From Hydrometeorology to River Water Quality: Can a Deep Learning Model Predict Dissolved Oxygen at the Continental Scale?

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.

Improving real-time forecasting of bay water quality by integrating in-situ monitoring, machining learning, and process-based modeling

Frequent occurrences of disasters such as red tides significantly threaten bay ecosystems, making near real-time water quality forecasting crucial for disaster warning and decision-making. Conventional techniques, such as process-based modeling and in-situ monitoring, struggle to achieve this for the entire bay region when applied independently. This study proposes a hybrid approach that integrates in-situ monitoring, process-based modeling, and machine learning (ML) to address this challenge. The feasibility of the approach was validated using Shenzhen Bay, a cross-boundary bay co-administered by Shenzhen and Hong Kong, as a testbed. ML models exhibited superior performance for location-specific forecasting, achieving Nash-Sutcliffe efficiency (NSE) values of 0.90, 0.84, 0.85, and 0.73 for dissolved oxygen, chlorophyll a, total nitrogen, and total phosphorus, respectively. Forecasting accuracy declined with longer lead times. Additionally, this study developed a dual-clustering method to optimize the selection of monitoring locations, minimizing the number of sites while effectively capturing the water quality across the entire bay. The results suggest that a monitoring network consisting of just two locations can adequately represent the overall water quality conditions within Shenzhen Bay. Using output from the Delft-3D model built for Shenzhen Bay, ML-based surrogate models successfully extended water quality forecasting from the two strategically selected locations to the entire bay area, with NSE values exceeding 0.8 in most regions. The hybrid approach provides a methodological foundation for achieving near real-time water quality forecasting across the entire bay areas, contributing to maintaining water security and preserving valuable ecosystem services in a rapidly changing environment.

Artificial intelligence based detection and control strategies for river water pollution: A comprehensive review

Water quality (WQ) is a metric for assessing the overall health and safety of water bodies like a river. Owing to the habitation of anthropogenic habitation around its basin, the rivers can become one of the most contaminated water sources globally. The solutions to prevent and remit the impact of river water pollution faces many challenges, one of these entails the management of nonlinear, nonstationary water related dataset. This paper provides a detailed overview of Artificial Intelligence (AI) based techniques and algorithms, highlighting their practical applications in the critical domain of river water pollution diction and control. This review shows models for river WQ simulation from 2019 to 2024, in which over 110 research articles from various databases are analyzed. Key advancements in Machine Learning (ML) and Deep Learning (DL) technologies, including Artificial Neural Networks (ANN), Deep Neural Networks (DNN), Long Short-Term Memory (LSTM), and Random Forest (RF), are highlighted. Besides that, the amalgamation of Internet of Things (IoT) technologies is tested, showing their role in enhancing real-time monitoring and predictive capabilities through continuous data collection and advanced ML/DL models. This review addresses critical challenges and identifies emerging opportunities for future research by showcasing the application of ML, DL, and IoT innovations in surface WQ modeling. It highlights the potential of leveraging advanced technologies to form strengthen solutions for sustainable water resource management and the protection of vital aquatic ecosystems.

Temporal and spatial feature extraction using graph neural networks for multi-point water quality prediction in river network areas

Deep learning methods have demonstrated strong capabilities in capturing nonlinear relationships for water quality prediction, yet existing studies predominantly focus on individual monitoring sites while neglecting pollutant spatial dynamics. To address this limitation, a Spatio-Temporal Feature Graph Neural Network (STF-GNN) was proposed, which integrated graph convolutional networks (GCN), gated recurrent units (GRU), and self-attention mechanisms to explicitly model multi-scale spatiotemporal dependencies among distributed monitoring stations. By representing stations as graph nodes with adjacency relationships, STF-GNN could simultaneously extract spatial topological features and temporal evolution patterns from multivariate time series data. Experimental results demonstrated superior performance in dissolved oxygen (DO) and total nitrogen (TN) prediction, achieving RMSE values of 0.233 (DO) and 0.033 (TN), outperforming baseline models by 36.54-161.47 % in accuracy. Cross-basin validations revealed robust generalization capabilities of the established model, maintaining maximum relative errors below 0.639 (DO) and 0.606 (TN) without site-specific customization. Notably, the model achieved 88 % peak-valley synchronization at untrained station, demonstrating strong anti-interference ability against unseen environmental variations. Ablation studies confirmed the necessity of both spatial and temporal modules, with their omission causing significant accuracy declines (12.07-19.25 %). These findings highlighted the critical roles of both spatial and temporal feature extraction in improving predictive performance of the model. The work can provide a theoretically grounded framework for spatially-aware water quality prediction, supporting enhanced environmental monitoring strategies.

Data-driven models for forecasting algal biomass in a large and deep reservoir

Early warning of algal biomass is important for the preservation and management of drinking water. However, accurate prediction of algal biomass in large and deep reservoirs remains a challenge. Here, we used six years of high-frequency observations (30 min/time) to train long short-term memory (LSTM) models for forecasting chlorophyll-a concentration (CChla) and column-integrated CChla (CIC) for a large and deep Chinese reservoir (Xin'anjiang Reservoir). Five LSTM-based algal biomass forecasting models were developed, including four CChla models for various forecasting scales (1-hour, 3-hour, 6-hour, and 24-hour) and a CIC model (forecasting scale: 1-day). The results showed that the trained LSTM-based models can accurately predict CChla and CIC at reservoir scale and the root mean square error (RSME) values are less than 1.1 and 14.9 μg/L, respectively. The proposed CChla LSTM model outperformed the MLP, CNN, CNN-LSTM, and RNN models, with the RMSE decreasing by 2.6%, 4.8%, 5.3%, and 9.3%, respectively. Similarly, the proposed CIC LSTM model surpassed the MLP, CNN, CNN-LSTM, and RNN models, resulting in a RMSE reduction of 36.1%, 46%, 50.3%, and 52.8%, respectively. With the time lag increase, the performance of the multistep-ahead forecasting model exhibits initial improvement followed by deterioration. The best performance of the multistep-ahead forecasting model was observed when the input time length is 6-8 times the forecasting time length. Spatially, the proposed models perform better at the sites with small variations in algal biomass. On the other hand, water temperature is the most important influential factor for predicting algal biomass. Our work provides an effective tool for managers to develop preemptive measures to control algal blooms.

Predicting Chlorophyll-a Concentrations in the World's Largest Lakes Using Kolmogorov-Arnold Networks

Accurate prediction of chlorophyll-a (Chl-a) concentrations, a key indicator of eutrophication, is essential for the sustainable management of lake ecosystems. This study evaluated the performance of Kolmogorov-Arnold Networks (KANs) along with three neural network models (MLP-NN, LSTM, and GRU) and three traditional machine learning tools (RF, SVR, and GPR) for predicting time-series Chl-a concentrations in large lakes. Monthly remote-sensed Chl-a data derived from Aqua-MODIS spanning September 2002 to April 2024 were used. The models were evaluated based on their forecasting capabilities from March 2024 to August 2024. KAN consistently outperformed others in both test and forecast (unseen data) phases and demonstrated superior accuracy in capturing trends, dynamic fluctuations, and peak Chl-a concentrations. Statistical evaluation using ranking metrics and critical difference diagrams confirmed KAN's robust performance across diverse study sites, further emphasizing its predictive power. Our findings suggest that the KAN, which leverages the KA representation theorem, offers improved handling of nonlinearity and long-term dependencies in time-series Chl-a data, outperforming neural network models grounded in the universal approximation theorem and traditional machine learning algorithms.

