Next Article in Journal
Scalability and Computational Performance of an Ecohydrological Model Using Machine Learning-Based Prediction
Previous Article in Journal
Machine Learning-Assisted In Situ Monitoring System for Identifying and Predicting Components, Concentrations, and Viscosities of Fracturing Flowback Wastewater
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Water
Volume 18
Issue 4
10.3390/w18040465
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

Trends in Maumee River Nitrogen Loads and Their Complex Relationship to Harmful Algal Blooms in Western Lake Erie

by
Nusrat N. Khan
1,
Hans W. Paerl
2,
Mark J. McCarthy
3,
Silvia E. Newell
4,
Noah Rudko
5,* and
Rebecca Logsdon Muenich
5
1
School of Sustainable Engineering and the Built Environment, Arizona State University, Tempe, AZ 85201, USA
2
Department of Earth, Marine and Environmental Sciences, Institute of Marine Sciences, University of North Carolina at Chapel Hill, Morehead City, NC 28557, USA
3
Chair of Hydrobiology and Fisheries, Estonian University of Life Sciences, 51006 Tartu, Estonia
4
School for the Environment and Sustainability, University of Michigan, Ann Arbor, MI 48108, USA
5
Biological & Agricultural Engineering, University of Arkansas, 790 W. Dickson St., Fayetteville, AR 72701, USA
*
Author to whom correspondence should be addressed.
Water 2026, 18(4), 465; https://doi.org/10.3390/w18040465
Submission received: 9 January 2026 / Revised: 26 January 2026 / Accepted: 2 February 2026 / Published: 11 February 2026
(This article belongs to the Section Water, Agriculture and Aquaculture)
Browse Figures
Versions Notes

Abstract

Nutrient runoff from agricultural activities in the watershed of Western Lake Erie (WLE) is a dominant driver of harmful algal blooms (HABs). While phosphorus (P) is a key factor causing these blooms and has been the focus for researchers and policymakers, the influence of nitrogen (N) on bloom dynamics has been overlooked. Total Kjeldahl N (TKN; organic N and ammonium N) has not been the focus of eutrophication research but was recently linked to bloom development in WLE. Here, monotonic and oscillatory statistical trend analyses were performed to interpret long-term (1982 to 2022) patterns of TKN in the Maumee River and were compared to algal biomass data as chlorophyll a. A predictive regression model used principal component analysis to estimate a chlorophyll-based index of HABs in WLE, and a systematic iterative process identified that TKN influences bloom dynamics along with soluble reactive phosphorus (SRP), total suspended solids (TSS) and flow. Although TKN loads exhibited a long-term decline, this decrease did not correspond to reduced HAB severity, reflecting the strong influence of flow-driven hydrologic variability on nutrient delivery and bloom response. The modeling results demonstrate that TKN, together with SRP, TSS, and flow, significantly contributes to predicting bloom magnitude. These findings highlight the need for dual-nutrient (N and P) management strategies and additional analyses of nutrient–hydrology interactions to improve HAB mitigation in WLE.

1. Introduction

Harmful algal blooms (HABs) driven by excessive nutrient inputs pose significant risks to water quality, ecosystem and human health, and local economies. These blooms can generate toxins, promote hypoxia, increase treatment costs, and impair fisheries and recreation [1,2]. The Lake Erie region, supporting approximately 13 million people [3], has experienced recurring cyanobacterial HABs, particularly in its shallow, warm, and agriculturally-influenced western basin [4,5]. Despite management efforts, HAB severity indices have exceeded target levels nearly every year since 2008 [6].
Phosphorus (P) has long been the primary focus of freshwater eutrophication management, particularly following the Great Lakes Water Quality Agreement (GLWQA), which established basin-wide P reduction goals [7]. Recent HAB resurgence, however, has occurred despite substantial P control efforts and the widespread adoption of agricultural conservation practices [8,9]. As a result, there is growing recognition that nitrogen (N) also plays a critical role in bloom magnitude, persistence, and toxicity, particularly in systems dominated by non-N-fixing cyanobacteria such as Microcystis spp. [10,11,12,13,14]. Recent modeling studies exploring nitrogen–phosphorus interactions have generated debate regarding management implications, and some predictions remain unvalidated by long-term interannual lake observations [15,16].
Within this broader N framework, Total Kjeldahl Nitrogen (TKN), the sum of organic N and ammonium, has emerged as an important but understudied component of nutrient cycling in the Maumee River watershed and Western Lake Erie (WLE). Although machine learning and field-based studies increasingly identify TKN as a strong predictor of bloom development, toxicity, or persistence [17,18], most watershed-scale nutrient trend analyses and management strategies continue to emphasize dissolved inorganic N or total N. Long-term patterns in TKN, and their relationship to HAB dynamics, remain poorly characterized.
The Maumee River contributes approximately half of the P load to Lake Erie while providing only ~5% of the total inflow [19]. With 73% of the watershed in agricultural use [20] and extensive subsurface drainage [21], the river delivers substantial loads of both N and P, which interact with changing climate, hydrology, and land management [22,23]. Previous long-term Maumee analyses (e.g., [24]) provide valuable insight into nutrient patterns, yet they do not include TKN and extend only through 2013, leaving an important gap during a period of unprecedented HAB severity.
This study advances the understanding of Maumee–Lake Erie nutrient dynamics by providing the first long-term (1982–2022) trend assessment of TKN loads in the Maumee River, extending a decade beyond previous analyses and capturing recent management interventions and hydrologic change. This study integrates multiple complementary trend detection methods including monotonic (Mann–Kendall, Sen’s slope) and oscillatory frameworks (Sequential MK, STL decomposition) to characterize nonlinear, inflection-point-driven behavior not captured by monotonic tests alone and compares TKN temporal patterns to TP, SRP, and satellite-derived chlorophyll indices to evaluate how different nutrients align with periods of bloom intensification and decline. Finally a PCA-informed multiple regression model is developed to evaluate whether TKN provides a predictive value for bloom extent alongside P and hydrologic drivers, addressing multicollinearity that complicates traditional regression approaches. Together, these contributions provide new empirical evidence that TKN patterns align with key bloom periods and that reduced N forms can contribute additional explanatory power alongside P. Given recent calls for dual N–P management in Lake Erie [14,25], a clearer understanding of TKN dynamics is essential for designing more effective mitigation strategies. The objectives of this research were to:
(1)
Evaluate long-term monotonic and oscillatory trends in TKN loads from the Maumee River;
(2)
Compare these trends with TP, SRP, and satellite-derived bloom extent (chlorophyll index) in WLE;
(3)
Develop a dimensionally-reduced regression model to test whether TKN contributes to predicting bloom extent when considered jointly with other nutrients and hydrologic variables.
By addressing these goals, this study provides new insight into the role of reduced N forms in driving HAB dynamics and supports emerging evidence that managing both N and P will be necessary to mitigate eutrophication in WLE.

