Predicting stream water quality under different urban development pattern scenarios with an interpretable machine learning approach

Long-term water quality simulation and driving factors identification within the watershed scale using machine learning

Understanding long-term trends and analyzing their driving factors are essential to effectively enhance water quality in watersheds. In China, although the overall quality of surface water continues to improve, significant issues remain in certain regions. The Liao River Basin, a critical industrial hub and key agricultural grain base in northeast China, continues to face unstable water quality conditions, despite over 20 years of management efforts. This study compared several data-driven models (random forest (RF), support vector machine regression (SVR), K-nearest neighbors (KNN), stacking, long short-term memory (LSTM), convolutional-long short-term memory (CNN-LSTM)), to accurately fill the water quality data gaps (i.e., total nitrogen (TN), ammonia nitrogen (NH3-N), total phosphorus (TP), chemical oxygen demand (CODCr), permanganate index (CODMn), electroconductibility (E)) from 1980 to 2022 in Liao River Basin. In addition, the SHapley Additive exPlanations (SHAP) model was employed to quantitatively assess the driving factors of water quality. The results showed that the RF model exhibited robust predictive capabilities. TN showed a steady increase of approximately 20 % from 1980 to 2022, while the other parameters were effectively controlled. Anthropogenic activities, especially in agriculture and urban areas, were found to significantly contribute to water quality deterioration. Additionally, climatic factors such as extreme rainfall, annual average precipitation, and extreme temperatures-along with geographical factors like soil properties and slope, were found to play crucial roles in influencing water quality.

Intensifying Nutrient Removal in Hybrid-Constructed Wetlands Treating Urban Streamwater

This study investigated the influence of hydraulic retention time (HRT) variation and the presence of macrophytes on the efficiency of two pilot-scale hybrid-constructed wetlands (HCWs) treating urban streamwater contaminated with nontreated sanitary sewage contributions from the surrounding communities. Each HCW comprises a vertical unit (VF) and a horizontal unit (HF) filled with sand and gravel. HCW-P was planted withEichornia crassipes onto the filtering media, and HCW-C was set up as a control unit with no macrophytes. The novelty of this study consists of evaluating the combination of these factors (HRT and macrophytes) in the operation of HCWs for removing organic matter and nutrients. The operation of the HCWs was divided into step I, with a hydraulic retention time (HRT) of 9 days for 133 days, and step II, with an HRT of 5 days for 131 days. Neither HRT variation (p-value = 0.7691) nor the presence of macrophytes (p-value = 0.0941) influenced the COD removal, as the HCWs achieved high removal efficiencies (>87%) during the operation. HCW-P achieved higher total nitrogen (TN) removal efficiencies in steps I and II (56% and 78%) compared to HCW-C (31% and 48%) during the operation, demonstrating the improvement in removing TN due to the presence of macrophytes (p-value ≤ 0.05). In addition, the shorter HRT promoted an increase of 22% in TN removal for HCW-P (p-value ≤ 0.05). The macrophytes and longer HRT enhanced total ammonia nitrogen (TAN) removal, as HCW-P (46% and 88%) achieved higher removal efficiencies than HCW-C (29% and 72%) in steps I and II, respectively (p-value ≤ 0.05). Regarding total phosphorus (TP), HCW-C and HCW-P achieved removal efficiencies of 63% and 89% in step I and 69% and 96% in step II, confirming the influence of HRT and macrophytes on TP removal. Finally, macrophytes demonstrated adaptability and resilience to the operational conditions, even when fixed in HCWs, which presented robustness in removing organic matter and nutrients from the urban streamwater via biofilm assimilation and adsorption under HRT variations.

A unified subregional framework for modeling stream water quality across watersheds of a hydrologic subregion

Modeling stream water quality is informed by knowledge about pertinent factors and processes. The models must be validated against water quality observations, which may exist sufficiently in some watersheds (data rich watersheds) but may be limited or lacking in other cases (i.e., ungauged and poorly gauged watersheds). Machine learning (ML) algorithms have been growingly applied for water quality modeling, but they are limited to the data used for their training and validation. The question arises whether an ML-based model developed in one watershed can be transferred to adjacent watersheds. Here, we developed a unified subregional framework (i.e., one single consistent model configuration and standardized input variables) for modeling daily in-stream concentrations of nutrients-total phosphorus (TP) and total nitrogen (TN)-fecal coliform (FC) and dissolved oxygen (DO) in watersheds of a hydrologic subregion. The watersheds differ in their characteristics in terms of dominant land use/land cover (LULC) and topography. The framework was presented in the Peace-Tampa Bay subregion located in Southwest Florida. We found that the unified framework can be successfully developed for the watershed-scale modeling of DO and TP (Nash Sutcliffe Efficiency [NSE] > 0.75), and to a lesser extent for TN and FC (NSE > 0.49). The influence of dominant LULC was most prominent in modeling FC and TP, while the effect of topography was more pronounced for FC and TN than TP and DO. We also observed that longer-term antecedent conditions were more influential in modeling FC and TP, while shorter term saturation was more influential for modeling TN and DO. Insights from this study can be used to develop similarity criteria based on watershed characteristics, which support development of transferable models for predicting stream water quality in ungauged and poorly gauged watersheds.

Enhancing pollution management in watersheds: A critical review of total maximum daily load (TMDL) implementation

The total maximum daily load (TMDL) program is a regulatory tool to ensure that water bodies meet quality standards by calculating the maximum pollutant load that a water body can assimilate while meeting water quality criteria. Implementing the TMDL program poses significant challenges for water quality management, especially in the context of climate change. This review highlights the application of TMDL in water quality management through various case studies and field applications, demonstrating its practical implementation. TMDL programs have been extensively utilized for water quality management, from pollutant reduction to the adoption of best management practices. A comprehensive analysis of several models, covering watershed, economic, machine learning, and simple frameworks, is systematically discussed to examine their strengths, limitations, effectiveness, and adaptability within the TMDL framework. Criteria for model selection are emphasized, balancing factors such as data availability, model complexity and accuracy, as well as time and cost considerations. We demonstrate that emerging machine learning techniques, simplified modeling approaches, and margin of safety estimation methods can help address data limitations and mitigate uncertainties in water quality assessment during the total pollutant load control process. Additionally, we address current challenges and future research directions in TMDL, particularly regarding modeling complex pollutant interactions and enhancing regulatory frameworks. This comprehensive review serves as an invaluable resource for environmental scientists, water quality managers, and policymakers seeking to implement and apply the TMDL framework effectively. It provides a clear understanding of the TMDL process, its practical applications, and the selection of appropriate modeling tools for successful water quality management.

Landscape fragmentation of built-up land significantly impact on water quality in the Yellow River Basin

Urbanization development often leads to significant changes in the extent in area and fragmentation of built-up land landscape pattern (BLLP) in river basins, which greatly impact the processes of rainfall runoff and pollutant migration. Understanding the spatial scale effects and driving mechanisms of BLLP changes on water quality in large river basins is a challenging research topic and an international frontier in the interdisciplinary fields of geography and environment. This study analyzes the spatial variations of BLLP and water quality throughout the Yellow River Basin (YRB) during the rainy seasons from 2019 to 2021 (4 h scale). Utilized the random forest model to quantitatively separates the contributions of rainfall processes to surface runoff and water pollution, revealing the scale effects and non-linear driving mechanisms of BLLP impacts on water environment changes. The results indicate that: 1) The YRB exhibits great spatial heterogeneity in terms of both BLLP and water quality, with places with lower water quality displaying bigger areas and higher degrees of BLLP fragmentation. 2) The patch density and built-up land area (PD.B and CA.B) have a major impact on changes in water quality in the YRB, with notable impacts noted in circular buffer zones with radii of 20 km and 5 km, respectively. 3) PD.B is sensitive to water quality in the YRB, explaining 39.1%-49.5% of the variance under different rainfall conditions, and exhibits a significant non-linear relationship, with an impact threshold of 0.38 (n/100 ha). The study suggests that for large-scale regions like the YRB, the degree of BLLP fragmentation is more likely to lead to degradation of water environmental quality compared to its area. BLLP fragmentation due to higher PD.B and CA.B disrupts the original ecosystem and hydrological connectivity, resulting in poorer retention and filtration of pollutants carried by rainfall runoff, while increasing the export of other pollutants. However, once urbanization surpasses a certain threshold, the BLLP fragmentation can enhance water quality by reducing the impermeable surface connectivity, as they are no longer impacted by expanding areas. To achieve ecologically sustainable development, it is necessary to apply rational landscape management and water resource management policies that consider the dual process of how BLLP fragmentation affects the water environment.