Enhanced predictive modeling of dissolved oxygen concentrations in riverine systems using novel hybrid temporal pattern attention deep neural networks

Monitoring water quality and river ecosystems is vital for maintaining public health and environmental sustainability. Over the past decade, data-driven methods have been extensively used for river water quality modeling, including dissolved oxygen (DO) concentrations. Despite advancements, challenges persist regarding accuracy, scalability, and adaptability of data-driven models to diverse environmental conditions. Previous studies primarily employed singular models or basic combinations of machine learning techniques, lacking advanced integration of adaptive mechanisms to process complex and evolving datasets. The current study introduces innovative hybrid models that integrate temporal pattern attention (TPA) mechanisms with advanced neural networks, including feed-forward neural networks (FFNNs) and long short-term memory networks (LSTMs). This approach leverages the synergistic strengths of individual models, significantly enhancing the accuracy of DO predictions. The models were rigorously tested against water quality data obtained from two distinct riverine environments, the Illinois River (ILL) and Des Plaines River (DP). Daily measured water quality data, including DO, chlorophyll-a, nitrate plus nitrite, water temperature, specific conductance, and pH, from 2017 to 2024 provided a robust foundation for comprehensive analysis of DO dynamics in these rivers. We conducted 10 scenarios with different model inputs, wherein the hybrid TPACWRNN-LSTM-10 model particularly excelled, achieving coefficient of determination values of 0.993 and 0.965, and root mean squared errors of 0.241 mg/L and 0.450 mg/L for DO predictions at the ILL and DP stations, respectively. The model's reliability was further confirmed by Willmott's index values of 0.998 and 0.992 and Nash-Sutcliffe efficiency values of 0.990 and 0.961 at the ILL and DP stations, respectively. Additionally, Shapley additive explanations (SHAP) values were utilized to interpret each predictor's contribution, revealing key drivers of DO predictions. We believe the novel hybrid modeling approach presented in this study could benefit utilities and water resource management systems for predicting water quality in complex systems.

Assessment of Chicken Fecal Contamination Using Microbial Source Tracking (MST) and Environmental DNA (eDNA) Profiling in Silway River, Philippines

The Silway River has historically failed to meet safe fecal coliform levels due to improper waste disposal. The river mouth is located in General Santos City, the tuna capital of the Philippines and a leading producer of hogs, cattle, and poultry. The buildup of contaminants due to direct discharge of waste from chicken farms and existing water quality conditions has led to higher fecal matter in the Silway River. While there were technical reports in the early 2000s about poultry farming, this is the first study where fecal coliform from poultry farming was detected in the Silway River using highly sensitive protocols like qPCR. This study characterized the effect of flow velocity and physicochemical water quality parameters on chicken fecal contamination. Gene markers such as Ckmito and ND5-CD were used to detect and quantify poultry manure contamination through microbial source tracking (MST) and environmental DNA (eDNA) profiling. The results of this study showed the presence of chicken fecal bacteria in all stations along the Silway River. The results revealed that normal levels of water quality parameters such as temperature, pH, and high TSS concentrations create favorable conditions for chicken fecal coliforms to thrive. Multiple regression analysis showed that flow velocity and DO significantly affect chicken fecal contamination. A lower cycle threshold (Ct) value indicated higher concentration of the marker ND5-CD, which means higher fecal contamination. It was found that there was an inverse relationship between the Ct value and both velocity (R2 = 0.55, p = 0.01) and DO (R2 = 0.98, p = 0.2), suggesting that low flow velocity and low DO can lead to higher fecal contamination. Findings of fecal contamination could negatively impact water resources, the health of nearby residents, and surrounding farms and industries, as well as the health and growth of fish.

Unravelling nitrate transformation mechanisms in karst catchments through the coupling of high-frequency sensor data and machine learning

Nitrate dynamics within a catchment are critical to the earth's system process, yet the intricate details of its transport and transformation at high resolutions remain elusive. Hydrological effects on nitrate dynamics in particular have not been thoroughly assessed previously and this knowledge gap hampers our understanding and effective management of nitrogen cycling in watersheds. Here, machine learning (ML) models were employed to reconstruct the annual variation trend in nitrate dynamics and isotopes within a typical karst catchment. Random forest model demonstrates promising potential in predicting nitrate concentration and its isotopes, surpassing other ML models (including Long Short-term Memory, Convolutional Neural Network, and Support Vector Machine) in performance. The ML-modeled NO3--N concentrations, δ15N-NO3-, and δ18O-NO3- values were in close agreement with field data (NSE values of 0.95, 0.80, and 0.53, respectively), which are notably challenging to achieve for process models. During the transition from dry to wet period, approximately 23.0 % of the annual precipitation (∼269.1 mm) was identified as the threshold for triggering a rapid response in the wet period. The modeled nitrate isotope values were significantly supported by the field data, suggesting seasonal variations of nitrogen sources, with precipitation as the primary driving force for fertilizer sources. Mixing of multiple sources appeared to be the main control of the transport and transformation of nitrate during the rising limb in the wet period, whereas process control (denitrification) took precedence during the falling limb, and the fate of nitrate was controlled by biogeochemical processes during the dry period.

Increasing phosphorus loss despite widespread concentration decline in US rivers

The loss of phosphorous (P) from the land to aquatic systems has polluted waters and threatened food production worldwide. Systematic trend analysis of P, a nonrenewable resource, has been challenging, primarily due to sparse and inconsistent historical data. Here, we leveraged intensive hydrometeorological data and the recent renaissance of deep learning approaches to fill data gaps and reconstruct temporal trends. We trained a multitask long short-term memory model for total P (TP) using data from 430 rivers across the contiguous United States (CONUS). Trend analysis of reconstructed daily records (1980-2019) shows widespread decline in concentrations, with declining, increasing, and insignificantly changing trends in 60%, 28%, and 12% of the rivers, respectively. Concentrations in urban rivers have declined the most despite rising urban population in the past decades; concentrations in agricultural rivers however have mostly increased, suggesting not-as-effective controls of nonpoint sources in agriculture lands compared to point sources in cities. TP loss, calculated as fluxes by multiplying concentration and discharge, however exhibited an overall increasing rate of 6.5% per decade at the CONUS scale over the past 40 y, largely due to increasing river discharge. Results highlight the challenge of reducing TP loss that is complicated by changing river discharge in a warming climate.

A novel coupling interpretable machine learning framework for water quality prediction and environmental effect understanding in different flow discharge regulations of hydro-projects

Machine learning models (MLMs) have been increasingly used to forecast water pollution. However, the "black box" characteristic for understanding mechanism processes still limits the applicability of MLMs for water quality management in hydro-projects under complex and frequently artificial regulation. This study proposes an interpretable machine learning framework for water quality prediction coupled with a hydrodynamic (flow discharge) scenario-based Random Forest (RF) model with multiple model-agnostic techniques and quantifies global, local, and joint interpretations (i.e., partial dependence, individual conditional expectation, and accumulated local effects) of environmental factor implications. The framework was applied and verified to predict the permanganate index (CODMn) under different flow discharge regulation scenarios in the Middle Route of the South-to-North Water Diversion Project of China (MRSNWDPC). A total of 4664 sampling cases data matrices, including water quality, meteorological, and hydrological indicators from eight national stations along the main canal of the MRSNWDPC, were collected from May 2019 to December 2020. The results showed that the RF models were effective in forecasting CODMn in all flow discharge scenarios, with a mean square error, coefficient of determination, and mean absolute error of 0.006-0.026, 0.481-0.792, and 0.069-0.104, respectively, in the testing dataset. A global interpretation indicated that dissolved oxygen, flow discharge, and surface pressure are the three most important variables of CODMn. Local and joint interpretations indicated that the RF-based prediction model provides a basic understanding of the physical mechanisms of environmental systems. The proposed framework can effectively learn the fundamental environmental implications of water quality variations and provide reliable prediction performance, highlighting the importance of model interpretability for trustworthy machine learning applications in water management projects. This study provides scientific references for applying advanced data-driven MLMs to water quality forecasting and a reliable methodological framework for water quality management and similar hydro-projects.

The impact of anthropogenic pressures on microbial diversity and river multifunctionality relationships on a global scale

Conserving biodiversity is crucial for maintaining essential ecosystem functions, as indicated by the positive relationships between biodiversity and ecosystem functioning. However, the impacts of declining biodiversity on ecosystem functions in response to mounting human pressures remain uncertain. This uncertainty arises from the complexity of trade-offs among human activities, climate change, river properties, and biodiversity, which have not been comprehensively addressed collectively. Here, we provide evidence that river biodiversity was significantly and positively associated with multifunctionality and contributed to key ecosystem functions such as microbially driven water purification, leaf litter decomposition and pathogen control. However, human pressure led to abrupt changes in microbial diversity and river multifunctionality relationships at a human pressure value of 0.5. In approximately 30 % (N = 58) of countries globally, the ratio of area above this threshold exceeded the global average (∼11 %), especially in Europe. Results show that human pressure affected ecosystem functions through direct effects and interactive effects. We provide more direct evidence that the nonadditive effects triggered by prevailing human pressure impact the multifunctionality of rivers globally. Under high levels of human stress, the beneficial effects of biodiversity on nutrient cycling, carbon storage, gross primary productivity, leaf litter decomposition, and pathogen control tend to diminish. Our findings highlight that considering interactions between human pressure and local abiotic and biotic factors is key for understanding the fate of river ecosystems under climate change and increasing human pressure.

