E-Locus - Institutional Repository of the University of Crete

Home Collections Type of Work Post-graduate theses

Post-graduate theses

Search command : Author="Στεφανίδης" And Author="Κωνσταντίνος"

Current Record: 4 of 6610

[Add to Basket]

Identifier

000465799

Title

Homogenization of the network of air pollution observations in Crete

Alternative Title

Ομογενοποίηση του δικτύου μετρήσεων ατμοσφαιρικής ρύπανσης στην Κρήτη

Author

Παρασκευαΐδου, Σοφία Β.

Thesis advisor

Κανακίδου, Μαρία

Reviewer

Καλυβίτης, Νικόλαος
Χρηστάκης, Νικόλαος

Abstract

Air quality monitoring is critical in addressing environmental and public health issues. Low-cost microsensors offer a potential, cost-effective method for measuring particulate matter (PM) and other hazardous atmospheric pollutants. However, their effectiveness is heavily influenced by environmental factors such as high humidity and soil dust occurrences, requiring precise field and laboratory evaluation of their performance. This thesis studies the use of low-cost microsensors, specifically the commercially available Bettair, Kunak, and PurpleAir devices, for atmospheric pollutant monitoring, emphasizing the importance of calibration adapted to individual settings. To improve sensor accuracy, we created computational modifications utilizing Machine Learning (ML) methods, which turned out to be the most effective calibration method. After treating the data and evaluating their correlations (Pearson and Spearman’s Rank Correlation Coefficients) we started applying the ML algorithms. The models we ultimately used were the XGBoost (EXtreme Gradient Boosting) and Random Forest (RF) regressors, which are supervised learning algorithms. From this study, we concluded that the RF regressor was the one giving better predictions, after training with each microsensor’s corresponding measurements, for the concentrations values from the reference station. However, in order to have an accurate and efficient model, large amount of data is needed and with more features. Our findings from both field trials and statistical analysis show that machine learning- based calibration greatly enhances sensor performance, making low-cost microsensors a semi-viable alternative for monitoring air quality across a wide range of environments.

Language

English

Subject

Air Quality networks

Calibration

Machine learning models

Microsensors

PM particles

Δίκτυα ατμοσφαιρικού ελέγχου

Μικροαισθητήρες

Μοντέλα μηχανικής μάθησης

Ρύθμιση/προσαρμογή