Prediction of leachate quantity and quality from a landfill site by the long short-term memory model

Abstract

The long short-term memory (LSTM) model was first applied in this study for the prediction of the leachate quantity and quality at a real landfill site. In our LSTM model, in the learning phase from July 2003 to March 2018, three input data items consisting of the daily precipitation (DP), the daily average temperature (DAT), and the accumulated amount of landfilled waste presented the quantity of leachate generated with high accuracy. The DAT was important for the landfill site, particularly in a snow area because it contributes to the leachate generated during the spring thaw with low precipitation. In the testing phase from April 2018 to March 2019, our LSTM model predicted the leachate generated with a mean absolute percentage error (MAPE) of 26.2%. The concentrations of biological oxygen demand, chemical oxygen demand, total nitrogen, calcium ion and chloride ion in leachate were presented in the learning phase by six input data items: DP, DAT, and the daily amount of landfilled waste (incineration residue, incombustible waste, business waste, and combustible waste) with high R² values. In the testing phase, the quality of leachate was predicted with the MAPE between 11.8% and 30.2%. Another year data from April 2019 to March 2020 was used to verify accuracy of our model with no overfitting. This study showed the possibility of applying the LSTM model to future predictions of leachate quantity and quality from landfill sites with an acceptable error for daily operation.

Keywords: Future prediction; Landfill sites; Long short-term memory model; Quantity and quality of leachate; Snow area.