Active Learning-enhanced Deep Neural Network (AL-DNN) for uncertainty analysis of landfill leakage risk.

Landfills serve both as essential infrastructure for solid waste disposal and as potential sources of significant environmental risk. Leachate leakage has become a global concern due to its adverse impact on groundwater quality and associated public health threats. Accurate assessment of concealed leakage and its uncertainty is critical for effective risk management. However, traditional analytical-Monte Carlo coupled methods often fail to capture the complexity of leakage-related uncertainties and are computationally expensive. To address these challenges, we propose an Active Learning-enhanced Deep Neural Network (AL-DNN) model. This surrogate model is constructed based on datasets generated by a high-performance Groundwater Simulation Numerical Model (GSNM), providing a computationally efficient alternative for simulating contaminant transport. Firstly, a deep neural network is used to replicate simulation results of groundwater contamination under uncertain parameters. Secondly, an active learning strategy is introduced to identify and select informative samples, improving prediction accuracy while minimizing the number of required labeled data. Results show that the AL-DNN model achieves comparable accuracy to traditional methods using only 60 samples, leading to a 90 % reduction in computation time. Finally, the model is applied to a simulated landfill leakage scenario to predict COD concentration distributions and assess contamination risks. The results indicate a maximum exceedance probability of 0.76 at Monitoring Well 1. These findings highlight the effectiveness and practicality of the proposed method in supporting rapid, reliable groundwater contamination risk assessments under uncertainty.