pond 2 where the highest CVs for MC were seen in the winter months, December through February. Pond 3 was not sampled in the winter months. At Pond 3, the highest CVs for MC were seen in June and September. The variability in TEMP, pH, SPC, FDOM, and DO was generally very low for all three ponds throughout the sampling periods.
The cumulative probability distribution functions of MC are shown in Figure 3. For each pond, a distribution is shown for each sampling year, interior samples, nearshore samples, and all samples combined. The Kolmogorov–Smirnov test indicated significant (p < 0.05) differences in the distributions between years 2022 and 2023 at both Georgia Ponds (Pond 1 < 0.001 and Pond 2 = 0.003). Distributions between nearshore and interior sampling locations were not significantly different in the Georgia ponds. At Pond 3 (Maryland), distributions between nearshore and interior sampling locations were significantly different (p = 0.023). Additionally, at Pond 3, a comparison of distributions for the 2022–2024 sampling seasons indicated that the distributions of 2023 compared to 2024 (p < 0.001) and 2022 compared to 2024 (p = 0.003) were significantly different, but not the distributions of 2022 compared to 2023 (p = 0.154).

3.2. Spearman Correlations

The highest reported Spearman rank correlations for each measured variable in all three ponds for both nearshore and interior samples are shown in Figure 4 and Figure 5, respectively. The Spearman rank correlations for each individual pond’s interior and nearshore samples can be seen in Supplemental Figures S2–S4, for Ponds 1, 2, and 3, respectively. In general, Pond 2 had the strongest correlations between measured parameters for both interior and nearshore samples. Across all three ponds, the strongest correlations for MC were with CHL (Pond 1), BGA (Pond 2 and 3), and NTU (Pond 2 and 3) for both interior and nearshore sample sets. Phytoplankton pigments (CHL and BGA) had strong positive correlations with each other (r = 0.702, Pond 3) and with NTU (CHL r = 0.676, Pond 3; BGA r = 0.867, Pond 2) for interior samples. BGA also correlated strongly with DO (r = 0.651, Pond 2) for interior samples. For nearshore samples, CHL was only moderately correlated with BGA (r = 0.406, Pond 3), and correlations with other measured parameters were less (r < 0.400). Nearshore BGA samples were highly correlated with NTU, pH, and DO (r > 0.600, Pond 2).

3.3. Random Forest Applications

3.3.1. R2 Differences in Training and Testing Datasets

Training and testing datasets for all, interior, and nearshore data were used to assess RF model performance in using water quality variables to predict MC concentrations. Performance (R2) of the RF models for all three ponds is reported in Table 2 and RMSE values are reported in Supplemental Table S4. The plots showing predicted MC concentrations and measured MC concentrations using the three models for each of the ponds are shown in Supplemental Figure S5. Training performance varied between ponds and datasets (all, interior, nearshore). The R2 values for training datasets ranged from 0.358 to 0.727, and the R2 values for the testing datasets ranged from 0.388 to 0.695. The RMSE values for training datasets ranged from 0.64 to 3.71 and the RMSE values for the testing datasets ranged from 0.50 to 3.97. In all cases, R2 values for interior models had better performance metrics than nearshore models and entire-pond models. Similarly, the lowest RMSE values were for the interior models compared to the nearshore and entire pond models. Pond 2 had the highest overall average performance when compared to Pond 1, and Pond 3, with an average training R2 of 0.660 and testing R2 of 0.650. Pond 3 had the lowest performance of the three ponds, with an average training R2 of 0.462 and an average testing R2 of 0.474. In general, testing R2 values were very similar to the training R2 (Training R2 ± 0.05), indicating that the RF model was not overfitting [89,90].

3.3.2. Similarities and Differences in Important Variables

Important predictors for MC concentrations varied among ponds. However, TEMP was an important predictor in approximately 30% of instances. Other common important predictors were CHL (10%), BGA (19%), NTU (19%), and FDOM (10%). For Pond 1, the top three predictors were always CHL, TEMP, and FDOM, although in varying orders of importance depending on which model was run. With the exception of the Pond 3 nearshore model, the most important predictor for each model was either a photosynthetic pigment (CHL or BGA) or TEMP. The only input parameter to not show up as a predictor in any model was DO. Additionally, SPC was only listed as an important predictor once and was listed as the third most important predictor. NTU was only an important predictor for Pond 2 and Pond 3, and FDOM was only an important predictor for Pond 1.

