Search

Low-cost sensors (LCSs) used for measuring air quality have become popular because of their portability, affordability, and ease of operation. However, LCS data often have accuracy and bias issues that need to be addressed before using them for research. LCSs are, therefore, collocated with reference-grade instruments, and various statistical and machine learning (ML) approaches are used to correct the observed bias in data. In this study, collocation experiments were conducted in Bengaluru, India, for about nine months (December 2021 to August 2022). We used nine PM2.5 LCSs that were collocated with a beta attenuation monitor (BAM), which is certified by the United States Environmental Protection Agency (USEPA). Hourly averaged data from LCSs and BAM were used to train various ML correction models. The LCSs included in the study—Airveda, Atmos, Prana Air, BlueSky, Aurassure, Aerogram, PurpleAir, and Prkruti—are widely available in the Indian market. The ML models include support vector regression (SVR), decision tree (DT), random forest (RF), and eXtreme gradient boosting (XGBoost). For the LCSs used in the study, a total of 170 ML models were built to identify the best-performing correction model for each sensor. Model performances were evaluated based on the following metrics: mean absolute error (MAE), root mean square error (RMSE), and normalised RMSE (NRMSE). During the study period, the average hourly BAM concentration was ~32 µg/m3. Hourly averaged PM2.5 from LCSs and BAM exhibited a linear relationship. The NRMSE values of the raw (uncorrected) LCSs PM2.5 with respect to BAM PM2.5 varied between 0.26 and 0.89 across various sensors. The Plantower-based LCSs (Atmos I, PurpleAir, and Aerogram) performed better, characterised by the lowest RMSE/NRMSE values. SVR was found to be the best-performing model for most of the sensors in correcting raw LCSs PM2.5 data. The NRMSE of the ML models’ corrected LCSs PM2.5 was reduced by 46% to 74% across various sensors compared to the uncorrected LCSs PM2.5. As a case study, we also added black carbon (BC) data to our ML models, but no significant change (improvement by 6% RMSE) in performance was observed.

High-resolution spatial maps of air pollution can be useful for air quality management. In low- and middle-income countries, regulatory measurements of criteria pollutants are typically insufficient to generate spatial maps, due to the sparsely located monitoring stations. Alternatively, high-resolution spatial maps of air pollution can be achieved by dispersion (physics-based) and statistical regression (training-based) modelling. Resolutions of up to ~25 meters can be achieved by Land Use Regression (LUR) modelling based spatial predictions. In this study, we leveraged the high density of regulatory monitors located in New Delhi, India, and developed LUR models for all the major criteria pollutants (PM10, PM2.5, SO2, NO2, and Ozone). New Delhi is one of the most heavily polluted cities in the world. We used data from 40 continuous ambient air quality monitoring stations’ (CAAQMS) for the year 2019 to develop seasonal and annual LUR models, following the ESCAPE (European Study of Cohorts for Air Pollution Effects) stepwise supervised regression method. Model predictors included land use parameters, road lengths, rail track lengths, population, satellite pollution, and NDVI data, along with air pollution point source location data and reanalysis meteorology. The models were validated using leave one station out (LOSO) and 10-fold cross validations (CV). The model adjusted R2 values varied between 0.08 and 0.64. Particle pollutant models (PM2.5 and PM10) performed better than those of gaseous pollutants. Further, ozone models performed the least. Across seasons, summer models performed the best (least) for PM (gaseous pollutants). Models with adjusted R2 were used for spatial predictions at 50-m resolution for the Delhi National Capital Territory region. Spatio-seasonal characteristics of air pollution were studied using the generated high-resolution maps.

We report on the long-term performance of a popular low-cost PM2.5 sensor, the PurpleAir PA-II, at multiple sites in India, with the aim of identifying robust calibration protocols. We established 3 distinct sites in India (North India: Delhi, Hamirpur; South India: Bangalore), where we collocated PA-II with reference beta-attenuation monitors to characterize sensor performance and to model calibration relationships between PA-IIs and reference monitors for hourly data. Our sites remained in operation across all major seasons of India. Without calibration, the PA-IIs had high precision (Normalized Root Mean Square Error [NRMSE] among replicate sensors ≤ 15 %) and tracked the overall seasonal and diurnal signals from the reference instruments well (Pearson’s r ≥ 0.9) but were inaccurate (NRMSE ≥ 40 %). We used a comprehensive feature selection process to create optimized site-specific calibrations. Relative to the uncalibrated data, parsimonious least-squares long-term calibration models improved PA-II performance at all sites (cross-validated NRMSE: 20–30 %, R2: 0.82–0.95), particularly by reducing seasonal and diurnal biases. Because aerosol properties and meteorology vary regionally, the form of these long-term models differed by site. Likewise, using a moving-window calibration, we find a calibration scheme using seasonally specific information somewhat improves performance relative to a static long-term calibration model. In contrast, we demonstrate that a successful short-term calibration exercise for one season may not transfer reliably to other seasons. Overall, we demonstrate how the PA-II, when paired with a careful calibration scheme, can provide actionable information on PM2.5 in India with only modest irreducible uncertainty.

Mobile monitoring can supplement regulatory measurements, particularly in low-income countries where stationary monitors’ density is low. Here, we report results from a ~year-long mobile monitoring campaign of on-road black carbon (BC) mass concentration, ultrafine particle (UFP) number concentration, and carbon dioxide (CO2) in Bengaluru, India. The study route included ~150 unique kms covering urban and peri-urban residential neighborhoods and the central business district (CBD); ~22 repeat measurements were made per monitored road-segment, covering most seasons. After cleaning the data for known instrument artifacts and sensitivities, we generated 30 m high-resolution stable ‘data only’ spatial maps of BC, UFP, and CO2 for the study route. For the urban residential areas, the mean BC levels for residential roads, arterials, and highways were ~10, 22, and 56 µgm-³, respectively. A similar pattern (highways being characterized by highest pollution levels) was also observed for the UFP and CO2. Using the data from repeat measurements, we carried out a Monte Carlo subsampling analysis to understand the minimum number of repeat measures to generate stable maps of pollution in the city. Leveraging the simultaneous nature of the measurements, we also mapped the quasi-emission factors (QEF) of the pollutants under investigation. Finally, the results are discussed in the context of technical aspects of the campaign, limitations, and their policy relevance.