Environmental determinants of aerobic methane oxidation in a tropical river network

Aerobic methane oxidation (MOX) significantly reduces methane (CH4) emissions from inland water bodies and is, therefore, an important determinant of global CH4 budget. Yet, the magnitude and controls of MOX rates in rivers - a quantitatively significant natural source of atmospheric CH4 - are poorly constrained. Here, we conducted a series of incubation experiments to understand the magnitude and environmental controls of MOX rates in tropical fluvial systems. We observed a large variability in MOX rate (0.03 - 3.45 μmol l-1d-1) shaped by a suit of environmental variables. Accordingly, we developed an empirical model for MOX that incorporate key environmental drivers, including temperature, CH4, total phosphorus, and dissolved oxygen (O2) concentrations, based on the results of our incubation experiments. We show that temperature dependency of MOX (activation energy: 0.66 ± 0.18 eV) is lower than that of sediment methanogenesis (0.71 ± 0.21 eV) in the studied tropical fluvial network. Furthermore, we observed a non-linear relationship between O2 concentration and MOX, with the highest MOX rate occuring ∼135 μmol O2l-1, above or below this "optimal O2" concentration, MOX rate shows a gradual decline. Together, our results suggest that the relatively lower temperature response of MOX compared to methanogenesis along with the projected decrease of O2 concentration due to organic pollution may cause elevated CH4 emission from tropical southeast Asian rivers. Since estimation of CH4 oxidation is often neglected in routine CH4 monitoring programs, the model developed here may help to integrate MOX rate into process-based models for fluvial CH4 budget.

Which riverine water quality parameters can be predicted by meteorologically-driven deep learning?

River water quality has been significantly impacted by climate change and extreme weather events worldwide. Despite increasing studies on deep learning techniques for river water quality management, understanding which riverine water quality parameters can be well predicted by meteorologically-driven deep learning still requires further investigation. Here we explored the prediction performance of a traditional Recurrent Neural Network, a Long Short-Term Memory network (LSTM), and a Gated Recurrent Unit (GRU) using meteorological conditions as inputs in the Dahei River basin. We found that deep learning models (i.e., LSTM and GRU) demonstrated remarkable effectiveness in predicting multiple water quality parameters at daily scale, including water temperature, dissolved oxygen, electrical conductivity, chemical oxygen demand, ammonia nitrogen, total phosphorous, and total nitrogen, but not turbidity. The GRU model performed best with an average determination coefficient of 0.94. Compared to the daily-average prediction, the GRU model exhibited limited error increment of 10-40 % for most water quality parameters when predicting daily extreme values (i.e., the maximum and minimum). Moreover, deep learning showed superior performance in collective prediction for multiple water quality parameters than individual ones, enabling a more comprehensive understanding of the river water quality dynamics from meteorological data. This study holds the promise of applying meteorologically-driven deep learning techniques for water quality prediction to a broader range of watersheds, particularly in chemically ungauged areas.

Pollution loads in the middle-lower Yangtze river by coupling water quality models with machine learning

Pollution control and environmental protection of the Yangtze River have received major attention in China. However, modeling the river's pollution load remains challenging due to limited monitoring and unclear spatiotemporal distribution of pollution sources. Specifically, anthropogenic activities' contribution to the pollution have been underestimated in previous research. Here, we coupled a hydrodynamic-based water quality (HWQ) model with a machine learning (ML) model, namely attention-based Gated Recurrent Unit, to decipher the daily pollution loads (i.e., chemical oxygen demand, COD; total phosphorus, TP) and their sources in the Middle-Lower Yangtze River from 2014 to 2018. The coupled HWQ-ML model outperformed the standalone ML model with KGE values ranging 0.77-0.91 for COD and 0.47-0.64 for TP, while also reducing parameter uncertainty. When examining the relative contributions at the Middle Yangtze River Hankou cross-section, we observed that the main stream and tributaries, lateral anthropogenic discharges, and parameter uncertainty contributed 15, 66, and 19% to COD, and 58, 35, and 7% to TP, respectively. For the Lower Yangtze River Datong cross-section, the contributions were 6, 69, and 25% for COD and 41, 42, and 17% for TP. According to the attention weights of the coupled model, the primary drivers of lateral anthropogenic pollution sources, in descending order of importance, were temperature, date, and precipitation, reflecting seasonal pollution discharge, industrial effluent, and first flush effect and combined sewer overflows, respectively. This study emphasizes the synergy between physical modeling and machine learning, offering new insights into pollution load dynamics in the Yangtze River.

Analyzing spatio-temporal dynamics of dissolved oxygen for the River Thames using superstatistical methods and machine learning

By employing superstatistical methods and machine learning, we analyze time series data of water quality indicators for the River Thames (UK). The indicators analyzed include dissolved oxygen, temperature, electrical conductivity, pH, ammonium, turbidity, and rainfall, with a specific focus on the dynamics of dissolved oxygen. After detrending, the probability density functions of dissolved oxygen fluctuations exhibit heavy tails that are effectively modeled using q-Gaussian distributions. Our findings indicate that the multiplicative Empirical Mode Decomposition method stands out as the most effective detrending technique, yielding the highest log-likelihood in nearly all fittings. We also observe that the optimally fitted width parameter of the q-Gaussian shows a negative correlation with the distance to the sea, highlighting the influence of geographical factors on water quality dynamics. In the context of same-time prediction of dissolved oxygen, regression analysis incorporating various water quality indicators and temporal features identify the Light Gradient Boosting Machine as the best model. SHapley Additive exPlanations reveal that temperature, pH, and time of year play crucial roles in the predictions. Furthermore, we use the Transformer, a state-of-the-art machine learning model, to forecast dissolved oxygen concentrations. For long-term forecasting, the Informer model consistently delivers superior performance, achieving the lowest Mean Absolute Error (0.15) and Symmetric Mean Absolute Percentage Error (21.96%) with the 192 historical time steps that we used. This performance is attributed to the Informer's ProbSparse self-attention mechanism, which allows it to capture long-range dependencies in time-series data more effectively than other machine learning models. It effectively recognizes the half-life cycle of dissolved oxygen, with particular attention to critical periods such as morning to early afternoon, late evening to early morning, and key intervals between the 16th and 26th quarter-hours of the previous half-day. Our findings provide valuable insights for policymakers involved in ecological health assessments, aiding in accurate predictions of river water quality and the maintenance of healthy aquatic ecosystems.

An integrated deep learning approach for modeling dissolved oxygen concentration at coastal inlets based on hydro-climatic parameters

Climate change has a significant impact on dissolved oxygen (DO) concentrations, particularly in coastal inlets where numerous human activities occur. Due to the various water quality (WQ), hydrological, and climatic parameters that influence this phenomenon, predicting and modeling DO variation is a challenging process. Accordingly, this study introduces an innovative Deep Learning Neural Network (DLNN) methodology to model and predict DO concentrations for the Egyptian Rashid coastal inlet, leveraging field-recorded WQ and hydroclimatic datasets. Initially, statistical and exploratory data analyses are performed to provide a thorough understanding of the relationship between DO fluctuations and associated WQ and hydroclimatic stressors. As an initial step towards developing an effective DO predictive model, conventional Machine Learning (ML) approaches such as Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Decision Tree Regressor (DTR) are employed. Subsequently, a DLNN approach is utilized to validate the prediction capabilities of the investigated conventional ML approaches. Finally, a sensitivity analysis is conducted to evaluate the impact of WQ and hydroclimatic parameters on predicted DO. The outcomes demonstrate that DLNN significantly improves DO prediction accuracy by 4% compared to the best-performing ML approach, achieving a Correlation Coefficient of 0.95 with a root mean square error (RMSE) of 0.42 mg/l. Solar radiation (SR), pH, water levels (WL), and atmospheric pressure (P) emerge as the most significant hydroclimatic parameters influencing DO fluctuations. Ultimately, the developed models could serve as effective indicators for coastal authorities to monitor DO changes resulting from accelerated climate change along the Egyptian coast.