4. Discussion

CyanoHABs in freshwater systems can be closely tied to certain water quality variables [47,49]. However, understanding how these environmental conditions influence cyanoHABs and microcystin concentrations are less straightforward. While much work has been conducted to try to elucidate these relationships in bodies of water that are primarily used as drinking water sources and for recreational activities, much less work has been performed on agricultural waters. In this study, we used RF to examine the relationship between water quality parameters and microcystin concentrations from two agricultural irrigation ponds in Georgia and one in Maryland that are known to experience blooms of Microcystis species [31,37]. RF was selected as our machine learning model due to its predictive power despite complex and nonlinear ecological relationships [91], (see also performance statistics in Supplemental Table S1) and previous successes in using RF to model cyanobacteria populations [66,92], including within small agricultural irrigation ponds [93]. Use of multiple ML models was not considered in this study due to the small size of agricultural ponds studied (~1 ha), but it should be noted that waterbody size can impact model performance due to the co-mingling of biologic and hydrologic factors.
Since microcystins are direct exudates of cyanobacteria, it was expected that the pigments CHL and BGA would be top predictors for MC. All cyanobacteria species contain the photosynthetic pigments, chlorophyll and phycocyanin, the latter being a pigment known to be only produced by cyanobacteria. Phycocyanin has been utilized as a bioindicator for cyanoHABs, since higher concentrations of phycocyanin often correlate with increases in cyanobacteria biomass, and therefore, a higher risk for elevated microcystin concentrations can be inferred [94,95,96]. Chlorophyll is routinely used as a bioindicator of total phytoplankton biomass and has been useful in determining the presence of cyanobacteria blooms in large bodies of water [53,97,98]. While neither phycocyanin nor chlorophyll measurements can directly inform about cyanotoxin concentrations [53,95], correlations between phycocyanin, chlorophyll and microcystin concentrations can help inform resource managers about health risks. Previously, this research group found high correlations between CHL, BGA, and MC concentrations in agricultural irrigation waters in Georgia, both temporally and spatially [31]. The RF modelling performed here confirmed these earlier findings and expanded this relationship to include Maryland waters.
Among the top predictors, TEMP was one of the most influential for the prediction of MC concentrations. There are plenty of studies indicating that water temperature plays a large role in the growth of cyanobacteria [99,100] and the production or release of microcystin [101,102]. The strength of this model is illustrated by strong correlations between MC concentrations and TEMP in Pond 1 and Pond 2. As described in Smith et al. [31], high MC concentrations were detected in the summer months in Pond 1; however, in Pond 2, the highest MC concentrations were detected in the winter. This rules out seasonality as a driving factor behind these Microcystis blooms. Instead, the model is capturing an elevated water temperature-driven bloom event typical of summer [100], as well as a winter event during which pond waters warmed above 10 °C, the temperature at which winter benthic Microcystis populations can increase growth and toxin production activities [103]. Year-round sampling in Maryland agricultural ponds would be needed to test the robustness of this RF model to predict winter cyanoHAB and their toxins due to the inherent climate differences between the mid-Atlantic region and the southeast United States [104] that could necessitate regional tuning of the model.
In general, our RF interior models performed better than nearshore models in predicting MC concentrations, which is likely due to the fact the nearshore regions tended to be more variable at these ponds [31]. While studies are limited for modelling MC in nearshore and interior locations, there are examples of modelling cyanobacteria populations. One study in Lake Okeechobee, Florida, modelled cyanobacteria blooms using a coupled physical–biogeochemical model and optical data from SeaPrism [105] and noted that most of the nearshore blooms were missed by the model [106]. In another study by Lofton et al. [107], where cyanobacteria density was predicted using near-term forecasts at Lake Sunapee in New Hampshire, it was stated that the uncertainties of cyanobacteria density could be reduced if more nearshore samples were collected to better characterize the physical drivers and processes of the nearshore cyanobacteria populations. Model performance based on nearshore or interior locations has different implications depending on the waterbody’s intended use. The interior models performed better, which could lead to better risk predictions for microcystins potentially being transported to nearby crops and fields during irrigation practices. However, with the nearshore model’s poorer performance, there are reduced safeguards for assessing microcystins that could potentially be ingested by livestock that use the waterbody for drinking and cooling purposes.
While the training and testing R2 values are relatively close to each other, indicating that the models are not overfitting, and despite consistent model performance between the testing and training data, the R2 values themselves are not particularly high (average R2 value of 0.585). This suggests that the model is only capturing a limited portion of the variability in MC concentrations. One reason for these lower R2 values might be that the dataset is missing key explanatory variables that are more directly related to MC presence and concentration than what was measured in this study. Microcystin levels are influenced by a complex set of biological, chemical, and environmental factors [3,15]. While our current models rely on measurements made in the field with a hand-held sonde (i.e., TEMP, NTU, CHL, BGA, pH, DO, SPC, and fDOM), these measurements may not fully capture the dynamics driving MC variability. Important drivers such as nutrient concentrations (i.e., nitrogen and phosphorus) and algal activity, including diel vertical migration and toxin gene expression, that require labor intensive laboratory analyses were not included in this dataset. These types of laboratory analyses might explain MC variability better than easily measured in situ water quality data and, therefore, may produce a model with better performance for predicting MC concentrations. Our models are limited by the input data; although the use of near instantaneous field measurements makes data collection and prediction more efficient, it also limits the model’s ability to make more accurate predictions. This tradeoff between ease of measurement and predictive capability is important to acknowledge.
While limited data are available for microcystins in agricultural irrigation and livestock watering ponds, the microcystin concentrations in the agricultural ponds of this study were comparable to those reported for other fresh waterbodies. Specifically, Pond 1 and Pond 2 microcystin ranges were similar to those reported for eutrophic lakes [43,108]. In comparison, Pond 3 was more representative of the microcystin ranges seen for oligotrophic or mesotrophic lakes [109,110] and reservoirs [111,112]. Earlier work on Pond 1 and 2 showed that the microcystin concentrations in these ponds varied spatially and temporally throughout the study period [31]. Despite MC concentrations being highly variable throughout this 17-month study period, the models were able to accurately (R2 values between 0.567 and 0.727 for both ponds) capture and predict MC concentrations. Additionally, at Pond 3 in Maryland, earlier work showed that MC concentrations exhibited a seasonal trend throughout the growing season from May to September [37]. While the Pond 3 model proved to be less accurate (R2 values between 0.358 and 0.570) than Ponds 1 and 2, the interior and entire-pond models still were able to capture and predict MC concentrations within a smaller season rather than over the course of a year.
Due to the site-specific and non-linear relationships between water quality, cyanobacteria blooms, and cyanotoxins [17,45,70], it is unknown whether a universal model with high-level generalization can be developed for predicting microcystin occurrence and concentrations. Therefore, it is important to collect and test spatially and temporally rigorous data on a regional scale. The major obstacle for this assessment, and numerous others [15,65,113,114], is that there are biological and chemical components which can strongly impact microcystin production and concentration that are not easy or cost-effective to monitor. Measuring both known strong drivers such as nutrients, and potentially influential drivers such as organic forms of DOM, phytoplankton population structure, and weather, may lead to model accuracy improvements and potentially a universal model. Future work should aim to determine the limit of accuracy for models given the available data pool from ponds with different natural conditions and geographical locations.

5. Conclusions

Machine learning, implemented through the RF algorithm enhanced with repeated cross validations, was successful in the prediction of MC concentrations within three agricultural ponds located in Georgia and Maryland, USA. Predictions for interior models were more accurate than those for the whole pond and nearshore locations in the ponds. The models created for the Georgia ponds were more accurate than the models created for the Maryland pond. Water quality parameters such as water temperature and phytoplankton pigments (phycocyanin and chlorophyll), collected using a hand-held sonde, were the most influential for predictions. Models using these types of in situ water quality parameters, when optimized regionally, may allow water resource managers and farmers to reduce the reliance on more time-consuming and costly monitoring methods for microcystins in agricultural ponds, thus reducing risks to human and animal health. Since many of the same water quality parameters collected to assess microbial water quality per FSMA guidelines [25] have been demonstrated here to be predictive of microcystins concentrations, these time- and cost-saving measures provide dual benefits to the agriculture industry.

Supplementary Materials

The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/w17162361/s1. Figure S1: Time series data for coefficient of variance on each sampling date for all measured water quality parameters; Figure S2: Pond 1 Spearman Rank Correlations; Figure S3: Pond 2 Spearman Rank Correlations; Figure S4: Pond 3 Spearman Rank Correlations; Figure S5: Testing data for all models and ponds. Predicted microcystin concentrations compared to the actual measured microcystin concentrations. Supplemental Table S1. Performance metrics (R2) of the random forest (RF), Extreme Gradient Boosting (XGB), Stochastic Gradient Boosting (SGB), Support Vector Machines (SVM), and K-nearest neighbors (KNN) models. Supplemental Table S2. Average rank of models ran based on R2 values. The higher the rank the higher the R2 values across the ponds and locations. Supplemental Table S3. Performance metrics R2 of the random forest (RF) and RF run with repeated cross validations. Supplemental Table S4. Performance metrics (average RMSE) of the random forest models.

Author Contributions

Conceptualization: J.E.S., M.D.S., R.L.H. and Y.P.; methodology: J.E.S. and M.D.S.; formal analysis: J.E.S.; investigation: J.E.S. and J.A.W.; data curation: J.E.S. and J.A.W.; resources: J.L.W.; writing—original draft preparation, J.E.S. and Y.P.; writing—review and editing: J.E.S., Y.P., R.L.H. and J.L.W.; project administration: Y.P. and R.L.H.; funding acquisition: Y.P. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded through the USDA’s Agricultural Research Service, project number 8042-42610-001-000-D.

