Laboratory: IMT Nord Europe, CERI Systèmes Numériques, Lille, France
Duration: 6 months
Supervisors: Christelle Garnier and Anne Savard
Context:
The water quality in watersheds and rivers is a crucial issue for the environment, human health, and economic development. The main sources of water pollution are varied: agricultural activities, industrial discharges, urban wastewater, climate change… Chemicals such as pesticides and fertilizers, or microplastics can accumulate in rivers, threatening aquatic biodiversity and the health of people who depend on these water resources.
Monitoring river water quality through a network of measurement points enables communities and institutions to develop efficient, sustainable management strategies and ensure safe water for both people and ecosystems. Various data can be used to measure water quality, such as turbidity, dissolved oxygen, chlorophyll, sediment concentrations… Among these data, turbidity caused by suspended particles is considered as the most relevant indicator for water quality. High turbidity reduces light penetration, harms aquatic plants and animals, and indicates the presence of pollutants or sediments.
Accurately forecasting turbidity and other water-quality indicators is crucial for informed decision-making, risk prevention, water resource management, and climate-change adaptation, as it enables accurate anomaly detection, i.e. unusual events causing physical-chemical imbalance and increase in pollutant concentration.
Objectives:
The main purpose of this internship is to develop novel machine learning-based forecasting techniques to detect anomalies in water quality time series data focusing on the following two steps:
1) Dataset consolidation: The goal is to exploit all available measurement sources on a watershed to enhance data quality, correct potential biases, and fill in missing values.
2) Time series forecasting: The goal is to implement and compare several time series forecasting approaches, such as pattern-based prediction like Facebook Prophet, and neural network-based models, as in our previous works [1, 2] on air quality forecasting. Comparisons will be multi-criteria, emphasizing reliability enhancement, computational complexity and confidence quantification of the predicted values.
How to apply:
The applicants should have a strong background in either data science or applied mathematics. Interested candidates have to send their detailed CV, academic records, at least one academic referee and a short motivation letter via email to the contacts below.
Applications will be considered as they arrive, therefore early application is highly encouraged.
Contacts:
Anne Savard
Email: anne.savard@imt-nord-europe.fr
Webpage: https://recherche.imt-nord-europe.fr/personnel/savard-anne/
Christelle Garnier
Email: christelle.garnier@imt-nord-europe.fr
Webpage: https://recherche.imt-nord-europe.fr/personnel/garnier-christelle/
References:
[1] S. Masmoudi, C. Garnier, A. Savard, V. Itier, S. Sauvage, F. Bulot, P. Kaluzny, « Enhanced sensor environment graph based deep learning approach for air quality anomaly detection », EUSIPCO, Lyon, France, 2024.
[2] S. Masmoudi, A. Savard, C. Garnier, V. Itier, P. Kaluzny, S. Sauvage, « Graph-based air quality anomaly detection: seasonality effects », to be submitted to IEEE Sensors, 2025.
