Large scale prediction of groundwater nitrate concentrations from spatial data using machine learning

A multidisciplinary approach using hydrogeochemistry, δ15NNO3 isotopes, land use, and statistical tools in evaluating nitrate pollution sources and biochemical processes in Costa Rican volcanic aquifers

Nitrate pollution threatens the Barva and Colima multi-aquifer system, the primary drinking water source in the Greater Metropolitan Area of Costa Rica. In addressing nitrate contamination dynamics, this study proposes an integrated approach by combining multivariate statistical analyses, hydrochemical parameters, sewage discharge, and regional land-use and land-cover patterns to assess the extent and degree of contamination, dominant biogeochemical processes, and refine the interpretation of nitrate sources previously derived solely from δ15NNO3 information. Over seven years (2015-2022), 714 groundwater samples from 43 sites were analyzed for nitrate and major ions, including two sampling campaigns for dissolved organic and inorganic carbon, nitrite, ammonium, FeTotal, MnTotal, and δ15NNO3 analyses. The findings presented elevated nitrate concentrations in urban and agricultural/urban areas, surpassing the Maximum Concentration Levels on several occasions, and oxidizing conditions favoring mineralization and nitrification processes in unconfined Barva and locally confined Upper Colima/Lower Colima aquifers. Similar nitrate contents and spatial patterns in agricultural and urban zones in the shallow Barva aquifer suggest comparable contributions from nitrogen fertilizers and urban wastewaters despite the gradual increase in urban land cover and the reduction of agricultural areas. Isotopic analyses and dissolved organic carbon (DOC) indicate a shift in nitrate sources from agricultural to urban areas in both Barva and Colima aquifers. Principal Component and Hierarchical Cluster Analyses link land use, nitrate sources, and water quality. Three distinct sample clusters aligned with forest/grassland, agricultural/urban, and urban land use, emphasizing the impact of anthropogenic activities on groundwater quality, even in the deeper Colima aquifers. The study challenges nitrate isotope mixing models, enhancing accuracy in identifying pollution sources and assessing the spatial extent of contamination by incorporating DOC and other hydrochemical parameters. Similar outcomes, with and without the use of nitrate isotopes, reinforce the usefulness of the integrated approach, providing a practical and cost-effective alternative.

Advancing groundwater quality predictions: Machine learning challenges and solutions

Machine learning (ML) is revolutionizing groundwater quality research by enhancing predictive accuracy and management strategies for contamination. This comprehensive review explores the evolution of ML technologies and their integration into environmental science, assessing 230 papers to understand the advancements and challenges in groundwater quality research. It reveals that a substantial portion of the research neglects critical preprocessing steps, crucial for model accuracy, with 83 % of the studies overlooking this phase. Furthermore, while model optimization is more commonly addressed, being implemented in 65 % of the papers, there is a noticeable gap in model interpretability, with only 15 % of the research providing explanations for model outcomes. Comparative evaluation of ML algorithms and careful selection of evaluation metrics are deemed essential for determining model fitness and reliability. The review underscores the need for interdisciplinary collaboration, methodological rigor, and continuous innovation to advance ML in groundwater management. By addressing these challenges and implementing solutions, the full potential of ML can be harnessed to tackle complex environmental issues and ensure sustainable groundwater management. This comprehensive and critical review paper can serve as a guiding framework to establish minimum standards for developing ML in groundwater quality studies.

Identifying the spatial pattern and driving factors of nitrate in groundwater using a novel framework of interpretable stacking ensemble learning

Groundwater nitrate contamination poses a potential threat to human health and environmental safety globally. This study proposes an interpretable stacking ensemble learning (SEL) framework for enhancing and interpreting groundwater nitrate spatial predictions by integrating the two-level heterogeneous SEL model and SHapley Additive exPlanations (SHAP). In the SEL model, five commonly used machine learning models were utilized as base models (gradient boosting decision tree, extreme gradient boosting, random forest, extremely randomized trees, and k-nearest neighbor), whose outputs were taken as input data for the meta-model. When applied to the agricultural intensive area, the Eden Valley in the UK, the SEL model outperformed the individual models in predictive performance and generalization ability. It reveals a mean groundwater nitrate level of 2.22 mg/L-N, with 2.46% of sandstone aquifers exceeding the drinking standard of 11.3 mg/L-N. Alarmingly, 8.74% of areas with high groundwater nitrate remain outside the designated nitrate vulnerable zones. Moreover, SHAP identified that transmissivity, baseflow index, hydraulic conductivity, the percentage of arable land, and the C:N ratio in the soil were the top five key driving factors of groundwater nitrate. With nitrate threatening groundwater globally, this study presents a high-accuracy, interpretable, and flexible modeling framework that enhances our understanding of the mechanisms behind groundwater nitrate contamination. It implies that the interpretable SEL framework has great promise for providing valuable evidence for environmental management, water resource protection, and sustainable development, particularly in the data-scarce area.

Modeling groundwater redox conditions at national scale through integration of sediment color and water chemistry in a machine learning framework

Redox conditions play a crucial role in determining the fate of many contaminants in groundwater, impacting ecosystem services vital for both the aquatic environment and human water supply. Geospatial machine learning has previously successfully modelled large-scale redox conditions. This study is the first to consolidate the complementary information provided by sediment color and water chemistry to enhance our understanding of redox conditions in Denmark. In the first step, the depth to the first redox interface is modelled using sediment color from 27,042 boreholes. In the second step, the depth of the first redox interface is compared against water chemistry data at 22,198 wells to classify redox complexity. The absence of nitrate containing water below the first redox interface is referred to as continuous redox conditions. In contrast, discontinuous redox conditions are identified by the presence of nitrate below the first redox interface. Both models are built using 20 covariate maps, encompassing diverse hydrologically relevant information. The first redox interface is modelled with a mean error of 0.0 m and a root-mean-squared error of 8.0 m. The redox complexity model attains an accuracy of 69.8 %. Results indicate a mean depth to the first redox interface of 8.6 m and a standard deviation of 6.5 m. 60 % of Denmark is classified as discontinuous, indicating complex redox conditions, predominantly collocated in clay rich glacial landscapes. Both maps, i.e., first redox interface and redox complexity are largely driven by the water table and hydrogeology. The developed maps contribute to our understanding of subsurface redox processes, supporting national-scale land-use and water management.

A machine learning framework for spatio-temporal vulnerability mapping of groundwaters to nitrate in a data scarce region in Lenjanat Plain, Iran

The temporal aspect of groundwater vulnerability to contaminants such as nitrate is often overlooked, assuming vulnerability has a static nature. This study bridges this gap by employing machine learning with Detecting Breakpoints and Estimating Segments in Trend (DBEST) algorithm to reveal the underlying relationship between nitrate, water table, vegetation cover, and precipitation time series, that are related to agricultural activities and groundwater demand in a semi-arid region. The contamination probability of Lenjanat Plain has been mapped by comparing random forest (RF), support vector machine (SVM), and K-nearest-neighbors (KNN) models, fed with 32 input variables (dem-derived factors, physiography, distance and density maps, time series data). Also, imbalanced learning and feature selection techniques were investigated as supplementary methods, adding up to four scenarios. Results showed that the RF model, integrated with forward sequential feature selection (SFS) and SMOTE-Tomek resampling method, outperformed the other models (F1-score: 0.94, MCC: 0.83). The SFS techniques outperformed other feature selection methods in enhancing the accuracy of the models with the cost of computational expenses, and the cost-sensitive function proved more efficient in tackling imbalanced data issues than the other investigated methods. The DBEST method identified significant breakpoints within each time series dataset, revealing a clear association between agricultural practices along the Zayandehrood River and substantial nitrate contamination within the Lenjanat region. Additionally, the groundwater vulnerability maps created using the candid RF model and an ensemble of the best RF, SVM, and KNN models predicted mid to high levels of vulnerability in the central parts and the downhills in the southwest.

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.

Assessing nitrate groundwater hotspots in Europe reveals an inadequate designation of Nitrate Vulnerable Zones

Monitoring networks show that the European Union Nitrates Directive (ND) has had mixed success in reducing nitrate concentrations in groundwater. By combining machine learning and monitored nitrate concentrations (1992-2019), we estimate the total area of nitrate hotspots in Europe to be 401,000 km2, with 47% occurring outside of Nitrate Vulnerable Zones (NVZs). We also found contrasting increasing or decreasing trends, varying per country and time periods. We estimate that only 5% of the 122,000 km2 of hotspots in 2019 will meet nitrate quality standards by 2040 and that these may be offset by the appearance of new hotspots. Our results reveal that the effectiveness of the ND is limited by both time-lags between the implementation of good practices and pollution reduction and an inadequate designation of NVZs. Substantial improvements in the designation and regulation of NVZs are necessary, as well as in the quality of monitoring stations in terms of spatial density and information available concerning sampling depth, if the objectives of EU legislation to protect groundwater are to be achieved.

