Groundwater quality modeling: On the analogy between integrative PSO and MRFO mathematical and machine learning models

Mohammad Zounemat-Kermani, Amin Mahdavi-Meymand, Marzieh Fadaee, Okke Batelaan, Reinhard Hinkelmann

    Research output: Contribution to journalArticlepeer-review

    Abstract

    Reliable and accurate modeling of groundwater quality is an important element of sustainable groundwater management of productive aquifers. In this research, specific conductance (SC) of groundwater is predicted based on different individual and integrative machine learning, adaptive neuro-fuzzy inference system (ANFIS), and nonlinear mathematical models. For developing the integrative models, the well-known particle swarm optimization (PSO) and novel manta ray foraging optimization (MRFO) heuristic algorithms are embedded in the models. Presenting different univariate, bivariate, and multivariate input scenarios, the parameters used to develop and validate the models include groundwater level, salinity, and water temperature at an observation well near Florida City. The findings reveal that applying more independent parameters (multivariate scenario) enhances the performance of both the mathematical and machine learning models. Even though the mathematical models present an acceptable performance for the prediction of SC (index of agreement, IA, equals 0.933), the ANFIS models provide the most accurate SC predictions (IA = 0.943). Both the PSO and MRFO algorithms improved the prediction capability of the ANFIS models with, respectively, 13% and 5% for the RMSE.

    Original languageEnglish
    Number of pages11
    JournalEnvironmental Quality Management
    Early online date7 Jun 2021
    DOIs
    Publication statusE-pub ahead of print - 7 Jun 2021

    Keywords

    • Aquifer
    • artificial intelligence
    • MRFO
    • PSO
    • specific conductance

    Fingerprint

    Dive into the research topics of 'Groundwater quality modeling: On the analogy between integrative PSO and MRFO mathematical and machine learning models'. Together they form a unique fingerprint.

    Cite this