Low-Cost Air Quality Sensor Evaluation and Calibration in Contrasting Aerosol Environments
Low-Cost Air Quality Sensor Evaluation and Calibration in Contrasting Aerosol Environments
复制标题
对比气溶胶环境中的低成本空气质量传感器评估和校准
DOI：
--
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
K. Ganesan
中科院分区：
文献类型：
--
作者：
Pawan Gupta;P. Doraiswamy;Jashwanth Reddy;Palak Balyan;S. Dey;R. Chartier;Adeel Khan;Karmann Riter;B. Feenstra;R. Levy;Nhu Nguyen Minh Tran;O. Pikelnaya;Kurinji L. Selvaraj;T. Ganguly;K. Ganesan
.在空气质量监测中使用低成本传感器（LCS）已经引起了社会各界的兴趣，包括社区和公民科学家，学术研究团体，环境机构和私营部门。由监管机构执行的传统空气监测涉及昂贵的监管级设备，并需要持续的维护和质量控制检查。低价格标签，最小的运营成本，易用性和开放的数据访问是LCS流行背后的主要驱动因素。本研究讨论了PM 2.5传感器在监测空气质量中的作用和相关挑战。我们提出的PurpleAir（PA.）的评价结果。PA-II LCS针对法规级PM 2.5联邦等效方法（FEM）和传感器校准算法的开发。LCS校准于2019年12月至2020年1月在北卡罗来纳州罗利和印度德里进行了2至4周，以评估不同气溶胶负载和环境条件下的数据质量。本练习旨在开发一个强大的校准模型，该模型使用PA测量参数（即，PM 2.5、温度、相对湿度）作为输入，并以每小时的规模提供偏差校正后的30个PM 2.5输出。因此，在校准模型开发过程中，校准模型依赖于通过FEM对PM 2.5的同时测量作为目标输出。我们应用各种统计和机器学习方法来实现区域校准模型。从我们的研究结果表明，通过适当的校准，我们可以实现偏差校正的PM 2.5数据使用PA传感器在12%的百分比内平均绝对偏差在每小时和6%内的日平均值。我们的研究还表明，应在当地或区域范围内对PA 35传感器进行部署前校准，以纠正来自现场的数据，用于科学数据分析。操作。Wi-Fi配置并注册到PurpleAir，这是一个传感器报告的工厂指定的传感器和路由器的电源，需要单独控制。路由器在使用前充满电。为了避免潜在的电池损坏和火灾风险，路由器的电源在充满电后关闭，并在大约8小时后手动重新打开。将95个传感器沿着与Wi-Fi热点单元一起设置是一个挑战。由于路由器和/或延长线的问题，40个传感器的数据不可用或大部分丢失。因此，本研究使用了其余55个探头的数据。两个区域对FEM的不同性能，主要是由于PM 2.5负载、颗粒类型和操作条件的差异。罗利的典型PM 2.5 20值小于20 µg m-3，这在德里很少观察到。在罗利进行的PA与FEM比较显示，每小时平均值存在较大的分散性，R值为0.34，RMSE为4.4 µg m-3。在对24小时内的数据进行平均之后（即，日平均值），相关性几乎增加了一倍（R = 0.66），RMSE减少了一半（2.2 µg m-3）。罗利的平均偏倚仍为负值，并且在小时和日平均基础上大致相同（~ -0.8 µg m-3或约5%至11%），表明PA在清洁条件下（PM 2.5 < 10 µg m-3）总体低估。相比之下，25德里的PA传感器经常高估每小时和每日平均PM 2.5浓度，平均偏差为正（~35至37 µg m 3，或23.8%至23.7%）。德里的PA测量结果与FEM显示出非常高的相关性（R ≥0.88），但每小时和每天的RMSE分别为60.75 µg m-3和48.13 µg m-3。因此，PA与FEM的比较证明了德里（高估但高度相关）和罗利（低估和低相关）的完全不同的传感器行为。这表明需要不同的校正系数或模型来校正不同PM 2.5负荷下的30 PA数据。同样重要的是要注意，德里和罗利的颗粒物化学成分预计会有所不同。在冬季，德里的颗粒物主要由含碳气溶胶和灰尘混合物组成（Shiva Nambra和Khare，2019），而罗利的PM主要由典型的城市硫酸盐和硝酸盐气溶胶组成（Cheng和Wang-Li，2019）。0.96至1.01小时平均值和0.78至0.93的R2，每日平均值的比率为0.99至1使用训练指标作为等效比较点，利用将RH和T效应R2合并为0.94（罗利）至0.96（印度）的平均偏差为4.2±18.9%（罗利），或比率（校正/F.E.M.）1.04±0.19）2.0±13.2%（印度，或比率为1.02±0.13），用于每小时的培训部分，（2018年）
. The use of low-cost sensors (LCS) in air quality monitoring has been gaining interest across all walks of society, including community and citizen scientists, academic research groups, environmental agencies, and the private sector. Traditional air monitoring, performed by regulatory agencies, involves expensive regulatory-grade equipment and requires ongoing maintenance and quality control checks. The low-price tag, minimal operating cost, ease of use, and open data access are the primary driving factors behind the popularity of LCS. This study discusses the role and associated challenges of PM 2.5 25 sensors in monitoring air quality. We present the results of evaluations of the PurpleAir (PA.) PA-II LCS against regulatory-grade PM 2.5 federal equivalent methods (FEM) and the development of sensor calibration algorithms. The LCS calibration was performed for 2 to 4 weeks during December 2019-January 2020 in Raleigh, NC, and Delhi, India, to evaluate the data quality under different aerosols loadings and environmental conditions. This exercise aims to develop a robust calibration model that uses PA measured parameters (i.e., PM 2.5 , temperature, relative humidity) as input and provides bias-corrected 30 PM 2.5 output at an hourly scale. Thus, the calibration model relies on simultaneous measurements of PM 2.5 by FEM as target output during the calibration model development process. We applied various statistical and machine learning methods to achieve a