Analysing agricultural plant protection product concentrations in groundwater in Germany: Nationwide database with site and compound insights

Pesticides from agricultural practices are among the most pressing reasons why groundwater sources do not reach the good chemical status standards as required by the European Water Framework directive. Complementary to previous federal pesticide reports, we analysed groundwater-monitoring data from 13 German Länder assembled in a database consisting of 26.192 groundwater measuring sites sampled between 1973 and 2021 of in total 521 parent compounds and metabolites. This study focuses on agricultural plant protection products. The monitored substance spectrum and site density developed over time and differs between Länder. More than 95 % of all samples lie below the respective (multiple) limits of quantification (LOQ). We thus report the frequency of exceedance above concentration thresholds, which allows to compare measurements temporally and spatially. Pesticide detections were found in all aquifer types, land uses and well screen depths. Most detections of higher concentrations were found in agricultural areas, at shallow screen depth in porous aquifers. Karst aquifers showed also a higher percentage of samples in higher concentration classes. Metabolites with high mobility and persistence were found in higher concentration ranges. Herbicides and metabolites thereof dominate the top 20 of pesticides that most frequently exceed 0.1 μg L-1. The ranking for 2010-2019 includes both authorised and banned compounds and their occurrence is discussed in the context of their mobility, persistence and underlying monitoring density. Yearly exceedance frequencies above 0.05, 0.1 μg L-1 and higher thresholds of metazachlor and its esa-metabolite, and national sales data of the parent compound did not show a temporal correlation in subsequent years. This study stresses the need for the harmonisation of heterogeneous pesticide data. Further, a characterisation of the groundwater data used to analyse pesticide occurrence in selected concentration ranges for relevant site factors and compound properties and provides a pesticide ranking based on exceedance frequencies is provided.

Development of a basin-scale total nitrogen prediction model by integrating clustering and regression methods

Nutrient runoff into rivers caused by human activity has led to global eutrophication issues. The Nakdong River in South Korea is currently facing significant challenges related to eutrophication and harmful algal blooms, underscoring the critical importance of managing total nitrogen (T-N) levels. However, traditional methods of indoor analysis, which depend on sampling, are labor-intensive and face limitations in collecting high-frequency data. Despite advancements in sensor allowing for the measurement of various parameters, sensors still cannot directly measure T-N, necessitating surrogate regression methods. Therefore, we conducted T-N predictions using a water quality dataset collected from 2018 to 2022 at 157 observatories within the Nakdong River basin. To account for the water quality characteristics of each location, we employed a clustering technique to divide the basin and compared a Gaussian mixture model with K-means clustering. Moreover, optimal regressor for each cluster was selected by comparing multiple linear regression (MLR), random forest, and XGBoost. The results showed that forming four clusters via K-means clustering was the most suitable approach and MLR was reasonably accurate for all clusters. Subsequently, recursive feature elimination cross-validation was used to identify suitable parameters for T-N prediction, thus leading to the construction of high-accuracy T-N prediction models. Clustering was useful not only for improving the regressors but also for spatially analyzing the water quality characteristics of the Nakdong River. The MLR model can reveal causal relationships and thus is useful for decision-making. The results of this study revealed that the combination of a simple linear regression model and clustering method can be applied to a wide watershed. The clustering-based regression model showed potential for accurately predicting T-N at the basin level and is expected to contribute to nationwide water quality management through future applications in various fields.

Mapping specific groundwater nitrate concentrations from spatial data using machine learning: A case study of chongqing, China

Groundwater resources is not only important essential water resources but also imperative connectors within the intricate framework of the ecological environment. High nitrate concentrations in groundwater can exerting adverse impacts on human health. It is imperative to accurately delineate the distribution characteristics of groundwater nitrate concentrations. Four different machine learning models (Gradient Boosting Regression (GB), Random Forest Regression (RF), Extreme Gradient Boosting Regression (XG) and Adaptive Boosting Regression (AD)) which combine spatial environmental data and different radius contributing area was developed to predict the distribution of nitrate concentration in groundwater. The models use 595 groundwater samples and included topography, remote sensing, hydrogeological and hydrological, climate, nitrate input, and socio-economic predictor. Gradient Boosting Regression model outperforms the other models (R2 = 0.627, MAE = 0.529, RMSE = 0.705, PICP = 0.924 for test dataset) under 500 m radius contributing area. A high-resolution (1 km) groundwater nitrate concentration distribution map reveal in the majority of the study area, groundwater nitrate concentrations are below 1 mg/L and high nitrate concentration (>10 mg/L) proportion in southeast, northeast and central main urban area karst valley regions is 1.89%, 0.91%, and 0.38% respectively. In study area, hydrogeological conditions, soil parameters, nitrogen input factors, and percentage of arable land are among the most influential explanatory factors. This work, serving as the inaugural application of utilizing effective spatial methods for predicting groundwater nitrate concentrations in Chongqing city, furnish decision-making support for the prevention and control of groundwater pollution, particularly in areas primarily dependent on groundwater for water supply and holds profound significance as a milestone achievement.

Revealing the drivers and genesis of NO3-N pollution classification in shallow groundwater of the Shaying River Basin by explainable machine learning and pathway analysis method

Nitrate (NO3-N), as one of the ubiquitous contaminants in groundwater worldwide, has posed a serious threat to public health and the ecological environment. Despite extensive research on its genesis, little is known about the differences in the genesis of NO3-N pollution across different concentrations. Herein, a study of NO3-N pollution concentration classification was conducted using the Shaying River Basin as a typical area, followed by examining the genesis differences across different pollution classifications. Results demonstrated that three classifications (0-9.98 mg/L, 10.14-27.44 mg/L, and 28.34-136.30 mg/L) were effectively identified for NO3-N pollution using Jenks natural breaks method. Random forest exhibited superior performance in describing NO3-N pollution and was thereby affirmed as the optimal explanatory method. With this method coupling SEMs, the genesis of different NO3-N pollution classifications was proven to be significantly different. Specifically, strongly reducing conditions represented by Mn2+, Eh, and NO2-N played a dominant role in causing residual NO3-N at low levels. Manure and sewage (represented by Cl-) leaching into groundwater through precipitation is mainly responsible for NO3-N in the 10-30 mg/L classification, with a cumulative contribution rate exceeding 80 %. NO3-N concentrations >30 mg/L are primarily caused by the anthropogenic loads stemming from manure, sewage, and agricultural fertilization (represented by Cl- and K+) infiltrating under precipitation in vulnerable hydrogeological conditions. Pathway analysis based on standardized effect and significance further confirmed the rationality and reliability of the above results. The findings will provide more accurate information for policymakers in groundwater resource management to implement effective strategies to mitigate NO3-N pollution.

Spatial distribution pattern and health risk of groundwater contamination by cadmium, manganese, lead and nitrate in groundwater of an arid area

Combining the results of base models to create a meta-model is one of the ensemble approaches known as stacking. In this study, stacking of five base learners, including eXtreme gradient boosting, random forest, feed-forward neural networks, generalized linear models with Lasso or Elastic Net regularization, and support vector machines, was used to study the spatial variation of Mn, Cd, Pb, and nitrate in Qom-Kahak Aquifers, Iran. The stacking strategy proved to be an effective substitute predictor for existing machine learning approaches due to its high accuracy and stability when compared to individual learners. Contrarily, there was not any best-performing base model for all of the involved parameters. For instance, in the case of cadmium, random forest produced the best results, with adjusted R2 and RMSE of 0.108 and 0.014, as opposed to 0.337 and 0.013 obtained by the stacking method. The Mn and Cd showed a tight link with phosphate by the redundancy analysis (RDA). This demonstrates the effect of phosphate fertilizers on agricultural operations. In order to analyze the causes of groundwater pollution, spatial methodologies can be used with multivariate analytic techniques, such as RDA, to help uncover hidden sources of contamination that would otherwise go undetected. Lead has a larger health risk than nitrate, according to the probabilistic health risk assessment, which found that 34.4% and 6.3% of the simulated values for children and adults, respectively, were higher than HQ = 1. Furthermore, cadmium exposure risk affected 84% of children and 47% of adults in the research area.

