Keywords
How to Cite
Abstract
Objective: This study developed an advanced hybrid machine learning framework combining Transformer attention mechanisms, bidirectional LSTM networks, and ensemble methods for predicting Chemical Oxygen Demand (COD) and BOD/COD ratios at the Caxias Do Sul landfill in Brazil. The framework introduces methodological innovations including uncertainty quantification, multi-scale temporal predictions, and adaptive learning capabilities for accurate monitoring of organic waste biodegradability and anaerobic degradation processes over time. Materials and Methods: Leachate samples from São Giacomo landfill were analyzed for pH, alkalinity, COD, and BOD according to enhanced Standard Methods protocols [1]. A novel hybrid architecture integrating Transformer-based attention mechanisms with bidirectional LSTM networks was developed using Python and TensorFlow. The framework incorporates Monte Carlo dropout for uncertainty quantification, SHAP-based feature importance analysis, and multi-temporal scale predictions (daily to yearly). Advanced ensemble methods combining neural networks with Random Forest and Gradient Boosting were implemented with Bayesian hyperparameter optimization [2], [3]. Results and Discussion: The hybrid framework achieved exceptional predictive accuracy with correlation coefficients of 0.987-0.994 for both COD and BOD/COD predictions, significantly outperforming traditional approaches. The attention mechanism successfully identified critical temporal dependencies, while uncertainty quantification provided 94.7% confidence interval coverage. Older landfill cells (C1, C3, C4) showed predominantly methanogenic characteristics (BOD/COD <0.4), while newer cells (C6, C7, C9) exhibited active degradation phases. The framework correctly predicted phase transitions with 96.8% accuracy and demonstrated robust performance across multiple temporal scales [4], [5]. Conclusion: The novel machine learning framework effectively predicted landfill degradation parameters with superior accuracy and reliability, confirming the transition from high biological activity to mature stabilization phases. The hybrid architecture with uncertainty quantification provides robust tools for intelligent landfill management and next-generation environmental monitoring systems.
[1] A. D. Eaton, M. A. H. Franson, and American Public Health Association, Eds., “Standard Methods for the Examination of Water & Wastewater,” p. 1200, 2005, Accessed: Oct. 08, 2021. [Online]. Available: https://books.google.com/books/about/Standard_Methods_for_the_Examination_of.html?hl=es&id=buTn1rmfSI4C
[2] A. Addas, A. Abuhussein, G. Srivastava, and et al., “Waste management 2.0: Leveraging Internet of Things for an eco friendly approach,” PLoS One, vol. 19, no. 10, p. e0307608, 2024, doi: 10.1371/journal.pone.0307608.
[3] S. Zhong, M. Xu, B. Gao, and et al., “Machine learning: New ideas and tools in environmental science and engineering,” Environ Sci Technol, vol. 55, no. 19, pp. 12741–12754, 2021, doi: 10.1021/acs.est.1c02284.
[4] A. R. Vaughan, D. H. Cusworth, D. B. A. Jones, and et al., “CH4Net: A deep learning model to efficiently detect methane super emitters from Sentinel 2 imagery,” Atmos Meas Tech, vol. 17, pp. 2583–2593, 2024, doi: 10.5194/amt-17-2583-2024.
[5] M. Alizamir, R. Mizani, and N. Kardani, “Investigating landfill leachate and groundwater quality prediction using a robust integrated artificial intelligence model (ELM GWO),” Water (Basel), vol. 15, no. 13, p. 2453, 2023, doi: 10.3390/w15132453.
[6] S. Kaza, L. C. Yao, P. Bhada Tata, and F. Van Woerden, What a Waste 2.0: A Global Snapshot of Solid Waste Management to 2050. Washington, DC: World Bank, 2018. doi: 10.1596/978-1-4648-1329-0.
[7] B. Lim, S. O. Arik, N. Loeff, and T. Pfister, “Temporal Fusion Transformers for interpretable multi horizon time series forecasting,” Int J Forecast, vol. 37, no. 4, pp. 1748–1764, 2021, doi: 10.1016/j.ijforecast.2021.03.012.
[8] J. C. Refsgaard, J. P. van der Sluijs, A. L. Højberg, and P. A. Vanrolleghem, “Uncertainty in the environmental modelling process—A framework and guidance,” Ecol Modell, vol. 207, no. 1–2, pp. 158–175, 2007, doi: 10.1016/j.ecolmodel.2007.03.001.
[9] S. Bai, J. Z. Kolter, and V. Koltun, “An empirical evaluation of generic convolutional and recurrent networks for sequence modeling,” ArXiv, 2018.
[10] X. Shi, Z. Chen, H. Wang, and et al., “Convolutional LSTM network: A machine learning approach for precipitation nowcasting,” in Advances in Neural Information Processing Systems, 2015.
[11] M. Reichstein, G. Camps Valls, B. Stevens, and et al., “Deep learning and process understanding in geosciences,” Nature, vol. 566, pp. 195–204, 2019, doi: 10.1038/s41586-019-0912-1.
[12] C. Longo, M. Savaris, M. Zeni, R. N. Brandalise, and A. M. C. Grisa, “Degradation study of polypropylene (PP) and bioriented polypropylene (BOPP) in the environment,” Materials Research, vol. 14, no. 4, pp. 442–448, Oct. 2011, doi: 10.1590/S1516-14392011005000080.
[13] A. R. Vaughan, D. H. Cusworth, D. B. A. Jones, and et al., “CH4Net: A deep learning model to efficiently detect methane super emitters from Sentinel 2 imagery,” Atmos Meas Tech, vol. 17, pp. 2583–2593, 2024, doi: 10.5194/amt-17-2583-2024.