Overlooked Roles and Transformation of Carbon-Centered Radicals Produced from Natural Organic Matter in a Thermally Activated Persulfate System

The presence and induced secondary reactions of natural organic matter (NOM) significantly affect the remediation efficacy of in situ chemical oxidation (ISCO) systems. However, it remains unclear how this process relates to organic radicals generated from reactions between the NOM and oxidants. The study, for the first time, reported the vital roles and transformation pathways of carbon-centered radicals (CCR•) derived from NOM in activated persulfate (PS) systems. Results showed that both typical terrestrial/aquatic NOM isolates and collected NOM samples produced CCR• by scavenging activated PS and greatly enhanced the dehalogenation performance under anoxic conditions. Under oxic conditions, newly formed CCR• could be oxidized by O2 and generate organic peroxide intermediates (ROO•) to catalytically yield additional •OH without the involvement of PS. Nuclear magnetic resonance (NMR) and Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR-MS) results indicated that CCR• predominantly formed from carboxyl and aliphatic structures instead of aromatics within NOM through hydrogen abstraction and decarboxylation reactions by SO4•- or •OH. Specific anoxic reactions (i.e., dehalogenation and intramolecular cross-coupling reactions) further promoted the transformation of CCR• to more unsaturated and polymerized/condensed compounds. In contrast, oxic propagation of ROO• enhanced bond breakage/ring cleavage and degradation of CCR• due to the presence of additional •OH and self-decomposition. This study provides novel insights into the role of NOM and O2 in ISCO and the development of engineered strategies for creating organic radicals capable of enhancing the remediation of specific contaminants and recovering organic carbon.

Multi-step ahead forecasting of electrical conductivity in rivers by using a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model enhanced by Boruta-XGBoost feature selection algorithm

Electrical conductivity (EC) is widely recognized as one of the most essential water quality metrics for predicting salinity and mineralization. In the current research, the EC of two Australian rivers (Albert River and Barratta Creek) was forecasted for up to 10 days using a novel deep learning algorithm (Convolutional Neural Network combined with Long Short-Term Memory Model, CNN-LSTM). The Boruta-XGBoost feature selection method was used to determine the significant inputs (time series lagged data) to the model. To compare the performance of Boruta-XGB-CNN-LSTM models, three machine learning approaches-multi-layer perceptron neural network (MLP), K-nearest neighbour (KNN), and extreme gradient boosting (XGBoost) were used. Different statistical metrics, such as correlation coefficient (R), root mean square error (RMSE), and mean absolute percentage error, were used to assess the models' performance. From 10 years of data in both rivers, 7 years (2012-2018) were used as a training set, and 3 years (2019-2021) were used for testing the models. Application of the Boruta-XGB-CNN-LSTM model in forecasting one day ahead of EC showed that in both stations, Boruta-XGB-CNN-LSTM can forecast the EC parameter better than other machine learning models for the test dataset (R = 0.9429, RMSE = 45.6896, MAPE = 5.9749 for Albert River, and R = 0.9215, RMSE = 43.8315, MAPE = 7.6029 for Barratta Creek). Considering the better performance of the Boruta-XGB-CNN-LSTM model in both rivers, this model was used to forecast 3-10 days ahead of EC. The results showed that the Boruta-XGB-CNN-LSTM model is very capable of forecasting the EC for the next 10 days. The results showed that by increasing the forecasting horizon from 3 to 10 days, the performance of the Boruta-XGB-CNN-LSTM model slightly decreased. The results of this study show that the Boruta-XGB-CNN-LSTM model can be used as a good soft computing method for accurately predicting how the EC will change in rivers.

Water quality prediction based on sparse dataset using enhanced machine learning

Water quality in surface bodies remains a pressing issue worldwide. While some regions have rich water quality data, less attention is given to areas that lack sufficient data. Therefore, it is crucial to explore novel ways of managing source-oriented surface water pollution in scenarios with infrequent data collection such as weekly or monthly. Here we showed sparse-dataset-based prediction of water pollution using machine learning. We investigated the efficacy of a traditional Recurrent Neural Network alongside three Long Short-Term Memory (LSTM) models, integrated with the Load Estimator (LOADEST). The research was conducted at a river-lake confluence, an area with intricate hydrological patterns. We found that the Self-Attentive LSTM (SA-LSTM) model outperformed the other three machine learning models in predicting water quality, achieving Nash-Sutcliffe Efficiency (NSE) scores of 0.71 for CODMn and 0.57 for NH3N when utilizing LOADEST-augmented water quality data (referred to as the SA-LSTM-LOADEST model). The SA-LSTM-LOADEST model improved upon the standalone SA-LSTM model by reducing the Root Mean Square Error (RMSE) by 24.6% for CODMn and 21.3% for NH3N. Furthermore, the model maintained its predictive accuracy when data collection intervals were extended from weekly to monthly. Additionally, the SA-LSTM-LOADEST model demonstrated the capability to forecast pollution loads up to ten days in advance. This study shows promise for improving water quality modeling in regions with limited monitoring capabilities.

A hybrid deep learning approach to predict hourly riverine nitrate concentrations using routine monitored data

With high-frequency data of nitrate (NO3-N) concentrations in waters becoming increasingly important for understanding of watershed system behaviors and ecosystem managements, the accurate and economic acquisition of high-frequency NO3-N concentration data has become a key point. This study attempted to use coupled deep learning neural networks and routine monitored data to predict hourly NO3-N concentrations in a river. The hourly NO3-N concentration at the outlet of the Oyster River watershed in New Hampshire, USA, was predicted through neural networks with a hybrid model architecture coupling the Convolutional Neural Networks and the Long Short-Term Memory model (CNN-LSTM). The routine monitored data (the river depth, water temperature, air temperature, precipitation, specific conductivity, pH and dissolved oxygen concentrations) for model training were collected from a nested high-frequency monitoring network, while the high-frequency NO3-N concentration data obtained at the outlet were not included as inputs. The whole dataset was separated into training, validation, and testing processes according to the ratio of 5:3:2, respectively. The hybrid CNN-LSTM model with different input lengths (1d, 3d, 7d, 15d, 30d) displayed comparable even better performance than other studies with lower frequencies, showing mean values of the Nash-Sutcliffe Efficiency 0.60-0.83. Models with shorter input lengths demonstrated both the higher modeling accuracy and stability. The water level, water temperature and pH values at monitoring sites were main controlling factors for forecasting performances. This study provided a new insight of using deep learning networks with a coupled architecture and routine monitored data for high-frequency riverine NO3-N concentration forecasting and suggestions about strategies about variable and input length selection during preprocessing of input data.

Deep learning for cross-region streamflow and flood forecasting at a global scale

Streamflow and flood forecasting remains one of the long-standing challenges in hydrology. Traditional physically based models are hampered by sparse parameters and complex calibration procedures particularly in ungauged catchments. We propose a novel hybrid deep learning model termed encoder-decoder double-layer long short-term memory (ED-DLSTM) to address streamflow forecasting at global scale for all (gauged and ungauged) catchments. Using historical datasets, ED-DLSTM yields a mean Nash-Sutcliffe efficiency coefficient (NSE) of 0.75 across more than 2,000 catchments from the United States, Canada, Central Europe, and the United Kingdom, highlighting improvements by the state-of-the-art machine learning over traditional hydrologic models. Moreover, ED-DLSTM is applied to 160 ungauged catchments in Chile and 76.9% of catchments obtain NSE >0 in the best situation. The interpretability of cross-region capacities of ED-DLSTM are established through the cell state induced by adding a spatial attribute encoding module, which can spontaneously form hydrological regionalization effects after performing spatial coding for different catchments. The study demonstrates the potential of deep leaning methods to overcome the ubiquitous lack of hydrologic information and deficiencies in physical model structure and parameterization.

