Low-cost particulate matter (PM) sensors are promising tools for supplementing existing air quality monitoring networks. However, the performance of the new generation of low-cost PM sensors under field conditions is not well understood. In this study, we characterized the performance capabilities of a new low-cost PM sensor model (Plantower model PMS3003) for measuring PM_2.5 at 1 min, 1 h, 6 h, 12 h, and 24 h integration times. We tested the PMS3003 sensors in both low-concentration suburban regions (Durham and Research Triangle Park (RTP), NC, US) with 1 h PM_2.5 (mean ± SD) of 9±9 and 10±3 µg m⁻³, respectively, and a high-concentration urban location (Kanpur, India) with 1 h PM_2.5 of 36±17 and 116±57 µg m⁻³ during monsoon and post-monsoon seasons, respectively. In Durham and Kanpur, the sensors were compared to a research-grade instrument (environmental β attenuation monitor, E-BAM) to determine how these sensors perform across a range of PM_2.5 concentrations and meteorological factors (e.g., temperature and relative humidity, RH). In RTP, the sensors were compared to three Federal Equivalent Methods (FEMs) including two Teledyne model T640s and a Thermo Scientific model 5030 SHARP to demonstrate the importance of the type of reference monitor selected for sensor calibration. The decrease in 1 h mean errors of the calibrated sensors using univariate linear models from Durham (201 %) to Kanpur monsoon (46 %) and post-monsoon (35 %) seasons showed that PMS3003 performance generally improved as ambient PM_2.5 increased. The precision of reference instruments (T640: ±0.5 µg m⁻³ for 1 h; SHARP: ±2 µg m⁻³ for 24 h, better than the E-BAM) is critical in evaluating sensor performance, and β-attenuation-based monitors may not be ideal for testing PM sensors at low concentrations, as underscored by (1) the less dramatic error reduction over averaging times in RTP against optically based T640 (from 27 % for 1 h to 9 % for 24 h) than in Durham (from 201 % to 15 %); (2) the lower errors in RTP than the Kanpur post-monsoon season (from 35 % to 11 %); and (3) the higher T640–PMS3003 correlations (R² ≥ 0.63) than SHARP–PMS3003 (R² ≥ 0.25). A major RH influence was found in RTP (1 h RH = 64±22 %) due to the relatively high precision of the T640 measurements that can explain up to ∼ 30 % of the variance in 1 min to 6 h PMS3003 PM_2.5 measurements. When proper RH corrections are made by empirical nonlinear equations after using a more precise reference method to calibrate the sensors, our work suggests that the PMS3003 sensors can measure PM_2.5 concentrations within ∼ 10 % of ambient values. We observed that PMS3003 sensors appeared to exhibit a nonlinear response when ambient PM_2.5 exceeded ∼ 125 µg m⁻³ and found that the quadratic fit is more appropriate than the univariate linear model to capture this nonlinearity and can further reduce errors by up to 11 %. Our results have substantial implications for how variability in ambient PM_2.5 concentrations, reference monitor types, and meteorological factors can affect PMS3003 performance characterization.