Data Availability Statement

The datasets generated and analyzed during the current study are not publicly available as of the submission date. The material is part of ongoing Ph.D. programs of J.E.S. at the University of Maryland and J.A.W. at the University of Georgia. Data will become available after completion of their degree programs and can be requested from the corresponding author, J.E.S.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Toxic Cyanobacteria in Water: A Guide to Their Public Health Consequences, Monitoring and Management, 2nd ed.; Chorus, I., Welker, M., Eds.; CRC Press: London, UK, 2021; ISBN 978-1-003-08144-9. [Google Scholar]
  2. Martinez i Quer, A.; Larsson, Y.; Johansen, A.; Arias, C.A.; Carvalho, P.N. Cyanobacterial Blooms in Surface Waters—Nature-Based Solutions, Cyanotoxins and Their Biotransformation Products. Water Res. 2024, 251, 121122. [Google Scholar] [CrossRef]
  3. Paerl, H.W.; Otten, T.G.; Joyner, A.R. Moving towards Adaptive Management of Cyanotoxin-Impaired Water Bodies. Microb. Biotechnol. 2016, 9, 641–651. [Google Scholar] [CrossRef]
  4. Bensalem, M.; Amrani, A.; Zaidi, H.; Sedrati, F.; Laouar, O.; Wang, Z.; Nasri, H. Impact of Long-Term Cyanotoxin Exposure on Cattle: Biochemical, Histological, and Oxidative Stress Assessment. Vet. World 2025, 189–201. [Google Scholar] [CrossRef]
  5. Stewart, I.; Seawright, A.A.; Shaw, G.R. Cyanobacterial Poisoning in Livestock, Wild Mammals and Birds — An Overview. In Cyanobacterial Harmful Algal Blooms: State of the Science and Research Needs; Hudnell, H.K., Ed.; Springer: New York, NY, USA, 2008; pp. 613–637. ISBN 978-0-387-75865-7. [Google Scholar]
  6. Mohamed, Z.A.; Mostafa, Y.; Alamri, S.; Hashem, M. Accumulation of Microcystin Toxin in Irrigation Water and Alfalfa (Medicago sativa) Forage Plant, and Assessing the Potential Risk to Animal Health. Chemosphere 2024, 364, 143248. [Google Scholar] [CrossRef] [PubMed]
  7. Redouane, E.M.; Tazart, Z.; Lahrouni, M.; Mugani, R.; Elgadi, S.; Zine, H.; Zerrifi, S.E.A.; Haida, M.; Martins, J.C.; Campos, A.; et al. Health Risk Assessment of Lake Water Contaminated with Microcystins for Fruit Crop Irrigation and Farm Animal Drinking. Env. Sci. Pollut. Res. 2023, 30, 80234–80244. [Google Scholar] [CrossRef]
  8. Faulkner, S.; Sweetman, C.; Hutson, J.; Soole, K.; Hobson, P.; Fallowfield, H. Uptake of the Cyanobacterial Toxin Microcystin by Crop Plants Irrigated with Contaminated Wastewater: A Review. Rev. Env. Sci. Biotechnol. 2025, 24, 217–238. [Google Scholar] [CrossRef]
  9. Shingai, Q.K.; Wilkinson, G.M. Microcystin as a Biogeochemical Cycle: Pools, Fluxes, and Fates of the Cyanotoxin in Inland Waters. Limnol. Oceanogr. Lett. 2023, 8, 406–418. [Google Scholar] [CrossRef]
  10. Zhang, Y.; Husk, B.R.; Duy, S.V.; Dinh, Q.T.; Sanchez, J.S.; Sauvé, S.; Whalen, J.K. Quantitative Screening for Cyanotoxins in Soil and Groundwater of Agricultural Watersheds in Quebec, Canada. Chemosphere 2021, 274, 129781. [Google Scholar] [CrossRef]
  11. Haida, M.; El Khalloufi, F.; Mugani, R.; Essadki, Y.; Campos, A.; Vasconcelos, V.; Oudra, B. Microcystin Contamination in Irrigation Water and Health Risk. Toxins 2024, 16, 196. [Google Scholar] [CrossRef]
  12. Van Hassel, W.H.R.; Tardy, E.; Cottyn, B.; Andjelkovic, M.; Decombel, A.; Van Wichelen, J.; Masquelier, J.; Rajkovic, A. Irrigation-Dependent Accumulation of Microcystin in Different Crops under Mid-Scale Greenhouse Conditions. J. Agric. Food Res. 2025, 20, 101753. [Google Scholar] [CrossRef]
  13. Chaffin, J.D.; Westrick, J.A.; Reitz, L.A.; Bridgeman, T.B. Microcystin Congeners in Lake Erie Follow the Seasonal Pattern of Nitrogen Availability. Harmful Algae 2023, 127, 102466. [Google Scholar] [CrossRef]
  14. Jones, M.R.; Pinto, E.; Torres, M.A.; Dörr, F.; Mazur-Marzec, H.; Szubert, K.; Tartaglione, L.; Dell’Aversano, C.; Miles, C.O.; Beach, D.G.; et al. CyanoMetDB, a Comprehensive Public Database of Secondary Metabolites from Cyanobacteria. Water Res. 2021, 196, 117017. [Google Scholar] [CrossRef] [PubMed]
  15. MacKeigan, P.W.; Zastepa, A.; Taranu, Z.E.; Westrick, J.A.; Liang, A.; Pick, F.R.; Beisner, B.E.; Gregory-Eaves, I. Microcystin Concentrations and Congener Composition in Relation to Environmental Variables across 440 North-Temperate and Boreal Lakes. Sci. Total Environ. 2023, 884, 163811. [Google Scholar] [CrossRef] [PubMed]
  16. Zhang, K.; Gu, Y.; Cheng, C.; Xue, Q.; Xie, L. Changes in Microcystin Concentration in Lake Taihu, 13 Years (2007–2020) after the 2007 Drinking Water Crisis. Environ. Res. 2024, 241, 117597. [Google Scholar] [CrossRef]
  17. Golshan, A.; Evans, C.; Geary, P.; Morrow, A.; Maeder, M.; Tauler, R. Patterns of Cyanobacterial Abundance in a Major Drinking Water Reservoir: What 3 Years of Comprehensive Monitoring Data Reveals? Env. Monit. Assess. 2020, 192, 113. [Google Scholar] [CrossRef] [PubMed]
  18. Xue, Q.; Xie, L.; Yang, J.R.; Yang, J.; Su, X. The Prevalence and Persistence of Microcystin in Seven Subtropical Reservoirs in China and Associated Potential Risks for Human Health. Environ. Technol. Innov. 2024, 33, 103476. [Google Scholar] [CrossRef]
  19. Chernova, E.; Sidelev, S.; Russkikh, I.; Korneva, L.; Solovyova, V.; Mineeva, N.; Stepanova, I.; Zhakovskaya, Z. Spatial Distribution of Cyanotoxins and Ratios of Microcystin to Biomass Indicators in the Reservoirs of the Volga, Kama and Don Rivers, the European Part of Russia. Limnologica 2020, 84, 125819. [Google Scholar] [CrossRef]
  20. Graham, J.L.; Dubrovsky, N.M.; Foster, G.M.; King, L.R.; Loftin, K.A.; Rosen, B.H.; Stelzer, E.A. Cyanotoxin Occurrence in Large Rivers of the United States. Inland. Waters 2020, 10, 109–117. [Google Scholar] [CrossRef]
