Development of artificial intelligence models for well groundwater quality simulation: Different modeling scenarios

Groundwater is one of the most important freshwater resources, especially in arid and semi-arid regions where the annual amounts of precipitation are small with frequent drought durations. Information on qualitative parameters of these valuable resources is very crucial as it might affect its applicability from agricultural, drinking, and industrial aspects. Although geo-statistics methods can provide insight about spatial distribution of quality factors, applications of advanced artificial intelligence (AI) models can contribute to produce more accurate results as robust alternative for such a complex geo-science problem. The present research investigates the capacity of several types of AI models for modeling four key water quality variables namely electrical conductivity (EC), sodium adsorption ratio (SAR), total dissolved solid (TDS) and Sulfate (SO₄) using dataset obtained from 90 wells in Tabriz Plain, Iran; assessed by k-fold testing. Two different modeling scenarios were established to make simulations using other quality parameters and the geographical information. The obtained results confirmed the capabilities of the AI models for modeling the well groundwater quality variables. Among all the applied AI models, the developed hybrid support vector machine-firefly algorithm (SVM-FFA) model achieved the best predictability performance for both investigated scenarios. The introduced computer aid methodology provided a reliable technology for groundwater monitoring and assessment.

Developing a framework for classifying water lead levels at private drinking water systems: A Bayesian Belief Network approach

The presence of lead in drinking water creates a public health crisis, as lead causes neurological damage at low levels of exposure. The objective of this research is to explore modeling approaches to predict the risk of lead at private drinking water systems. This research uses Bayesian Network approaches to explore interactions among household characteristics, geological parameters, observations of tap water, and laboratory tests of water quality parameters. A knowledge discovery framework is developed by integrating methods for data discretization, feature selection, and Bayes classifiers. Forward selection and backward selection are explored for feature selection. Discretization approaches, including domain-knowledge, statistical, and information-based approaches, are tested to discretize continuous features. Bayes classifiers that are tested include General Bayesian Network, Naive Bayes, and Tree-Augmented Naive Bayes, which are applied to identify Directed Acyclic Graphs (DAGs). Bayesian inference is used to fit conditional probability tables for each DAG. The Bayesian framework is applied to fit models for a dataset collected by the Virginia Household Water Quality Program (VAHWQP), which collected water samples and conducted household surveys at 2,146 households that use private water systems, including wells and springs, in Virginia during 2012 and 2013. Relationships among laboratory-tested water quality parameters, observations of tap water, and household characteristics, including plumbing type, source water, household location, and on-site water treatment are explored to develop features for predicting water lead levels. Results demonstrate that Naive Bayes classifiers perform best based on recall and precision, when compared with other classifiers. Copper is the most significant predictor of lead, and other important predictors include county, pH, and on-site water treatment. Feature selection methods have a marginal effect on performance, and discretization methods can greatly affect model performance when paired with classifiers. Owners of private wells remain disadvantaged and may be at an elevated level of risk, because utilities and governing agencies are not responsible for ensuring that lead levels meet the Lead and Copper Rule for private wells. Insight gained from models can be used to identify water quality parameters, plumbing characteristics, and household variables that increase the likelihood of high water lead levels to inform decisions about lead testing and treatment.

Feature selection approaches for predictive modelling of groundwater nitrate pollution: An evaluation of filters, embedded and wrapper methods

Recognising the various sources of nitrate pollution and understanding system dynamics are fundamental to tackle groundwater quality problems. A comprehensive GIS database of twenty parameters regarding hydrogeological and hydrological features and driving forces were used as inputs for predictive models of nitrate pollution. Additionally, key variables extracted from remotely sensed Normalised Difference Vegetation Index time-series (NDVI) were included in database to provide indications of agroecosystem dynamics. Many approaches can be used to evaluate feature importance related to groundwater pollution caused by nitrates. Filters, wrappers and embedded methods are used to rank feature importance according to the probability of occurrence of nitrates above a threshold value in groundwater. Machine learning algorithms (MLA) such as Classification and Regression Trees (CART), Random Forest (RF) and Support Vector Machines (SVM) are used as wrappers considering four different sequential search approaches: the sequential backward selection (SBS), the sequential forward selection (SFS), the sequential forward floating selection (SFFS) and sequential backward floating selection (SBFS). Feature importance obtained from RF and CART was used as an embedded approach. RF with SFFS had the best performance (mmce=0.12 and AUC=0.92) and good interpretability, where three features related to groundwater polluted areas were selected: i) industries and facilities rating according to their production capacity and total nitrogen emissions to water within a 3km buffer, ii) livestock farms rating by manure production within a 5km buffer and, iii) cumulated NDVI for the post-maximum month, being used as a proxy of vegetation productivity and crop yield.