Nitrate prediction in groundwater of data scarce regions: The futuristic fresh-water management outlook

Nitrate contamination in groundwater poses a significant threat to water quality and public health, especially in regions with limited data availability. This study addresses this challenge by employing machine learning (ML) techniques to predict nitrate (NO3--N) concentrations in Mexico's groundwater. Four ML algorithms-Extreme Gradient Boosting (XGB), Boosted Regression Trees (BRT), Random Forest (RF), and Support Vector Machines (SVM)-were executed to model NO3--N concentrations across the country. Despite data limitations, the ML models achieved robust predictive performances. XGB and BRT algorithms demonstrated superior accuracy (0.80 and 0.78, respectively). Notably, this was achieved using ∼10 times less information than previous large-scale assessments. The novelty lies in the first-ever implementation of the 'Support Points-based Split Approach' during data pre-processing. The models considered initially 68 covariates and identified 13-19 significant predictors of NO3--N concentration spanning from climate, geomorphology, soil, hydrogeology, and human factors. Rainfall, elevation, and slope emerged as key predictors. A validation incorporated nationwide waste disposal sites, yielding an encouraging correlation. Spatial risk mapping unveiled significant pollution hotspots across Mexico. Regions with elevated NO3--N concentrations (>10 mg/L) were identified, particularly in the north-central and northeast parts of the country, associated with agricultural and industrial activities. Approximately 21 million people, accounting for 10 % of Mexico's population, are potentially exposed to elevated NO3--N levels in groundwater. Moreover, the NO3--N hotspots align with reported NO3--N health implications such as gastric and colorectal cancer. This study not only demonstrates the potential of ML in data-scarce regions but also offers actionable insights for policy and management strategies. Our research underscores the urgency of implementing sustainable agricultural practices and comprehensive domestic waste management measures to mitigate NO3--N contamination. Moreover, it advocates for the establishment of effective policies based on real-time monitoring and collaboration among stakeholders.

Nitrate leaching potential from arable land in Germany: Identifying most relevant factors

Diffuse nitrogen losses from agriculture in Germany continue to cause regionally increased nitrate concentrations in groundwater. Groundwater quality monitoring cannot be a timely indicator of the effects of mitigation measures being applied in agriculture, due to frequently long transport routes and high residence times of the leachate. Instead, nitrate leaching potential is often determined at field and farm scale by monitoring soil mineral nitrogen contents at 0-90 cm depth in autumn (SMNa), i.e. before the start of the annual leachate period. In this study, we developed an understanding of the controls on the soil mineral nitrogen content at the start of winter. In an on-farm approach, extensive data was collected from 48 farms in five nitrate-sensitive regions in Germany from 2017 to 2020. From this data set, 25 management and site factors were evaluated with regard to their significance for SMNa by means of a random forest model. With the random forest regression, we identified the role of the factors on SMNa with an acceptable model accuracy with R2 = 0.56. The results show that the cultivated crop is the most important factor influencing SMNa. Potatoes, oilseed rape and maize produced the highest SMNas, whereas SMNas were lowest after spring barley, sugar beet and winter barley. Among site factors, soil type and texture as well as precipitation in October were most decisive. The effects of N fertilisation parameters such as rate and timing were masked by these site factors. The results show that the reduction of nitrogen-intensive crops in crop sequences can be a promising measure for the reduction of nitrate loads. On the other hand, our analysis makes clear that soil-related factors controlling nitrogen release and risk of leaching, as well as weather, can significantly mask the effect of cultivation.

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.

Improving groundwater nitrate concentration prediction using local ensemble of machine learning models

Groundwater is one of the most important water resources around the world, which is increasingly exposed to contamination. As nitrate is a common pollutant of groundwater and has negative effects on human health, predicting its concentration is of particular importance. Ensemble machine learning (ML) algorithms have been widely employed for nitrate concentration prediction in groundwater. However, existing ensemble models often overlook spatial variation by combining ML models with conventional methods like averaging. The objective of this study is to enhance the spatial accuracy of groundwater nitrate concentration prediction by integrating the outputs of ML models using a local approach that accounts for spatial variation. Initially, three widely used ML models including random forest regression (RFR), k-nearest neighbor (KNN), and support vector regression (SVR) were employed to predict groundwater nitrate concentration of Qom aquifer in Iran. Subsequently, the output of these models were integrated using geographically weighted regression (GWR) as a local model. The findings demonstrated that the ensemble of ML models using GWR resulted in the highest performance (R2 = 0.75 and RMSE = 9.38 mg/l) compared to an ensemble model using averaging (R2 = 0.68 and RMSE = 10.56 mg/l), as well as individual models such as RFR (R2 = 0.70 and RMSE = 10.16 mg/l), SVR (R2 = 0.59 and RMSE = 11.95 mg/l), and KNN (R2 = 0.57 and RMSE = 12.19 mg/l). The resulting prediction map revealed that groundwater nitrate contamination is predominantly concentrated in urban areas located in the northwestern regions of the study area. The insights gained from this study have practical implications for managers, assisting them in preventing nitrate pollution in groundwater and formulating strategies to improve water quality.

Prediction of groundwater quality index to assess suitability for drinking purpose using averaged neural network and geospatial analysis

Groundwater quality management is pivotal for ensuring public health and ecological resilience. However, the conventional water quality indices often face challenges related to parameter selection, geographic coverage, and scalability. The integration of machine learning and spatial analysis represents a promising methodological shift, allowing for high accuracy and adaptive management strategies. The Safe Groundwater Project in Unsupplied Areas (2017-2020) employed a comprehensive Groundwater Quality Index (GQI) to evaluate potable groundwater quality across South Korea, utilizing a large dataset comprising 28 water quality parameters and 3552 wells. This study revealed that over 50 % of the evaluated wells (Total 8326 wells) were inappropriate as sources of drinking water, indicating a pressing need for policy revision. The averaged neural network model achieved a high predictive accuracy of approximately 95 % for GQI grades, outperforming other classification models. The introduction of 2D spatial analysis in conjunction with machine learning algorithms notably increased the predictive accuracy for unevenly distributed groundwater samples. Moreover, this combined approach enabled the intuitive visualization of groundwater vulnerability across various regions, which can inform targeted interventions for effective resource allocation and management. This research represents a methodologically robust, interdisciplinary approach that holds significant implications for a framework for future groundwater quality management and vulnerability assessment.

Application of machine learning in groundwater quality modeling - A comprehensive review

Groundwater is a crucial resource across agricultural, civil, and industrial sectors. The prediction of groundwater pollution due to various chemical components is vital for planning, policymaking, and management of groundwater resources. In the last two decades, the application of machine learning (ML) techniques for groundwater quality (GWQ) modeling has grown exponentially. This review assesses all supervised, semi-supervised, unsupervised, and ensemble ML models implemented to predict any groundwater quality parameter, making this the most extensive modern review on this topic. Neural networks are the most used ML model in GWQ modeling. Their usage has declined in recent years, giving rise to more accurate or advanced techniques such as deep learning or unsupervised algorithms. Iran and the United States lead the world in areas modeled, with a wealth of historical data available. Nitrate has been modeled most exhaustively, targeted by nearly half of all studies. Advancements in future work will be made with further implementation of deep learning and explainable artificial intelligence or other cutting-edge techniques, application of these techniques for sparsely studied variables, the modeling of new or unique study areas, and the implementation of ML techniques for groundwater quality management.

Prediction of phosphorus concentrations in shallow groundwater in intensive agricultural regions based on machine learning

Excessive accumulation of phosphorus in soil profiles has become the main source of phosphorus in groundwater due to the application of phosphorus fertilizers in intensive agricultural regions (IARs). Elevated phosphorus concentrations in groundwater have become a global phenomenon, which places enormous pressure on the safe use of water resources and the safety of the aquatic environment. Currently, the prediction of pollutant concentrations in groundwater mainly focuses on nitrate nitrogen, while research on phosphorus prediction is limited. Taking the IARs approximately 8 plateau lakes in the Yunnan-Guizhou Plateau as an example, 570 shallow groundwater samples and 28 predictor variables were collected and measured, and a machine learning approach was used to predict phosphorus concentrations in groundwater. The performance of three machine learning algorithms and different sets of variables for predicting phosphorus concentrations in shallow groundwater was evaluated. The results showed that after all variables were introduced into the model, the R², RMSE and MAE of support vector machine (SVM), random forest (RF) and neural network (NN) were 0.52-0.60, 0.101-0.108 and 0.074-0.081, respectively. Among them, the SVM model had the best prediction effect. The clay content and water-soluble phosphorus in soil and soluble organic carbon in groundwater had a high contribution to the prediction accuracy of the model. The prediction accuracy of the model with reduced number of variables showed that when the number of variables was equal to 6, the RF model had R², RMSE and MAE values of 0.53, 0.108 and 0.074, respectively, and the number of variables increased again; there were small changes in R², RMSE and MAE. Compared with the SVM and NN models, the RF model can achieve higher accuracy by inputting fewer variables.