  21. Burch, M.D. Effective Doses, Guidelines & Regulations. Adv. Exp. Med. Biol. 2008, 619, 831–853. [Google Scholar] [CrossRef]
  22. World Health Organization Guidelines for Drinking-Water Quality; World Health Organization: Geneva, Switzerland, 2002; ISBN 978-92-4-154535-8.
  23. Yeager, N.; Carpenter, A. State Approaches to Addressing Cyanotoxins in Drinking Water. AWWA Water Sci. 2019, 1, e1121. [Google Scholar] [CrossRef]
  24. US Environmental Protection Agency Drinking Water Health Advisories for Two Cyanobacterial Toxins; EPA: Washington, DC, USA, 2015; p. 3.
  25. FDA (US Food and Drug Administration) The New FDA Food Modernization Act (FSMA). Preventative Controls and Produce Safety Rules. Available online: https://www.fda.gov/food/guidance-regulation-food-and-dietary-supplements/food-safety-modernization-act-fsma (accessed on 4 June 2025).
  26. Loftin, K.A.; Dietze, J.E.; Meyer, M.T.; Graham, J.L.; Maksimowicz, M.M.; Toyne, K.D. Total Cylindrospermopsins, Microcystins/Nodularins, and Saxitoxins Data for the 2007 United States Environmental Protection Agency National Lake Assessment; U.S. Geological Survey: Reston, VA, USA, 2016.
  27. Pollard, A.I.; Hampton, S.E.; Leech, D.M. The Promise and Potential of Continental-Scale Limnology Using the U.S. Environmental Protection Agency’s National Lakes Assessment. Limnol. Oceanogr. Bull. 2018, 27, 36–41. [Google Scholar] [CrossRef]
  28. Guan, M.; Zhang, Y.; Li, W.; Li, N.; Qi, L.; Shi, K.; Zhang, Y.; Qin, B.; Huang, C. Quantifying the Effects of Wind Wave on Cyanobacterial Blooms in Large Shallow Lake from 10 Years High Frequency Satellite Observation. Hydrobiologia 2025, 852, 891–908. [Google Scholar] [CrossRef]
  29. Zastepa, A.; Taranu, Z.E.; Kimpe, L.E.; Blais, J.M.; Gregory-Eaves, I.; Zurawell, R.W.; Pick, F.R. Reconstructing a Long-Term Record of Microcystins from the Analysis of Lake Sediments. Sci. Total Environ. 2017, 579, 893–901. [Google Scholar] [CrossRef]
  30. Chen, H.; Zhu, W.; Wang, R.; Feng, G.; Xue, Z. Rapid Horizontal Accumulation and Bloom Formation of the Cyanobacterium Microcystis under Wind Stress. Hydrobiologia 2023, 850, 123–135. [Google Scholar] [CrossRef]
  31. Smith, J.E.; Widmer, J.A.; Wolny, J.L.; Dunn, L.L.; Stocker, M.D.; Hill, R.L.; Pisani, O.; Coffin, A.W.; Pachepsky, Y. Persistence of Microcystin in Three Agricultural Ponds in Georgia, USA. Toxins 2024, 16, 482. [Google Scholar] [CrossRef]
  32. Chaffin, J.D.; Westrick, J.A.; Furr, E.; Birbeck, J.A.; Reitz, L.A.; Stanislawczyk, K.; Li, W.; Weber, P.K.; Bridgeman, T.B.; Davis, T.W.; et al. Quantification of Microcystin Production and Biodegradation Rates in the Western Basin of Lake Erie. Limnol. Oceanogr. 2022, 67, 1470–1483. [Google Scholar] [CrossRef]
  33. Palagama, D.S.W.; Baliu-Rodriguez, D.; Snyder, B.K.; Thornburg, J.A.; Bridgeman, T.B.; Isailovic, D. Identification and Quantification of Microcystins in Western Lake Erie during 2016 and 2017 Harmful Algal Blooms. J. Great Lakes Res. 2020, 46, 289–301. [Google Scholar] [CrossRef]
  34. Scherer, P.I.; Raeder, U.; Geist, J.; Zwirglmaier, K. Influence of Temperature, Mixing, and Addition of Microcystin-LR on Microcystin Gene Expression in Microcystis Aeruginosa. MicrobiologyOpen 2017, 6, e00393. [Google Scholar] [CrossRef]
  35. Lind, O.; Dávalos-Lind, L.; López, C.; López, M.; Dyble Bressie, J. Seasonal Morphological Variability in an in Situ Cyanobacteria Monoculture: Example from a Persistent Cylindrospermopsis Bloom in Lake Catemaco, Veracruz, Mexico. J. Limnol. 2016, 75, 66–80. [Google Scholar] [CrossRef]
  36. Hayes, N.M.; Vanni, M.J. Microcystin Concentrations Can Be Predicted with Phytoplankton Biomass and Watershed Morphology. Inland. Waters 2018, 8, 273–283. [Google Scholar] [CrossRef]
  37. Smith, J.E. Multiyear Spatial and Temporal Variation of Microcystin Concentrations in an Irrigation Pond in Maryland, USA. In Proceedings of the European Geosciences Union General Assembly 2024 (EGU24), Vienna, Austria, 14–19 April 2014. [Google Scholar]
  38. Guo, C.; Zhu, G.; Paerl, H.W.; Zhu, M.; Yu, L.; Zhang, Y.; Liu, M.; Zhang, Y.; Qin, B. Extreme Weather Event May Induce Microcystis Blooms in the Qiantang River, Southeast China. Env. Sci. Pollut. Res. 2018, 25, 22273–22284. [Google Scholar] [CrossRef] [PubMed]
  39. Wood, S.A.; Borges, H.; Puddick, J.; Biessy, L.; Atalah, J.; Hawes, I.; Dietrich, D.R.; Hamilton, D.P. Contrasting Cyanobacterial Communities and Microcystin Concentrations in Summers with Extreme Weather Events: Insights into Potential Effects of Climate Change. Hydrobiologia 2017, 785, 71–89. [Google Scholar] [CrossRef]
  40. Rastogi, R.P.; Madamwar, D.; Incharoensakdi, A. Bloom Dynamics of Cyanobacteria and Their Toxins: Environmental Health Impacts and Mitigation Strategies. Front. Microbiol. 2015, 6, 1254. [Google Scholar] [CrossRef]
  41. Gobler, C.J.; Davis, T.W.; Coyne, K.J.; Boyer, G.L. Interactive Influences of Nutrient Loading, Zooplankton Grazing, and Microcystin Synthetase Gene Expression on Cyanobacterial Bloom Dynamics in a Eutrophic New York Lake. Harmful Algae 2007, 6, 119–133. [Google Scholar] [CrossRef]
  42. Hayes, N.M.; Haig, H.A.; Simpson, G.L.; Leavitt, P.R. Effects of Lake Warming on the Seasonal Risk of Toxic Cyanobacteria Exposure. Limnol. Oceanogr. Lett. 2020, 5, 393–402. [Google Scholar] [CrossRef]
  43. Jacquemin, S.J.; Doll, J.C.; Johnson, L.T.; Newell, S.E. Exploring Long-Term Trends in Microcystin Toxin Values Associated with Persistent Harmful Algal Blooms in Grand Lake St Marys. Harmful Algae 2023, 122, 102374. [Google Scholar] [CrossRef] [PubMed]