2. Materials and Methods

2.1. Study Area

Among the Great Lakes, Lake Erie has the highest influx of agricultural runoff from its watershed, including the Maumee, Sandusky, Cuyahoga, and Grand rivers in Ohio, as well as the Huron and Raisin rivers from Michigan and the Grand River from Ontario. The Maumee River is the largest tributary and drainage basin in the watershed (17,000 km2), discharging water from northwest Ohio, northeastern Indiana, and southeastern Michigan. The Maumee River discharges approximately 5% of the total water flow, and around 50% of the P load, into WLE [19], and is the focus of this study. According to the 2021 National Land Cover Database (NLCD), the Maumee River watershed (Figure 1) has 73% of its land under agricultural cultivation [20], with corn and soybeans being the major crops. Although mostly agricultural, there are urban areas in the watershed including cities like Toledo, Ohio [26]. Tables S1–S3 summarize land cover distribution, counties, and rivers in the Maumee watershed. The Maumee River watershed has poorly drained soil [27] and low slopes. The watershed spans the Eastern Corn Belt Plains (ECBP) and Huron–Erie Lake Plains (HELP) ecoregions and has a humid continental climate classified as Köppen Dfa, characterized by warm summers, cold winters, and precipitation distributed throughout the year [28]. Mean annual precipitation is approximately 875 mm, while annual potential evapotranspiration is lower, resulting in a positive climatic water balance and runoff-dominated hydrologic conditions. Runoff generation in the Maumee River watershed is primarily controlled by precipitation, antecedent soil moisture, and extensive tile drainage, reflecting a humid climate with a positive precipitation–evapotranspiration balance [29,30]. Catchment–lake interactions strongly control nutrient delivery to WLE, as discharge timing and magnitude, together with internal lake processes, mediate how watershed nutrient loads translate into bloom response [5].

2.2. Data Collection and Preprocessing

TKN concentrations and flow for the Maumee River at Waterville (USGS gaging station 4193500) were obtained from the National Center for Water Quality Research (NCWQR) at Heidelberg University [31]. This water quality monitoring station has provided near-daily, long-term nutrient concentration and flow data for the Maumee River since 1975 [31,32]. The focus of this study was on TKN, but concentrations for SRP, TP, nitrate plus nitrite (NOx), and total suspended solids (TSS) were also analyzed. The dataset used here ranged from January 1982 through December 2022. Concentrations were converted to daily loads using streamflow observations. Following the computation method of [24], when multiple entries existed for a single day, daily loads were averaged. Daily loads were aggregated to monthly means to support trend and decomposition analyses. TKN loads were strongly skewed, with very few extreme outliers (Figure S1, Table S4). Missing data were rare (<1% of the record) and occurred only when an entire month lacked sufficient daily observations to compute a monthly mean. In these cases, the missing monthly value was imputed using the long-term mean for that calendar month calculated from the remaining 39 years, preserving the seasonal structure while minimizing distortion of interannual variability.
Average monthly precipitation and temperature data were obtained from NOAA National Centers for Environmental Information (NCEI) global summaries for months that included no missing values [33]. Due to the large watershed area, five meteorological stations were considered to determine regional variations in precipitation and temperature: (1) Fort Wayne International Airport (USW00014827); (2) Findlay Airport (USW00014825); (3) Toledo Express Airport (USW00094830); (4) Adrian 2 NNE (USC00200032); and (5) Bowling Green WWTP (USC00330862). The Thiessen polygon method was applied to obtain area-weighted precipitation and temperature values for the entire basin (Figure S2, Table S5).
Since the Severity Index from NOAA is a measure of bloom biomass, 10-day composite chlorophyll-a Index (CI) values were sourced from [34] as a proxy for bloom extent. CI values were reported from June to October for 2002 to 2019 and were extracted from 1 km resolution National Center for Coastal Ocean Science (NCCOS) satellite images. Satellite images from the Medium Resolution Imaging Spectrometer (2002–2011) and the Moderate Resolution Imaging Spectroradiometer (2012–2019) were used to extract CI values. The highest CI value from each 10-day period was selected to create a 10-day composite CI value, which was visually represented as heat maps. Red–green–blue (RGB) values from the heat maps were converted back to numerical CI values and then summed across all pixels to obtain a composite value for the entire WLE. The resulting dataset was compared with previous data [35] from 2002 to 2015 and showed a high correlation (R2 = 0.955), despite some discrepancies. Although NOAA NCCOS also produces a Cyanobacterial Index product, that metric differs conceptually and temporally from the chlorophyll-based index used here and is not available as a long-term basin-aggregated time series suitable for this analysis.

2.3. Trend Analysis

To determine trends and patterns for TKN, water quality parameters (SRP, TSS, NOx, TN, and TP) and climatic drivers, non-parametric statistical methods applied included: (1) Mann–Kendall trend test (MK); (2) Sen’s Slope (SS); (3) Sequential Forward and Backward (SFB) trend analysis; and (4) Seasonal-Trend decomposition using Locally Estimated Scatterplot Smoothing (LOESS; STL) (Figure 2). Different statistical methods were compiled to examine monotonic and oscillatory trends, as well as to compare them with previous studies (e.g., [24]. MK and SS are monotonic trends that provide one-directional, increasing or decreasing trends. In contrast, SFB and STL were used to observe oscillatory trends over time. All trend analyses were performed in R Version 2023 using “stats”, “kendall”, “trend”, and “trendchange” packages.
The MK trend test [36,37] is widely used, with adaptability to handle missing data based on the assumption that no serial correlation exists without a trend [38]. MK trend analysis was applied to TKN, SRP, TP, NOx, and TSS monthly mean loads from 1982 through 2022 for the Maumee River. Along with the MK test, the SS method was applied to determine monotonic directions of trends in the data. SS is also known as the Theil–Sen Estimator and is robust to outliers and non-parametric trend analysis, which is based on assumptions of independence of the observations [38,39]. To provide serially uncorrelated time series, a pre-whitening [40] was performed in R. After obtaining monotonic trends using MK and SS analyses, a sequential [41] approach was implemented on all data. Sequential MK revealed both backward and forward trends, identifying distinct change points. SFB was applied on yearly average loads of TKN, SRP, and TP for the period of analysis. For CI data derived from [34], SFB was applied to identify continuous, oscillatory, forward and backward trends.
In a similar analysis of trends from the Maumee River into Lake Erie [24], STL was applied to study long-term and seasonal changes in flow, nutrient concentrations, and loads. STL is used to explore time-series data by segregation into various components, such as smoothed trend, seasonal, and residual. LOESS fits curves to the data through iterative computation [42]. Since the previous analysis [24] was only performed in the years 1975–2013 and did not include TKN, we aimed to compare and build upon this previous work by applying these methods to monthly average loads.

2.4. Multiple Linear Regression

The second goal of this work was to develop a predictive model for CI, since chlorophyll-a is an indicator of phytoplankton biomass. Many previous studies have applied a methodology consisting of dimensionality reduction through Principal Component Analysis (PCA) and stepwise regression to predict chlorophyll-a in China [43,44,45], Korea [46,47], and Turkey [48]. Trend analysis showed that TKN patterns were changing over time in the Maumee River; how and if TKN could be used to predict HABs in Lake Erie were also examined. To serve this goal, we performed Stepwise Multiple Linear Regression (SMLR) using nine predictor variables (TKN, TN, TP, NOx, SRP, TSS, flow, precipitation, temperature) to predict CI. The SMLR model was evaluated using the R-squared value. Figure 3 shows the steps of detecting and removing multicollinearity and moving towards SMLR.
Due to multicollinearity detected using Variance Inflation Factors (VIF), first PCA [49,50] was performed by transforming highly correlated variables into uncorrelated variables (i.e., principal components or PCs). PCA can reduce redundancy and retain variance in the nine predictor variables, addressing dimensionality and multicollinearity. To perform PCA, monthly total loads of TKN, TSS, TN, TP, SRP, and NOx, total precipitation, sum of flow, and the average temperature of the five months March through July for each year from 2002 to 2019 were calculated. The March–July period was selected to capture the dominant watershed nutrient-loading window associated with spring runoff, fertilizer application, and high discharge that precede and condition summer cyanobacterial bloom development; later months primarily reflect in-lake biological and physical processes rather than external nutrient export [51]. To interpret the PCs, the variable loadings for each component were examined. Loadings represent the correlation between the original variables and the PCs. Following common practice in multivariate, a variable to contribute meaningfully to a component if its absolute loading was ≥ 0.5 was considered. The dependent variable was the average CI for August from 2002 through 2019. Standardization was performed on independent variables. From the nine PC scores, the combination of statistically significant PCs to incorporate into the regression equation was developed by an iterative stepwise process. The SMLR model started with including all predictors, then excluding the least significant predictors through an automated model selection based on Akaike Information Criterion (AIC). The final model was selected when no further improvement in AIC was possible.

