Purple air sensors are low-cost sensors which measure particulate matter (PM) and are being extensively utilized by a wide group of people including individuals, local communities as well as state and local air quality monitoring agencies. PurpleAir sensors comprise of a massive global network. Previous performance was mostly short-term and covereda limited number of PurpleAir sensors in smaller geographic areas. This study is extensive covering the state of California comprising of a huge number of sensors (give count of sensors) and for a longer time period of 4 years (2016-2020). PurpleAir sensors overpredict aerosol concentrations for both short term and long-term in case of daily averaged PM2.5 concentrations but importantly do capture the high peaks and temporal variability of long-term PM2.5 concentrations. In total 4000 collocated sensors over the years (2016-2020) were considered approximately for the whole of State of California and quality checked against Federal Reference Method (FRM) monitors.
Geo-statistical interpolation techniques such as InverseDistance Weight (IDW) and Kriging have been applied to predict PM2.5 concentrations at reference monitors using sensor data and the t test shows significant differences between interpolated and reference values for long term daily averaged PM2.5 concentrations because of over predictions. The RMSE varies between 8 and 16 µg/m3 and mean normalized bias (MNB) varied between 6 to 23 percent consistent with previous studies. With regression model predicted output RMSE varied between 3 to 10 µg/m3. Short-tern wild fire episodes were evaluated in this study. It is seen sensors do a better job of capturing the episode peaks of concentrations.
This study aims to provide satisfactory estimations of PM2.5 concentrations with high spatiotemporal resolutions based on statistical models critical at community levels. This can play a significant role in air pollution monitoring in some developing countries which have high aerosol concentrations and data is sparse at local levels since high values are captured well by the sensors as seen in this study.
.jpg)