  44. Patiño, R.; Christensen, V.G.; Graham, J.L.; Rogosch, J.S.; Rosen, B.H. Toxic Algae in Inland Waters of the Conterminous United States—A Review and Synthesis. Water 2023, 15, 2808. [Google Scholar] [CrossRef]
  45. Liu, X.; Lu, X.; Chen, Y. The Effects of Temperature and Nutrient Ratios on Microcystis Blooms in Lake Taihu, China: An 11-Year Investigation. Harmful Algae 2011, 10, 337–343. [Google Scholar] [CrossRef]
  46. Leigh, C.; Burford, M.A.; Roberts, D.T.; Udy, J.W. Predicting the Vulnerability of Reservoirs to Poor Water Quality and Cyanobacterial Blooms. Water Res. 2010, 44, 4487–4496. [Google Scholar] [CrossRef]
  47. Baer, M.M.; Godwin, C.M.; Johengen, T.H. The Effect of Single versus Dual Nutrient Decreases on Phytoplankton Growth Rates, Community Composition, and Microcystin Concentration in the Western Basin of Lake Erie. Harmful Algae 2023, 123, 102382. [Google Scholar] [CrossRef] [PubMed]
  48. Wagner, N.D.; Quach, E.; Buscho, S.; Ricciardelli, A.; Kannan, A.; Naung, S.W.; Phillip, G.; Sheppard, B.; Ferguson, L.; Allen, A.; et al. Nitrogen Form, Concentration, and Micronutrient Availability Affect Microcystin Production in Cyanobacterial Blooms. Harmful Algae 2021, 103, 102002. [Google Scholar] [CrossRef] [PubMed]
  49. Cui, X.; Yang, N.; Cui, H.; Yang, Q.; Wu, Z.; Shao, B.; Zhao, Y.; Tong, Y. Interspecific Competition Enhances Microcystin Production by Microcystis Aeruginosa under the Interactive Influences of Temperature and Nutrients. Water Res. 2024, 265, 122308. [Google Scholar] [CrossRef] [PubMed]
  50. Merder, J.; Harris, T.; Zhao, G.; Stasinopoulos, D.M.; Rigby, R.A.; Michalak, A.M. Geographic Redistribution of Microcystin Hotspots in Response to Climate Warming. Nat. Water 2023, 1, 844–854. [Google Scholar] [CrossRef]
  51. Cunha, D.G.F.; Dodds, W.K.; Loiselle, S.A. Factors Related to Water Quality and Thresholds for Microcystin Concentrations in Subtropical Brazilian Reservoirs. Inland. Waters 2018, 8, 368–380. [Google Scholar] [CrossRef]
  52. Te, S.H.; Gin, K.Y.-H. The Dynamics of Cyanobacteria and Microcystin Production in a Tropical Reservoir of Singapore. Harmful Algae 2011, 10, 319–329. [Google Scholar] [CrossRef]
  53. Fickas, K.C.; O’Shea, R.E.; Pahlevan, N.; Smith, B.; Bartlett, S.L.; Wolny, J.L. Leveraging Multimission Satellite Data for Spatiotemporally Coherent cyanoHAB Monitoring. Front. Remote Sens. 2023, 4, 1157609. [Google Scholar] [CrossRef]
  54. Hong, S.M.; Morgan, B.J.; Stocker, M.D.; Smith, J.E.; Kim, M.S.; Cho, K.H.; Pachepsky, Y.A. Using Machine Learning Models to Estimate Escherichia Coli Concentration in an Irrigation Pond from Water Quality and Drone-Based RGB Imagery Data. Water Res. 2024, 260, 121861. [Google Scholar] [CrossRef]
  55. Bunyon, C.L.; Fraser, B.T.; McQuaid, A.; Congalton, R.G. Using Imagery Collected by an Unmanned Aerial System to Monitor Cyanobacteria in New Hampshire, USA, Lakes. Remote Sens. 2023, 15, 2839. [Google Scholar] [CrossRef]
  56. Sharp, S.L.; Forrest, A.L.; Bouma-Gregson, K.; Jin, Y.; Cortés, A.; Schladow, S.G. Quantifying Scales of Spatial Variability of Cyanobacteria in a Large, Eutrophic Lake Using Multiplatform Remote Sensing Tools. Front. Environ. Sci. 2021, 9, 612934. [Google Scholar] [CrossRef]
  57. Ivey, J.E.; Wolny, J.L.; Heil, C.A.; Murasko, S.M.; Brame, J.A.; Parks, A.A. Urea Inputs Drive Picoplankton Blooms in Sarasota Bay, Florida, U.S.A. Water 2020, 12, 2755. [Google Scholar] [CrossRef]
  58. Hagh, S.F.; Amngostar, P.; Zylka, A.; Zimmerman, M.; Cresanti, L.; Karins, S.; O’Neil-Dunne, J.P.; Ritz, K.; Williams, C.J.; Morales-Williams, A.M.; et al. Autonomous UAV-Mounted LoRaWAN System for Real-Time Monitoring of Harmful Algal Blooms (HABs) and Water Quality. IEEE Sens. J. 2024, 24, 11414–11424. [Google Scholar] [CrossRef]
  59. Venkatachari, A.; Bourbonnais, A.; Salman, I.; Rekleitis, I.; Li, A.Q.; Cottingham, K.; Ewing, H.; Bruesewitz, D.; Arsenault, E.; Shingai, Q. Use of an Autonomous Surface Vehicle to Collect High Spatial Resolution Water Quality Data at Lake Wateree, SC. J. South Carol. Water Resour. 2024, 9, 10. [Google Scholar] [CrossRef]
  60. Bartelt, G.; You, J.; Hondzo, M. Remote Cyanobacteria Detection by Multispectral Drone Imagery. Lake Reserv. Manag. 2024, 40, 236–247. [Google Scholar] [CrossRef]
  61. Gómez, D.; Salvador, P.; Sanz, J.; Casanova, J.L. A New Approach to Monitor Water Quality in the Menor Sea (Spain) Using Satellite Data and Machine Learning Methods. Environ. Pollut. 2021, 286, 117489. [Google Scholar] [CrossRef] [PubMed]
  62. Kwon, Y.S.; Pyo, J.; Kwon, Y.-H.; Duan, H.; Cho, K.H.; Park, Y. Drone-Based Hyperspectral Remote Sensing of Cyanobacteria Using Vertical Cumulative Pigment Concentration in a Deep Reservoir. Remote Sens. Environ. 2020, 236, 111517. [Google Scholar] [CrossRef]
  63. Copetti, D.; Matarrese, R.; Bresciani, M.; Guzzella, L. Integration of In Situ and Remote Sensing Measurements for the Management of Harmful Cyanobacteria Blooms. A Lesson from a Strategic Multiple-Uses Reservoir (Lake Occhito, South Italy). Water 2021, 13, 2162. [Google Scholar] [CrossRef]
  64. Wilkinson, A.A.; Hondzo, M.; Guala, M. Vertical Heterogeneities of Cyanobacteria and Microcystin Concentrations in Lakes Using a Seasonal In Situ Monitoring Station. Glob. Ecol. Conserv. 2020, 21, e00838. [Google Scholar] [CrossRef]
  65. Harris, T.D.; Graham, J.L. Predicting Cyanobacterial Abundance, Microcystin, and Geosmin in a Eutrophic Drinking Water Reservoir Using a 14-Year Dataset. Lake Reserv. Manag. 2017, 33, 32–48. [Google Scholar] [CrossRef]
  66. Mellios, N.; Moe, S.J.; Laspidou, C. Machine Learning Approaches for Predicting Health Risk of Cyanobacterial Blooms in Northern European Lakes. Water 2020, 12, 1191. [Google Scholar] [CrossRef]
