Research scope:
Recent research reveals that the potent greenhouse gas (GHG) and ozone depleting substance nitrous oxide (N2O) dominates GHG emissions from
biological wastewater treatment. The development of effective operational strategies to reduce or avoid N2O generation requires in-depth understanding of
the N2O formation mechanisms.
This internship involves advanced data analysis using machine-learning (ML) techniques on long-term datasets from full-scale biological wastewater
treatment plants, with the goal of unravelling hidden relationships between operating conditions and N2O emissions.
Work plan:
Starting material: Clear overview of the existing data and its preprocessing a specific WWTP
Following tasks will be carried out during the internship (20 weeks – longer or shorter period is possible (min. 16 weeks)):
1. Getting familiar with the available data (2 weeks)
2. Screening analysis of ML models (4 weeks)
TASKS: (1) Creation of benchmark model (able to predict average N2O emissions), (2) generation of different ML models, (3) benchmarking based on
predictive accuracy, (4) keep two most promising models.
OUTPUT: Insights whether the dataset provides enough information for the estimation of N2O formation.
3. Optimization and training two best ML models (4 weeks)
TASKS: Optimization and training of the two most promising ML model.
OUTPUT: Two ML models which quantitatively represent the N2O emissions.
4. First insights in explainable machine learning (8 weeks)
TASKS: (1) Review of XAI tools and possibilities (2) Employ XAI tools: understand predictions of the ML models and if possible translate into mechanistic
building blocks
OUTPUT: First insights in using ML to support mechanistic N2O model building.
5. Reporting and buffer (2 weeks)
You will be supervised on a daily basis by a postdoctoral student. A workplace with a computer will be provided in Dübendorf.