Machine learning framework for irrigation water quality assessment in Kerala Rivers, India

Area of Science:

• Environmental Science
• Agricultural Engineering
• Data Science

Background:

• Sustainable agriculture relies on accurate irrigation water quality assessment.
• Conventional indices struggle with complex hydro-chemical relationships and extensive parameter needs.
• Freshwater resources face increasing pressure, necessitating efficient water quality monitoring.

Purpose of the Study:

• To develop an integrated machine learning framework for predicting and classifying the Irrigation Water Quality Index (FIWQI).
• To evaluate the performance of various machine learning models using physicochemical parameters from Indian rivers.
• To provide a scalable solution for irrigation water quality monitoring and decision support.

Main Methods:

• A Fuzzy Irrigation Water Quality Index (FIWQI) was computed using fuzzy analytic hierarchy process (Fuzzy-AHP).
• Machine learning models (tree-based ensemble, neural network, classical) were trained and evaluated.
• Data preprocessing included normalization, multicollinearity analysis, class balancing (SMOTE), and hyperparameter optimization via cross-validation.

Main Results:

• Tree-based ensemble models, especially CatBoost, significantly outperformed classical models (R²=0.996, RMSE <1).
• Classification accuracy exceeded 97% for tree-based models, a substantial improvement over classical methods.
• Boron, fluoride, and electrical conductivity were identified as key predictors for water quality classification.

Conclusions:

• The proposed machine learning framework offers a data-efficient and scalable solution for irrigation water quality assessment.
• The study highlights the effectiveness of advanced machine learning in capturing nonlinear hydro-chemical relationships.
• This approach supports informed decision-making for sustainable agricultural management in river-dependent regions.