Machine learning models for prediction of nutrient concentrations in surface water in an agricultural watershed

Prediction and quantification of nutrient concentrations in surface water has gained substantial attention during recent decades because excess nutrients released from agricultural and urban watersheds can significantly deteriorate surface water quality. Machine learning (ML) models are considered an effective tool for better understanding and characterization of nutrient release from agricultural fields to surface water. However, to date, no systematic investigations have examined the implementation of different classification and regression ML models in agricultural settings to predict nutrient concentrations in surface water using a group of input variables including climatological (e.g., precipitation), hydrological (e.g., stream flow) and field characteristics (i.e., land and crop use). In the current study, multiple classification (e.g., decision trees) and regression (e.g., regression trees) ML models were applied on a dataset pertaining to surface water quality in an agricultural watershed in southern Ontario, Canada (i.e., Upper Parkhill watershed). The target variables of these models were the nutrient concentrations in surface water including nitrate, total phosphorus, soluble reactive phosphorus, and total dissolved phosphorus. These target variables were predicted using physical and chemical water parameters of surface water (e.g., temperature and DO), climatological, hydrological, and field conditions as the input variables. The performance of these different models was assessed using various evaluation metrics such as classification accuracy (CA) and coefficient of determination (R2) for classification and regression models, respectively. In general, both the ensemble bagged trees and logistic regression (CA ≥ 0.72), and exponential Gaussian process regression (R2≥ 0.93) models were the optimal classification and regression ML algorithms, respectively, where they resulted in the highest prediction accuracy of the target variables. The insights and outcomes of the current study demonstrates that ML models can be employed to effectively predict and quantify the nutrient concentrations in surface waters to supplement field-collected monitoring data in agricultural watersheds, assisting in maintaining high quality of the available surface water resources.

Dynamic patterns and potential drivers of river water quality in a coastal city: Insights from a machine-learning-based framework and water management

River water quality continues to deteriorate under the coupled effects of climate change and human activities. Machine learning (ML) is a promising approach for analyzing water quality. Nevertheless, the spatiotemporal dynamics of river water quality and their potential mechanisms in changing environments remain incomprehensively understood through available ML-based researches. Here, we developed a ML-based framework integrating a self-organizing map (SOM) model with a random forest (RF) model. This framework was applied to simultaneously investigate the spatiotemporal patterns and potential drivers of river permanganate (CODMn), ammonia nitrogen (NH3-N), and total phosphorus (TP) dynamics across 34 sites from 2010 to 2020 in a coastal city threatened by deteriorating water environment in southeastern China. The sites were divided into two clusters in the spatial context with different water quality conditions. The year of 2015 for NH3-N and 2018 for CODMn and TP were identified as the key turning points of water quality variations. Features including sewage discharge, population dynamics, percentage of cultivated land, and fertilizer application contributed greatly to water quality deterioration. The increase in forest vegetation reflected by percentage of forest and leaf area index may improve water quality. The ML-based modeling framework demonstrated a promising way to address the spatiotemporal dynamics of river water quality, and provided insights for water management in a coastal city with intensifying human-nature interactions.

Machine learning approaches to identify spatial factors and their influential distances for heavy metal contamination in downstream sediment

Contaminated sediments can adversely affect aquatic ecosystems, making the identification and management of pollutant sources extremely important. In this study, we proposed machine learning approaches to reveal sources and their influential distances for heavy metal contamination of downstream sediment. We employed classification models with artificial neural networks (ANN) and random forest (RF), respectively, to predict the heavy metal contamination of stream sediments using upland environmental variables as input features. A comprehensive Korean nationwide monitoring database containing 1546 datasets was used to train and test the models. These datasets encompass the concentrations of eight heavy metals (Ar, Cd, Cr, Cu, Hg, Ni, Pb, and Zn) in sediment samples collected from 160 stream sites across the nation from 2014 to 2018. Model's prediction accuracy was evaluated for input feature sets from different influential upland areas defined by different buffer radii and the watershed boundary for each site. Although both ANN and RF models were unsatisfactory in predicting heavy metal quartile classes, RF-classifiers with adaptive synthetic oversampling (ORFC) showed reasonably well-predicted classes of the sediment samples based on the Canada's Sediment Quality Guidelines (accuracy ranged from 0.67 to 0.94). The best influential distance (i.e., buffer radius) was determined for each heavy metal based on the accuracy of ORFC. The results indicated that Cd, Cu and Pb had shorter influential distances (1.5-2.0 km) than the other heavy metals with little difference in accuracy for different influential distances. Feature importance calculation revealed that upland soil contamination was the primary factor for Hg and Ni, while residential areas and roads were significant features associated with Pb and Zn contamination. This approach offers information on major contamination sources and their influential areas to be prioritized for managing contaminated stream sediments.

Physics-informed machine learning algorithms for forecasting sediment yield: an analysis of physical consistency, sensitivity, and interpretability

The sediment transport, involving the movement of the bedload and suspended sediment in the basins, is a critical environmental concern that worsens water scarcity and leads to degradation of land and its ecosystems. Machine learning (ML) algorithms have emerged as powerful tools for predicting sediment yield. However, their use by decision-makers can be attributed to concerns regarding their consistency with the involved physical processes. In light of this issue, this study aims to develop a physics-informed ML approach for predicting sediment yield. To achieve this objective, Gaussian, Center, Regular, and Direct Copulas were employed to generate virtual combinations of physical of the sub-basins and hydrological datasets. These datasets were then utilized to train deep neural network (DNN), conventional neural network (CNN), Extra Tree, and XGBoost (XGB) models. The performance of these models was compared with the modified universal soil loss equation (MUSLE), which serves as a process-based model. The results demonstrated that the ML models outperformed the MUSLE model, exhibiting improvements in Nash-Sutcliffe efficiency (NSE) of approximately 10%, 18%, 32%, and 41% for the DNN, CNN, Extra Tree, and XGB models, respectively. Furthermore, through Sobol sensitivity and Shapley additive explanation-based interpretability analyses, it was revealed that the Extra Tree model displayed greater consistency with the physical processes underlying sediment transport as modeled by MUSLE. The proposed framework provides new insights into enhancing the accuracy and applicability of ML models in forecasting sediment yield while maintaining consistency with natural processes. Consequently, it can prove valuable in simulating process-related strategies aimed at mitigating sediment transport at watershed scales, such as the implementation of best management practices.

