A Priori Estimation of the Saturated Hydraulic Conductivity of Municipal Solid Waste Using Empirical and Machine Learning Methods

The saturated hydraulic conductivity (kS) of municipal solid waste (MSW) is important for understanding leachate movement in landfills. Previous prediction models for MSW kS were developed using a limited amount of data obtained from localized MSW but their accuracy must improve. In this study, an exhaustive data set covering 425 tests, including 39 conducted in this study and 386 reported in 60 prior studies, was constructed. It contains the kS and other related MSW properties, including composition, dry unit weight (γd), total porosity (nt), and drainable porosity (nd), and conditions such as effective vertical stress (σv′) and compaction level of MSW. The prediction models of MSW kS were then developed using semiempirical regression and a machine learning algorithm, multiexpression programming (MEP). The results showed that the kS-nd correlation, which was well-regressed using a power function (R2=0.71), was more accurate than the commonly used kS-γd and kS-nt correlations. The proposed MEP model, developed with nd, σv′, and MSW composition as input variables, achieved a higher R2 of 0.91. Furthermore, considering that nd is important in predicting kS and is difficult to measure in the field, an empirical method to estimate nd was proposed. The estimated nd (nd*) depended on the estimated nt (nt*), MSW composition, compaction level, σv′, and γd of MSW. The nd*-kS correlation (R2=0.72) was validated based on 183 tests in 36 studies. The two proposed methods herein exhibited higher accuracy in predicting MSW kS compared to the empirical methods previously reported in the literature. Either method should be used for the design and monitoring of MSW landfills based on the availability of nd.