Groundwater supply is crucial to both anthropogenic and industrial activities in Canada. In Saskatchewan, around 63 percent of municipalities rely on groundwater as their primary water supplies. Recent data suggested groundwater quality in Condie aquifer might have been compromised due to the operations of an unlined municipal landfill and the industrial area upstream. The complex geographical conditions, hydrochemistry reactions in subsurface environment, missing data and inconsistent sampling has made the assessment of groundwater difficult. Two studies on the groundwater quality near the Regina landfill are conducted. In the first study, groundwater quality is evaluated with respect to its ions correlations, soilwater processes, and suitability for drinking and irrigation. Unlike similar studies, geological locations of the water samples were explicitly considered. It is found that the abundance of cations in the groundwater was: calcium > magnesium > sodium > potassium > manganese (Ca2+ > Mg2+ > Na+ > K+ > Mn2+); and for anions: sulphate > bicarbonate > chloride (SO4 2- > HCO3

• Cl-). Correlation analysis and ion plots pointed to gypsum and halite dissolution being the main factors affecting groundwater chemistry. Principal component analysis yielded three principal components, covering 80.7% of the total variance. Boxplots which show the variation of indicators suggests possible groundwater contamination from landfill operation. Wilcox diagrams indicate groundwater near landfill was not suitable for irrigation. Dual-step multiple linear regression (MLR) results suggested that total hardness is statistically related to calcium, magnesium, and chloride ions. The model is able to accurately predict the concentration of total hardness with small errors (R2 = 0.995, slope = 0.995). In the second study, total dissolved solids (TDS) is selected and modeled using both conventional statistical approaches (Multiple Linear Regression, Hybrid PCR) and machine learning methods (Artificial Neural Network and Adaptive Neuro-Fuzzy Inference System). A total of twenty models are constructed and evaluated. Results show that Total Dissolved Solids is statistically related to parameters such as Ca, Mg, HCO3, SO4, Cl, EC, and pH. Multiple evaluation metrics such as Mean Absolute Error, Root Mean Squared Error, Coefficient of determination, and the Percentage error were used for model evaluation. It is found that Hybrid PCR provides model fit with high accuracy, and comparing observed with predicted values in testing stage, R2 varies from 0.96 to 0.980 Mean Absolute Error from 72.64 to 97.55 , and Root Mean Square Error from 97.95 to 123.55. The two machine learning methods produce larger variations from model to model.