3. Results

3.1. Monotonic Trends

The MK trend test applied to monthly mean TKN loads in the Maumee River at Waterville for the last four decades exhibited a steady, consistent, and monotonic increase in concentrations, without a significant p-value (0.54). SS supported the robustness of the MK results by providing similar trends. The Maumee River TKN load exhibited positive SS, indicating an upward trend over time (z-value= 0.67, p-value= 0.5). However, the low z-value and high p-value of this observed trend suggested that it was inconclusive, statistically (Figure S3). The absence of a significant monotonic trend is not unexpected given the time span of the dataset, and there were many known climatic and landscape changes occurring in the Maumee River watershed during this time [21]. TP and SRP had similar upward monotonic trends, but these were also not statistically significant.

3.2. Oscillatory Trends

3.2.1. Sequential Mann–Kendall Forward and Backward Trends

The sequential Mann–Kendall trend provided an oscillatory trend by sequentially applying MK in forward and backward directions (Figure 4).
Sequential Mann–Kendall trend analysis offered the advantage of detecting inflection points over the period of analysis. The x-axis represents year, and the y-axis represents the moving average of standardized MK statistics for trend detection. For TKN loads in the Maumee River, sequential MK trend analysis revealed shifts in 2003 via forward Mann–Kendall trend intersection. Although HABs were again observed in WLE starting in the mid–1990s, high Microcystis biovolumes became persistent in 2003 [52,53]. The increasing forward trend after 2003 continued until 2013, followed by a slower increase until 2018. These differences coincide with changes in spring discharge, agricultural management, and nutrient transport pathways in the Maumee River watershed. However, both the forward and backward trends were within the 95% confidence interval throughout the entire time series. Since the forward and backward Mann-Kendall Z-values remain within the ±1.96 bounds throughout the time series, the null hypothesis of no significant trend cannot be rejected. Therefore, there is no statistically significant increasing or decreasing trend at the 95% confidence level. Variations in the trend can thus be due to random fluctuations rather than consistent increasing or decreasing patterns.

3.2.2. Seasonal Trend Decomposition by LOESS (STL)

STL was employed to extract long-term trends for TKN loads from the Maumee River and to enable a consistent comparison with previous studies (e.g., [24]). TKN loads declined from the early 1980s until around 2000, then rose sharply to a peak around 2010, followed by a decline through 2020 (Figure 5). While this pattern may suggest multiple phases, the overall shape is more consistent with a sigmoidal trend than with true oscillations. The sequential Mann–Kendall test identified a notable shift in 2003, which aligns with the upward inflection seen in the STL trend. Additionally, the forward Mann–Kendall results support the recent decreasing phase in TKN loads, suggesting a potential shift in nutrient loading dynamics over the past decade.

3.3. Comparison of TKN Trends to TP and SRP

STL was also applied to SRP, TP, precipitation, temperature, and flow. TKN, SRP, and TP showed similar trends, i.e., increasing in the 1990s and declining after 2010 (Figure 6). SRP showed a similar increasing pattern as TKN when HAB recurrence was observed in WLE. The apparent changepoints in TKN, TP, and SRP may be influenced by concurrent shifts in river discharge (Figure S4), making it difficult to isolate the effects of discharge from those of other drivers.

3.4. Comparison of TKN Trends to Bloom Extent

Average August CI indices were considered light (0–4), mild (4–6.8), significant (6.8–15.2), and extreme (>15.2). Significant CIs were observed in 2008, 2009, and 2017, and extreme CIs occurred in 2011, 2013, and 2015 (Figure 7a). 2011 through 2015 marked a high bloom extent period (except for 2012), and STL trends for TKN, TP, and SRP loads indicated decreasing patterns influenced by flow but near peak values for the trends during this period.
Available data for CI were not sufficient to develop STL trends, so sequential MK trends were applied instead. An inflection point in 2008 showed a distinct shift in the trend (Figure 7b), which is supported by the extreme HABs in 2011 and 2015. The sequential forward and backward Mann–Kendall tests indicate a statistically significant increasing trend in the dataset, as the Z-values exceed the ±1.96 threshold at the 95% confidence level. The forward trend was increasing but decreased after 2017. The observed shape was characterized by an increase after 2008 and a decrease after 2017.

3.5. Predictive Modeling

3.5.1. Correlation

VIF was extremely high (>1; Table 1), indicating severe multicollinearity, as expected due to the interrelated shared sources and drivers of nutrients. Multicollinearity indicates strong correlation between predictor variables, making it difficult to identify the most important factors [54,55]. As a result, simple multiple linear regression was not suitable for this analysis. Figure 8 shows that, except for climatic drivers (precipitation and temperature), all nutrient loads and streamflow were highly correlated with each other.

3.5.2. Principal Component Analysis

To predict CI using MLR and to address multicollinearity, PCA was applied to the nine predictor variables to transform them into unrelated components while preserving variance. The loads (Table 2) were examined to interpret the contribution of each predictor variable to the principal components.
PC1 (+nutrients) had positive, and nearly equal, loadings from most of the variables except for precipitation and temperature and also explained most of data’s variation (Figure S5). PC2 (−climate) had high and negative loadings from temperature and precipitation. PC3 (+precipitation) had high positive loadings from precipitation. PC4 (−N) had high negative loadings from NOx, TN, and temperature. PC5 was interpreted as a dissolved phosphorus (Dis. P) axis, characterized by a high positive loading from SRP and a strong negative loading from TSS, indicating a gradient from particulate-bound to bioavailable phosphorus. PC6 (+SRP) had high positive loadings from SRP and high negative loadings from TKN and flow. PC7 (+TKN) had high positive loadings from TKN and less strong negative loading from flow, meaning PC7 can be interpreted as a “concentration-driven TKN” componentwhere elevated TKN occurs during low-flow conditions. PC8 (-TP) had a strong negative loading from TP and less strong positive loading from TSS. PC9 (+TN) had a strong positive loading from TN and a strong negative loading from NOx. Only PC1 had an eigen value more than 1. To determine how many PCs were explaining the most variance, a scree plot of cumulative explained variance versus number of components was developed (Figure 9, Table S6), which showed that 98% of the variance was explained by the first seven PCs.