regional calibration model. The results from our study indicate that, with proper calibration, we can achieve bias-corrected PM 2.5 data using PA sensors within 12% percentage mean absolute bias at hourly and within 6% for a daily average. Our study also suggests that pre-deployment calibrations developed at local or regional scales should be performed for the PA 35 sensors to correct data from the field for scientific data analysis. operational. Wi-Fi configured and registered with PurpleAir, a sensor-reported factory-specified the power to the sensors and the routers to be controlled separately. The routers were fully charged prior to use. To avoid potential battery damage and fire risk, the power to the routers was turned off once fully charged and manually turned back on after about 8 hours. Setting up a collocation of 95 sensors along with Wi-Fi hotspot units was a challenge. Due to issues with the router and/or the extension cords, data were either unavailable or largely missing for 40 sensors. Thus, data from the remaining 55 sensors were used in this study. different performances against FEM in the two regions, mainly due to differences in PM 2.5 loading, particle type, and operating conditions. The typical PM 2.5 20 values in Raleigh were less than 20 µg m -3 which were rarely observed in Delhi. The PA vs. FEM comparison in Raleigh showed a big scatter in hourly averages with a R value of 0.34 and RMSE of 4.4 µg m -3 . After averaging data over a 24-hour period (i.e., daily average), the correlation almost doubled (R = 0.66), and RMSE was reduced by half (2.2 µg m -3 ). The mean bias for Raleigh remained negative and about the same (~ -0.8 µg m -3 or about 5% to 11%) on both hourly and daily average basis, suggesting an overall underestimation by PA in clean conditions (PM 2.5 < 10 µg m -3 ). In contrast, the PA sensors in 25 Delhi often overestimated PM 2.5 concentrations at both hourly and daily averages with a positive mean bias (~35 to 37 µg m 3 , or 23.8% to 23.7%). PA measurements in Delhi showed a very high degree of correlation (R ≥0.88) with FEM but with a high RMSE of 60.75 µg m -3 and 48.13 µg m -3 on an hourly and daily basis, respectively. Thus, the comparison of PA with FEM demonstrated completely different sensor behavior for Delhi (overestimation but highly correlated) and Raleigh (underestimation and low correlation). This suggests the need for different calibration coefficients or models for correcting 30 PA data under different PM 2.5 loadings. It is also important to note that the chemical composition of particles in Delhi and Raleigh is expected to be different. The Delhi particles are dominated by carbonaceous aerosols with a mixture of dust during the winter period (Shiva Nagendra and Khare, 2019), whereas PM in Raleigh is dominated by typical urban sulfate and nitrate aerosols (Cheng and Wang-Li, 2019). of 0.96 to1.01 hourly average and R 2 of 0.78 to 0.93 with ratios of 0.99 to 1 for daily average Using metrics for the training as an equivalent point of comparison, utilizing an that incorporates RH and T effects R 2 of 0.94 (Raleigh) to 0.96 (India) a mean bias of 4.2±18.9% (Raleigh, or a ratio (corrected/F.E.M.) 1.04±0.19) 2.0±13.2% (India, or a ratio of 1.02±0.13) for training portion of hourly the et al., (2018)
影响因子：
8.9
作者：
Kelly, K. E.;Whitaker, J.;Petty, A.;Widmer, C.;Dybwad, A.;Sleeth, D.;Martin, R.;Butterfield, A.
通讯作者：
Butterfield, A.
影响因子：
8.3
作者：
Bi, Jianzhao;Stowell, Jennifer;Liu, Yang
通讯作者：
Liu, Yang
影响因子：
11.8
作者：
Zusman, Marina;Schumacher, Cooper S.;Sheppard, Lianne
通讯作者：
Sheppard, Lianne