Spatiotemporal variations, sources, pollution status and health risk assessment of dissolved trace elements in a major Arabian Sea draining river: insights from multivariate statistical and machine learning approaches

River Mahi drains through semi-arid regions (Western India) and is a major Arabian Sea draining river. As the principal surface water source, its water quality is important to the regional population. Therefore, the river water was sampled extensively (n = 64, 16 locations, 4 seasons and 2 years) and analyzed for 11 trace elements (TEs; Sr, V, Cu, Ni, Zn, Cd, Ba, Cr, Mn, Fe, and Co). Machine learning (ML) and multivariate statistical analysis (MVSA) were applied to investigate their possible sources, spatial-temporal-annual variations, evaluate multiple water quality parameters [heavy metal pollution index (HPI), heavy metal evaluation index (HEI)], and health indices [hazard quotient (HQ), and hazard index (THI)] associated with TEs. TE levels were higher than their corresponding world average values in 100% (Sr, V and Zn), 78%(Cu), 41%(Ni), 27%(Cr), 9%(Cd), 8%(Ba), 8%(Co), 6%(Fe), and 0%(Mn), of the samples. Three principal components (PCs) accounted for 74.5% of the TE variance: PC-1 (Fe, Co, Mn and Cu) and PC-2 (Sr and Ba) are contributed from geogenic sources, while PC-3 (Cr, Ni and Zn) are derived from geogenic and anthropogenic sources. HPI, HEI, HQ and THI all indicate that water quality is good for domestic purposes and poses little hazard. ML identified Random forest as the most suitable model for predicting HEI class (accuracy: 92%, recall: 92% and precision: 94%). Even with a limited dataset, the study underscores the potential application of ML to predictive classification modeling.

Machine learning-based water quality prediction using octennial in-situ Daphnia magna biological early warning system data

Biological early warning system (BEWS) has been globally used for surface water quality monitoring. Despite its extensive use, BEWS has exhibited limitations, including difficulties in biological interpretation and low alarm reproducibility. This study addressed these issues by applying machine learning (ML) models to eight years of in-situ BEWS data for Daphnia magna. Six ML models were adopted to predict contamination alarms from Daphnia behavioral parameters. The light gradient boosting machine model demonstrated the most significant improvement in predicting alarms from Daphnia behaviors. Compared with the traditional BEWS alarm index, the ML model enhanced the precision and recall by 29.50% and 43.41%, respectively. The speed distribution index and swimming speed were significant parameters for predicting water quality warnings. The nonlinear relationships between the monitored Daphnia behaviors and water physicochemical water quality parameters (i.e., flow rate, Chlorophyll-a concentration, water temperature, and conductivity) were identified by ML models for simulating Daphnia behavior based on the water contaminants. These findings suggest that ML models have the potential to establish a robust framework for advancing the predictive capabilities of BEWS, providing a promising avenue for real-time and accurate assessment of water quality. Thereby, it can contribute to more proactive and effective water quality management strategies.

Application of Machine Learning in Modeling the Relationship between Catchment Attributes and Instream Water Quality in Data-Scarce Regions

This research delves into the efficacy of machine learning models in predicting water quality parameters within a catchment area, focusing on unraveling the significance of individual input variables. In order to manage water quality, it is necessary to determine the relationship between the physical attributes of the catchment, such as geological permeability and hydrologic soil groups, and in-stream water quality parameters. Water quality data were acquired from the Iran Water Resource Management Company (WRMC) through monthly sampling. For statistical analysis, the study utilized 5-year means (1998-2002) of water quality data. A total of 88 final stations were included in the analysis. Using machine learning methods, the paper gives relations for 11 in-stream water quality parameters: Sodium Adsorption Ratio (SAR), Na+, Mg2+, Ca2+, SO42-, Cl-, HCO3-, K+, pH, conductivity (EC), and Total Dissolved Solids (TDS). To comprehensively evaluate model performance, the study employs diverse metrics, including Pearson's Linear Correlation Coefficient (R) and the mean absolute percentage error (MAPE). Notably, the Random Forest (RF) model emerges as the standout model across various water parameters. Integrating research outcomes enables targeted strategies for fostering environmental sustainability, contributing to the broader goal of cultivating resilient water ecosystems. As a practical pathway toward achieving a delicate balance between human activities and environmental preservation, this research actively contributes to sustainable water ecosystems.

Guiding the landscape patterns evolution is the key to mitigating river water quality degradation

Consensus has emerged that landscape pattern evolution significantly impacts the river environment. However, there remains unclear how the landscape pattern evolves possible to achieve a balance between land resource use and water conservation. Thus, simulating future landscape patterns under different scenarios to predict river eutrophication level is critical to propose targeted landscape planning programs and alleviate river water quality degradation. Here, we coupled five water quality parameters (TOC, TN, NO3--N, NH4+-N, TP), collected from October 2020 to September 2021, to construct the river eutrophication index (EI) to assess river water quality. Meanwhile, based on redundancy analysis, patch-generating land use simulation model, and stepwise multiple linear regression model comprehensively analyze the Fengyu River watershed landscape patterns evolution and their impact on river eutrophication. Results indicated that current rivers reach eutrophic levels, and EI reaches 40.7. The landscape patterns explain 88.2 % of river eutrophication variation, while the LPI_Con metric is critical and individually explained 21.5 %. Furthermore, eutrophication in the watershed will increase in 2040 under the natural development (ND) scenario, and the EI will reach 44.4. In contrast, farmland protection (FP) scenarios and environmental protection (EP) scenarios contribute to mitigating eutrophication, the EI values are 38.2 and 38.1, respectively. The results provide a potential mechanistic explanation that river eutrophication is a consequence of unreasonable landscape pattern evolution. Guiding the landscape patterns evolution based on critical driver factors from a planning perspective is conducive to mitigating river water quality degradation.

Application of classification machine learning algorithms for characterizing nutrient transport in a clay plain agricultural watershed

Excess nutrients in surface water and groundwater can lead to water quality deterioration in available water resources. Thus, the classification of nutrient concentrations in water resources has gained significant attention during recent decades. Machine learning (ML) algorithms are considered an efficient tool to describe nutrient loss from agricultural land to surface water and groundwater. Previous studies have applied regression and classification ML algorithms to predict nutrient concentrations in surface water and/or groundwater, or to categorize an output variable using a limited number of input variables. However, there have been no studies that examined the application of different ML classification algorithms in agricultural settings to classify various output variables using a wide range of input variables. In this study, twenty-four ML classification algorithms were implemented on a dataset from three locations within the Upper Parkhill watershed, an agricultural watershed in southern Ontario, Canada. Nutrient concentrations in surface water were classified using geochemical and physical water parameters of surface water and groundwater (e.g., pH), climate and field conditions as the input variables. The performance of these algorithms was evaluated using four evaluation metrics (e.g., classification accuracy) to identify the optimal algorithm for classifying the output variables. Ensemble bagged trees was found to be the optimal ML algorithm for classifying nitrate concentration in surface water (accuracy of 90.9%), while the weighted KNN was the most appropriate algorithm for categorizing the total phosphorus concentration (accuracy of 87%). The ensemble subspace discriminant algorithm gave the highest overall classification accuracy for the concentration of soluble reactive phosphorus and total dissolved phosphorus in surface water with an accuracy of 79.2% and 77.9%, respectively. This study exemplifies that ML algorithms can be used to signify exceedance of recommended concentrations of nutrients in surface waters in agricultural watersheds. Results are useful for decision makers to develop nutrient management strategies.

Deep learning based soft-sensor for continuous chlorophyll estimation on decentralized data

Monitoring the concentration of pigments like chlorophyll (Chl) in water-bodies is a key task to contribute to their conservation. However, with the existing sensor technology, measurement in real-time and with enough frequency to ensure proper risk management is not completely feasible. In this work, with the concept of data-driven soft-sensing, three hydrophysical features are used together with three meteorological ones to estimate the concentration of Chl in two tributaries of the River Thames. Data driven models, specifically neural networks, are used with three learning approaches: individual, centralized and federated. Data reduction scenarios are proposed in order to analyze the performance of each approach when less data is available. The best results in the training are usually obtained with the individual approach. However, the federated learning provides better generalization ability. It was also observed that in most of the cases the results of the federated learning approach improve those of the centralized one.