  67. Zeng, Q.; Liu, Y.; Zhao, H.; Sun, M.; Li, X. Comparison of Models for Predicting the Changes in Phytoplankton Community Composition in the Receiving Water System of an Inter-Basin Water Transfer Project. Environ. Pollut. 2017, 223, 676–684. [Google Scholar] [CrossRef]
  68. Zhang, Y.; Hou, J.; Gu, Y.; Zhu, X.; Xia, J.; Wu, J.; You, G.; Yang, Z.; Ding, W.; Miao, L. Spatiotemporal Variation Assessment and Improved Prediction Of Cyanobacteria Blooms in Lakes Using Improved Machine Learning Model Based on Multivariate Data. Environ. Manag. 2025, 75, 694–709. [Google Scholar] [CrossRef]
  69. Xu, Y.; Zhang, D.; Lin, J.; Peng, Q.; Lei, X.; Jin, T.; Wang, J.; Yuan, R. Prediction of Phytoplankton Biomass and Identification of Key Influencing Factors Using Interpretable Machine Learning Models. Ecol. Indic. 2024, 158, 111320. [Google Scholar] [CrossRef]
  70. Hwang, S.-Y.; Choi, B.-W.; Park, J.-H.; Shin, D.-S.; Chung, H.-S.; Son, M.-S.; Lim, C.-H.; Chae, H.-M.; Ha, D.-W.; Jung, K.-Y. Evaluating Statistical Machine Learning Algorithms for Classifying Dominant Algae in Juam Lake and Tamjin Lake, Republic of Korea. Water 2023, 15, 1738. [Google Scholar] [CrossRef]
  71. Drobnič, F.; Kos, A.; Pustišek, M. On the Interpretability of Machine Learning Models and Experimental Feature Selection in Case of Multicollinear Data. Electronics 2020, 9, 761. [Google Scholar] [CrossRef]
  72. Kuhn, M. Building Predictive Models in R Using the Caret Package. J. Stat. Softw. 2008, 28, 1–26. [Google Scholar] [CrossRef]
  73. Couronné, R.; Probst, P.; Boulesteix, A.-L. Random Forest versus Logistic Regression: A Large-Scale Benchmark Experiment. BMC Bioinform. 2018, 19, 270. [Google Scholar] [CrossRef]
  74. Probst, P.; Wright, M.N.; Boulesteix, A.-L. Hyperparameters and Tuning Strategies for Random Forest. WIREs Data Min. Knowl. Discov. 2019, 9, e1301. [Google Scholar] [CrossRef]
  75. Francy, D.S.; Brady, A.M.G.; Ecker, C.D.; Graham, J.L.; Stelzer, E.A.; Struffolino, P.; Dwyer, D.F.; Loftin, K.A. Estimating Microcystin Levels at Recreational Sites in Western Lake Erie and Ohio. Harmful Algae 2016, 58, 23–34. [Google Scholar] [CrossRef]
  76. Fu, X.; Zheng, M.; Su, J.; Xi, B.; Wei, D.; Wang, X. Spatiotemporal Patterns and Threshold of Chlorophyll-a in Lake Taihu Based on Microcystins. Env. Sci. Pollut. Res. 2023, 30, 49327–49338. [Google Scholar] [CrossRef]
  77. Koreivienė, J.; Anne, O.; Kasperovičienė, J.; Burškytė, V. Cyanotoxin Management and Human Health Risk Mitigation in Recreational Waters. Env. Monit. Assess. 2014, 186, 4443–4459. [Google Scholar] [CrossRef]
  78. Sakai, H.; Hao, A.; Iseri, Y.; Wang, S.; Kuba, T.; Zhang, Z.; Katayama, H. Occurrence and Distribution of Microcystins in Lake Taihu, China. Sci. World J. 2013, 2013, 838176. [Google Scholar] [CrossRef]
  79. Yu, L.; Kong, F.; Zhang, M.; Yang, Z.; Shi, X.; Du, M. The Dynamics of Microcystis Genotypes and Microcystin Production and Associations with Environmental Factors during Blooms in Lake Chaohu, China. Toxins 2014, 6, 3238–3257. [Google Scholar] [CrossRef]
  80. Sinang, S.C.; Reichwaldt, E.S.; Ghadouani, A. Spatial and Temporal Variability in the Relationship between Cyanobacterial Biomass and Microcystins. Env. Monit. Assess. 2013, 185, 6379–6395. [Google Scholar] [CrossRef]
  81. Duong, T.T.; Le, T.P.Q.; Dao, T.-S.; Pflugmacher, S.; Rochelle-Newall, E.; Hoang, T.K.; Vu, T.N.; Ho, C.T.; Dang, D.K. Seasonal Variation of Cyanobacteria and Microcystins in the Nui Coc Reservoir, Northern Vietnam. J. Appl. Phycol. 2013, 25, 1065–1075. [Google Scholar] [CrossRef]
  82. Singh, S.; Rai, P.K.; Chau, R.; Ravi, A.K.; Neilan, B.A.; Asthana, R.K. Temporal Variations in Microcystin-Producing Cells and Microcystin Concentrations in Two Fresh Water Ponds. Water Res. 2015, 69, 131–142. [Google Scholar] [CrossRef] [PubMed]
  83. Park, Y.; Lee, H.K.; Shin, J.-K.; Chon, K.; Kim, S.; Cho, K.H.; Kim, J.H.; Baek, S.-S. A Machine Learning Approach for Early Warning of Cyanobacterial Bloom Outbreaks in a Freshwater Reservoir. J. Environ. Manag. 2021, 288, 112415. [Google Scholar] [CrossRef] [PubMed]
  84. Swartz, T.M.; Miller, J.R. The American Pond Belt: An Untold Story of Conservation Challenges and Opportunities. Front. Ecol. Environ. 2021, 19, 501–509. [Google Scholar] [CrossRef]
  85. Ni, J.; Liu, R.; Li, Y.; Tang, G.; Shi, P. An Improved Transfer Learning Model for Cyanobacterial Bloom Concentration Prediction. Water 2022, 14, 1300. [Google Scholar] [CrossRef]
  86. Mu, W.; Kleter, G.A.; Bouzembrak, Y.; Dupouy, E.; Frewer, L.J.; Radwan Al Natour, F.N.; Marvin, H.J.P. Making Food Systems More Resilient to Food Safety Risks by Including Artificial Intelligence, Big Data, and Internet of Things into Food Safety Early Warning and Emerging Risk Identification Tools. Compr. Rev. Food Sci. Food Saf. 2024, 23, e13296. [Google Scholar] [CrossRef]
  87. Hammer, Ø.; Harper, D.A.T.; Ryan, P.D. PAST: Paleontological Statistics Software Package for Education and Data Analysis 2001. Palaeontol. Electron. 2001, 4, 1–9. [Google Scholar]
  88. George, D.; Mallery, P. SPSS for Windows Step by Step: A Simple Guide and Reference 18.0 Update, 11th ed.; Prentice Hall Press: Upper Saddle River, NJ, USA, 2010; ISBN 978-0-205-01124-7. [Google Scholar]
  89. Barreñada, L.; Dhiman, P.; Timmerman, D.; Boulesteix, A.-L.; Van Calster, B. Understanding Overfitting in Random Forest for Probability Estimation: A Visualization and Simulation Study. Diagn. Progn. Res. 2024, 8, 14. [Google Scholar] [CrossRef]
  90. Ying, X. An Overview of Overfitting and Its Solutions. J. Phys. Conf. Ser. 2019, 1168, 022022. [Google Scholar] [CrossRef]