Intelligent soft computational models integrated for the prediction of potentially toxic elements and groundwater quality indicators: a case study

Reports have shown that potentially toxic elements (PTEs) in air, water, and soil systems expose humans to carcinogenic and non-carcinogenic health risks. In southeastern Nigeria, works that have used data-driven algorithms in predicting PTEs in groundwater are scarce. In addition, only a few works have simulated water quality indices using machine learning modelling methods in the region. Therefore, in this study, physicochemical analyses were carried out on groundwater samples in southeastern Nigeria. The laboratory results were used to compute two water quality indices: pollution index of groundwater (PIG) and the water pollution index (WPI), to ascertain groundwater quality. In addition, the physicochemical parameters served as input variables for multiple linear regression (MLR) and artificial neural network (ANN) modelling and prediction of Cr, Fe, Ni, NO3−, Pb, Zn, WPI, and PIG. The results of WPI and PIG computation showed that about 30–35% of the groundwater samples were unsuitable for human consumption, whereas 65–70% of the samples were deemed suitable. The insights from the PIG and WPI model also revealed that lead (Pb) was the most influential PTE that degraded the quality of groundwater resources in the research area. The findings of the MLR and ANN models indicated strong positive prediction accuracies (R 2 = 0.856–1.000) with low modeling errors. The predictive MLR and ANN models of the PIG and WPI generally outperformed those of the PTEs. The models produced in this study predicted the PTEs better compared to previous studies. Thus, this work provides insights into effective water sustainability, management, and protection.

Predicting Potential Climate Change Impacts on Groundwater Nitrate Pollution and Risk in an Intensely Cultivated Area of South Asia

One of the potential impacts of climate change is enhanced groundwater contamination by geogenic and anthropogenic contaminants. Such impacts should be most evident in areas with high land-use change footprint. Here, we provide a novel documentation of the impact on groundwater nitrate (GWNO3 ) pollution with and without climate change in one of the most intensely groundwater-irrigated areas of South Asia (northwest India) as a consequence of changes in land use and agricultural practices at present and predicted future times. We assessed the probabilistic risk of GWNO3 pollution considering climate changes under two representative concentration pathways (RCPs), i.e., RCP 4.5 and 8.5 for 2030 and 2040, using a machine learning (Random Forest) framework. We also evaluated variations in GWNO3 distribution against a no climate change (NCC) scenario considering 2020 status quo climate conditions. The climate change projections showed that the annual temperatures would rise under both RCPs. The precipitation is predicted to rise by 5% under RCP 8.5 by 2040, while it would decline under RCP 4.5. The predicted scenarios indicate that the areas at high risk of GWNO3 pollution will increase to 49 and 50% in 2030 and 66 and 65% in 2040 under RCP 4.5 and 8.5, respectively. These predictions are higher compared to the NCC condition (43% in 2030 and 60% in 2040). However, the areas at high risk can decrease significantly by 2040 with restricted fertilizer usage, especially under the RCP 8.5 scenario. The risk maps identified the central, south, and southeastern parts of the study area to be at persistent high risk of GWNO3 pollution. The outcomes show that the climate factors may impose a significant influence on the GWNO3 pollution, and if fertilizer inputs and land uses are not managed properly, future climate change scenarios can critically impact the groundwater quality in highly agrarian areas.

Spatial patterns and driving factor analysis of recommended nitrogen application rate for the trade-off between economy and environment for maize in China

Appropriate nitrogen (N) application increases crop yield, while its unreasonable application results in environmental problem. Determining the appropriate N application rate is the key to sustainable development. Here, the denitrification-decomposition (DNDC) model was used to analyze the effects of N fertilizer on maize yields, economic benefits, nitrate leaching, and nitrous oxide emissions in China. The N application rate for the trade-off between economy and environment at the county scale was further determined. The geodetector model was used to identify the main driving factors and their interactions of the recommended N rate in each agricultural zone. The results showed that the recommended N rate was generally high in the northwest but low in the south, consistent with the spatial patterns of yield potential. However, clay soils with clay ratios greater than 34% in southern China and sandy soils with bulk densities greater than 1.5 g cm^-3 on the Huanghuaihai Plain experienced high N levels and low yields, and thus soils need to be improved. Potential grain yield was the main driving factor in most zones, yet its effects gradually weakened from north to south. The influence of soil characteristics increased from north to south. It was found that the current average N application rate of farmers in China was 249 kg N/ha, and 86.55% of counties had excessive N applications. Compared to the regional optimal N rate at a regional scale, a differentiated N application strategy at the county scale determined in this study increased maize yield and economic benefit by 10.51% and 10.85%, respectively, and reduced N₂O emissions and NO₃^- leaching by 28.72% and 33.60%, respectively. The current research provides a scientific basis for China to formulate a win-win N management strategy for economy and environment and provides a method reference for other countries.

Multiorder hydrologic Position for Europe - a Set of Features for Machine Learning and Analysis in Hydrology

The presented dataset EU-MOHP v013.1.1 provides multiscale information on the hydrologic position (MOHP) of a geographic point within its respective river network and catchment as gridded maps. More precisely, it comprises the three measures “divide to stream distance” (DSD) as sum of the distances to the nearest stream and catchment divide, “lateral position” (LP) as a relative measure of the position between the nearest stream and divide and “stream distance” (SD) as the distance to the nearest stream. These three measures are calculated for nine hydrologic orders to reflect different spatial scales from local to continental. Its spatial extent covers major parts of the European Economic Area (EEA39) which also largely coincides with physiographical Europe. Although there are multiple potential use cases, this dataset serves predominantly as valuable static environmental descriptor or predictor variable for hydrogeological and hydrological modelling such as mapping or forecasting tasks using machine learning. The generation of this dataset uses free open source software only and therefore can be transferred to other regions or input datasets.

Measurement(s)	divide to stream distance • lateral position • stream distance
Technology Type(s)	remote sensing
Sample Characteristic - Environment	drainage basin • groundwatershed • catchment
Sample Characteristic - Location	Europe

Definition of hot-spots to reduce the nitrogen losses from agricultural land to groundwater in Slovakia

Agriculture is a significant contributor to nitrate pollution of groundwater which in many cases serves as a source of drinking water. Therefore, targeted reduction of nitrogen leaching losses is fully justified to address this issue. The aim of the study was to define the areas of utilized agricultural land (UAL) in Slovakia, where a nitrogen surplus needs to be reduced. Using the average values of leachable nitrogen in the period 2015-2018 and the long-term amount of percolated water, the nitrate concentration in leachate was calculated. To ensure that agricultural activities will contribute to the gradual reduction of nitrate concentration in groundwater, the nitrate concentration in leachate of 40 mg L-1 was chosen as the target value. This concentration was exceeded at 11.7% of the UAL area. The average indicative amount of nitrogen in industrial fertilizers that needs to be reduced to achieve a stricter nitrate concentration in the leachate in these hot-spots is 16 kg ha-1 with the proviso that in two districts this value exceeds 30 kg ha-1.

Application and Research of Computer Intelligent Technology in Modern Agricultural Machinery Equipment

Every country, including China, is deeply concerned and interested in the topic of agricultural machinery automation. The world's population is growing at an astronomical rate, and as a result, the need of food is also growing at an astronomical rate. Farmers are forced to apply more toxic pesticides since traditional methods are not up to the task of meeting the rising demand. This has a major impact on agricultural practices, and in the long run, the land becomes barren and unproductive. Intelligent technologies such as Internet of Things, wireless communication, and machine learning can help with crop disease and pesticide storage management, as well as water management and irrigation. In this paper, we design and analyze an intelligent system that automatically predicts the agricultural land features for irrigation purpose. Initially, the dataset is collected and preprocessed using normalization. The features are extracted using principal component analysis (PCA). For automatic prediction by the equipment, we propose heterogeneous fuzzy-based artificial neural network (HF-ANN) with genetic quantum spider monkey optimization (GQ-SMO) algorithm. Analyses and comparisons are made between the proposed approach and current methodologies. The findings indicate the effectiveness of the proposed system.