3.5.3. Stepwise Linear Regression

Stepwise MLR was performed, using all PCs as candidate predictors, to identify the most effective predictors of CI. Model selection was based on a combination of Akaike Information Criterion (AIC), adjusted R2, and overall model performance rather than statistical significance of individual coefficients alone. The optimal model included PC1 (+nutrients), PC5 (+DIS. P), and PC7 (+TKN), yielding an overall model p-value of 0.0044. The derived equation was
CI = 1.697 PC1 + 12.109 PC5 − 8.12 PC7 + 8.953
where PC scores were calculated using the loadings reported in Table 2. PC1 represents overall nutrient loading, PC5 reflects dissolved phosphorus availability, and PC7 captures a low-flow nutrient regime characterized by elevated TKN and reduced discharge. Although the regression coefficient for PC7 was not individually significant at α = 0.05 (p = 0.20), its inclusion improved overall model performance and reduced AIC, indicating that it contributes explanatory value within the multivariate framework (Table 3). Detailed regression results are provided in Table S7.
The model performed moderately well given the absence of long-term data for bloom extent. The model’s multiple R-squared was 0.6, and adjusted R-squared was 0.51. Together, the model’s p-value was 0.004, which is less than our acceptable α = 0.05, supporting the usefulness of the model in predicting CI. The negative association between PC7 and CI suggests that elevated TKN under low-flow conditions does not independently increase bloom extent, but that nitrogen remains a necessary component interacting with phosphorus availability and hydrologic conditions to explain bloom dynamics.

4. Discussion

In this study, both monotonic and oscillatory long-term trends in Total Kjeldahl Nitrogen (TKN) were examined, and their relationship with bloom extent (CI) in Western Lake Erie was evaluated. TKN exhibited temporal patterns like soluble reactive phosphorus (SRP), reflecting that both represent highly bioavailable nutrient fractions. While extensive management efforts have focused on reducing phosphorus loads in the Western Lake Erie Basin, nitrogen has not been explicitly targeted, and phosphorus-oriented practices do not consistently reduce nitrogen losses.
Despite declining trends in TKN, TP, and SRP after approximately 2010, bloom extent increased markedly after 2008, with extreme events in 2011, 2013, 2015, and 2017. This decoupling indicates that bloom response is governed not by total annual nutrient loads alone, but by flow-mediated nutrient delivery, timing of availability, seasonal bloom dynamics and temperature-dependent internal nutrient recycling. NOAA bloom severity indices suggest less intense blooms after 2019 [6], although blooms remain problematic. During these periods, blooms were particularly problematic due to their large spatial extent, persistence, and frequent toxin production, which have caused drinking water advisories, ecosystem degradation, and economic impacts in Western Lake Erie. Interpretation of these nutrient trends requires careful consideration of hydrologic influences. Observed changepoints in TKN, TP, and SRP loads coincide closely with shifts in river discharge (Figure S3), indicating strong hydrologic control on nutrient delivery. Previous studies have shown that flow normalization can partially account for these effects in the Maumee River [56]. In this study, load-based metrics to represent nutrient export to Western Lake Erie were focused; however, this limits the ability to fully separate management-driven changes from discharge-driven variability. Consequently, identified changepoints likely reflect combined hydrologic and watershed influences rather than discrete responses.
Although nutrient stoichiometry (e.g., Redfield ratios) is commonly used to infer in-lake nutrient limitation, such ratios are not directly applicable to watershed-scale, flow-integrated nutrient load analyses and were therefore not included in this study. The predictive modeling framework explicitly incorporated phosphorus (SRP and TP), nitrogen (TKN, TN, NOx), suspended sediments, and hydrologic and climatic drivers using PCA and stepwise regression. Both SRP and TKN emerged as significant contributors within a multivariate, flow-dependent framework, demonstrating that phosphorus is included and mechanistically linked with nitrogen and hydrology. The negative association between the TKN-dominated component (PC7) and CI suggests that elevated TKN under low-flow conditions alone does not increase bloom intensity, but that nitrogen remains necessary in combination with phosphorus and flow to explain bloom dynamics. While PC7 has a relatively small contribution to total variance, it captures a distinct low-flow nutrient regime characterized by elevated TKN and reduced discharge. As such, PC7 should not be interpreted as representing TKN independently, but rather as reflecting an interaction between nitrogen availability and hydrologic conditions. This result indicates that TKN contributes to bloom dynamics in specific flow contexts, rather than acting as a dominant or standalone driver.
Our results are consistent with, but extend beyond, previous trend-based studies in the Maumee River and Lake Erie. Decreasing TP and SRP trends in earlier decades [57], stabilization of phosphorus loads after the early 2000s, and declining TKN after 2010 [56] align with our findings. By applying Sequential Mann–Kendall and STL analyses, multi-phase and oscillatory behavior was identified that is not captured by monotonic or flow-normalized approaches alone [58].
Differences in the timing of nutrient trends further support a mechanistic interpretation. SRP increased earlier than TKN and TP, likely due to its higher solubility and sensitivity to tile drainage, reduced tillage, and increased discharge [21]. The delayed decline in TP relative to SRP and TKN may reflect its dominance by nonpoint sources and erosion processes [59]. Increasing precipitation and flow identified through STL analysis (SI) further emphasize hydrologic control on nutrient transport.
Cyanobacterial blooms worldwide including Lakes Taihu and Dianchi [60,61], East African lakes [62], Lake Biwa [63], Lake Okeechobee [64], and Lake Taupo [65] demonstrate the combined roles of nitrogen and phosphorus in supporting blooms. In the Maumee River watershed, nitrogen losses primarily occur through leaching and tile drainage, while phosphorus losses are dominated by runoff and erosion. Conservation practices such as drainage water management can effectively reduce nitrogen [66] but have variable effects on phosphorus [67,68]. These contrasting pathways support recommendations for dual nutrient reductions, including proposed nitrogen reduction targets alongside phosphorus controls [14].
There are a few limitations to this work. Traditional monotonic trend tests showed limited significance, reflecting the complexity of long-term watershed-scale datasets. The moderate model performance (R2 = 0.60) and the lack of significance for some components likely reflect unmodeled internal nutrient recycling and data constraints. The use of satellite-derived CI data through 2019 further limits interpretation, and future work should incorporate updated bloom metrics and toxicity indicators. While explicit seasonal and physical process modeling is beyond the scope of this study, these factors represent important directions for future work. Future analyses separating bloom-season and non-bloom-season dynamics may further improve model performance and interpretation. Nevertheless, the results demonstrate that nitrogen, phosphorus, and flow jointly govern bloom dynamics, reinforcing the need for integrated, dual-nutrient management strategies rather than phosphorus-only controls.

5. Conclusions

The overall goal of this study was to investigate patterns of TKN loads from the Maumee River and compare them with TP and SRP patterns, as well as to examine the potential role of TKN as a predictive variable for HABs in Western Lake Erie. Our results show that TKN loads have long-term patterns like TP and SRP and then emerged as a predictor in statistical models of bloom dynamics. However, negative associations between PC7 (high TKN under low flow) and bloom size highlight the complexity of interpreting external nitrogen effects, particularly because of the strong role of internal N recycling during low-flow, high-biomass periods (e.g., [17,69]),but also shows that the combination of concentration-driven TKN plays an important role along with phosphorus and hydrology. These findings indicate the need for further analysis incorporating internal cycling and seasonal variability. Together with previous studies, our results contribute to the growing evidence that both nitrogen and phosphorus should be considered in efforts to develop more effective nutrient management strategies for reducing bloom occurrences in Lake Erie.

Supplementary Materials

The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/w18040465/s1, Figure S1: Exploratory Analysis of TKN; Figure S2: Thiessen polygon for 5 stations containing precipitation and temperature data; Figure S3: Mann Kendall and Sen’s Slope trend for TKN; Figure S4: Trends for Flow, Precipitation; Figure S5: PCA biplot of observations and variable loadings for PC1 (71.5%) and PC2 (11.1%); Table S1: National Land Cover Class area percentage for Maumee watershed 2; Table S2: List of Counties in Maumee Watershed 3; Table S3: List of Rivers 4; Table S4: Stations used for Thiessen Polygon 5; Table S5: Statistics for TKN load 7; Table S6: Properties of PCA 9; Table S7: Multiple Linear Regression.