Interpretable spatial identity neural network-based epidemic prediction

Epidemic spatial-temporal risk analysis, e.g., infectious number forecasting, is a mainstream task in the multivariate time series research field, which plays a crucial role in the public health management process. With the rise of deep learning methods, many studies have focused on the epidemic prediction problem. However, recent primary prediction techniques face two challenges: the overcomplicated model and unsatisfactory interpretability. Therefore, this paper proposes an Interpretable Spatial IDentity (ISID) neural network to predict infectious numbers at the regional weekly level, which employs a light model structure and provides post-hoc explanations. First, this paper streamlines the classical spatio-temporal identity model (STID) and retains the optional spatial identity matrix for learning the contagion relationship between regions. Second, the well-known SHapley Additive explanations (SHAP) method was adopted to interpret how the ISID model predicts with multivariate sliding-window time series input data. The prediction accuracy of ISID is compared with several models in the experimental study, and the results show that the proposed ISID model achieves satisfactory epidemic prediction performance. Furthermore, the SHAP result demonstrates that the ISID pays particular attention to the most proximate and remote data in the input sequence (typically 20 steps long) while paying little attention to the intermediate steps. This study contributes to reliable and interpretable epidemic prediction through a more coherent approach for public health experts.

Spatialization and driving factors of carbon budget at county level in the Yangtze River Delta of China

The county is the basic administrative unit of China, and the spatialization of carbon budget at the county scale plays an irreplaceable role in deepening the understanding of the carbon emission mechanism and spatial pattern. Yueqing County, an economically developed county in the Yangtze River Delta of China, was selected as the study area, the spatial pattern of the carbon budget and the optimal resolution of the spatialization at the county level were dissected on the basis of accurate accounting, and driving factors of carbon emissions were further identified using the geographically weighted regression model. The results indicated that (1) the carbon emissions were mainly generated from fossil fuel combustion related to energy, accounting for 98.8% of the total carbon budget in the study area; (2) the optimal resolution of spatialization was 200 m and carbon emissions were concentrated in the southeast of the study area; (3) energy intensity, energy structure, per capita GDP, and urbanization rate were positively correlated with carbon emissions, while population played a bidirectional role in carbon emissions. This study not only strengthens the understanding of the patterns and drivers of the carbon budget but also establishes a theoretical framework and operational tools for policymakers to formulate solutions to mitigate the carbon crisis.

Microbial source tracking to elucidate the impact of land-use and physiochemical water quality on fecal contamination in a mixed land-use watershed

Escherichia coli has been widely used as a fecal indicator bacterium (FIB) for monitoring water quality in drinking water sources and recreational water. However, fecal contamination sources remain difficult to identify and mitigate, as millions of cases of infectious diseases are reported yearly due to swimming and bathing in recreational water. The objective of this study was to apply molecular techniques for microbial source tracking (MST) to identify sources of fecal contamination in a representative mixed land-use watershed located in the Appalachian Mountains of the United States of America (USA). Monthly samples were collected over one year at 11 sites, including the confluence of key first-order streams in the study watershed representing distinct land-use types and anticipated fecal sources. Results indicated that coupled monitoring of host-specific MST markers with the FIB E. coli effectively identified sources and quantified fecal contamination in the study watershed. Human-associated MST markers were abundant primarily at developed sites, suggesting septic or sewer failure is a key source of fecal input to the watershed. Across the dataset, samples positive for E. coli and human MST markers were associated with a higher pH than those samples from which each target was not detected, thereby suggesting that acid mine drainage in the watershed likely contributed to inactivation or loss of culturability in E. coli. In addition, this research provides the first evidence that the BacCan-UCD marker is present in fox feces and can influence MST results in areas where substantial wildlife activity is present. Identifying the sources of fecal contamination and better understanding the impact of in-stream physiochemistry throughout this study will help to develop sustainable and effective watershed management plans to control fecal contamination to protect drinking water sources and recreational water.

Spatiotemporal analysis and prediction of water quality in Pearl River, China, using multivariate statistical techniques and data-driven model

Identifying spatiotemporal variation patterns and predicting future water quality are critical for rational and effective surface water management. In this study, an exploratory analysis and forecast workflow for water quality in Pearl River, Guangzhou, China, was established based on the 4-h interval dataset selected from 10 stations for water quality monitoring from 2019 to 2021. The multiple statistical techniques, such as cluster analysis (CA), principal component analysis (PCA), correlation analysis (CoA), and redundancy analysis (RDA), as well as data-driven model (i.e., gated recurrent unit (GRU)), were applied for assessing and predicting the water quality in the basin. The investigated sampling stations were classified into 3 categories based on differences in water quality, i.e., low, moderate, and high pollution regions. The average water quality indexes (WQI) values ranged from 38.43 to 92.63. Nitrogen was the most dominant pollutant, with high TN concentrations of 0.81-7.67 mg/L. Surface runoff, atmospheric deposition, and anthropogenic activities were the major contributors affecting the spatiotemporal variations in water quality. The decline in river water quality during the wet season was mainly attributed to increased surface runoff and extensive human activities. Furthermore, the short-term prediction of river water quality was achieved using the GRU model. The result indicated that for both DLCK and DTJ stations, the WQI for the 5-day lead time were predicted with accuracies of 0.82; for the LXH station, the WQI for the 3-day lead time was forecasted with an accuracy of 0.83. The finding of this study will shed a light on an effective reference and systematic support for spatio-seasonal variation and prediction patterns of water quality.

Global patterns and key drivers of stream nitrogen concentration: A machine learning approach

Anthropogenic loading of nitrogen to river systems can pose serious health hazards and create critical environmental threats. Quantification of the magnitude and impact of freshwater nitrogen requires identifying key controls of nitrogen dynamics and analyzing both the past and present patterns of nitrogen flows. To tackle this challenge, we adopted a machine learning (ML) approach and built an ML-driven representation that captures spatiotemporal variability in nitrogen concentrations at global scale. Our model uses random forests to regress a large sample of monthly measured stream nitrogen concentrations onto a set of 17 predictors with a spatial resolution of 0.5-degree over the 1990-2013, including observations within the pixel and upstream drivers. The model was validated with data from rivers outside the training dataset and was used to predict nitrogen concentrations in 520 major river basins of the world, including many with scarce or no observations. We predicted that the regions with highest median nitrogen concentrations in their rivers (in 2013) were: United States (Mississippi), Pakistan, Bangladesh, India (Indus, Ganges), China (Yellow, Yangtze, Yongding, Huai), and most of Europe (Rhine, Danube, Vistula, Thames, Trent, Severn). Other major hotspots were the river basins of the Sebou (Morroco), Nakdong (South Korea), Kitakami (Japan), and Egypt's Nile Delta. Our analysis showed that the rate of increase in nitrogen concentration between 1990s and 2000s was greatest in rivers located in eastern China, eastern and central parts of Canada, Baltic states, Pakistan, mainland southeast Asia, and south-eastern Australia. Using a new grouped variable importance measure, we also found that temporality (month of the year and cumulative month count) is the most influential predictor, followed by factors representing hydroclimatic conditions, diffuse nutrient emissions from agriculture, and topographic features. Our model can be further applied to assess strategies designed to reduce nitrogen pollution in freshwater bodies at large spatial scales.

Large-scale prediction of stream water quality using an interpretable deep learning approach

Deep learning methods, which have strong capabilities for mapping highly nonlinear relationships with acceptable calculation speed, have been increasingly applied for water quality prediction in recent studies. However, it is argued that the practicality of deep learning methods is limited due to the lack of physical mechanics to explain the prediction results of water quality changes. A knowledge gap exists in rationalizing the deep learning results for water quality predictions. To address this gap, an interpretable deep learning framework was established to predict the spatiotemporal variations of water quality parameters in a large spatial region. Mereological, land-use, and socioeconomic variables were adopted to predict the daily variations of stream water quality parameters across 138 sub-catchments in a total of over 575,250 km² in southern China. The coefficients of determination of chemical oxygen demand (COD), total phosphorus (TP), and total nitrogen (TN) predictions were over 0.80, suggesting a satisfactory prediction performance. The model performance in terms of prediction accuracy could be improved by involving land-use and socioeconomic predictors in addition to hydrological variables. The SHapley Additive exPlanations method used in this study was demonstrated to be effective for interpreting the prediction results by identifying the significant variables and reasoning their influencing directions on the variation of each water quality parameter. The air temperature, proportion of forest area, grain production, population density, and proportion of urban area in each sub-catchment as well as the accumulated rainfall within the previous 3 days were identified as the most significant variables affecting the variations of dissolved oxygen, COD, ammoniacal nitrogen(NH₃-N), TN, TP, and turbidity in the stream water in the case area, respectively.