Factors controlling groundwater radioactivity in arid environments: An automated machine learning approach

Groundwater resources in the Kingdom of Saudi Arabia (KSA) have high levels of natural radioactivity. Within the northwestern KSA, gross alpha (α) and gross beta (β) levels exceed national and international drinking-water limits. In this study, we developed and used an automated machine learning (AML) approach to quantify relationships between gross α and gross β activities and different geological, hydrogeological, and geochemical conditions. Two AML model groups (group I for gross α; group II for gross β) were constructed, using water samples collected from 360 irrigation and water supply wells, to define a robust model that explains the spatial variability in gross α and gross β activities, as well as variables that control the gross activities. Each group contained four model families: deep neural network (DNN), gradient boosting machine (GBM), generalized linear model (GLM), and distributed random forest (DRF). Model inputs include chemical compositions as well as geological and hydrogeological conditions. Three performance metrics were used to evaluate the models during training and testing: normalized root mean square error (NRMSE), Pearson's correlation coefficient (r), and Nash-Sutcliff efficiency (NSE) coefficient. Results indicate that (1) the GBM model outperformed (training: NRMSE: 0.37 ± 0.10; r: 0.92 ± 0.05; NSE: 0.85 ± 0.09; testing: NRMSE: 0.71 ± 0.08; r: 0.72 ± 0.08; NSE: 0.49 ± 0.12) the DNN, DRF, and GLM models when modelling gross α activities; (2) gross α activities are controlled by pH, stream density, nitrate, manganese, and vegetation index; (3) the DRF model outperformed (training: NRMSE: 0.41 ± 0.05; r: 0.92 ± 0.02; NSE: 0.83 ± 0.04; testing: NRMSE: 0.67 ± 0.09; r: 0.77 ± 0.07; NSE: 0.54 ± 0.12) the GBM, DNN, and GLM models when modelling gross β activities; (4) input variables that affect the gross β actives are pH, temperature, stream density, lithology, and nitrate; and (5) no single model could be used to model both gross α and gross β activities-instead, a combination of AML models should be used. Our computationally efficient approach provides a framework and insights for using AML techniques in water quality investigations and promotes more and improved use of different geological, hydrogeological, and geochemical datasets by the scientific community and decision makers to develop guidelines for mitigation.

Hybrid Machine Learning Models for Soil Saturated Conductivity Prediction

The hydraulic conductivity of saturated soil is a crucial parameter in the study of any engineering problem concerning groundwater. Hydraulic conductivity mainly depends on particle size distribution, soil compaction, and properties that influence aggregation and water retention. Generally, finding simple and accurate analytical equations between the hydraulic conductivity of soil and the characteristics on which it depends is a very hard task. Machine learning algorithms can provide excellent tools for tackling highly nonlinear regression problems. Additionally, hybrid models resulting from the combination of multiple machine learning algorithms can further improve the accuracy of predictions. Five different models were built to predict saturated hydraulic conductivity using a dataset extracted from the Soil Water Infiltration Global database. The models were based on different predictors. Seven variants of each model were compared, replacing the implemented algorithm. Three variants were based on individual models, while four variants were based on hybrid models. The employed individual machine learning algorithms were Multilayer Perceptron, Random Forest, and Support Vector Regression. The model based on the largest number of predictors led to the most accurate predictions. In addition, across all models, hybrid variants based on all three algorithms and hybridized variants of Random Forest and Support Vector Regression proved to be the most accurate (R2 values up to 0.829). However, all variants showed a tendency to overestimate conductivity in soils where it is very low.

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.

Machine Learning Algorithms for Modeling and Mapping of Groundwater Pollution Risk: A Study to Reach Water Security and Sustainable Development (Sdg) Goals in a Mediterranean Aquifer System

Groundwater pollution poses a severe threat and issue to the environment and humanity overall. That is why mitigative strategies are urgently needed. Today, studies mapping groundwater risk pollution assessment are being developed. In this study, five new hybrid/ensemble machine learning (ML) models are developed, named DRASTIC-Random Forest (RF), DRASTIC-Support Vector Machine (SVM), DRASTIC-Multilayer Perceptron (MLP), DRASTIC-RF-SVM, and DRASTIC-RF-MLP, for groundwater pollution assessment in the Saiss basin, in Morocco. The performances of these models are evaluated using the Receiver Operating Characteristic curve (ROC curve), precision, and accuracy. Based on the results of the ROC curve method, it is indicated that the use of hybrid/ensemble machine learning (ML) models improves the performance of the individual machine learning (ML) algorithms. In effect, the AUC value of the original DRASTIC is 0.51. Furthermore, both hybrid/ensemble models, DRASTIC-RF-MLP (AUC = 0.953) and DRASTIC-RF-SVM, (AUC = 0.901) achieve the best accuracy among the other models, followed by DRASTIC-RF (AUC = 0.852), DRASTIC-SVM (AUC = 0.802), and DRASTIC-MLP (AUC = 0.763). The results delineate areas vulnerable to pollution, which require urgent actions and strategies to improve the environmental and social qualities for the local population.

Modeling Groundwater Nitrate Contamination Using Artificial Neural Networks

The scope of the present study is the estimation of the concentration of nitrates (NO3−) in groundwater using artificial neural networks (ANNs) based on easily measurable in situ data. For the purpose of the current study, two feedforward neural networks were developed to determine whether including land use variables would improve the model results. In the first network, easily measurable field data were used, i.e., pH, electrical conductivity, water temperature, air temperature, and aquifer level. This model achieved a fairly good simulation based on the root mean squared error (RMSE in mg/L) and the Nash–Sutcliffe Model Efficiency (NSE) indicators (RMSE = 26.18, NSE = 0.54). In the second model, the percentages of different land uses in a radius of 1000 m from each well was included in an attempt to obtain a better description of nitrate transport in the aquifer system. When these variables were used, the performance of the model increased significantly (RMSE = 15.95, NSE = 0.70). For the development of the models, data from chemical and physical analyses of groundwater samples from wells located in the Kopaidian Plain and the wider area of the Asopos River Basin, both in Greece, were used. The simulation that the models achieved indicates that they are a potentially useful tools for the estimation of groundwater contamination by nitrates and may therefore constitute a basis for the development of groundwater management plans.

Identification and apportionment of shallow groundwater nitrate pollution in Weining Plain, northwest China, using hydrochemical indices, nitrate stable isotopes, and the new Bayesian stable isotope mixing model (MixSIAR)

Groundwater nitrate (NO₃^-) pollution is a worldwide environmental problem. Therefore, identification and partitioning of its potential sources are of great importance for effective control of groundwater quality. The current study was carried out to identify the potential sources of groundwater NO₃^- pollution and determine their apportionment in different land use/land cover (LULC) types in a traditional agricultural area, Weining Plain, in Northwest China. Multiple hydrochemical indices, as well as dual NO₃^- isotopes (δ¹⁵N-NO₃ and δ¹⁸O-NO₃), were used to investigate the groundwater quality and its influencing factors. LULC patterns of the study area were first determined by interpreting remote sensing image data collected from the Sentinel-2 satellite, then the Bayesian stable isotope mixing model (MixSIAR) was used to estimate proportional contributions of the potential sources to groundwater NO₃^- concentrations. Groundwater quality in the study area was influenced by both natural and anthropogenic factors, with anthropological impact being more important. The results of LULC revealed that the irrigated land is the dominant LULC type in the plain, covering an area of 576.6 km² (57.18% of the total surface study area of the plain). On the other hand, the results of the NO₃^- isotopes suggested that manure and sewage (M&S), as well as soil nitrogen (SN), were the major contributors to groundwater NO₃^-. Moreover, the results obtained from the MixSIAR model showed that the mean proportional contributions of M&S to groundwater NO₃^- were 55.5, 43.4, 21.4, and 78.7% in the forest, irrigated, paddy, and urban lands, respectively. While SN showed mean proportional contributions of 29.9, 43.4, 61.5, and 12.7% in the forest, irrigated, paddy, and urban lands, respectively. The current study provides valuable information for local authorities to support sustainable groundwater management in the study region.