Author Contributions

Conceptualization, N.N.K. and R.L.M.; Methodology, N.N.K., H.W.P., M.J.M., S.E.N., N.R. and R.L.M.; Formal analysis, N.N.K.; Resources, N.N.K. and R.L.M.; Data curation, N.N.K.; Writing—original draft, N.N.K.; Writing—review & editing, N.N.K., H.W.P., M.J.M., S.E.N., N.R. and R.L.M.; Supervision, R.L.M.; Project administration, R.L.M.; Funding acquisition, R.L.M. All authors have read and agreed to the published version of the manuscript.

Funding

This project was supported by the National Science Foundation (NSF) under the award numbers NSF DISES 2108917 and 2418066 and USA National Institutes of Health (NIH) (2P01ES028939-06). MJM was also supported during manuscript preparation by the Estonian Research Council (PRG1954). Support from the Open Access Publishing Fund administrated through the University of Arkansas Libraries was provided to publish this manuscript.

Data Availability Statement

The original data presented in the study are openly available as follows: Flow data were collected from USGS Water Data for Nation. https://waterdata.usgs.gov/nwis/inventory?agency_code=USGS&site_no=04193500 Accessed: 12 July 2023. National Center for Water Quality Research (NCWQR)(Data Portal-HTLP_DataPortal). Station Name: Maumee. Corresponding USGS Station: 04193500. Watershed (HUC 8): Lake Erie. Data Range: 1 January 1982 to 31 December 2022.