Adaptive prediction for effluent quality of wastewater treatment plant: Improvement with a dual-stage attention-based LSTM network

The accurate effluent prediction plays a crucial role in providing early warning for abnormal effluent and achieving the adjustment of feedforward control parameters during wastewater treatment. This study applied a dual-staged attention mechanism based on long short-term memory network (DA-LSTM) to improve the accuracy of effluent quality prediction. The results showed that input attention (IA) and temporal attention (TA) significantly enhanced the prediction performance of LSTM. Specially, IA could adaptively adjust feature weights to enhance the robustness against input noise, with R2 increased by 13.18%. To promote its long-term memory ability, TA was used to increase the memory span from 96 h to 168 h. Compared to a single LSTM model, the DA-LSTM model showed an improvement in prediction accuracy by 5.10%, 2.11%, 14.47% for COD, TP, and TN. Additionally, DA-LSTM demonstrated excellent generalization performance in new scenarios, with the R2 values for COD, TP, and TN increasing by 22.67%, 20.06%, and 17.14% respectively, while the MAPE values decreased by 56.46%, 63.08%, and 42.79%. In conclusion, the DA-LSTM model demonstrated excellent prediction performance and generalization ability due to its advantages of feature-adaptive weighting and long-term memory focusing. This has forward-looking significance for achieving efficient early warning of abnormal operating conditions and timely management of control parameters.

Analysis of dissolved oxygen influencing factors and concentration prediction using input variable selection technique: A hybrid machine learning approach

Accurate quantification of dissolved oxygen (DO) is critically important for the protection and management of aquatic ecosystems. Successful applications have utilized mechanistic and data-driven models to simulate DO content in aquatic ecosystems. However, mechanistic models present challenges due to their complex and difficult-to-solve conditions, making them less portable. Additionally, data-driven model predictions are hindered by the challenge of numerous input variables, impacting both the running speed and prediction performance of the model. To address these challenges, water quality data and meteorological data of the Tanjiang River were obtained. The maximum information coefficient (MIC) input variable selection technique was employed to identify primary environmental factors influencing DO changes. Furthermore, coupled with support vector regression (SVR), two models (SVR and MIC-SVR) were employed to estimate the DO concentration of the Tanjiang River, and the optimal model was established. The results indicated a shift in the primary pollution factor from ammonia nitrogen to total phosphorus after recent treatment in the Tanjiang River. In comparison with the SVR model, the root mean square error (RMSE) of the MIC-SVR model was reduced by 4.46%, and the Nash-efficiency coefficient (NSE) was improved by 45.85%. In addition, study of kernel function selection revealed that considering as many kernel functions as possible is necessary for improving the performance of the SVR model. Conclusively, the proposed MIC-SVR model serves as an effective tool to analyze the relationship between DO and environmental factors, identifying the primary causes of low DO, and accurately predict the DO concentration in the Tanjiang River (especially in its middle and lower reaches), thus providing a reference for governmental decision-making on water environmental protection and water resource management.

Performance evaluation of deep learning based stream nitrate concentration prediction model to fill stream nitrate data gaps at low-frequency nitrate monitoring basins

Accurate and frequent nitrate estimates can provide valuable information on the nitrate transport dynamics. The study aimed to develop a data-driven modeling framework to estimate daily nitrate concentrations at low-frequency nitrate monitoring sites using the daily nitrate concentration and stream discharge information of a neighboring high-frequency nitrate monitoring site. A Long Short-Term Memory (LSTM) based deep learning (DL) modeling framework was developed to predict daily nitrate concentrations. The DL modeling framework performance was compared with two well-established statistical models, including LOADEST and WRTDS-Kalman, in three selected basins in Iowa, USA: Des Moines, Iowa, and Cedar River. The developed DL model performed well with NSE >0.70 and KGE >0.70 for 67% and 79% nitrate monitoring sites, respectively. DL and WRTDS-Kalman models performed better than the LOADEST in nitrate concentration and load estimation for all low-frequency sites. The average NSE performance of the DL model in daily nitrate estimation is 20% higher than that of the WRTDS-Kalman model at 18 out of 24 sites (75%). The WRTDS-Kalman model showed unrealistic fluctuations in the estimated daily nitrate time series when the model received limited observed nitrate data (less than 50) for simulation. The DL model indicated superior performance in winter months' nitrate prediction (60% of cases) compared to WRTDS-Kalman models (33% of cases). The DL model also better represented the exceedance days from the USEPA maximum contamination level (MCL). Both the DL and WRTDS-Kalman models demonstrated similar performance in annual stream nitrate load estimation, and estimated values are close to actual nitrate loads.

Reliable water quality prediction and parametric analysis using explainable AI models

The consumption of water constitutes the physical health of most of the living species and hence management of its purity and quality is extremely essential as contaminated water has to potential to create adverse health and environmental consequences. This creates the dire necessity to measure, control and monitor the quality of water. The primary contaminant present in water is Total Dissolved Solids (TDS), which is hard to filter out. There are various substances apart from mere solids such as potassium, sodium, chlorides, lead, nitrate, cadmium, arsenic and other pollutants. The proposed work aims to provide the automation of water quality estimation through Artificial Intelligence and uses Explainable Artificial Intelligence (XAI) for the explanation of the most significant parameters contributing towards the potability of water and the estimation of the impurities. XAI has the transparency and justifiability as a white-box model since the Machine Learning (ML) model is black-box and unable to describe the reasoning behind the ML classification. The proposed work uses various ML models such as Logistic Regression, Support Vector Machine (SVM), Gaussian Naive Bayes, Decision Tree (DT) and Random Forest (RF) to classify whether the water is drinkable. The various representations of XAI such as force plot, test patch, summary plot, dependency plot and decision plot generated in SHAPELY explainer explain the significant features, prediction score, feature importance and justification behind the water quality estimation. The RF classifier is selected for the explanation and yields optimum Accuracy and F1-Score of 0.9999, with Precision and Re-call of 0.9997 and 0.998 respectively. Thus, the work is an exploratory analysis of the estimation and management of water quality with indicators associated with their significance. This work is an emerging research at present with a vision of addressing the water quality for the future as well.

Prediction of riverine daily minimum dissolved oxygen concentrations using hybrid deep learning and routine hydrometeorological data

Dissolved oxygen (DO) depletion is a severe threat to aquatic ecosystems. Hence, using easily available routine hydrometeorological variables without DO as inputs to predict the daily minimum DO concentration in rivers has huge practical significance in the watershed management. The daily minimum DO concentrations at the outlet of the Oyster River watershed in New Hampshire, USA, were predicted by a set of deep learning neural networks using meteorological data and high-frequency water level, water temperature, and specific conductance (CTD) data measured within the watershed. The dependent variable, DO concentration, was measured at the outlet. From April 2013 to March 2018, the dataset was separated into training, validation, and test portions with a ratio of 5:3:3. A Long Short-Term Memory (LSTM) model and a hybrid Convolutional Neural Networks (CNN-LSTM) model were trained and evaluated for predicting the daily minimum DO concentration. The hybrid CNN-LSTM model exhibited the better predictive stability but the comparable accuracy (the mean R2 value = 0.865) compared with the pure LSTM model (the mean R2 value = 0.839). The model performance (both the stability and accuracy) was improved by aggregating the input data frequency from 15 min of raw data to 24 h. Likewise, the modeling performance didn't benefit from including 'forecasted' meteorological data at the predicted time step in the input dataset. This study provided an efficient and low-cost approach to predict the water quality in rivers and streams to realize the scientific watershed management.

Single and combined effects of increased temperature and methylmercury on different stages of the marine rotifer Brachionus plicatilis

Rapid, anthropogenic activity-induced global warming is a severe problem that not only raises water temperatures but also shifts aquatic environments by increasing the bioavailability of heavy metals (HMs), with potentially complicated effects on aquatic organisms, including small aquatic invertebrates. For this paper, we investigated the combined effects of temperature (23 and 28 °C) and methylmercury (MeHg) by measuring physiological changes, bioaccumulation, oxidative stress, antioxidants, and the mitogen-activated protein kinase signaling pathway in the marine rotifer Brachionus plicatilis. High temperature and MeHg adversely affected the survival rate, lifespan, and population of rotifers, and bioaccumulation, oxidative stress, and biochemical reactions depended on the developmental stage, with neonates showing higher susceptibility than adults. These findings demonstrate that increased temperature enhances potentially toxic effects from MeHg, and susceptibility differs with the developmental stage. This study provides a comprehensive understanding of the combined effects of elevated temperature and MeHg on rotifers. ENVIRONMENTAL IMPLICATION: Methylmercury (MeHg) is a widespread and harmful heavy metal that can induce lethal effects on aquatic organisms in even trace amounts. The toxicity of metals can vary depending on various environmental conditions. In particular, rising temperatures are considered a major factor affecting bioavailability and toxicity by changing the sensitivity of organisms. However, there are few studies on the combinational effects of high temperatures and MeHg on aquatic animals, especially invertebrates. Our research would contribute to understanding the actual responses of aquatic organisms to complex aquatic environments.