Predictive modeling of groundwater nitrate pollution and evaluating its main impact factors using random forest

Groundwater quality in plains and basins of arid and semi-arid regions with increased agriculture and urbanization development faces severe nitrate pollution, which is affected by both climate and anthropogenic activities. Here, shallow groundwater nitrate concentrations in the Yinchuan Region in central Yinchuan Plain were modeled during 2000, 2005, 2010, and 2015 using random forest. Multiple spatial environment factors were taken as predictor variables. The relative importance of these factors was also calculated using the constructed model. Remote sensing and GIS methods were used to compile various environmental factors to generate training and test sets for training and validation of the random forest model. Mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) between the observed and predicted groundwater nitrate concentrations were used to measure the model performance. As indicated by these metrics, the random forest model for groundwater nitrate prediction was performed well. The relative importance of the predictor variables computed by the model indicated groundwater nitrate was mainly affected by the distance to the Yellow River, meteorological elements (precipitation, evaporation, and mean air temperature), and water level elevation. Additionally, urban and arable land were the two land use/land cover types that mainly influenced groundwater nitrate concentration in the Yinchuan Region, of which urban land was more influential than arable land as a result of intense expansion of urban land from 2000 to 2015. Overall, the current study provides an approach to integrate multiple environmental factors for groundwater quality study and is also significant for sustainable groundwater management in the Yinchuan Region.

Prediction of groundwater nitrate concentration in a semiarid region using hybrid Bayesian artificial intelligence approaches

Nitrate is a major pollutant in groundwater whose main source is municipal wastewater and agricultural activities. In the present study, Bayesian approaches such as Bayesian generalized linear model (BGLM), Bayesian regularized neural network (BRNN), Bayesian additive regression tree (BART), and Bayesian ridge regression (BRR) were used to model groundwater nitrate contamination in a semiarid region Marvdasht watershed, Fars province, Iran. Eleven groundwater (GW) nitrate conditioning factors have been taken as input parameters for predictive modeling. The results showed that the Bayesian models used in this study were all competent to model groundwater nitrate and the BART model with R2 = 0.83 was more efficient than the other models. The result of variable importance showed that potassium (K) has the highest importance in the models followed by rainfall, altitude, groundwater depth, and distance from the residential area. The results of the study can support the decision-making process to control and reduce the sources of nitrate pollution.

Predicting sustainable arsenic mitigation using machine learning techniques

This study evaluates state-of-the-art machine learning models in predicting the most sustainable arsenic mitigation preference. A Gaussian distribution-based Naïve Bayes (NB) classifier scored the highest Area Under the Curve (AUC) of the Receiver Operating Characteristic curve (0.82), followed by Nu Support Vector Classification (0.80), and K-Neighbors (0.79). Ensemble classifiers scored higher than 70% AUC, with Random Forest being the top performer (0.77), and Decision Tree model ranked fourth with an AUC of 0.77. The multilayer perceptron model also achieved high performance (AUC=0.75). Most linear classifiers underperformed, with the Ridge classifier at the top (AUC=0.73) and perceptron at the bottom (AUC=0.57). A Bernoulli distribution-based Naïve Bayes classifier was the poorest model (AUC=0.50). The Gaussian NB was also the most robust ML model with the slightest variation of Kappa score on training (0.58) and test data (0.64). The results suggest that nonlinear or ensemble classifiers could more accurately understand the complex relationships of socio-environmental data and help develop accurate and robust prediction models of sustainable arsenic mitigation. Furthermore, Gaussian NB is the best option when data is scarce.

Machine learning predictions of nitrate in groundwater used for drinking supply in the conterminous United States

Groundwater is an important source of drinking water supplies in the conterminous United State (CONUS), and presence of high nitrate concentrations may limit usability of groundwater in some areas because of the potential negative health effects. Prediction of locations of high nitrate groundwater is needed to focus mitigation and relief efforts. A three-dimensional extreme gradient boosting (XGB) machine learning model was developed to predict the distribution of nitrate. Nitrate was predicted at a 1 km resolution for two drinking water zones, each of variable depth, one for domestic supply and one for public supply. The model used measured nitrate concentrations from 12,082 wells and included predictor variables representing well characteristics, hydrologic conditions, soil type, geology, land use, climate, and nitrogen inputs. Predictor variables derived from empirical or numerical process-based models were also included to integrate information on controlling processes and conditions. The model provided accurate estimates at national and regional scales: the training (R² of 0.83) and hold-out (R² of 0.49) data fits compared favorably to previous studies. Predicted nitrate concentrations were less than 1 mg/L across most of the CONUS. Nationally, well depth, soil and climate characteristics, and the absence of developed land use were among the most influential explanatory factors. Only 1% of the area in either water supply zone had predicted nitrate concentrations greater than 10 mg/L; however, about 1.4 M people depend on groundwater for their drinking supplies in those areas. Predicted high concentrations of nitrate were most prevalent in the central CONUS. In areas of predicted high nitrate concentration, applied manure, farm fertilizer, and agricultural land use were influential predictor variables. This work represents the first application of XGB to a three-dimensional national-scale groundwater quality model and provides a significant milestone in the efforts to document nitrate in groundwater across the CONUS.

Predicting regional fluoride concentrations at public and domestic supply depths in basin-fill aquifers of the western United States using a random forest model

A random forest regression (RFR) model was applied to over 12,000 wells with measured fluoride (F) concentrations in untreated groundwater to predict F concentrations at depths used for domestic and public supply in basin-fill aquifers of the western United States. The model relied on twenty-two regional-scale environmental and surficial predictor variables selected to represent factors known to control F concentrations in groundwater. The testing model fit R² and RMSE were 0.52 and 0.78 mg/L. Comparisons of measured to predicted proportions of four F-concentrations categories (<0.7 mg/L, 0.7-2 mg/L, >2 mg/L - 4 mg/L, and > 4 mg/L) indicate that the model performed well at making regional-scale predictions. Differences between measured and predicted proportions indicate underprediction of measured F at values by between 4 and 20 mg/L, representing less than 1% of the regional scale predicted values. These residuals most often map to geographic regions where local-scale processes including evaporative discharge in closed basins or intermittent streams concentrate fluoride in shallow groundwater. Despite this, the RFR model provides spatially continuous F predictions across the basin-fill aquifers where discrete samples are missing. Further, the predictions capture documented areas that exceed the F maximum contaminant level for drinking water of 4 mg/L and areas that are below the oral-health benchmark of 0.7 mg/L. These predictions can be used to estimate fluoride concentrations in unmonitored areas and to aid in identifying geographic areas that may require further investigation at localized scales.

Electrical conductivity and dissolved oxygen as predictors of nitrate concentrations in shallow groundwater in Erhai Lake region

Elevated nitrogen (N) concentration in shallow groundwater is becoming increasingly problematic, putting water resources under pressure. For more effective management of such a resource, more precise predictors of N level in groundwater using smart monitoring networks are needed. However, external factors such as land use type, rainfall, and N loads from multiple sources (residential and agricultural) make it difficult to accurately predict the spatial and temporal variations of N concentration. In order to identify the key factors affecting spatial and temporal N concentration in shallow groundwater and develop a predictive model, 635 groundwater samples from drinking wells in residential areas and agricultural wells in croplands of a typical agricultural watershed in the Erhai Lake Basin, southwest China, in the period from 2018 to 2020, were collected and analyzed. The results showed that the type of land use and seasonal variations significantly affected the N forms and their concentrations in the shallow groundwater, as the ratios of ON and NO₃^--N to TN were 30%-39% and 52%-59% for the two land uses and 25%-44% and 46%-66% for seasonal changes. Their variations were reflected by electrical conductivity (EC) and redox environment. EC and dissolved oxygen (DO) had a positive non-linear relationship with the concentrations of total nitrogen (TN) and nitrate (NO₃^--N). The fitted non-linear quantitative models were established separately to predict TN and NO₃^--N concentrations in groundwater using easily available indictors (EC and DO). The high accuracy and performance of the models were investigated and approved by rRMSE, MAE, and 1:1 line. These findings can provide technical support for the rapid prediction and evaluation of N pollution in shallow groundwater through easily available indicators.

