Mixing data for multivariate statistical study of groundwater quality

Area of Science:

• Environmental Science
• Water Quality Assessment
• Statistical Modeling

Background:

• Water quality monitoring is crucial for public health and environmental management.
• Kozhikode City, Kerala, India, faces challenges in maintaining water quality due to various sources.
• Multivariate statistical methods offer powerful tools for analyzing complex water quality datasets.

Purpose of the Study:

• To apply and compare multivariate statistical models, including Multiple Linear Regression (MLR), Structural Equation Modeling (SEM), and Adaptive Neuro-Fuzzy Inference System (ANFIS), for water quality assessment.
• To develop a unified MLR model by combining water quality data from different locations and time periods.
• To evaluate the predictive performance of ANFIS against MLR for total dissolved solids (TDS).

Main Methods:

• Collected and combined water quality data from multiple sites and times in Kozhikode City.
• Developed and tested several MLR models with TDS as the dependent variable and various water quality parameters as independent variables.
• Constructed an SEM model using a combined dataset and compared its coefficients with the corresponding MLR model.
• Developed an ANFIS model using the combined dataset, with TDS as the output and other parameters as inputs.

Main Results:

• A unified MLR model was established by mixing datasets, showing comparable performance to unmixed models.
• SEM analysis yielded identical regression coefficients to the corresponding MLR model, likely due to increased sample size.
• The ANFIS model exhibited superior predictive accuracy for TDS on an external dataset compared to the MLR model.
• Key water quality parameters identified as significant predictors for TDS included calcium, magnesium, nitrate, sodium, chloride, potassium, total hardness, and sulfate.

Conclusions:

• Combining datasets can enhance the robustness of statistical models like MLR.
• ANFIS is a more effective modeling approach than MLR for predicting total dissolved solids (TDS) in complex water quality scenarios.
• The findings provide valuable insights for water resource management and pollution control strategies in Kozhikode City.