Deep learning for water quality

Understanding and predicting the quality of inland waters are challenging, particularly in the context of intensifying climate extremes expected in the future. These challenges arise partly due to complex processes that regulate water quality, and arduous and expensive data collection that exacerbate the issue of data scarcity. Traditional process-based and statistical models often fall short in predicting water quality. In this Review, we posit that deep learning represents an underutilized yet promising approach that can unravel intricate structures and relationships in high-dimensional data. We demonstrate that deep learning methods can help address data scarcity by filling temporal and spatial gaps and aid in formulating and testing hypotheses via identifying influential drivers of water quality. This Review highlights the strengths and limitations of deep learning methods relative to traditional approaches, and underscores its potential as an emerging and indispensable approach in overcoming challenges and discovering new knowledge in water-quality sciences.

High-frequency data significantly enhances the prediction ability of point and interval estimation

Accurate prediction of dissolved oxygen (DO) dynamics is crucial for understanding the influence of environmental factors on the stability of aquatic ecosystem. However, limited research has been conducted to determine the optimal frequency of water quality monitoring that ensures continuous assessment of water health while minimizing costs. To address these challenges, the present study developed a hybrid stochastic hydrological model (i.e., ARIMA-GARCH hybrid model) and machine learning (ML) models. The objective of this study is to identify the best-performing model and establish the optimal monitoring frequency. Results revealed that high-frequency DO monitoring data exhibit greater variability compared to low-frequency data. Moreover, the ARIMA-GARCH model demonstrates promising potential in predicting DO concentrations for low-frequency monitoring data, surpassing ML models in performance. Furthermore, increasing the monitoring frequency significantly improves the prediction accuracy of models, regardless of whether point (with lower R2 values of 0.64 and 0.51 for daily detection than these of every 15 min (0.96 and 0.99) at CHQ and LHT, respectively) or interval predictions (with RIW higher values of 2.00 and 1.55 for daily detection higher than these of 0.02 and 0.16 in every 15 min at CHQ and LHT, respectively) are considered. Additionally, a 4 hourly monitoring frequency was found to be optimal for water quality assessment using each model. These findings identify the superior performing of the ARIMA-GARCH model and highlight the crucial role of monitoring frequency in enhancing DO prediction and improving model performance.

Predicting dissolved oxygen level using Young's double-slit experiment optimizer-based weighting model

Accurate prediction of the dissolved oxygen level (DOL) is important for enhancing environmental conditions and facilitating water resource management. However, the irregularity and volatility inherent in DOL pose significant challenges to achieving precise forecasts. A single model usually suffers from low prediction accuracy, narrow application range, and difficult data acquisition. This study proposes a new weighted model that avoids these problems, which could increase the prediction accuracy of the DOL. The weighting constructs of the proposed model (PWM) included eight neural networks and one statistical method and utilized Young's double-slit experimental optimizer as an intelligent weighting tool. To evaluate the effectiveness of PWM, simulations were conducted using real-world data acquired from the Tualatin River Basin in Oregon, United States. Empirical findings unequivocally demonstrated that PWM outperforms both the statistical model and the individual machine learning models, and has the lowest mean absolute percentage error among all the weighted models. Based on two real datasets, the PWM can averagely obtain the mean absolute percentage errors of 1.0216%, 1.4630%, and 1.7087% for one-, two-, and three-step predictions, respectively. This study shows that the PWM can effectively integrate the distinctive merits of deep learning methods, neural networks, and statistical models, thereby increasing forecasting accuracy and providing indispensable technical support for the sustainable development of regional water environments.

Deep learning in water protection of resources, environment, and ecology: achievement and challenges

The breathtaking economic development put a heavy toll on ecology, especially on water pollution. Efficient water resource management has a long-term influence on the sustainable development of the economy and society. Economic development and ecology preservation are tangled together, and the growth of one is not possible without the other. Deep learning (DL) is ubiquitous in autonomous driving, medical imaging, speech recognition, etc. The spectacular success of deep learning comes from its power of richer representation of data. In view of the bright prospects of DL, this review comprehensively focuses on the development of DL applications in water resources management, water environment protection, and water ecology. First, the concept and modeling steps of DL are briefly introduced, including data preparation, algorithm selection, and model evaluation. Finally, the advantages and disadvantages of commonly used algorithms are analyzed according to their structures and mechanisms, and recommendations on the selection of DL algorithms for different studies, as well as prospects for the application and development of DL in water science are proposed. This review provides references for solving a wider range of water-related problems and brings further insights into the intelligent development of water science.

Machine Learning-Assisted, Process-Based Quality Control for Detecting Compromised Environmental Sensors

Machine learning (ML) techniques promise to revolutionize environmental research and management, but collecting the necessary volumes of high-quality data remains challenging. Environmental sensors are often deployed under harsh conditions, requiring labor-intensive quality assurance and control (QAQC) processes. The need for manual QAQC is a major impediment to the scalability of these sensor networks. Existing techniques for automated QAQC make strong assumptions about noise profiles in the data they filter that do not necessarily hold for broadly deployed environmental sensors, however. Toward the goal of increasing the volume of high-quality environmental data, we introduce an ML-assisted QAQC methodology that is robust to low signal-to-noise ratio data. Our approach embeds sensor measurements into a dynamical feature space and trains a binary classification algorithm (Support Vector Machine) to detect deviation from expected process dynamics, indicating whether a sensor has become compromised and requires maintenance. This strategy enables the automated detection of a wide variety of nonphysical signals. We apply the methodology to three novel data sets produced by 136 low-cost environmental sensors (stream level, drinking water pH, and drinking water electroconductivity), deployed by our group across 250,000 km2 in Michigan, USA. The proposed methodology achieved accuracy scores of up to 0.97 and consistently outperformed state-of-the-art anomaly detection techniques.

Remote sensing-enabled machine learning for river water quality modeling under multidimensional uncertainty

Two fundamental problems have inhibited progress in the simulation of river water quality under climate (and other) uncertainty: 1) insufficient data, and 2) the inability of existing models to account for the complexity of factors (e.g., hydro-climatic, basin characteristics, land use features) affecting river water quality. To address these concerns this study presents a technique for augmenting limited ground-based observations of water quality variables with remote-sensed surface reflectance data by leveraging a machine learning model capable of accommodating the multidimensionality of water quality influences. Total Suspended Solids (TSS) can serve as a surrogate for chemical and biological pollutants of concern in surface water bodies. Historically, TSS data collection in the United States has been limited to the location of water treatment plants where state or federal agencies conduct regularly-scheduled water sampling. Mathematical models relating riverine TSS concentration to the explanatory factors have therefore been limited and the relationships between climate extremes and water contamination events have not been effectively diagnosed. This paper presents a method to identify these issues by utilizing a Long Short-Term Memory Network (LSTM) model trained on Moderate Resolution Imaging Spectroradiometer (MODIS) satellite reflectance data, which is calibrated to TSS data collected by the Ohio River Valley Water Sanitation Commission (ORSANCO). The methodology developed enables a thorough empirical analysis and data-driven algorithms able to account for spatial variability within the watershed and provide effective water quality prediction under uncertainty.

Sampling frequency optimization of the water quality monitoring network in São Paulo State (Brazil) towards adaptive monitoring in a developing country

Water quality monitoring networks (WQMNs) that capture both the temporal and spatial dimensions are essential to provide reliable data for assessing water quality trends in surface waters, as well as for supporting initiatives to control anthropogenic activities. Meeting these monitoring goals as efficiently as possible is crucial, especially in developing countries where the financial resources are limited and the water quality degradation is accelerating. Here, we asked if sampling frequency could be reduced while maintaining the same degree of information as with bimonthly sampling in the São Paulo State (Brazil) WQMN. For this purpose, we considered data from 2004 to 2018 for 56 monitoring sites distributed into four out of 22 of the state's water resources management units (UGRHIs, "Unidades de Gerenciamento de Recursos Hídricos"). We ran statistical tests for identifying data redundancy among two-month periods in the dry and wet seasons, followed by objective criteria to develop a sampling frequency recommendation. Our results showed that the reduction would be feasible in three UGRHIs, with the number of annual samplings ranging from two to four (instead of the original six). In both seasons, dissolved oxygen and Escherichia coli required more frequent sampling than the other analyzed parameters to adequately capture variability. The recommendation was compatible with flexible monitoring strategies observed in well-structured WQMNs worldwide, since the suggested sampling frequencies were not the same for all UGRHIs. Our approach can contribute to establishing a methodology to reevaluate WQMNs, potentially resulting in less costly and more adaptive strategies in São Paulo State and other developing areas with similar challenges.