  91. Allen Akselrud, C.I. Random Forest Regression Models in Ecology: Accounting for Messy Biological Data and Producing Predictions with Uncertainty. Fish. Res. 2024, 280, 107161. [Google Scholar] [CrossRef]
  92. Suikkanen, S.; Uusitalo, L.; Lehtinen, S.; Lehtiniemi, M.; Kauppila, P.; Mäkinen, K.; Kuosa, H. Diazotrophic Cyanobacteria in Planktonic Food Webs. Food Webs 2021, 28, e00202. [Google Scholar] [CrossRef]
  93. Smith, J.E.; Wolny, J.L.; Hill, R.L.; Stocker, M.D.; Pachepsky, Y. Examining the Relationship between Phytoplankton Community Structure and Water Quality Measurements in Agricultural Waters: A Machine Learning Application. Environments 2022, 9, 142. [Google Scholar] [CrossRef]
  94. Moldaenke, C.; Fang, Y.; Yang, F.; Dahlhaus, A. Early Warning Method for Cyanobacteria Toxin, Taste and Odor Problems by the Evaluation of Fluorescence Signals. Sci. Total Environ. 2019, 667, 681–690. [Google Scholar] [CrossRef] [PubMed]
  95. McQuaid, N.; Zamyadi, A.; Prévost, M.; Bird, D.F.; Dorner, S. Use of in Vivo Phycocyanin Fluorescence to Monitor Potential Microcystin -Producing Cyanobacterial Biovolume in a Drinking Water Source. J. Environ. Monit. 2011, 13, 455–463. [Google Scholar] [CrossRef]
  96. Mchau, G.J.; Makule, E.; Machunda, R.; Gong, Y.Y.; Kimanya, M. Phycocyanin as a Proxy for Algal Blooms in Surface Waters: Case Study of Ukerewe Island, Tanzania. Water Pract. Technol. 2019, 14, 229–239. [Google Scholar] [CrossRef]
  97. Schürmann, Q.J.F.; Visser, P.M.; Sollie, S.; Kardinaal, W.E.A.; Faassen, E.J.; Lokmani, R.; van der Oost, R.; Van de Waal, D.B. Risk Assessment of Toxic Cyanobacterial Blooms in Recreational Waters: A Comparative Study of Monitoring Methods. Harmful Algae 2024, 138, 102683. [Google Scholar] [CrossRef]
  98. Li, J.; Persson, K.M.; Pekar, H.; Jansson, D. Evaluation of Indicators for Cyanobacterial Risk in 108 Temperate Lakes Using 23 Years of Environmental Monitoring Data. Env. Sci. Eur. 2021, 33, 54. [Google Scholar] [CrossRef]
  99. Griffith, A.W.; Gobler, C.J. Harmful Algal Blooms: A Climate Change Co-Stressor in Marine and Freshwater Ecosystems. Harmful Algae 2020, 91, 101590. [Google Scholar] [CrossRef]
  100. Paerl, H.W.; Huisman, J. Blooms Like It Hot. Science 2008, 320, 57–58. [Google Scholar] [CrossRef]
  101. Mowe, M.A.D.; Porojan, C.; Abbas, F.; Mitrovic, S.M.; Lim, R.P.; Furey, A.; Yeo, D.C.J. Rising Temperatures May Increase Growth Rates and Microcystin Production in Tropical Microcystis Species. Harmful Algae 2015, 50, 88–98. [Google Scholar] [CrossRef]
  102. Walls, J.T.; Wyatt, K.H.; Doll, J.C.; Rubenstein, E.M.; Rober, A.R. Hot and Toxic: Temperature Regulates Microcystin Release from Cyanobacteria. Sci. Total Environ. 2018, 610–611, 786–795. [Google Scholar] [CrossRef] [PubMed]
  103. Ma, J.; Qin, B.; Paerl, H.W.; Brookes, J.D.; Hall, N.S.; Shi, K.; Zhou, Y.; Guo, J.; Li, Z.; Xu, H.; et al. The Persistence of Cyanobacterial (Icrocystis spp.) Blooms throughout Winter in Lake Taihu, China. Limnol. Oceanogr. 2016, 61, 711–722. [Google Scholar] [CrossRef]
  104. Šeparović, L.; Alexandru, A.; Laprise, R.; Martynov, A.; Sushama, L.; Winger, K.; Tete, K.; Valin, M. Present Climate and Climate Change over North America as Simulated by the Fifth-Generation Canadian Regional Climate Model. Clim. Dyn. 2013, 41, 3167–3201. [Google Scholar] [CrossRef]
  105. Zibordi, G.; Mélin, F.; Berthon, J.-F.; Holben, B.; Slutsker, I.; Giles, D.; D’Alimonte, D.; Vandemark, D.; Feng, H.; Schuster, G.; et al. AERONET-OC: A Network for the Validation of Ocean Color Primary Products. J. Atmos. Ocean. Technol. 2009, 26, 1634–1651. [Google Scholar] [CrossRef]
  106. Jiang, M.; Brereton, A.; Beckler, J.; Moore, T.; Brewton, R.A.; Hu, C.; Lapointe, B.E.; McFarland, M.N. Modeling Water Quality and Cyanobacteria Blooms in Lake Okeechobee: I. Model Descriptions, Seasonal Cycles, and Spatial Patterns. Ecol. Model. 2025, 502, 111018. [Google Scholar] [CrossRef]
  107. Lofton, M.E.; Brentrup, J.A.; Beck, W.S.; Zwart, J.A.; Bhattacharya, R.; Brighenti, L.S.; Burnet, S.H.; McCullough, I.M.; Steele, B.G.; Carey, C.C.; et al. Using Near-Term Forecasts and Uncertainty Partitioning to Inform Prediction of Oligotrophic Lake Cyanobacterial Density. Ecol. Appl. 2022, 32, e2590. [Google Scholar] [CrossRef] [PubMed]
  108. Beversdorf, L.J.; Miller, T.R.; McMahon, K.D. Long-Term Monitoring Reveals Carbon–Nitrogen Metabolism Key to Microcystin Production in Eutrophic Lakes. Front. Microbiol. 2015, 6, 456. [Google Scholar] [CrossRef] [PubMed]
  109. Garland, D.; Koehler, H.; McGirr, S.; Parkes, R.; E Lucy, F.; Touzet, N. Seasonal Community Dynamics and Toxicity Potential of Cyanobacteria in Lough Arrow, an Oligo-Mesotrophic Lake in the North-West of Ireland. Limnologica 2023, 103, 126124. [Google Scholar] [CrossRef]
  110. Tekanova, E.; Sidelev, S.; Kalinkina, N.; Chernova, E.; Barinova, S.; Sharov, A.; Smirnova, V. Toxigenic Cyanobacteria and Microcystins in a Large Northern Oligotrophic Lake Onego, Russia. Toxins 2024, 16, 457. [Google Scholar] [CrossRef]
  111. Mariani, M.A.; Padedda, B.M.; Kaštovský, J.; Buscarinu, P.; Sechi, N.; Virdis, T.; Lugliè, A. Effects of Trophic Status on Microcystin Production and the Dominance of Cyanobacteria in the Phytoplankton Assemblage of Mediterranean Reservoirs. Sci. Rep. 2015, 5, 17964. [Google Scholar] [CrossRef] [PubMed]
  112. Moraes, M.d.A.B.; Rodrigues, R.d.A.M.; Podduturi, R.; Jørgensen, N.O.G.; Calijuri, M.d.C. Prediction of Cyanotoxin Episodes in Freshwater: A Case Study on Microcystin and Saxitoxin in the Lobo Reservoir, São Paulo State, Brazil. Environments 2023, 10, 143. [Google Scholar] [CrossRef]