Real-time streamflow forecasting in a reservoir-regulated river basin using explainable machine learning and conceptual reservoir module

Real-time streamflow forecasting is essential to manage water resources effectively in a reservoir-regulated basin. However, forecasting becomes challenging without weather and upstream reservoir outflows forecasts in real-time. In this context, a novel hybrid approach is proposed in this study to forecast the streamflows and reservoir outflows in real-time. In this approach, the Explainable Machine Learning model is embedded with a conceptual reservoir module for forecasting streamflows using short-term weather forecasts. Long Short Term Memory (LSTM), a Machine Learning model, is used in this study to predict the streamflow, and the model's explainability is examined by Shapley additive explanations method (SHAP). Panchet reservoir catchment, which contains Tenughat and Konar reservoirs, is selected as a study area. The LSTM model performance is excellent in predicting the streamflows of Tenughat, Konar and Panchet catchments with NSE values of 0.93, 0.87, and 0.96, respectively. The SHAP method identified the high-impact variables as streamflows and precipitation of 1-day lag. In forecasting, bias-corrected Global Forecast System data is used with the LSTM model to forecast the streamflows in three catchments. The inflows are forecasted well up to a 3-day lead in Tenughat and Konar reservoirs with NSE values above 0.88 and 0.87, respectively. The reservoir module performance in forecasting Tenughat and Konar reservoirs' outflows with the inflow forecasts is also promising up to a 3-day lead with NSE values above 0.88 for both reservoirs. The inflows forecasting to Panchet reservoir with reservoirs' outflows as additional inputs is excellent up to 5-day lead (NSE = 0.96-0.88). However, the forecasting error increased from 77 m³/s to 134 m³/s with the lead time. This approach could provide an efficient way to reduce flood risks in the reservoir-regulated basin.

Screening priority pesticides for drinking water quality regulation and monitoring by machine learning: Analysis of factors affecting detectability

Proper selection of new contaminants to be regulated or monitored prior to implementation is an important issue for regulators and water supply utilities. Herein, we constructed and evaluated machine learning models for predicting the detectability (detection/non-detection) of pesticides in surface water as drinking water sources. Classification and regression models were constructed for Random Forest, XGBoost, and LightGBM, respectively; of these, the LightGBM classification model had the highest prediction accuracy. Furthermore, its prediction performance was superior in all aspects of Recall, Precision, and F-measure compared to the detectability index method, which is based on runoff models from previous studies. Regardless of the type of machine learning model, the number of annual measurements, sales quantity of pesticide for rice-paddy field, and water quality guideline values were the most important model features (explanatory variables). Analysis of the impact of the features suggested the presence of a threshold (or range), above which the detectability increased. In addition, if a feature (e.g., quantity of pesticide sales) acted to increase the likelihood of detection beyond a threshold value, other features also synergistically affected detectability. Proportion of false positives and negatives varied depending on the features used. The superiority of the machine learning models is their ability to represent nonlinear and complex relationships between features and pesticide detectability that cannot be represented by existing risk scoring methods.

Stream water quality optimized prediction based on human activity intensity and landscape metrics with regional heterogeneity in Taihu Basin, China

The driving effects of landscape metrics on water quality have been acknowledged widely, however, the guiding significance of human activity intensity and landscape metrics based on reference conditions for water environment management remains to be explored. Thus, we used the self-organized map, long- and short-term memory (LSTM) algorithm, and geographic detectors to simulate the response of human activity intensity and landscape metrics to water quality in Taihu Lake Basin, China. Fitting results of LSTM displayed that the accuracy was acceptable, and scenario 2 (regional heterogeneity) was more efficient than scenario 1 (regional consistent) in the improvement of water quality. In the driving analysis for the reference conditions, clusters I and II (urban agglomeration areas) were mainly affected by the amount of production wastewater per unit of developed land and the amount of livelihood wastewater per unit of developed land, respectively. Their optimal values were 0.09 × 10³ t/km² (reduction of 35.71%) and 0.2 × 10³ t/km² (reduction of 4.76%). Cluster III (agricultural production areas) was mainly affected by interference intensity, and the optimal value was 2.17 (increased 38.22%), and cluster IV (ecological forest areas) was mainly affected by the fragmentation of cropland, and the optimal value was 1.14 (reduction of 1.72%). The research provides a reference for the prediction of water quality response and presents an ecological and economic sustainability way for watershed governance.

Investigating water quality sensitivity to climate variability and its influencing factors in four Lake Erie watersheds

Climate change alters weather patterns and hydrological cycle, thus potentially aggravating water quality impairment. However, the direct relationships between climate variability and water quality are complicated by a multitude of hydrological and biochemical mechanisms dominate the process. Thus, little is known regarding how water quality responds to climate variability in the context of changing meteorological conditions and human activities. Here, a longitudinal study was conducted using trend, correlation, and redundancy analyses to explore stream water quality sensitivity to temperature, precipitation, streamflow, and how the sensitivity was affected by watershed climate, land cover percentage, landscape configuration, fertilizer application, and tillage types. Specifically, daily pollutant concentration data of suspended solid (SS), total phosphorus (TP), soluble reactive phosphorus (SRP), total Kjeldahl nitrogen (TKN), nitrate and nitrite (NO_x), and chloride (Cl) were used as water quality indicators in four Lake Erie watersheds from 1985 to 2017, during which the average temperature has increased 0.5 °C and the total precipitation has increased 9%. Results show that precipitation and flow were positively associated with SRP, NO_x, TKN, TP, and SS, except for SRP and NO_x in the urban basin. The rising temperatures led to increasing concentrations of SS, TKN, and TP in the urban basin. SRP and NO_x sensitivity to precipitation was higher in the years with more precipitation and higher precipitation seasonality, and the basins with more spatially aggregated cropland. No-tillage and reduced tillage management could decrease both precipitation and temperature sensitivity for most pollutants. As one of the first studies leveraging multiple watershed environmental variables with long-term historical climate and water quality data, this study can assist target land use planning and management policy to mitigate future climate change effects on surface water quality.

Assessing the impact of watershed characteristics and management on nutrient concentrations in tropical rivers using a machine learning method

Excessive loadings of terrestrial nitrogen and phosphorus, as well as their imbalances with silicon, have been recognized as one of the major causes of water quality and ecosystem deterioration in receiving waters. In this study, a periodic water quality monitoring was conducted in the rivers and streams of a tropical island (Ishigaki Island, Japan) to identify the factors controlling the concentrations of dissolved inorganic nitrogen (DIN), total phosphorus (TP) and dissolved silicon (DSi) with a special focus on the catchment characteristics (e.g., land use, surface geology, topography). Random Forest (RF) machine learning algorithm was employed to develop predictive models for nutrient concentrations from the catchment properties. The developed models could predict nutrient concentrations with sufficient accuracy, demonstrating that the studied nutrients are strongly affected by catchment properties. Agricultural land uses (e.g., livestock barn, sugarcane field) were ranked as the most important parameters for DIN and TP, while broadleaf forest was the most influential factor for DSi. Using the RF models, the contributions of DIN originating from sugarcane fields (i.e., fertilizers) and barns (i.e., manure) to riverine DIN were estimated, which were up to 60% in total in the studied river basins. Furthermore, the yield of DIN from sugarcane fields, calculated as the concentration of DIN derived from sugarcane fields divided by the percent area of sugarcane fields, strongly positively correlated with the areal coverage of limestone, suggesting that fertilizer-derived DIN is more prone to leaching out from cropland soil to groundwater and rivers in catchments with a higher dominance of calcareous geology. These results, including the methodology employed, have implications for water quality assessment and management in inland and coastal waters not only at the study site but also other regions.