Application of machine learning in river water quality management: a review

Machine learning (ML), a branch of artificial intelligence (AI), has been increasingly used in environmental engineering due to the ability to analyze complex nonlinear problems (such as ones connected with water quality management) through a data-driven approach. This study provides an overview of different ML algorithms applied for monitoring and predicting river water quality. Different parameters could be monitored or predicted, such as dissolved oxygen (DO), biological and chemical oxygen demand (BOD and COD), turbidity levels, the concentration of different ions (such as Mg2+ and Ca2+), heavy metal or other pollutant's concentration, pH, temperature, and many more. Although many algorithms have been investigated for the prediction of river water quality, there are several which are most commonly used in engineering practice. These models mostly include so-called supervised learning algorithms, such as artificial neural network (ANN), support vector machine (SVM), random forest (RF), decision tree (DT), and deep learning (DL). To further enhance prediction power, novel hybrid algorithms, could be used. However, the quality of prediction is not only dependent on the applied algorithm but also on the availability of previously mentioned water quality parameters, their selection, and the combination of input data used to train the ML model.

A deep learning-based novel approach to generate continuous daily stream nitrate concentration for nitrate data-sparse watersheds

High-frequency stream nitrate concentration provides critical insights into nutrient dynamics and can help to improve the effectiveness of management decisions to maintain a sustainable ecosystem. However, nitrate monitoring is conventionally conducted through lab analysis using in situ water samples and is typically at coarse temporal resolution. In the last decade, many agencies started collecting high-frequency (5-60 min intervals) nitrate data using optical sensors. The hypothesis of the study is that the data-driven models can learn the trend and temporal variability in nitrate concentration from high-frequency sensor-based nitrate data in the region and generate continuous nitrate data for unavailable data periods and data-limited locations. A Long Short-Term Memory (LSTM) model-based framework was developed to estimate continuous daily stream nitrate for dozens of gauge locations in Iowa, USA. The promising results supported the hypothesis; the LSTM model demonstrated median test-period Nash-Sutcliffe efficiency (NSE) = 0.75 and RMSE = 1.53 mg/L for estimating continuous daily nitrate concentration in 42 sites, which are unprecedented performance levels. Twenty-one sites (50 % of all sites) and thirty-four sites (76 % of all sites) demonstrated NSE > 0.75 and 0.50, respectively. The average nitrate concentration of neighboring sites was identified as a crucial determinant of continuous daily nitrate concentration. Seasonal model performance evaluation showed that the model performed effectively in the summer and fall seasons. About 26 sites showed correlations >0.60 between estimated nitrate concentration and discharge. The concentration-discharge (c-Q) relationship analysis showed that the study watersheds had four dominant nitrate transport patterns from landscapes to streams with increasing discharge, including the flushing pattern being the most dominant one. Stream nitrate estimation impedes due to data inadequacy. The modeling framework can be used to generate temporally continuous nitrate at nitrate data-limited regions with a nearby sensor-based nitrate gauge. Watershed planners and policymakers could utilize the continuous nitrate data to gain more information on the regional nitrate status and design conservation practices accordingly.

A novel four-dimensional prediction model of soil heavy metal pollution: Geographical explanations beyond artificial intelligence "black box"

The current artificial intelligence (AI)-based prediction approaches of soil pollutants are inadequate in estimating the geospatial source-sink processes and striking a balance between the interpretability and accuracy, resulting in poor spatial extrapolation and generalization. In this study, we developed and tested a geographically interpretable four-dimensional AI prediction model for soil heavy metal (Cd) contents (4DGISHM) in Shaoguan city of China from 2016 to 2030. The 4DGISHM approach characterized spatio-temporal changes in source-sink processes of soil Cd by estimating spatio-temporal patterns and the effects of drivers and their interactions of soil Cd at local to regional scales using TreeExplainer-based SHAP and parallel ensemble AI algorithms. The results demonstrate that the prediction model achieved MSE and R² values of 0.012 and 0.938, respectively, at a spatial resolution of 1 km. The predicted areas exceeding the risk control values for soil Cd across Shaoguan from 2022 to 2030 increased by 22.92% at the baseline scenario. By 2030, enterprise and transportation emissions (SHAP values 0.23 and 0.12 mg/kg, respectively) were the major drivers. The influence of driver interactions on soil Cd was marginal. Our approach surpasses the limitations of the AI "black box" by integrating spatio-temporal source-sink explanation and accuracy. This advancement enables geographically precise prediction and control of soil pollutants.

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.

A spectral learning path for simultaneous multi-parameter detection of water quality

Water quality parameters (WQP) are the most intuitive indicators of the environmental quality of water body. Due to the complexity and variability of the chemical environment of water body, simple and rapid detection of multiple parameters of water quality becomes a difficult task. In this paper, spectral images (named SPIs) and deep learning (DL) techniques were combined to construct an intelligent method for WQP detection. A novel spectroscopic instrument was used to obtain SPIs, which were converted into feature images of water chemistry and then combined with deep convolutional neural networks (CNNs) to train models and predict WQP. The results showed that the method of combining SPIs and DL has high accuracy and stability, and good prediction results with average relative error of each parameter (anions and cations, TOC, TP, TN, NO₃^--N, NH₃-N) at 1.3%, coefficient of determination (R²) of 0.996, root mean square error (RMSE) of 0.1, residual prediction deviation (RPD) of 16.2, and mean absolute error (MAE) of 0.067. The method can achieve rapid and accurate detection of high-dimensional water quality multi-parameters, and has the advantages of simple pre-processing and low cost. It can be applied not only to the intelligent detection of environmental waters, but also has the potential to be applied in chemical, biological and medical fields.

Recent global decline in rainfall interception loss due to altered rainfall regimes

Evaporative loss of interception (Ei) is the first process occurring during rainfall, yet its role in large-scale surface water balance has been largely underexplored. Here we show that Ei can be inferred from flux tower evapotranspiration measurements using physics-informed hybrid machine learning models built under wet versus dry conditions. Forced by satellite and reanalysis data, this framework provides an observationally constrained estimate of Ei, which is on average 84.1 ± 1.8 mm per year and accounts for 8.6 ± 0.2% of total rainfall globally during 2000-2020. Rainfall frequency regulates long-term average Ei changes, and rainfall intensity, rather than vegetation attributes, determines the fraction of Ei in gross precipitation (Ei/P). Rain events have become less frequent and more intense since 2000, driving a global decline in Ei (and Ei/P) by 4.9% (6.7%). This suggests that ongoing rainfall changes favor a partitioning towards more soil moisture and runoff, benefiting ecosystem functions but simultaneously increasing flood risks.

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.

An Outlook for Deep Learning in Ecosystem Science

Rapid advances in hardware and software, accompanied by public- and private-sector investment, have led to a new generation of data-driven computational tools. Recently, there has been a particular focus on deep learning—a class of machine learning algorithms that uses deep neural networks to identify patterns in large and heterogeneous datasets. These developments have been accompanied by both hype and scepticism by ecologists and others. This review describes the context in which deep learning methods have emerged, the deep learning methods most relevant to ecosystem ecologists, and some of the problem domains they have been applied to. Deep learning methods have high predictive performance in a range of ecological contexts, leveraging the large data resources now available. Furthermore, deep learning tools offer ecosystem ecologists new ways to learn about ecosystem dynamics. In particular, recent advances in interpretable machine learning and in developing hybrid approaches combining deep learning and mechanistic models provide a bridge between pure prediction and causal explanation. We conclude by looking at the opportunities that deep learning tools offer ecosystem ecologists and assess the challenges in interpretability that deep learning applications pose.