Application of novel framework approach for prediction of nitrate concentration susceptibility in coastal multi-aquifers, Bangladesh

This study aims to construct a novel framework approach for predicting and mapping nitrate concentration susceptibility in the coastal multi-aquifers of Bangladesh by coupling the K-fold cross-validation method and novel ensemble learning algorithms, including Boosting, Bagging and Random Forest (RF). In total, 286 nitrate sampling sites were employed in the model work. The dataset was demarcated into a 75:25 ratio for model construction (75% 3-fold ≅ 214 sites) and (25% 1-fold ≅ 72 sites) for model validation using the 4-fold cross-validation schemes. A total of 14 groundwater causative factors including salinity, depth, pH, EC, As, HCO₃^-, F^-, Cl^-, SO₄^2-, PO₄^2-, Na⁺, K⁺, Mg²⁺, and Ca²⁺ were adopted for the construction of the proposed models. OneR relative importance model was employed to choose and rank critical factors for spatial nitrate modeling. The results showed that depth, pH and As are the most influential causative factors in the elevated nitrate concentration in groundwater. Based on the model assessment criteria such as receiver operating characteristic (ROC)'s AUC (area under curve), sensitivity, specificity, accuracy, precession, F score, and Kappa coefficient, the Boosting model outperforms others (r = 0.92, AUG ≥ 0.90) in mapping nitrate concentration susceptibility, followed by Bagging and RF models. The results of mapping nitrate concentration also demonstrated that the south-central and western regions had an elevated amount of nitrate content than other regions due to depth variation in the study area. During our sampling campaign, we observed hundreds of fish hatcheries operation, a fish landing center and aquaculture farms which are the reasons for overexploitation and excessive agrochemicals used in the study area. Thus, the dependability of ensemble learning modeling verifies the effectiveness and applicability of the proposed novel approach for decision-makers in groundwater pollution management at the local and regional levels.

Assessing Nitrate Contamination Risks in Groundwater: A Machine Learning Approach

Groundwater is one of the primary sources for the daily water requirements of the masses, but it is subjected to contamination due to the pollutants, such as nitrate, percolating through the soil with water. Especially in built-up areas, groundwater vulnerability and contamination are of major concern, and require appropriate consideration. The present study develops a novel framework for assessing groundwater nitrate contamination risk for the area along the Karakoram Highway, which is a part of the China Pakistan Economic Corridor (CPEC) route in northern Pakistan. A groundwater vulnerability map was prepared using the DRASTIC model. The nitrate concentration data from a previous study were used to formulate the nitrate contamination map. Three machine learning (ML) models, i.e., Support Vector Machine (SVM), Multivariate Discriminant Analysis (MDA), and Boosted Regression Trees (BRT), were used to analyze the probability of groundwater contamination incidence. Furthermore, groundwater contamination probability maps were obtained utilizing the ensemble modeling approach. The models were calibrated and validated through calibration trials, using the area under the receiver operating characteristic curve method (AUC), where a minimum AUC threshold value of 80% was achieved. Results indicated the accuracy of the models to be in the range of 0.82–0.87. The final groundwater contamination risk map highlights that 34% of the area is moderately vulnerable to groundwater contamination, and 13% of the area is exposed to high groundwater contamination risk. The findings of this study can facilitate decision-making regarding the location of future built-up areas properly in order to mitigate the nitrate contamination that can further reduce the associated health risks.

A novel methodology for Groundwater Flooding Susceptibility assessment through Machine Learning techniques in a mixed-land use aquifer

Many areas around the world are affected by Groundwater Level rising (GWLr). One of the most severe consequences of this phenomenon is Groundwater Flooding (GF), with serious impacts for the human and natural environment. In Europe, GF has recently received specific attention with Directive 2007/60/EC, which requires Member States to map GF hazard and propose measures for risk mitigation. In this paper a methodology has been developed for Groundwater Flooding Susceptibility (GFS) assessment, using for the first time Spatial Distribution Models. These Machine Learning techniques connect occurrence data to predisposing factors (PFs) to estimate their distributions. The implemented methodology employs aquifer type, depth of piezometric level, thickness and hydraulic conductivity of unsaturated zone, drainage density and land-use as PFs, and a GF observations inventory as occurrences. The algorithms adopted to perform the analysis are Generalized Boosting Model, Artificial Neural Network and Maximum Entropy. Ensemble Models are carried out to reduce the uncertainty associated with each algorithm and increase its reliability. GFS is mapped by choosing the ensemble model with the best predictivity performance and dividing occurrence probability values into five classes, from very low to very high susceptibility, using Natural Breaks classification. The methodology has been tested and statistically validated in an area of 14,3 km² located in the Metropolitan City of Naples (Italy), affected by GWLr since 1990 and GF in buildings and agricultural soils since 2007. The results of modeling show that about 93% of the inventoried points fall in the high and very high GFS classes, and piezometric level depth, thickness of unsaturated zone and drainage density are the most influencing PFs, in accordance with field observations and the triggering mechanism of GF. The outcomes provide a first step in the assessment of GF hazard and a decision support tool to local authorities for GF risk management.

Spatial Prediction of Groundwater Potentiality in Large Semi-Arid and Karstic Mountainous Region Using Machine Learning Models

The drinking and irrigation water scarcity is a major global issue, particularly in arid and semi-arid zones. In rural areas, groundwater could be used as an alternative and additional water supply source in order to reduce human suffering in terms of water scarcity. In this context, the purpose of the present study is to facilitate groundwater potentiality mapping via spatial-modelling techniques, individual and ensemble machine-learning models. Random forest (RF), logistic regression (LR), decision tree (DT) and artificial neural networks (ANNs) are the main algorithms used in this study. The preparation of groundwater potentiality maps was assembled into 11 ensembles of models. Overall, about 374 groundwater springs was identified and inventoried in the mountain area. The spring inventory data was randomly divided into training (75%) and testing (25%) datasets. Twenty-four groundwater influencing factors (GIFs) were selected based on a multicollinearity test and the information gain calculation. The results of the groundwater potentiality mapping were validated using statistical measures and the receiver operating characteristic curve (ROC) method. Finally, a ranking of the 15 models was achieved with the prioritization rank method using the compound factor (CF) method. The ensembles of models are the most stable and suitable for groundwater potentiality mapping in mountainous aquifers compared to individual models based on success and prediction rate. The most efficient model using the area under the curve validation method is the RF-LR-DT-ANN ensemble of models. Moreover, the results of the prioritization rank indicate that the best models are the RF-DT and RF-LR-DT ensembles of models.

Development of a groundwater contamination index based on the agricultural hazard and aquifer vulnerability: Application to Portugal

Reducing nitrate leaching may not result in a significant improvement of groundwater quality. The amount of nitrate reaching groundwater depends not only on the hazard related to agricultural activities but also on-site specific groundwater vulnerability. Using national databases and other compiled datasets, the agricultural hazard was calculated as the ratio of (i) the nitrate leached estimated from the N surplus, and (ii) the water surplus, a proxy of the percolating water below the root zone. By combining the hazard with a multi-parameter groundwater vulnerability, a spatially explicit groundwater contamination risk, developed for mainland Portugal, was computed for 1999 and 2009. Results show an increase from 8,800 to 82,679 ha of the territory rated with a very high contamination risk. The priority areas were successfully screened by the Index, coinciding with the current Vulnerable Zones, although additional hotspots were detected in southern Portugal. Percolation, including both irrigation activity and precipitation, was found to be a key driver for the groundwater contamination risk due to its opposite effects in the hazard and in the vulnerability. Reducing nitrogen leaching may be insufficient to reduce the risk of nitrate contamination if there is a relatively larger reduction in precipitation. This index is particularly useful when applied to contrasting situations of vulnerability and hazard, which require distinct mitigation measures to mitigate groundwater contamination.

Exploration of E. coli contamination drivers in private drinking water wells: An application of machine learning to a large, multivariable, geo-spatio-temporal dataset

Groundwater resources are under increasing threats from contamination and overuse, posing direct threats to human and environmental health. The purpose of this study is to better understand drivers of, and relationships between, well and aquifer characteristics, sampling frequencies, and microbiological contamination indicators (specifically E. coli) as a precursor for improving knowledge and tools to assess aquifer vulnerability and well contamination within Ontario, Canada. A dataset with 795, 023 microbiological testing observations over an eight-year period (2010 to 2017) from 253,136 unique wells across Ontario was employed. Variables in this dataset include date and location of test, test results (E. coli concentration), well characteristics (well depth, location), and hydrogeological characteristics (bottom of well stratigraphy, specific capacity). Association rule analysis, univariate and bivariate analyses, regression analyses, and variable discretization techniques were utilized to identify relationships between E. coli concentration and the other variables in the dataset. These relationships can be used to identify drivers of contamination, their relative importance, and therefore potential public health risks associated with the use of private wells in Ontario. Key findings are that: i) bedrock wells completed in sedimentary or igneous rock are more susceptible to contamination events; ii) while shallow wells pose a greater risk to consumers, deep wells are also subject to contamination events and pose a potentially unanticipated risk to health of well users; and, iii) well testing practices are influenced by results of previous tests. Further, while there is a general correlation between months with the greatest testing frequencies and concentrations of E. coli occurring in samples, an offset in this timing is observed in recent years. Testing remains highest in July while peaks in adverse results occur up to three months later. The realization of these trends prompts a need to further explore the bases for such occurrences.