Predicting in-stream water quality constituents at the watershed scale using machine learning

Predicting in-stream water quality is necessary to support the decision-making process of protecting healthy waterbodies and restoring impaired ones. Data-driven modeling is an efficient technique that can be used to support such efforts. Our objective was to determine if in-stream concentrations of contaminants, nutrients-total phosphorus (TP) and total nitrogen (TN) -total suspended solids (TSS), dissolved oxygen (DO), and fecal coliform bacteria (FC) can be predicted satisfactorily using machine learning (ML) algorithms based on publicly available datasets. To achieve this objective, we evaluated four modeling scenarios, differing in terms of the required inputs (i.e., publicly available datasets (e.g., land-use/land cover)), antecedent conditions, and additional in-stream water quality observations (e.g., pH and turbidity). We implemented five ML algorithms-Support Vector Machines, Random Forest (RF), eXtreme Gradient Boost (XGB), ensemble RF-XGB, and Artificial Neural Network (ANN) -and demonstrated our modeling framework in an inland stream-Bullfrog Creek, located near Tampa, Florida. The results showed that, while including additional water quality drivers improved overall model performance for all target constituents, TP, TN, DO, and TSS could still be predicted satisfactorily using only publicly available datasets (Nash-Sutcliffe efficiency [NSE] > 0.75 and percent bias [PBIAS] < 10%), whereas FC could not (NSE < 0.49 and PBIAS >25%). Additionally, antecedent conditions slightly improved predictions and reduced the predictive uncertainty, particularly when paired with other water quality observations (6.9% increase in NSE for FC, and 2.7% for TP, TN, DO, and TSS). Also, comparable model performances of all water quality constituents in wet and dry seasons suggest minimal season-dependence of the predictions (<4% difference in NSE and < 10% difference in PBIAS). Our developed modeling framework is generic and can serve as a complementary tool for monitoring and predicting in-stream water quality constituents.

Nitrate concentrations predominantly driven by human, climate, and soil properties in US rivers

Nitrate is one of the most widespread and persistent pollutants in our time. Our understanding of nitrate dynamics has advanced substantially in the past decades, although its predominant drivers across gradients of climate, land use, and geology have remained elusive. Here we collated nitrate data from 2061 rivers along with 32 watershed characteristic indexes and developed machine learning models to reconstruct long-term mean (multi-year average) nitrate concentrations in the contiguous United States (CONUS). The trained models show similarly satisfactory model performance and can predict nitrate concentrations in chemically-ungauged places with about 70% accuracy. Further analysis revealed that five (out of 32) indexes (drivers) can explain about 70% of spatial variations in mean nitrate concentrations. The five influential drivers are nitrogen application rates N_rate and urban area A_urban% (human drivers), mean annual precipitation and temperature (climate drivers), and sand percent Sand% (soil property driver). Nitrate concentrations in undeveloped sites are primarily modulated by climate and soil property; they decrease with increasing mean discharge and Sand%. Nitrate concentrations in agriculture and urban sites increase with N_rate and A_urban% until reaching their apparent maxima around 10,000 kg/km²/yr and around 25%, respectively. Results indicate that nitrate concentrations may remain similar or increase with growing human population. In addition, nitrate concentrations can increase even without human input, as warming escalates water demand and reduces mean discharge in many places. These results allude to a conceptual model that highlights the impacts of distinct drivers: while human drivers predominate nitrogen input to land and rivers, climate drivers and soil properties modulate its transport and transformation, the balance of which determine long-term mean concentrations. Such mechanism-based insights and forecasting capabilities are essential for water management as we expect changing climate and growing agriculture and urbanization.

Random forest-based modeling of stream nutrients at national level in a data-scarce region

Nutrient runoff from agricultural production is one of the main causes of water quality deterioration in river systems and coastal waters. Water quality modeling can be used for gaining insight into water quality issues in order to implement effective mitigation efforts. Process-based nutrient models are very complex, requiring a lot of input parameters and computationally expensive calibration. Recently, ML approaches have shown to achieve an accuracy comparable to the process-based models and even outperform them when describing nonlinear relationships. We used observations from 242 Estonian catchments, amounting to 469 yearly TN and 470 TP measurements covering the period 2016-2020 to train random forest (RF) models for predicting annual N and P concentrations. We used a total of 82 predictor variables, including land cover, soil, climate and topography parameters and applied a feature selection strategy to reduce the number of dependent features in the models. The SHAP method was used for deriving the most relevant predictors. The performance of our models is comparable to previous process-based models used in the Baltic region with the TN and TP model having an R² score of 0.83 and 0.52, respectively. However, as input data used in our models is easier to obtain, the models offer superior applicability in areas, where data availability is insufficient for process-based approaches. Therefore, the models enable to give a robust estimation for nutrient losses at national level and allows to capture the spatial variability of the nutrient runoff which in turn enables to provide decision-making support for regional water management plans.

Using Random Forest, a machine learning approach to predict nitrogen, phosphorus, and sediment event mean concentrations in urban runoff

Estimating pollutant loads from developed watersheds is vitally important to reduce nonpoint source pollution from urban areas, as a key tool in meeting water quality goals is the implementation of Stormwater Control Measures (SCMs). SCMs are selected and sized based on influent pollutant loads. A common method used to estimate pollutant loads in urban runoff is the Event Mean Concentration (EMC) method. In this study, we develop and apply data-driven models using Random Forest (RF), a machine learning approach, to predict Total Nitrogen (TN), Total Phosphorus (TP), Total Suspended Solids (TSS), and Ortho-Phosphorus (Ortho-P) EMCs in urban runoff. The parameters considered in this study were climatological characteristics (i.e., Antecedent Dry Period or ADP, Precipitation Depth or P, Duration or D, and Intensity or I) and catchment characteristics including land use-related parameters including Imperviousness or Imp, Saturated Hydraulic Conductivity or K_sat, and Available Water Capacity or AWC), and site-specific parameters including Slope (S), and Catchment Size (A). Stormwater quality data for this study were obtained from the National Stormwater Quality Database (NSQD), which is the largest repository of stormwater quality data in the U.S. Results demonstrate that land use-related characteristics (i.e., Imp, K_sat, and AWC) were the most effective variables for predicting all EMCs. For TP, TSS, and Ortho-P, site-specific characteristics (S and A) had a greater effect than climatological characteristics (i.e., ADP, P, D, and I). However, for TN, climatological characteristics had a greater effect than site-specific characteristics (S and A). In addition, for TN, TP, and TSS, precipitation characteristics (P, D, and I) were found to be more effective parameters for estimating EMCs than ADP. This study highlights the most influential parameters affecting EMCs which can be used by stakeholders and SCMs designers to improve estimates of nutrients and sediment EMCs. The selection and design of the highest performing SCMs is essential in achieving effective treatment of stormwater, attaining water quality goals, and protecting downstream waterbodies.

Simulation of regional groundwater levels in arid regions using interpretable machine learning models

Regional groundwater level forecasting is critical to water resource management, especially for arid regions which require effective management of groundwater resources to meet human and ecosystem needs. In this study Machine Learning and Deep Learning approaches - Support Vector Machine, Generalized Regression Neural Network, Decision Tree, Random Forest (RF), Convolutional Neural Network, Long Short Term Memory and Gated Recurrent Network- have been used to simulate the groundwater levels in the lower Tarim River basin (LTRB) which is an extreme dryland. The results showed that models developed here with easily available input data such as relative humidity, flow volume and distance to the riverbank can fully utilize spatiotemporally inconsistent groundwater monitoring data to predict the spatiotemporal variation of the groundwater system in arid regions where exist intermittent flow. The shapely additive explanations method was used to interpret the RF model and discover the effect of meteorological, hydrological and environmental variables on the regional groundwater. These explanations showed that the flow volume, the distance to the river channel and reservoir have a critical impact on groundwater changes. Within 300 m distance to the riverbank, groundwater would mainly be influenced by the flow volume and the distance to the reservoir. While far from the riverbank, these effects decreased gradually further away from the river course. The RF prediction results showed that in the next three years (2021-2023), the groundwater level on average may decline to -6.4 m, and the suitable areas for natural vegetation growth would be limited to 39% if no water conveyance in the LTRB. To guarantee the stability of ecosystems in the LTRB, it is necessary to convey the water annually. These results can support spatiotemporal predictions of groundwater levels predominantly recharged by intermittent flow, and form a scientific basis for sustainable water resources management in arid regions.