Ecological risk and machine learning based source analyses of trace metals in typical surface water

Surface water is threatened by trace metal pollution due to increasing anthropogenic activities. Therefore, an appropriate source identification was essential to reduce the ecological risk posed by the given pollutants. In this study, shallow and deep learning approaches trained by a registered environmental dataset of discharge sources were employed to classify the potential emission sources of trace metals in the Elbe River, Germany. The results showed that the overall concentration rank of the given metals was Zn (226.5 ± 526.5 μg·L^-1) > Ni (5.6 ± 4.7 μg·L^-1) > Cu (5.3 ± 5.8 μg·L^-1) > As (3.3 ± 3.7 μg·L^-1) > Pb (2.9 ± 5.2 μg·L^-1) > Cr (1.8 ± 2.5 μg·L^-1) > Cd (1.3 ± 3.1 μg·L^-1) in seven tributaries and the mainstream of the Elbe River, among which the tributary Triebisch had the highest risk quotient over 86. Random Forest outperformed other algorithms with the highest Kappa median values of 0.59 and the lowest Hamming-loss values of 0.22 in extraction of the majority voted class. Then, the source apportionment conducted by random forest suggested that wastewater disposal and metal industrial emissions were the source contributors in the tributary Triebisch (probabilities: 0.39, 0.3), upstream segment (0.45, 0.25), and downstream segment (0.32, 0.23) of the given river. Additional sources of mineral industry emissions were found in the upstream segment (0.21) and downstream segment (0.22). The data provided herein suggest that random forest would be an effective approach to identify pollutants in aquatic environments and could assist source-oriented adaptive management.

Prediction of Dichloroethene Concentration in the Groundwater of a Contaminated Site Using XGBoost and LSTM

Chlorinated aliphatic hydrocarbons (CAHs) are widely used in agriculture and industries and have become one of the most common groundwater contaminations. With the excellent performance of the deep learning method in predicting, LSTM and XGBoost were used to forecast dichloroethene (DCE) concentrations in a pesticide-contaminated site undergoing natural attenuation. The input variables included BTEX, vinyl chloride (VC), and five water quality indicators. In this study, the predictive performances of long short-term memory (LSTM) and extreme gradient boosting (XGBoost) were compared, and the influences of variables on models’ performances were evaluated. The results indicated XGBoost was more likely to capture DCE variation and was robust in high values, while the LSTM model presented better accuracy for all wells. The well with higher DCE concentrations would lower the model’s accuracy, and its influence was more evident in XGBoost than LSTM. The explanation of the SHapley Additive exPlanations (SHAP) value of each variable indicated high consistency with the rules of biodegradation in the real environment. LSTM and XGBoost could predict DCE concentrations through only using water quality variables, and LSTM performed better than XGBoost.

Predicting Dynamic Riverine Nitrogen Export in Unmonitored Watersheds: Leveraging Insights of AI from Data-Rich Regions

Terrestrial export of nitrogen is a critical Earth system process, but its global dynamics remain difficult to predict at a high spatiotemporal resolution. Here, we use deep learning (DL) to model daily riverine nitrogen export in response to hydrometeorological and anthropogenic drivers. Long short-term memory (LSTM) models for the daily concentration and flux of dissolved inorganic nitrogen (DIN) were built in a coastal watershed in southeastern China with a typical subtropical monsoon climate. The DL models exhibited excellent accuracy for both DIN concentration and flux, with Nash-Sutcliffe efficiency coefficients (NSEs) up to 0.67 and 0.92, respectively, a performance unlikely to be achieved by generic process-based models with comparable data quality. The flux model ensemble, without retraining, performed well (mean NSE = 0.32-0.84) in seven distinct watersheds in Asia, Europe, and North America, and retraining with multi-watershed data further improved the lowest NSE from 0.32 to 0.68. DL interpretation confirmed that interbasin consistency of riverine nitrogen export exists across different continents, which stems from the similarities in rainfall-runoff relationships. The multi-watershed flux model projects 0.60-12.4% increases in the nitrogen export to oceans from the studied watersheds under a 20% increase in fertilizer consumption, which rises to 6.7-20.1% with a 10% increase in runoff, indicating the synergistic effect of human activities and climate change. The DL-based method represents a successful case of explainable artificial intelligence in environmental science, providing a potential shortcut to a consistent understanding of the global daily-resolution dynamics of riverine nitrogen export under the currently limited data conditions.

Forecasting Water Temperature in Cascade Reservoir Operation-Influenced River with Machine Learning Models

Water temperature (WT) is a critical control for various physical and biochemical processes in riverine systems. Although the prediction of river water temperature has been the subject of extensive research, very few studies have examined the relative importance of elements affecting WT and how to accurately estimate WT under the effects of cascaded dams. In this study, a series of potential influencing variables, such as air temperature, dew temperature, river discharge, day of year, wind speed and precipitation, were used to forecast daily river water temperature downstream of cascaded dams. First, the permutation importance of the influencing variables was ranked in six different machine learning models, including decision tree (DT), random forest (RF), gradient boosting (GB), adaptive boosting (AB), support vector regression (SVR) and multilayer perceptron neural network (MLPNN) models. The results showed that day of year (DOY) plays the most important role in each model for the prediction of WT, followed by flow and temperature, which are two commonly important factors in unregulated rivers. Then, combinations of the three most important inputs were used to develop the most parsimonious model based on the six machine learning models, where their performance was compared according to statistical metrics. The results demonstrated that GB3 and RF3 gave the most accurate forecasts for the training dataset and the test dataset, respectively. Overall, the results showed that the machine learning model could be effectively applied to predict river water temperature under the regulation of cascaded dams.

Water quality forecasting based on data decomposition, fuzzy clustering and deep learning neural network

Water quality forecasting can provide useful information for public health protection and support water resources management. In order to forecast water quality more accurately, this paper proposes a novel hybrid model by combining data decomposition, fuzzy C-means clustering and bidirectional gated recurrent unit. Firstly, the original water quality data is decomposed into several subseries by empirical wavelet transform, and then, the decomposed subseries are recombined by fuzzy C-means clustering. Next, for each clustered series, bidirectional gated recurrent unit is applied to develop prediction model. Finally, the forecast result is obtained by the summation of the predictions for the subseries. The proposed forecast model is evaluated by the water quality data of Poyang Lake, China. Results show that the proposed forecast model provides highly accurate forecast result for all of the six water quality data: the average of MAPE of the forecast results for the six water quality datasets is 4.59% for 7 day ahead prediction. Furthermore, our model shows better forecast performance than the other models. Particularly, compared with the single BiGRU model, MAPE decreased by 32.86% in average. Results demonstrate that the proposed forecast model can be used effectively for water quality forecasting.

Deep learning model based on urban multi-source data for predicting heavy metals (Cu, Zn, Ni, Cr) in industrial sewer networks

The high concentrations of heavy metals in municipal industrial sewer networks will seriously impact the microorganisms of the activated sludge in the wastewater treatment plant (WWTP), thus deteriorating the effluent quality and destroying the stability of sewage treatment. Therefore, timely prediction and early warning of heavy metal concentrations in industrial sewer networks is crucial. However, due to the complex sources of heavy metals in industrial sewer networks, traditional physical modeling and linear methods cannot establish an accurate prediction model. Herein, we developed a Gated Recurrent Unit (GRU) neural network model based on a deep learning algorithm for predicting the concentrations of heavy metals in industrial sewer networks. To train the GRU model, we used low-cost and easy-to-obtain urban multi-source data, including socio-environmental indicator data, air environmental indicator data, water quantity indicator data, and easily measurable water quality indicator data. The model was applied to predict the concentrations of heavy metals (Cu, Zn, Ni, and Cr) in the sewer networks of an industrial area in southern China. The results are compared with the commonly used Artificial Neural Network (ANN) model. In this study, it was shown that the GRU had better prediction performance for Cu, Zn, Ni, and Cr concentrations, with the average R² significantly increased by 12.35%, 11.94%, 9.21%, and 8.13%, respectively, compared to ANN predictions. The sensitivity analysis based on Shapley (SHAP) values revealed that conductivity (σ), temperature (T), pH, and sewage flow (Flow) contributed significantly to the prediction results of the model. Furthermore, the three input variables including air pressure (AP), land area (A), and population (Pop.) were removed without affecting the prediction performance of the model, which maximized the modeling efficiency and reduced the operational cost. This study provides an economical and feasible technical method for early warning of abnormal heavy metal concentrations in urban industrial sewer networks.

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.