Conflicts of Interest

The authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

References

  1. Ansari, A.A.; Gill, S.S.; Khan, F.A. Eutrophication: Threat to Aquatic Ecosystems. In Eutrophication: Causes, Consequences and Control; Ansari, A.A., Singh Gill, S., Lanza, G.R., Rast, W., Eds.; Springer: Dordrecht, The Netherlands, 2011; pp. 143–170. [Google Scholar]
  2. Dodds, W.K.; Smith, V.H. Nitrogen, Phosphorus, and Eutrophication in Streams. Inland Waters 2016, 6, 155–164. [Google Scholar] [CrossRef]
  3. Cargnelli, L.; Clement, S.; Drouin, R. Lake Erie Lakewide Action and Management Plan, 2019–2023. Available online: https://www.epa.gov/greatlakes/lakewide-action-and-management-plans-great-lakes (accessed on 17 May 2023).
  4. Steinman, A.D.; Cardinale, B.J.; Munns, W.R.; Ogdahl, M.E.; Allan, J.D.; Angadi, T.; Bartlett, S.; Brauman, K.; Byappanahalli, M.; Doss, M.; et al. Ecosystem Services in the Great Lakes. J. Gt. Lakes Res. 2017, 43, 161–168. [Google Scholar] [CrossRef]
  5. Watson, S.B.; Miller, C.; Arhonditsis, G.; Boyer, G.L.; Carmichael, W.; Charlton, M.N.; Confesor, R.; Depew, D.C.; Höök, T.O.; Ludsin, S.A.; et al. The Re-Eutrophication of Lake Erie: Harmful Algal Blooms and Hypoxia. Harmful Algae 2016, 56, 44–66. [Google Scholar] [CrossRef] [PubMed]
  6. NCCOS. Western Lake Erie HAB Seasonal Assessment 2023. Available online: https://nccospublicstor.blob.core.windows.net/hab-data/bulletins/lake-erie/2023/FinalAssessment_2023_10.pdf (accessed on 26 August 2024).
  7. GLWQA. Recommended Phosphorus Loading Targets for Lake Erie. 2015. Available online: https://www.epa.gov/sites/default/files/2015-06/documents/report-recommended-phosphorus-loading-targets-lake-erie-201505.pdf (accessed on 26 December 2025).
  8. Kalcic, M.M.; Kirchhoff, C.; Bosch, N.; Muenich, R.L.; Murray, M.; Griffith Gardner, J.; Scavia, D. Engaging Stakeholders To Define Feasible and Desirable Agricultural Conservation in Western Lake Erie Watersheds. Environ. Sci. Technol. 2016, 50, 8135–8145. [Google Scholar] [CrossRef]
  9. Scavia, D.; Kalcic, M.; Muenich, R.L.; Read, J.; Aloysius, N.; Bertani, I.; Boles, C.; Confesor, R.; DePinto, J.; Gildow, M.; et al. Multiple Models Guide Strategies for Agricultural Nutrient Reductions. Front. Ecol. Environ. 2017, 15, 126–132. [Google Scholar] [CrossRef]
  10. Chaffin, J.D.; Bridgeman, T.B.; Bade, D.L. Nitrogen Constrains the Growth of Late Summer Cyanobacterial Blooms in Lake Erie. Adv. Microbiol. 2013, 3, 126–132. [Google Scholar] [CrossRef]
  11. Gobler, C.J.; Burkholder, J.M.; Davis, T.W.; Harke, M.J.; Johengen, T.; Stow, C.A.; Van de Waal, D.B. The Dual Role of Nitrogen Supply in Controlling the Growth and Toxicity of Cyanobacterial Blooms. Harmful Algae 2016, 54, 87–97. [Google Scholar] [CrossRef] [PubMed]
  12. Newell, S.E.; Davis, T.W.; Johengen, T.H.; Gossiaux, D.; Burtner, A.; Palladino, D.; McCarthy, M.J. Reduced Forms of Nitrogen Are a Driver of Non-Nitrogen-Fixing Harmful Cyanobacterial Blooms and Toxicity in Lake Erie. Harmful Algae 2019, 81, 86–93. [Google Scholar] [CrossRef]
  13. Paerl, H.W.; Scott, J.T.; McCarthy, M.J.; Newell, S.E.; Gardner, W.S.; Havens, K.E.; Hoffman, D.K.; Wilhelm, S.W.; Wurtsbaugh, W.A. It Takes Two to Tango: When and Where Dual Nutrient (N & P) Reductions Are Needed to Protect Lakes and Downstream Ecosystems. Environ. Sci. Technol. 2016, 50, 10805–10813. [Google Scholar] [CrossRef] [PubMed]
  14. Paerl, H.W.; Chaffin, J.D.; Cheshire, J.H.; Plaas, H.E.; Barnard, M.A.; Goerlitz, L.B.; Braddy, J.S.; Sabo, A.; Nelson, L.M.; Yue, L. Dual Phosphorus and Nitrogen Nutrient Reduction Will Be More Effective than a Phosphorus-Only Reduction in Mitigating Diatom and Cyanobacterial Blooms in Lake Erie, USA–Canada. Limnol. Oceanogr. 2024, 69, 2913–2928. [Google Scholar] [CrossRef]
  15. Stow, C.A.; Stumpf, R.P.; Rowe, M.D.; Johnson, L.T.; Carrick, H.J.; Yerubandi, R. Model Assumptions Limit Implications for Nitrogen and Phosphorus Management. J. Gt. Lakes Res. 2022, 48, 1735–1737. [Google Scholar] [CrossRef]
  16. Huisman, J.; Dittmann, E.; Fastner, J.; Schuurmans, J.M.; Scott, J.T.; Van de Waal, D.B.; Visser, P.M.; Welker, M.; Chorus, I. Comment on “Models Predict Planned Phosphorus Load Reduction Will Make Lake Erie More Toxic”. Science 2022, 378, eadd9959. [Google Scholar] [CrossRef]
  17. Hoffman, D.K.; McCarthy, M.J.; Boedecker, A.R.; Myers, J.A.; Newell, S.E. The Role of Internal Nitrogen Loading in Supporting Non-N-Fixing Harmful Cyanobacterial Blooms in the Water Column of a Large Eutrophic Lake. Limnol. Oceanogr. 2022, 67, 2028–2041. [Google Scholar] [CrossRef]
  18. 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]
  19. Ohio EPA. Fact Sheet Maumee Watershed Nutrient; Ohio EPA: Columbus, OH, USA, 2023. Available online: https://dam.assets.ohio.gov/image/upload/epa.ohio.gov/Portals/35/tmdl/MaumeeNutrient/Maumee-TMDL-FS.pdf (accessed on 26 August 2024).
  20. Dewitz, J. National Land Cover Database (NLCD) 2021 Products; U.S. Geological Survey: Reston, VA, USA, 2023. [CrossRef]
  21. Jarvie, H.P.; Johnson, L.T.; Sharpley, A.N.; Smith, D.R.; Baker, D.B.; Bruulsema, T.W.; Confesor, R. Increased Soluble Phosphorus Loads to Lake Erie: Unintended Consequences of Conservation Practices? J. Environ. Qual. 2017, 46, 123–132. [Google Scholar] [CrossRef] [PubMed]
  22. Williams, M.R.; King, K.W. Changing Rainfall Patterns Over the Western Lake Erie Basin (1975–2017): Effects on Tributary Discharge and Phosphorus Load. Water Resour. Res. 2020, 56, e2019WR025985. [Google Scholar] [CrossRef]
  23. Wang, R.; Ma, Y.; Zhao, G.; Zhou, Y.; Shehab, I.; Burton, A. Investigating Water Quality Sensitivity to Climate Variability and Its Influencing Factors in Four Lake Erie Watersheds. J. Environ. Manag. 2023, 325, 116449. [Google Scholar] [CrossRef] [PubMed]
  24. Stow, C.A.; Cha, Y.; Johnson, L.T.; Confesor, R.; Richards, R.P. Long-Term and Seasonal Trend Decomposition of Maumee River Nutrient Inputs to Western Lake Erie. Environ. Sci. Technol. 2015, 49, 3392–3400. [Google Scholar] [CrossRef]
  25. 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]
  26. Czajkowski, K.; Lawrence, P.L. GIS and Remote Sensing Applications for Watershed Planning in the Maumee River Basin, Ohio. In Geotechnologies and the Environment; Springer: Dordrecht, The Netherlands, 2013; Volume 7. [Google Scholar] [CrossRef]
  27. Soil Survey Staff, Natural Resources Conservation Service, United States Department of Agriculture Web Soil Survey 2024. Available online: https://websoilsurvey.nrcs.usda.gov/app/ (accessed on 26 August 2024).
  28. Peel, M.C.; Finlayson, B.L.; McMahon, T.A. Updated World Map of the Köppen-Geiger Climate Classification. Hydrol. Earth Syst. Sci. 2007, 11, 1633–1644. [Google Scholar] [CrossRef]
  29. Pike, J.G. The Estimation of Annual Run-off from Meteorological Data in a Tropical Climate. J. Hydrol. 1964, 2, 116–123. [Google Scholar] [CrossRef]
  30. King, K.W.; Williams, M.R.; Fausey, N.R. Contributions of Systematic Tile Drainage to Watershed-Scale Phosphorus Transport. J. Environ. Qual. 2015, 44, 486–494. [Google Scholar] [CrossRef]
  31. NCWQR. Heidelberg Tributary Loading Program (HTLP) Dataset; Zenodo: Geneva, Switzerland, 2022. [Google Scholar] [CrossRef]
  32. Roerdink, A. Water Quality in Ohio Rivers and Streams; NCQWR: Tiffin, OH, USA, 2017; Available online: https://ncwqr.org/wp-content/uploads/2019/04/ncwqr-study-plan-20170127-with-addendums.pdf (accessed on 21 January 2025).
  33. Lawrimore, J.; Ray, R.; Applequist, S.; Korzeniewski, B.; Menne, M.J. Global Summary of the Month (GSOM), Version 1; NOAA National Centers for Environmental Information: Asheville, NC, USA, 2016. [CrossRef]
  34. Ai, H.; Zhang, K.; Sun, J.; Zhang, H. Short-Term Lake Erie Algal Bloom Prediction by Classification and Regression Models. Water Res. 2023, 232, 119710. [Google Scholar] [CrossRef]
  35. Stumpf, R.P.; Johnson, L.T.; Wynne, T.T.; Baker, D.B. Forecasting Annual Cyanobacterial Bloom Biomass to Inform Management Decisions in Lake Erie. J. Gt. Lakes Res. 2016, 42, 1174–1183. [Google Scholar] [CrossRef]
  36. Kendall, M.G. Rank Correlation Methods; Rank correlation methods; Griffin: Oxford, England, 1948. [Google Scholar]
  37. Mann, H.B. Nonparametric Tests Against Trend. Econometrica 1945, 13, 245–259. [Google Scholar] [CrossRef]
  38. Gilbert, R.O. Statistical Methods for Environmental Pollution Monitoring; Van Nostrand Reinhold Company: New York, NY, USA, 1987. [Google Scholar]
  39. Sen, P.K. Estimates of the Regression Coefficient Based on Kendall’s Tau. J. Am. Stat. Assoc. 1968, 63, 1379–1389. [Google Scholar] [CrossRef]
  40. Von Storch, H.; Navarra, A. (Eds.) Analysis of Climate Variability; Springer: Berlin/Heidelberg, Germany , 1995. [Google Scholar]
  41. Sneyers, R. On the Statistical Analysis of Series of Observations; Technical Note No. 143; World Meteorological Organization (WMO): Geneva, Switzerland, 1990; 192p. [Google Scholar] [CrossRef]
  42. Cleveland, R.; Cleveland, W.; McRay, J.; Terpenning, I. STL: A Seasonal-Trend Decomposition Procedure Based on LOESS. J. Off. Stat. 1990, 6, 3–73. Available online: https://www.proquest.com/openview/cc5001e8a0978a6c029ae9a41af00f21/1?pq-origsite=gscholar%26cbl=105444 (accessed on 12 January 2025).
  43. Cho, K.H.; Kang, J.-H.; Ki, S.J.; Park, Y.; Cha, S.M.; Kim, J.H. Determination of the Optimal Parameters in Regression Models for the Prediction of Chlorophyll-a: A Case Study of the Yeongsan Reservoir, Korea. Sci. Total Environ. 2009, 407, 2536–2545. [Google Scholar] [CrossRef]
  44. Liu, Y.; Guo, H.; Yang, P. Exploring the Influence of Lake Water Chemistry on Chlorophyll a: A Multivariate Statistical Model Analysis. Ecol. Model. 2010, 221, 681–688. [Google Scholar] [CrossRef]
  45. Xiaobo, L.; Fei, D.; Guojian, H.; Jingling, L. Use of PCA-RBF Model for Prediction of Chlorophyll-a in Yuqiao Reservoir in the Haihe River Basin, China. Water Supply 2013, 14, 73–80. [Google Scholar] [CrossRef]
  46. Lee, Y.; Ha, S.-Y.; Park, H.-K.; Han, M.-S.; Shin, K.-H. Identification of Key Factors Influencing Primary Productivity in Two River-Type Reservoirs by Using Principal Component Regression Analysis. Environ. Monit. Assess. 2015, 187, 213. [Google Scholar] [CrossRef]
  47. Qiuhua, L.; Lihai, S.; Tingjing, G.; Lei, Z.; Teng, O.; Guojia, H.; Chuan, C.; Cunxiong, L. Use of Principal Component Scores in Multiple Linear Regression Models for Simulation of Chlorophyll-a and Phytoplankton Abundance at a Karst Deep Reservoir, Southwest of China. Acta Ecol. Sin. 2014, 34, 72–78. [Google Scholar] [CrossRef]
  48. Çamdevýren, H.; Demýr, N.; Kanik, A.; Keskýn, S. Use of Principal Component Scores in Multiple Linear Regression Models for Prediction of Chlorophyll-a in Reservoirs. Ecol. Model. 2005, 181, 581–589. [Google Scholar] [CrossRef]
  49. Hotelling, H. Analysis of a Complex of Statistical Variables into Principal Components. J. Educ. Psychol. 1933, 24, 417–441. [Google Scholar] [CrossRef]
  50. Maćkiewicz, A.; Ratajczak, W. Principal Components Analysis (PCA). Comput. Geosci. 1993, 19, 303–342. [Google Scholar] [CrossRef]