Environmental impacts of autonomous vehicles: A review of the scientific literature

Autonomous vehicles (AVs) may have significant environmental impacts although there are still few studies focusing solely on these effects. A vast majority of articles address environmental issues as a secondary outcome and, above all, emissions are the main topic. As the notion of environmental impacts concerns many aspects than just air pollution, this paper aims to explore and show the findings and flaws of current research with a wider vision. For that purpose, a systematic review of the scientific literature was carried out broadening the scope to land, water, noise, and light pollution in addition to air. The results reveal potential benefits of AVs due to technical improvements, new possibilities in design and traffic flow enhancement, but the benefits depend on penetration levels, shared mobility acceptance and the interaction with other modes of transport. On the other hand, negative effects are also identified related to the decrease in the value of trip time and user tendencies. Among other potential impacts, changes in land use are increasingly being studied. These changes can lead to significative impacts on emissions as well as on soil and water although the latter have not yet been considered. Lastly, the likely improvements in noise and light pollution are scarcely explored. Given the lack of study of some of the environmental outcomes of AVs, it is not possible to draw a precise conclusion on their overall impact, calling for more comprehensive studies that enable to identify all the measures to be taken to achieve a sustainable future.

Machine learning-based estimation of riverine nutrient concentrations and associated uncertainties caused by sampling frequencies

Accurate and sufficient water quality data is essential for watershed management and sustainability. Machine learning models have shown great potentials for estimating water quality with the development of online sensors. However, accurate estimation is challenging because of uncertainties related to models used and data input. In this study, random forest (RF), support vector machine (SVM), and back-propagation neural network (BPNN) models are developed with three sampling frequency datasets (i.e., 4-hourly, daily, and weekly) and five conventional indicators (i.e., water temperature (WT), hydrogen ion concentration (pH), electrical conductivity (EC), dissolved oxygen (DO), and turbidity (TUR)) as surrogates to individually estimate riverine total phosphorus (TP), total nitrogen (TN), and ammonia nitrogen (NH₄⁺-N) in a small-scale coastal watershed. The results show that the RF model outperforms the SVM and BPNN machine learning models in terms of estimative performance, which explains much of the variation in TP (79 ± 1.3%), TN (84 ± 0.9%), and NH₄⁺-N (75 ± 1.3%), when using the 4-hourly sampling frequency dataset. The higher sampling frequency would help the RF obtain a significantly better performance for the three nutrient estimation measures (4-hourly > daily > weekly) for R² and NSE values. WT, EC, and TUR were the three key input indicators for nutrient estimations in RF. Our study highlights the importance of high-frequency data as input to machine learning model development. The RF model is shown to be viable for riverine nutrient estimation in small-scale watersheds of important local water security.

A review of the application of machine learning in water quality evaluation

With the rapid increase in the volume of data on the aquatic environment, machine learning has become an important tool for data analysis, classification, and prediction. Unlike traditional models used in water-related research, data-driven models based on machine learning can efficiently solve more complex nonlinear problems. In water environment research, models and conclusions derived from machine learning have been applied to the construction, monitoring, simulation, evaluation, and optimization of various water treatment and management systems. Additionally, machine learning can provide solutions for water pollution control, water quality improvement, and watershed ecosystem security management. In this review, we describe the cases in which machine learning algorithms have been applied to evaluate the water quality in different water environments, such as surface water, groundwater, drinking water, sewage, and seawater. Furthermore, we propose possible future applications of machine learning approaches to water environments.

Imputation of Ammonium Nitrogen Concentration in Groundwater Based on a Machine Learning Method

Ammonium is one of the main inorganic pollutants in groundwater, mainly due to agricultural, industrial and domestic pollution. Excessive ammonium can cause human health risks and environmental consequences. Its temporal and spatial distribution is affected by factors such as meteorology, hydrology, hydrogeology and land use type. Thus, a groundwater ammonium analysis based on limited sampling points produces large uncertainties. In this study, organic matter content, groundwater depth, clay thickness, total nitrogen content (TN), cation exchange capacity (CEC), pH and land-use type were selected as potential contributing factors to establish a machine learning model for fitting the ammonium concentration. The Shapley Additive exPlanations (SHAP) method, which explains the machine learning model, was applied to identify the more significant influencing factors. Finally, the machine learning model established according to the more significant influencing factors was used to impute point data in the study area. From the results, the soil organic matter feature was found to have a substantial impact on the concentration of ammonium in the model, followed by soil pH, clay thickness and groundwater depth. The ammonium concentration generally decreased from northwest to southeast. The highest values were concentrated in the northwest and northeast. The lowest values were concentrated in the southeast, southwest and parts of the east and north. The spatial interpolation based on the machine learning imputation model established according to the influencing factors provides a reliable groundwater quality assessment and was not limited by the number and the geographical location of samplings.

Measuring the zonal responses of nitrogen output to landscape pattern in a flatland with river network: a case study in Taihu Lake Basin, China

Landscape pattern changes induced by rapid urbanization and intensified agricultural activities have exerted great pressure on regional water purification services. Relationship between landscape metrics and nitrogen-related ecosystem services has been a major concern of many scholars and has been widely used for guidance for land use and cover (LULC) management. However, clear zonal differences may exist, especially in highly developed reticular river network area, thus limiting our understanding of nitrogen output (NOP) to landscape pattern in the details. The spatial distribution of regional NOP was obtained based on the InVEST model. The zonal responses of NOP to landscape patter were examined under hydraulic subregions and subbasin scale. The results show that the unit value of average NOP in the Taihu Lake Basin (TLB) was 146.14 (kg/km²), and the total output reached 23677.92 t in 2020. The simulation NOP showed reasonable agreement with verified water quality observations in the lake inlet stations, with an R² of 0.76. In terms of space composition, merely cropland have significant effects on NOP in the whole basin scale, while the explanatory variables include cropland and developed land in Pudong (PD), Puxi (PX), Wuchengxiyu (WC), and Hangjiahu (HJ) regions. In Huxi (HX) and Yangchengdianmao  (YC) regions, cropland and forest are the significant impact types, while in (Zhexi) ZX region, cropland, developed land, and forest are significant impact types. In the space configuration, the percentage of landscape (PLAND) or largest patch index (LPI) of cropland showed positive effects about NOP, whether in the whole basin or the hydraulic subregions. Edge density (ED) (-3.48), number of patches (NP) (-3.91), and percentage of like adjacencies (PLAND) (-2.80) of the forest exhibit negative correlations with NOP, in the HX, ZX, and YC region, respectively. It displays diversiform in the response of NOP to the landscape metric of developed land, which speculate that the heterogeneity of developed land can also have a constraint on NOP, in the highly urbanized areas with less forest area. In addition, the total nitrogen output of the TLB needs to be controlled, especially in HJ region which was identified as the sensitive area of pollution sources with the largest NOP and should be paid more attention to. Compared with the administrative management unit, it is more reasonable to control and manage the pollution sources by referring to the hydraulic subregions and subbasin units. Senior managers are required to strengthen communication and cooperation with hydraulic subregions across administrative regions. However, when managing NOP through the landscape modifications, measures should be taken to reduce the aggregation of nitrogen sources and increase the fragmentation of nitrogen sinks. As for high aggregation developed and agricultural land regions, the types of land used should be enriched to help the sustainable development.

