Low-cost sensor networks and land-use regression: interpolating nitrogen dioxide concentration at high temporal and spatial resolution in Southern California

TL;DR

This study leverages low-cost sensors and land-use regression with machine learning to estimate high-resolution nitrogen dioxide levels in Southern California, enhancing air quality monitoring capabilities.

Contribution

It introduces a novel approach combining low-cost sensor data with random forest models for high-resolution, hourly air quality estimation and deviation detection.

Findings

Effective high-resolution nitrogen dioxide mapping achieved

Deviations identify specific conditions affecting air quality

Model aligns with general expectations of pollution patterns

Abstract

The development of low-cost sensors and novel calibration algorithms offer new opportunities to supplement existing regulatory networks to measure air pollutants at a high spatial resolution and at hourly and sub-hourly timescales. We use a random forest model on data from a network of low-cost sensors to describe the effect of land use features on local-scale air quality, extend this model to describe the hourly-scale variation of air quality at high spatial resolution, and show that deviations from the model can be used to identify particular conditions and locations where air quality differs from the expected land-use effect. The conditions and locations under which deviations were detected conform to expectations based on general experience.