Estimating PM2.5 from photographs. (January 2020)

• Record Type: Journal Article
• Authors: Pudasaini, Batsal; Kanaparthi, Mark; Scrimgeour, Jan; Banerjee, Natasha; Mondal, Sumona; Skufca, Joseph; Dhaniyala, Suresh
• Is Part Of: Atmospheric environment. Volume 5(2020)
• Publication Date: 2020-01
• DOI: 10.1016/j.aeaoa.2020.100063
• Languages: English
• ISSNs: 2590-1621

Abstract

Abstract: Accurate determination of the impact of aerosol particles on human health is made challenging by the lack of high-resolution air quality data. Globally, PM2.5 monitoring sites are limited in number, as they require extensive manpower and equipment to operate. Here, we take a physics-based approach to estimate PM2.5 from analysis of photographs at different locations. We develop a governing equation that relates camera signal to the properties of aerosol, the incident light, and the object being imaged. From inversion of this integral equation, we establish an expression for turbidity. Analyzing 3-years of images captured from a camera at a fixed location (downtown Chicago), we calculate turbidity and compare the values against PM2.5 values estimated from nearby EPA monitoring sites. Our calculated photograph-based turbidity values are seen to have a statistically significant positive linear correlation to the estimated PM2.5 values with p-value less than 0.01 and an average binned R 2 greater than 0.50. We show that calculation of turbidity is possible from a single photograph, when it contains objects with the same albedo and that are identically illuminated. Our analysis suggests that camera-based PM monitoring could be a relatively low-cost, alternative approach to obtain global air quality data. Highlights: A physics-based model to find the relationship between pixel intensity & air quality. Analysis of 3 years of data to compare observations with theoretical predictions. Mean PM2.5 correlates well with turbidity calculated relative to a clean day picture. Established an approach to determined turbidity from a single photograph. Need data from additional sites for full validation of the technique.