Learning hierarchical Bayesian networks to assess the interaction effects of controlling factors on spatiotemporal patterns of fecal pollution in streams

The dynamics of fecal indicator bacteria, such as fecal coliforms (FC) in streams, are influenced by the interactions of a myriad of factors. To predict complex spatiotemporal patterns of FC in streams and assess the relative importance of numerous controlling factors, the adoption of a hierarchical Bayesian network (HBN) was proposed in this study. By introducing latent variables correlated to the observed variables into a Bayesian network, the HBN can represent causal relationships among a large set of variables with a multilevel hierarchy. The study area encompasses 215 sites across the watersheds of the four major rivers in South Korea. The monitoring data collected during the 2012-2019 period included 32 input variables pertaining to meteorology, geography, soil characteristics, land cover, urbanization index, livestock density, and point sources. As model endpoints, the exceedance probability of the FC standard concentration as well as two pollution characteristics (i.e., pollution degree and type), derived from FC load duration curves were used. The probability of exceeding an FC threshold value (200 CFU/100 mL) showed spatiotemporal variations, whereas pollution degree and type showed spatial variations that represent long-term severity and relative dominance of nonpoint and point source fecal pollution, respectively. The conceptual model was validated using structural equation modeling to develop the HBN. The results demonstrate that the HBN effectively simplified the model structure, while showing strong model performance (AUC = 0.81, accuracy = 0.74). The results of the sensitivity analysis indicate that land cover is the most important factor in predicting the probability of exceedance and pollution degree, whereas the urbanization index explains most of the variability in pollution type. Furthermore, the results of the scenario analysis suggest that the HBN provides an interpretable framework in which the interaction of controlling factors has causal relationships at different levels that can be identified and visualized.

Interpretable tree-based ensemble model for predicting beach water quality

Tree-based machine learning models based on environmental features offer low-cost and timely solutions for predicting microbial fecal contamination in beach water to inform the public of the health risk. However, many of these models are black boxes that are difficult for humans to understand, which may cause severe consequences such as unexplained decisions and failure in accountability. To develop interpretable predictive models for beach water quality, we evaluate five tree-based models, namely classification tree, random forest, CatBoost, XGBoost, and LightGBM, and employ a state-of-the-art explanation method SHAP to explain the models. When tested on the Escherichia coli (E. coli) concentration data collected from three beach sites along Lake Erie shores, LightGBM, followed by XGBoost, achieves the highest averaged precision and recall scores. For all three sites, both models suggest lake turbidity as the most important predictor, and elucidate the crucial role of accurate local data of wave height and rainfall in the model development. Local SHAP values further reveal the robustness of the importance of lake turbidity as its SHAP value increases nearly monotonically with its value and is minimally affected by other environmental factors. Moreover, we found an intriguing interaction between lake turbidity and day-of-year. This work suggests that the combination of LightGBM and SHAP has a promising potential to develop interpretable models for predicting microbial water quality in freshwater lakes.

Prediction of E. coli Concentrations in Agricultural Pond Waters: Application and Comparison of Machine Learning Algorithms

The microbial quality of irrigation water is an important issue as the use of contaminated waters has been linked to several foodborne outbreaks. To expedite microbial water quality determinations, many researchers estimate concentrations of the microbial contamination indicator Escherichia coli (E. coli) from the concentrations of physiochemical water quality parameters. However, these relationships are often non-linear and exhibit changes above or below certain threshold values. Machine learning (ML) algorithms have been shown to make accurate predictions in datasets with complex relationships. The purpose of this work was to evaluate several ML models for the prediction of E. coli in agricultural pond waters. Two ponds in Maryland were monitored from 2016 to 2018 during the irrigation season. E. coli concentrations along with 12 other water quality parameters were measured in water samples. The resulting datasets were used to predict E. coli using stochastic gradient boosting (SGB) machines, random forest (RF), support vector machines (SVM), and k-nearest neighbor (kNN) algorithms. The RF model provided the lowest RMSE value for predicted E. coli concentrations in both ponds in individual years and over consecutive years in almost all cases. For individual years, the RMSE of the predicted E. coli concentrations (log10 CFU 100 ml-1) ranged from 0.244 to 0.346 and 0.304 to 0.418 for Pond 1 and 2, respectively. For the 3-year datasets, these values were 0.334 and 0.381 for Pond 1 and 2, respectively. In most cases there was no significant difference (P > 0.05) between the RMSE of RF and other ML models when these RMSE were treated as statistics derived from 10-fold cross-validation performed with five repeats. Important E. coli predictors were turbidity, dissolved organic matter content, specific conductance, chlorophyll concentration, and temperature. Model predictive performance did not significantly differ when 5 predictors were used vs. 8 or 12, indicating that more tedious and costly measurements provide no substantial improvement in the predictive accuracy of the evaluated algorithms.

Machine Learning-Based Prediction of Chlorophyll-a Variations in Receiving Reservoir of World’s Largest Water Transfer Project—A Case Study in the Miyun Reservoir, North China

Although water transfer projects can alleviate the water crisis, they may cause potential risks to water quality safety in receiving areas. The Miyun Reservoir in northern China, one of the receiving reservoirs of the world’s largest water transfer project (South-to-North Water Transfer Project, SNWTP), was selected as a case study. Considering its potential eutrophication trend, two machine learning models, i.e., the support vector machine (SVM) model and the random forest (RF) model, were built to investigate the trophic state by predicting the variations of chlorophyll-a (Chl-a) concentrations, the typical reflection of eutrophication, in the reservoir after the implementation of SNWTP. The results showed that compared with the SVM model, the RF model had higher prediction accuracy and more robust prediction ability with abnormal data, and was thus more suitable for predicting Chl-a concentration variations in the receiving reservoir. Additionally, short-term water transfer would not cause significant variations of Chl-a concentrations. After the project implementation, the impact of transferred water on the water quality of the receiving reservoir would have gradually increased. After a 10-year implementation, transferred water would cause a significant decline in the receiving reservoir’s water quality, and Chl-a concentrations would increase, especially from July to August. This led to a potential risk of trophic state change in the Miyun Reservoir and required further attention from managers. This study can provide prediction techniques and advice on water quality security management associated with eutrophication risks resulting from water transfer projects.

An interpretable machine learning method for supporting ecosystem management: Application to species distribution models of freshwater macroinvertebrates

Species distribution models (SDMs), in which species occurrences are related to a suite of environmental variables, have been used as a decision-making tool in ecosystem management. Complex machine learning (ML) algorithms that lack interpretability may hinder the use of SDMs for ecological explanations, possibly limiting the role of SDMs as a decision-support tool. To meet the growing demand of explainable MLs, several interpretable ML methods have recently been proposed. Among these methods, SHaply Additive exPlanation (SHAP) has drawn attention for its robust theoretical justification and analytical gains. In this study, the utility of SHAP was demonstrated by the application of SDMs of four benthic macroinvertebrate species. In addition to species responses, the dataset contained 22 environmental variables monitored at 436 sites across five major rivers of South Korea. A range of ML algorithms was employed for model development. Each ML model was trained and optimized using 10-fold cross-validation. Model evaluation based on the test dataset indicated strong model performance, with an accuracy of ≥0.7 in all evaluation metrics for all MLs and species. However, only the random forest algorithm showed a behavior consistent with the known ecology of the investigated species. SHAP presents an integrated framework in which local interpretations that incorporate local interaction effects are combined to represent the global model structure. Consequently, this framework offered a novel opportunity to assess the importance of variables in predicting species occurrence, not only across sites, but also for individual sites. Furthermore, removing interaction effects from variable importance values (SHAP values) clearly revealed non-linear species responses to variations in environmental variables, indicating the existence of ecological thresholds. This study provides guidelines for the use of a new interpretable method supporting ecosystem management.