Simulating Complex Relationships Between Pollutants and the Environment Using Regression Splines: A Case Study for Landfill Leachate

Abstract

Pollutant concentrations in the environment are functions of source variables, environmental variables, and time. Relationships among these variables are complex and unknown, therefore multiple linear regression may not be applicable. One solution is use of Multivariate Adaptive Regression Splines (MARS). MARS is a flexible statistical method, used for machine learning, but it is not typically appropriate for sustainability applications due to limited data. While most machine learning algorithms require a large amount of data and may not yield interpretable results, by contrast, MARS in conjunction with design of experiments has the ability to find important interpretable relationships in limited data. Laboratory data were collected under a controlled environment to discover how leachate composition changes as functions of ambient temperature, rainfall intensities, and waste composition. The MARS application to landfill leachate demonstrates a promise in capturing complex relationships, illustrated via MARS 3D interaction plots.