Wireless low-cost particulate matter sensor networks (WLPMSNs) are transforming air quality monitoring by providing particulate matter (PM) information at finer spatial and temporal resolutions. However, large-scale WLPMSN calibration and maintenance remain a challenge. The manual labor involved in initial calibration by collocation and routine recalibration is intensive. The transferability of the calibration models determined from initial collocation to new deployment sites is questionable, as calibration factors typically vary with the urban heterogeneity of operating conditions and aerosol optical properties. Furthermore, the stability of low-cost sensors can drift or degrade over time. This study presents a simultaneous Gaussian process regression (GPR) and simple linear regression pipeline to calibrate and monitor dense WLPMSNs on the fly by leveraging all available reference monitors across an area without resorting to pre-deployment collocation calibration. We evaluated our method for Delhi, where the PM_2.5 measurements of all 22 regulatory reference and 10 low-cost nodes were available for 59 d from 1 January to 31 March 2018 (PM_2.5 averaged 138±31 µg m⁻³ among 22 reference stations), using a leave-one-out cross-validation (CV) over the 22 reference nodes. We showed that our approach can achieve an overall 30 % prediction error (RMSE: 33 µg m⁻³) at a 24 h scale, and it is robust as it is underscored by the small variability in the GPR model parameters and in the model-produced calibration factors for the low-cost nodes among the 22-fold CV. Of the 22 reference stations, high-quality predictions were observed for those stations whose PM_2.5 means were close to the Delhi-wide mean (i.e., 138±31 µg m⁻³), and relatively poor predictions were observed for those nodes whose means differed substantially from the Delhi-wide mean (particularly on the lower end). We also observed washed-out local variability in PM_2.5 across the 10 low-cost sites after calibration using our approach, which stands in marked contrast to the true wide variability across the reference sites. These observations revealed that our proposed technique (and more generally the geostatistical technique) requires high spatial homogeneity in the pollutant concentrations to be fully effective. We further demonstrated that our algorithm performance is insensitive to training window size as the mean prediction error rate and the standard error of the mean (SEM) for the 22 reference stations remained consistent at ∼30 % and ∼3 %–4 %, respectively, when an increment of 2 d of data was included in the model training. The markedly low requirement of our algorithm for training data enables the models to always be nearly the most updated in the field, thus realizing the algorithm's full potential for dynamically surveilling large-scale WLPMSNs by detecting malfunctioning low-cost nodes and tracking the drift with little latency. Our algorithm presented similarly stable 26 %–34 % mean prediction errors and ∼3 %–7 % SEMs over the sampling period when pre-trained on the current week's data and predicting 1 week ahead, and therefore it is suitable for online calibration. Simulations conducted using our algorithm suggest that in addition to dynamic calibration, the algorithm can also be adapted for automated monitoring of large-scale WLPMSNs. In these simulations, the algorithm was able to differentiate malfunctioning low-cost nodes (due to either hardware failure or under the heavy influence of local sources) within a network by identifying aberrant model-generated calibration factors (i.e., slopes close to zero and intercepts close to the Delhi-wide mean of true PM_2.5). The algorithm was also able to track the drift of low-cost nodes accurately within 4 % error for all the simulation scenarios. The simulation results showed that ∼20 reference stations are optimum for our solution in Delhi and confirmed that low-cost nodes can extend the spatial precision of a network by decreasing the extent of pure interpolation among only reference stations. Our solution has substantial implications in reducing the amount of manual labor for the calibration and surveillance of extensive WLPMSNs, improving the spatial comprehensiveness of PM evaluation, and enhancing the accuracy of WLPMSNs.

Satellite-based rapid sweeping screening of localized PM_2.5 hotspots at fine-scale local neighborhood levels is highly desirable. This motivated us to develop a random forest–convolutional neural network–local contrast normalization (RF–CNN–LCN) pipeline that detects local PM_2.5 hotspots at a 300 m resolution using satellite imagery and meteorological information. The RF–CNN joint model in the pipeline uses three meteorological variables and daily 3 m/pixel resolution PlanetScope satellite imagery to generate daily 300 m ground-level PM_2.5 estimates. The downstream LCN processes the estimated PM_2.5 maps to reveal local PM_2.5 hotspots. The RF–CNN joint model achieved a low normalized root mean square error for PM_2.5 of within ~31% and normalized mean absolute error of within ~19% on the holdout samples in both Delhi and Beijing. The RF–CNN–LCN pipeline reasonably predicts urban PM_2.5 local hotspots and coolspots by capturing both the main intra-urban spatial trends in PM_2.5 and the local variations in PM_2.5 with urban landscape, with local hotspots relating to compact urban spatial structures and coolspots being open areas and green spaces. Based on 20 sampled representative neighborhoods in Delhi, our pipeline revealed an annual average 9.2 ± 4.0 μg m⁻³ difference in PM_2.5 between the local hotspots and coolspots within the same community. In some cases, the differences were much larger; for example, at the Indian Gandhi International Airport, the increase was 20.3 μg m⁻³ from the coolest spot (the residential area immediately outside the airport) to the hottest spot (airport runway). This work provides a possible means of automatically identifying local PM_2.5 hotspots at 300 m in heavily polluted megacities and highlights the potential existence of substantial health inequalities in long-term outdoor PM_2.5 exposures even within the same local neighborhoods between local hotspots and coolspots.

It is challenging to guide neural network (NN) learning with prior knowledge. In contrast, many known properties, such as spatial smoothness or seasonality, are straightforward to model by choosing an appropriate kernel in a Gaussian process (GP). Many deep learning applications could be enhanced by modeling such known properties. For example, convolutional neural networks (CNNs) are frequently used in remote sensing, which is subject to strong seasonal effects. We propose to blend the strengths of deep learning and the clear modeling capabilities of GPs by using a composite kernel that combines a kernel implicitly defined by a neural network with a second kernel function chosen to model known properties (e.g., seasonality). We implement this idea by combining a deep network and an efficient mapping based on the Nystrom approximation, which we call Implicit Composite Kernel (ICK). We then adopt a sample-then-optimize approach to approximate the full GP posterior distribution. We demonstrate that ICK has superior performance and flexibility on both synthetic and real-world data sets. We believe that ICK framework can be used to include prior information into neural networks in many applications.