  51. Scavia, D.; David Allan, J.; Arend, K.K.; Bartell, S.; Beletsky, D.; Bosch, N.S.; Brandt, S.B.; Briland, R.D.; Daloğlu, I.; DePinto, J.V.; et al. Assessing and Addressing the Re-Eutrophication of Lake Erie: Central Basin Hypoxia. J. Gt. Lakes Res. 2014, 40, 226–246. [Google Scholar] [CrossRef]
  52. Bridgeman, T.B.; Chaffin, J.D.; Filbrun, J.E. A Novel Method for Tracking Western Lake Erie Microcystis Blooms, 2002–2011. J. Gt. Lakes Res. 2013, 39, 83–89. [Google Scholar] [CrossRef]
  53. Rinta-Kanto, J.M.; Ouellette, A.J.A.; Boyer, G.L.; Twiss, M.R.; Bridgeman, T.B.; Wilhelm, S.W. Quantification of Toxic Microcystis spp. during the 2003 and 2004 Blooms in Western Lake Erie Using Quantitative Real-Time PCR. Environ. Sci. Technol. 2005, 39, 4198–4205. [Google Scholar] [CrossRef]
  54. Kline, R.B. Principles and Practice of Structural Equation Modeling, 4th ed.; The Guilford Press: New York, NY, USA, 2016; Available online: https://dl.icdst.org/pdfs/files4/befc0f8521c770249dd18726a917cf90.pdf (accessed on 12 January 2025).
  55. Young, D.S. Handbook of Regression Methods; Chapman and Hall/CRC: New York, NY, USA, 2018. [Google Scholar] [CrossRef]
  56. Rowland, F.E.; Stow, C.A.; Johnson, L.T.; Hirsch, R.M. Lake Erie Tributary Nutrient Trend Evaluation: Normalizing Concentrations and Loads to Reduce Flow Variability. Ecol. Indic. 2021, 125, 107601. [Google Scholar] [CrossRef]
  57. Richards, R.P.; Baker, D.B. Trends in Nutrient and Suspended Sediment Concentrations in Lake Erie Tributaries, 1975–1990. J. Gt. Lakes Res. 1993, 19, 200–211. [Google Scholar] [CrossRef]
  58. Choquette, A.F.; Hirsch, R.M.; Murphy, J.C.; Johnson, L.T.; Confesor, R.B. Tracking Changes in Nutrient Delivery to Western Lake Erie: Approaches to Compensate for Variability and Trends in Streamflow. J. Gt. Lakes Res. 2019, 45, 21–39. [Google Scholar] [CrossRef]
  59. Maccoux, M.J.; Dove, A.; Backus, S.M.; Dolan, D.M. Total and Soluble Reactive Phosphorus Loadings to Lake Erie: A Detailed Accounting by Year, Basin, Country, and Tributary. J. Gt. Lakes Res. 2016, 42, 1151–1165. [Google Scholar] [CrossRef]
  60. Li, S.; Liu, C.; Sun, P.; Ni, T. Response of Cyanobacterial Bloom Risk to Nitrogen and Phosphorus Concentrations in Large Shallow Lakes Determined through Geographical Detector: A Case Study of Taihu Lake, China. Sci. Total Environ. 2022, 816, 151617. [Google Scholar] [CrossRef]
  61. Liang, Z.; Xu, Y.; Qiu, Q.; Liu, Y.; Lu, W.; Wagner, T. A Framework to Develop Joint Nutrient Criteria for Lake Eutrophication Management in Eutrophic Lakes. J. Hydrol. 2021, 594, 125883. [Google Scholar] [CrossRef]
  62. Yao, X.; Zhao, Z.; Wang, J.; Kimirei, I.A.; Sekadende, B.C.; Mgana, H.F.; Zhang, L. Microbial Nitrogen Nutrition Links to Dissolved Organic Matter Properties in East African Lakes. Sci. Total Environ. 2024, 950, 175197. [Google Scholar] [CrossRef]
  63. Gao, B.; Chen, M.; Hao, H.; Yamashiki, Y.A.; Ishikawa, K.; Jiao, C.; Cai, J.; Ismail, S. Recent Mechanisms of Surface Ecological Changes Driven by Climate Change and Human Activities in Lake Biwa, Japan. Front. Ecol. Evol. 2023, 11, 1335990. [Google Scholar] [CrossRef]
  64. Mazzei, V.; Sullivan, K.L.; Loftin, K. Phytoplankton Assemblage Structure, Drivers, and Thresholds with a Focus on Harmful Algal Bloom Ecology in the Lake Okeechobee System, Florida, USA. Harmful Algae 2024, 142, 102744. [Google Scholar] [CrossRef]
  65. Abell, J.M.; Özkundakci, D.; Hamilton, D.P.; van Dam-Bates, P.; Mcdowell, R.W. Quantifying the Extent of Anthropogenic Eutrophication of Lakes at a National Scale in New Zealand. Environ. Sci. Technol. 2019, 53, 9439–9452. [Google Scholar] [CrossRef]
  66. Wang, Z.; Shao, G.; Lu, J.; Zhang, K.; Gao, Y.; Ding, J. Effects of Controlled Drainage on Crop Yield, Drainage Water Quantity and Quality: A Meta-Analysis. Agric. Water Manag. 2020, 239, 106253. [Google Scholar] [CrossRef]
  67. Saadat, S.; Bowling, L.; Frankenberger, J.; Kladivko, E. Nitrate and Phosphorus Transport through Subsurface Drains under Free and Controlled Drainage. Water Res. 2018, 142, 196–207. [Google Scholar] [CrossRef] [PubMed]
  68. Sanchez Valero, C.; Madramootoo, C.A.; Stämpfli, N. Water Table Management Impacts on Phosphorus Loads in Tile Drainage. Agric. Water Manag. 2007, 89, 71–80. [Google Scholar] [CrossRef]
  69. Boedecker, A.R.; Niewinski, D.N.; Newell, S.E.; Chaffin, J.D.; McCarthy, M.J. Evaluating Sediments as an Ecosystem Service in Western Lake Erie via Quantification of Nutrient Cycling Pathways and Selected Gene Abundances. J. Gt. Lakes Res. 2020, 46, 920–932. [Google Scholar] [CrossRef]
Water 18 00465 g001
Figure 1. Location of the Maumee River watershed and its major tributaries, along with the location of the water quality monitoring station used in this study.
Figure 1. Location of the Maumee River watershed and its major tributaries, along with the location of the water quality monitoring station used in this study.
Water 18 00465 g001
Water 18 00465 g002
Figure 2. Trend analysis framework applied in this study.
Figure 2. Trend analysis framework applied in this study.
Water 18 00465 g002
Water 18 00465 g003
Figure 3. Workflow for regression.
Figure 3. Workflow for regression.
Water 18 00465 g003
Water 18 00465 g004
Figure 4. Sequential Mann–Kendall forward and backward Z statistics for monthly TKN loads in the Maumee River (1982–2022). Intersections between forward and backward curves indicate potential trend inflection points, while values remaining within ±1.96 indicate no statistically significant monotonic trend at the 95% confidence level.
Figure 4. Sequential Mann–Kendall forward and backward Z statistics for monthly TKN loads in the Maumee River (1982–2022). Intersections between forward and backward curves indicate potential trend inflection points, while values remaining within ±1.96 indicate no statistically significant monotonic trend at the 95% confidence level.
Water 18 00465 g004
Water 18 00465 g005
Figure 5. STL long-term trend component of monthly TKN loads for the Maumee River at Waterville (1982–2022). Note that the y-axis is compressed to better visualize the trend.
Figure 5. STL long-term trend component of monthly TKN loads for the Maumee River at Waterville (1982–2022). Note that the y-axis is compressed to better visualize the trend.
Water 18 00465 g005
Water 18 00465 g006
Figure 6. STL long-term trend components for (a) soluble reactive phosphorus (SRP) and (b) total phosphorus (TP) monthly mean loads in the Maumee River at Waterville (1982–2022). The STL trend for Total Kjeldahl Nitrogen (TKN) is shown on a secondary axis (dashed line) for comparison. The y-axis ranges are intentionally compressed to emphasize multi-decadal variability.
Figure 6. STL long-term trend components for (a) soluble reactive phosphorus (SRP) and (b) total phosphorus (TP) monthly mean loads in the Maumee River at Waterville (1982–2022). The STL trend for Total Kjeldahl Nitrogen (TKN) is shown on a secondary axis (dashed line) for comparison. The y-axis ranges are intentionally compressed to emphasize multi-decadal variability.
Water 18 00465 g006
Water 18 00465 g007
Figure 7. (a) Annual average Chlorophyll Index (CI), derived from 10-day composite satellite observations for the Western Lake Erie basin (2002–2019), used as a proxy for bloom extent. CI reflects spatially aggregated surface chlorophyll-a and does not represent toxicity. (b) Sequential Mann–Kendall forward and backward Z-statistics applied to annual CI values. Dashed horizontal lines denote the ±1.96 threshold corresponding to the 95% confidence level. Intersections and divergence between forward and backward series indicate periods of changing bloom dynamics rather than a single monotonic trend.
Figure 7. (a) Annual average Chlorophyll Index (CI), derived from 10-day composite satellite observations for the Western Lake Erie basin (2002–2019), used as a proxy for bloom extent. CI reflects spatially aggregated surface chlorophyll-a and does not represent toxicity. (b) Sequential Mann–Kendall forward and backward Z-statistics applied to annual CI values. Dashed horizontal lines denote the ±1.96 threshold corresponding to the 95% confidence level. Intersections and divergence between forward and backward series indicate periods of changing bloom dynamics rather than a single monotonic trend.
Water 18 00465 g007
Water 18 00465 g008
Figure 8. Pearson correlation matrix for nutrient loads, hydrologic, and climatic variables used in the predictive modeling framework. Strong positive correlations among nutrient variables reflect shared sources and discharge-driven transport, while negative correlations with temperature highlight seasonal and hydrologic contrasts. The high degree of multicollinearity among predictors motivated the use of Principal Component Analysis prior to regression modeling.
Figure 8. Pearson correlation matrix for nutrient loads, hydrologic, and climatic variables used in the predictive modeling framework. Strong positive correlations among nutrient variables reflect shared sources and discharge-driven transport, while negative correlations with temperature highlight seasonal and hydrologic contrasts. The high degree of multicollinearity among predictors motivated the use of Principal Component Analysis prior to regression modeling.
Water 18 00465 g008
Water 18 00465 g009
Figure 9. Scree plot of cumulative explained variance versus number of components.
Figure 9. Scree plot of cumulative explained variance versus number of components.
Water 18 00465 g009
Table 1. VIF values for predictor variables.
Table 1. VIF values for predictor variables.
TKNSRPTSSNOxTNTPPTFlow
1.8 × 10410342 × 1053.3 × 1051102.72.5215.3
Table 2. Loadings of predictor variables in each PC. Large positive loads (>0.5) are indicated in darker grey, while large negative loads (<−0.5) are indicated in light grey and bold. Distinguishing names are provided for each PC based on major loadings.
Table 2. Loadings of predictor variables in each PC. Large positive loads (>0.5) are indicated in darker grey, while large negative loads (<−0.5) are indicated in light grey and bold. Distinguishing names are provided for each PC based on major loadings.
VariablesPC1
(+Nutrients)		PC2
(−Climate)		PC3
(+Precipitation)		PC4
(−N)		PC5
(+DIS. P)		PC6
(+SRP)		PC7
(+TKN)		PC8
(−TP)		PC9
(+TN)
TKN0.37−0.12−0.16−0.080.02−0.430.730.24−0.18
SRP0.360.09−0.140.000.590.670.150.110.00
TSS0.36−0.04−0.330.14−0.640.28−0.210.450.00
NOx0.350.120.25−0.58−0.080.01−0.26−0.15−0.61
TN0.360.060.16−0.47−0.06−0.09−0.03−0.070.77
TP0.37−0.06−0.270.31−0.170.020.06−0.810.00
Precipitation0.23−0.580.700.30−0.050.130.040.040.00
Temperature−0.18−0.78−0.42−0.390.110.06−0.12−0.060.00
Flow0.36−0.05−0.140.280.42−0.50−0.560.180.00
Table 3. Final predictive model coefficient estimates and p-values.
Table 3. Final predictive model coefficient estimates and p-values.
Estimatesp-Values
Intercept8.953<0.001
PC1 (+nutrients)1.6970.007
PC5 (+DIS. P)12.1090.008
PC7 (+TKN)−8.120.2
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

