Distinguishing Sensor Errors from Environmental Events: A Spatio-temporal Analysis of Outlier Detection During Wildfire Pollution in Athens

Identificador:  imarina:9633053

Handle: https://hdl.handle.net/20.500.11797/imarina9633053

Autores:  Zafeirelli S; Batsaris M; Roussou O; Sigró J; Kavroudakis D

Resumen:
Low-cost environmental sensors in smart cities play a critical role in monitoring the environment, offering real-time information for urban management. The reliability of smart sensors remains uncertain, since sensors may report outliers when malfunctioning, or due to anomalies in the environment or extreme occurrences, which might skew the analysis if not treated carefully. The study seeks to support the distinction between likely sensor anomalies and spatially coherent environmental events by comparing different outlier detection methods: Interquartile Range (IQR), Local Outlier, and Global Outlier. The IQR method identifies temporal outliers based on historical data, whereas Local and Global methods use the spatial dimension, calculating the deviations from the local and global averages, respectively. During wildfire incidents near Athens, Greece, in August 2021, the methods were applied on environmental data from the PurpleAir sensor platform, which measures PM1.0, PM2.5, PM10.0, temperature and relative humidity. The IQR approach performed well in depicting short-term pollution peaks temporally associated with the wildfire period. The Local Outlier approach identifies a higher rate of local extreme values, thus suggesting sensitivity to localized environmental variability, while the Global Outlier method is more appropriate for widespread events.