Machine Learning Predictions of pH in the Glacial Aquifer System, Northern USA

A boosted regression tree model was developed to predict pH conditions in three dimensions throughout the glacial aquifer system of the contiguous United States using pH measurements in samples from 18,386 wells and predictor variables that represent aspects of the hydrogeologic setting. Model results indicate that the carbonate content of soils and aquifer materials strongly controls pH and, when coupled with long flowpaths, results in the most alkaline conditions. Conversely, in areas where glacial sediments are thin and carbonate-poor, pH conditions remain acidic. At depths typical of drinking-water supplies, predicted pH >7.5-which is associated with arsenic mobilization-occurs more frequently than predicted pH <6-which is associated with water corrosivity and the mobilization of other trace elements. A novel aspect of this model was the inclusion of numerically based estimates of groundwater flow characteristics (age and flowpath length) as predictor variables. The sensitivity of pH predictions to these variables was consistent with hydrologic understanding of groundwater flow systems and the geochemical evolution of groundwater quality. The model was not developed to provide precise estimates of pH at any given location. Rather, it can be used to more generally identify areas where contaminants may be mobilized into groundwater and where corrosivity issues may be of concern to prioritize areas for future groundwater monitoring.

Delineation of groundwater quality locations suitable for target end-use purposes through deep neural network models

Groundwater is the main source of water for beverages, and its quality varies depending on extraction location; this is particularly the case in regions with complex geology, topography, and multiple forms of land use. Thus, it is important to determine a suitable groundwater extraction location based on intended water use and the related water quality standards. In this study, deep neural network (DNN) models and GIS data relating to groundwater quality were applied to estimate potential maps of Gangwon Province in South Korea, where groundwater is frequently extracted for drinking purposes. These maps specify areas where the groundwater quality is conducive for being used as mineral water and water for brewing coffee (hereafter referred as "coffee water"). Sensitivity analysis identified how inputs were sensitive to model estimation and showed that land-use variables were the most sensitive. The importance of each variable quantified how good or bad its region is for the desired groundwater. The overall features of importance were similar between mineral water and coffee water. However, with differences in hydrogeological units, carbonate rock was a variable of high positive importance for mineral water; metamorphic rock was its equivalent for coffee water. Our results offer a potential map of desired groundwater quality in the absence of a detailed understanding of the underlying hydrochemical processes governing groundwater quality. Additionally, the development of such a potential mapping model can help to determine the appropriate development area of groundwater for their respective purposes.

Comparative Analysis of Artificial Intelligence Models for Accurate Estimation of Groundwater Nitrate Concentration

Prediction of the groundwater nitrate concentration is of utmost importance for pollution control and water resource management. This research aims to model the spatial groundwater nitrate concentration in the Marvdasht watershed, Iran, based on several artificial intelligence methods of support vector machine (SVM), Cubist, random forest (RF), and Bayesian artificial neural network (Baysia-ANN) machine learning models. For this purpose, 11 independent variables affecting groundwater nitrate changes include elevation, slope, plan curvature, profile curvature, rainfall, piezometric depth, distance from the river, distance from residential, Sodium (Na), Potassium (K), and topographic wetness index (TWI) in the study area were prepared. Nitrate levels were also measured in 67 wells and used as a dependent variable for modeling. Data were divided into two categories of training (70%) and testing (30%) for modeling. The evaluation criteria coefficient of determination (R²), mean absolute error (MAE), root mean square error (RMSE), and Nash–Sutcliffe efficiency (NSE) were used to evaluate the performance of the models used. The results of modeling the susceptibility of groundwater nitrate concentration showed that the RF (R² = 0.89, RMSE = 4.24, NSE = 0.87) model is better than the other Cubist (R² = 0.87, RMSE = 5.18, NSE = 0.81), SVM (R² = 0.74, RMSE = 6.07, NSE = 0.74), Bayesian-ANN (R² = 0.79, RMSE = 5.91, NSE = 0.75) models. The results of groundwater nitrate concentration zoning in the study area showed that the northern parts of the case study have the highest amount of nitrate, which is higher in these agricultural areas than in other areas. The most important cause of nitrate pollution in these areas is agriculture activities and the use of groundwater to irrigate these crops and the wells close to agricultural areas, which has led to the indiscriminate use of chemical fertilizers by irrigation or rainwater of these fertilizers is washed and penetrates groundwater and pollutes the aquifer.

Susceptibility Prediction of Groundwater Hardness Using Ensemble Machine Learning Models

Groundwater resources, unlike surface water, are more vulnerable to disturbances and contaminations, as they take a very long time and significant cost to recover. So, predictive modeling and prevention strategies can empower policymakers for efficient groundwater governance through informed decisions and recommendations. Due to the importance of groundwater quality modeling, the hardness susceptibility mapping using machine learning (ML) models has not been explored. For the first time, the current research aimed to predict groundwater hardness susceptibility using the ML models. The performance of two ensemble models of boosted regression trees (BRT) and random forest (RF) is investigated through the arrangement of a comparative study with multivariate discriminant analysis (MDA). According to the hardness values in 135 groundwater quality monitoring wells, the hard and soft water are determined; then, 11 predictor variables including distance from the sea (DFS), land use, elevation, distance from the river (DFR), depth to groundwater (DTGW), pH, precipitation (PCP), evaporation (E), groundwater level (GWL), curvature, and lithology are used for predicting the groundwater hardness susceptibility map. Results indicated that the variables of DFR, DTGW, elevation, and DFS had a higher contribution to the modeling process. So, the high harness areas are mostly related to low elevations, low DTGW, and proximity to river and sea, which facilitate the percolation conditions for minerals containing calcium or magnesium into groundwater.

Optimised neural network model for river-nitrogen prediction utilizing a new training approach

In the past few decades, there has been a rapid growth in the concentration of nitrogenous compounds such as nitrate-nitrogen and ammonia-nitrogen in rivers, primarily due to increasing agricultural and industrial activities. These nitrogenous compounds are mainly responsible for eutrophication when present in river water, and for ‘blue baby syndrome’ when present in drinking water. High concentrations of these compounds in rivers may eventually lead to the closure of treatment plants. This study presents a training and a selection approach to develop an optimum artificial neural network model for predicting monthly average nitrate-N and monthly average ammonia-N. Several studies have predicted these compounds, but most of the proposed procedures do not involve testing various model architectures in order to achieve the optimum predicting model. Additionally, none of the models have been trained for hydrological conditions such as the case of Malaysia. This study presents models trained on the hydrological data from 1981 to 2017 for the Langat River in Selangor, Malaysia. The model architectures used for training are General Regression Neural Network (GRNN), Multilayer Neural Network and Radial Basis Function Neural Network (RBFNN). These models were trained for various combinations of internal parameters, input variables and model architectures. Post-training, the optimum performing model was selected based on the regression and error values and plot of predicted versus observed values. Optimum models provide promising results with a minimum overall regression value of 0.92.

An investigation into seasonal variations of groundwater nitrate by spatial modelling strategies at two levels by kriging and co-kriging models

Nitrate pollution of groundwater through spatial models is investigated in this paper by using a sample of nitrate values at monitoring wells using the data from four seasons of a year, in which data are sparse. Two spatial modelling strategies are formulated at two levels, in which Strategy 1 comprises: three variations of kriging-based models (ordinary kriging, simple kriging and universal kriging), which are constructed at Level 1 to predict nitrate concentrations; and a Multiple Co-Kriging (MCoK) model is used at Level 2 to enhance the accuracy of the predictions. Strategy 2 is also at two levels but employs Indicator Kriging (IK) at Level 1 as a probabilistic spatial model to predict areas at risk of exceeding two thresholds of 37.5 mg/L and 50 mg/L of nitrate concentration, and Multiple Co-Indicator Kriging (MCoIK) at Level 2 for a better accuracy. The improvements at Level 2 for both strategies are remarkable and hence they are used to gain an insight into inherent problems. The results of a study delineate areas with excessive nitrate concentrations, which are in the vicinity of urban areas and hence reflect poor planning practices since the 1990s. The results further reveal the patterns on sensitivities to seasonal variations driven by aquifer recharge and strong dilution processes in spring times; and on the role of pumpage impacting aquifers giving rise to possible hotspots of nitrate concentrations.