Khan, N.N.; Paerl, H.W.; McCarthy, M.J.; Newell, S.E.; Rudko, N.; Muenich, R.L. Trends in Maumee River Nitrogen Loads and Their Complex Relationship to Harmful Algal Blooms in Western Lake Erie. Water 2026, 18, 465. https://doi.org/10.3390/w18040465

AMA Style

Khan NN, Paerl HW, McCarthy MJ, Newell SE, Rudko N, Muenich RL. Trends in Maumee River Nitrogen Loads and Their Complex Relationship to Harmful Algal Blooms in Western Lake Erie. Water. 2026; 18(4):465. https://doi.org/10.3390/w18040465

Chicago/Turabian Style

Khan, Nusrat N., Hans W. Paerl, Mark J. McCarthy, Silvia E. Newell, Noah Rudko, and Rebecca Logsdon Muenich. 2026. "Trends in Maumee River Nitrogen Loads and Their Complex Relationship to Harmful Algal Blooms in Western Lake Erie" Water 18, no. 4: 465. https://doi.org/10.3390/w18040465

APA Style

Khan, N. N., Paerl, H. W., McCarthy, M. J., Newell, S. E., Rudko, N., & Muenich, R. L. (2026). Trends in Maumee River Nitrogen Loads and Their Complex Relationship to Harmful Algal Blooms in Western Lake Erie. Water, 18(4), 465. https://doi.org/10.3390/w18040465

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