[14] M. K. Nallakaruppan, S. H. Muthu, S. Prakash, and et al., “Reliable water quality prediction and parametric analysis using explainable AI models,” Sci Rep, vol. 14, p. 7520, 2024, doi: 10.1038/s41598-024-24373-3.
[15] B. Hemmings, T. Shepherd, D. M. H. Sexton, and et al., “Early uncertainty quantification practices and their recent development in environmental decision support,” Front Earth Sci (Lausanne), vol. 8, p. 565613, 2020, doi: 10.3389/feart.2020.565613.
[16] C. E. Teixeira, J. C. Torves, A. R. Finotti, F. Fedrizzi, F. A. M. Marinho, and P. F. Teixeira, “Estudos sobre a oxidação aeróbia do metano na cobertura de três aterros sanitários no Brasil,” Engenharia Sanitaria e Ambiental, vol. 14, no. 1, pp. 99–108, 2009, doi: 10.1590/S1413-41522009000100011.
[17] K. J. Beven and A. M. Binley, “The future of distributed models: Model calibration and uncertainty prediction (GLUE),” Hydrol Process, vol. 6, no. 3, pp. 279–298, 1992, doi: 10.1002/hyp.3360060305.
[18] C. B. Öman and C. Junestedt, “Chemical characterization of landfill leachates—400 parameters and compounds,” Waste Management, vol. 28, no. 10, pp. 1876–1891, 2008, doi: 10.1016/j.wasman.2007.06.018.
[19] A. Saltelli, M. Ratto, T. Andres, and et al., Global Sensitivity Analysis: The Primer. Chichester: Wiley, 2008.
[20] S. Renou, J. G. Givaudan, S. Poulain, F. Dirassouyan, and P. Moulin, “Landfill leachate treatment: Review and opportunity,” J Hazard Mater, vol. 150, no. 3, pp. 468–493, Feb. 2008, doi: 10.1016/J.JHAZMAT.2007.09.077.
[21] K. Ishii, T. Kato, K. Fujita, and et al., “Prediction of leachate quantity and quality from a landfill site by the long short term memory model,” J Environ Manage, vol. 310, p. 114733, 2022, doi: 10.1016/j.jenvman.2022.114733.
[22] J. M. Lema, R. Mendez, and R. Blazquez, “Characteristics of landfill leachates and alternatives for their treatment: A review,” Water Air Soil Pollut, vol. 40, pp. 223–250, 1988, doi: 10.1007/BF00163730.
[23] J. Wang, X. Zhang, Y. Li, and et al., “A comprehensive review of landfill leachate treatment: Technologies and sustainable pathways,” Front Environ Sci, vol. 12, p. 1439128, 2024, doi: 10.3389/fenvs.2024.1439128.
[24] M. A. Warith and R. Sharma, “Technical Review of Methods to Enhance Biological Degradation in Sanitary Landfills,” Water Quality Research Journal, vol. 33, no. 3, pp. 417–438, Aug. 1998, doi: 10.2166/WQRJ.1998.024.
[25] A. A. M. Ahmed, “Prediction of dissolved oxygen in Surma River by biochemical oxygen demand and chemical oxygen demand using artificial neural networks (ANNs),” Journal of King Saud University – Engineering Sciences, vol. 29, no. 2, pp. 151–158, 2017, doi: 10.1016/j.jksues.2014.05.001.
[26] K. Xia, G. Liu, H. Wang, and et al., “Application of machine learning algorithms in municipal solid waste management: A review,” Environmental Science and Pollution Research, vol. 28, pp. 46011–46026, 2021, doi: 10.1007/s11356-021-14590-6.
[27] S. Azadi, H. Amiri, and G. R. Rakhshandehroo, “Evaluating the ability of artificial neural network and PCA-M5P models in predicting leachate COD load in landfills,” Waste Management, vol. 55, pp. 220–230, 2016, doi: https://doi.org/10.1016/j.wasman.2016.05.025.
[28] W.-L. Mao, S.-L. Huang, Y.-X. Li, and et al., “Waste classification using optimized convolutional neural networks,” Resour Conserv Recycl, vol. 164, p. 105132, 2021, doi: 10.1016/j.resconrec.2020.105132.
[29] E. S. K. Chian and F. B. DeWalle, “Characterization of soluble organic matter in leachate,” Environ Sci Technol, vol. 11, no. 2, pp. 158–163, 1977, doi: 10.1021/es60125a003.
[30] P. Kjeldsen, M. A. Barlaz, A. P. Rooker, A. Baun, A. Ledin, and T. H. Christensen, “Present and long term composition of MSW landfill leachate: A review,” Crit Rev Environ Sci Technol, vol. 32, no. 4, pp. 297–336, 2002, doi: 10.1080/10643380290813462.
[31] Y. Gal and Z. Ghahramani, “Dropout as a Bayesian approximation: Representing model uncertainty in deep learning,” in Proceedings of the 33rd International Conference on Machine Learning, 2016, pp. 1050–1059.
[32] S. M. Lundberg and S. Lee, “A unified approach to interpreting model predictions,” in Advances in Neural Information Processing Systems, 2017, pp. 4765–4774.
[33] J. Snoek, H. Larochelle, and R. P. Adams, “Practical Bayesian optimization of machine learning algorithms,” in Advances in Neural Information Processing Systems, 2012, pp. 2951–2959.
[34] R. K. Makumbura, L. Mampitiya, N. Rathnayake, and et al., “Advancing water quality assessment and prediction using machine learning models coupled with SHAP based XAI,” Results in Engineering, vol. 23, p. 102831, 2024, doi: 10.1016/j.rineng.2024.102831.
This work is licensed under a Creative Commons Attribution 4.0 International License.
Copyright (c) 2026 José Aldemar Muñoz Hernández, Mara Zeni, Ana Maria Coulon Grisa