  113. Wilkinson, G.M.; Walter, J.A.; Albright, E.A.; King, R.F.; Moody, E.K.; Ortiz, D.A. An Evaluation of Statistical Models of Microcystin Detection in Lakes Applied Forward under Varying Climate Conditions. Harmful Algae 2024, 137, 102679. [Google Scholar] [CrossRef] [PubMed]
  114. Taranu, Z.E.; Pick, F.R.; Creed, I.F.; Zastepa, A.; Watson, S.B. Meteorological and Nutrient Conditions Influence Microcystin Congeners in Freshwaters. Toxins 2019, 11, 620. [Google Scholar] [CrossRef]
Water 17 02361 g001
Figure 1. Sampling locations (yellow dots) in Pond 1 (A) and Pond 2 (B) over the observation period from 2022 to 2023. Pond 1 is a livestock watering pond which is occasionally used for irrigation, and Pond 2 is an agricultural irrigation pond, both located in southeastern Georgia.
Figure 1. Sampling locations (yellow dots) in Pond 1 (A) and Pond 2 (B) over the observation period from 2022 to 2023. Pond 1 is a livestock watering pond which is occasionally used for irrigation, and Pond 2 is an agricultural irrigation pond, both located in southeastern Georgia.
Water 17 02361 g001
Water 17 02361 g002
Figure 2. Sampling locations (yellow dots) for Pond 3 over the observation period during the growing seasons of 2022 (A), 2023 (B), and 2024 (B). Pond 3 is an agricultural irrigation pond located in central Maryland.
Figure 2. Sampling locations (yellow dots) for Pond 3 over the observation period during the growing seasons of 2022 (A), 2023 (B), and 2024 (B). Pond 3 is an agricultural irrigation pond located in central Maryland.
Water 17 02361 g002
Water 17 02361 g003
Figure 3. Cumulative probability distribution of microcystin (MC) concentrations at Pond 1, 2, and 3.
Figure 3. Cumulative probability distribution of microcystin (MC) concentrations at Pond 1, 2, and 3.
Water 17 02361 g003
Water 17 02361 g004
Figure 4. The highest reported Spearman rank correlations for all three ponds’ interior samples only. 0.XXX1—Pond 1, 0.XXX2—Pond 2, 0.XXX3—Pond 3.
Figure 4. The highest reported Spearman rank correlations for all three ponds’ interior samples only. 0.XXX1—Pond 1, 0.XXX2—Pond 2, 0.XXX3—Pond 3.
Water 17 02361 g004
Water 17 02361 g005
Figure 5. The highest reported Spearman rank correlations for all three ponds’ nearshore samples only. 0.XXX1—Pond 1, 0.XXX2—Pond 2, 0.XXX4—Pond 3.
Figure 5. The highest reported Spearman rank correlations for all three ponds’ nearshore samples only. 0.XXX1—Pond 1, 0.XXX2—Pond 2, 0.XXX4—Pond 3.
Water 17 02361 g005
Table 1. Descriptive statistics for measured water quality parameters of Pond 1, Pond 2, and Pond 3.
Table 1. Descriptive statistics for measured water quality parameters of Pond 1, Pond 2, and Pond 3.
Pond 1MinMaxMedMeanStd DevStd ErrSkewKurt
MC0.7427.514.457.096.060.361.381.28
CHL0.9440.958.4111.678.340.491.110.77
FDOM2.9767.8750.2750.508.670.51−0.511.97
DO2.9613.959.439.341.600.09−0.400.90
SPC7.10151.50120.70125.0815.250.90−1.5910.80
BGA1.1821.848.848.534.200.250.27−0.32
NTU3.70182.2254.9361.7331.381.851.201.79
pH5.499.708.408.200.960.06−0.23−1.21
TEMP13.5933.2827.0623.866.120.36−0.31−1.51
Pond 2MinMaxMedMeanStd DevStd ErrSkewKurt
MC0.5020.035.466.314.460.280.910.34
CHL0.2210.703.814.171.970.120.780.39
FDOM0.7870.3933.7434.9313.600.860.880.79
DO6.6719.3911.0611.452.890.180.870.04
SPC2.20282.70216.45215.7231.501.98−1.137.78
BGA0.3734.184.545.834.590.291.886.05
NTU4.02118.0527.3931.5919.211.211.262.05
pH4.0210.409.128.940.940.06−1.374.60
TEMP13.4331.6626.0424.894.940.31−0.55−0.72
Pond 3MinMaxMedMeanStd DevStd ErrSkewKurt
MC0.005.960.721.121.160.071.612.19
CHL0.01115.561.964.338.960.558.0590.68
FDOM6.7042.8824.1725.225.610.340.421.09
DO4.7414.079.369.351.930.120.07−0.33
SPC85.60197.40165.00168.1812.480.77−1.489.59
BGA0.0151.881.131.953.540.2211.12151.91
NTU0.01143.799.4512.3013.390.825.6244.16
pH6.2510.139.068.800.830.05−0.86−0.28
TEMP17.1432.9226.4926.233.280.20−0.570.24
Table 2. Performance metrics of the random forest models and important predictors.
Table 2. Performance metrics of the random forest models and important predictors.
PondModelTraining R2Testing R2Imp Pred #1Imp Pred #2Imp Pred #3
Pond 1All0.6420.589CHLTEMPFDOM
Interior0.6690.676CHLTEMPFDOM
Nearshore0.5760.596TEMPCHLFDOM
Pond 2All0.6310.642BGATEMPSPC
Interior0.7270.695TEMPNTUBGA
Nearshore0.6220.614BGATEMPNTU
Pond 3All0.4620.474TEMPNTUPH
Interior0.5450.570BGANTUTEMP
Nearshore0.3580.388NTUPHBGA
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Smith, J.E.; Widmer, J.A.; Stocker, M.D.; Wolny, J.L.; Hill, R.L.; Pachepsky, Y. Investigating the Relationship Between Microcystin Concentrations and Water Quality Parameters in Three Agricultural Irrigation Ponds Using Random Forest. Water 2025, 17, 2361. https://doi.org/10.3390/w17162361

AMA Style

Smith JE, Widmer JA, Stocker MD, Wolny JL, Hill RL, Pachepsky Y. Investigating the Relationship Between Microcystin Concentrations and Water Quality Parameters in Three Agricultural Irrigation Ponds Using Random Forest. Water. 2025; 17(16):2361. https://doi.org/10.3390/w17162361

Chicago/Turabian Style

Smith, Jaclyn E., James A. Widmer, Matthew D. Stocker, Jennifer L. Wolny, Robert L. Hill, and Yakov Pachepsky. 2025. "Investigating the Relationship Between Microcystin Concentrations and Water Quality Parameters in Three Agricultural Irrigation Ponds Using Random Forest" Water 17, no. 16: 2361. https://doi.org/10.3390/w17162361

APA Style

Smith, J. E., Widmer, J. A., Stocker, M. D., Wolny, J. L., Hill, R. L., & Pachepsky, Y. (2025). Investigating the Relationship Between Microcystin Concentrations and Water Quality Parameters in Three Agricultural Irrigation Ponds Using Random Forest. Water, 17(16), 2361. https://doi.org/10.3390/w17162361

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop