You are here:
Impacts of Data Completeness on Hourly Averaged PurpleAir PM2.5 Concentrations During Smoke Events
Citation:
Frederick, S., K. Johnson, A. Holder, AND A. Clements. Impacts of Data Completeness on Hourly Averaged PurpleAir PM2.5 Concentrations During Smoke Events. American Association of Aerosol Research (AAAR) Annual Conference, Research Triangle Park, NC, October 06 - 09, 2020.
Impact/Purpose:
Amid growing concern about air quality due to prevalent wildfire events in recent years, air sensors have experienced increasing use across the United States. Despite rising interest in this technology, its adoption remains limited due to uncertainty in measurement and variation in the quality of data produced, including reductions to the completeness of continuous pollutant measurement due events which interrupt data logging. This work determines the impact of reduced data completeness for hourly averaged PurpleAir PM2.5 measurements recorded during wildfire conditions and investigates AQI category prediction for sensor datasets simulating lowered completeness using the NowCast calculation for hourly averages. These results are important for ensuring the quality and practicability of sensor data as users develop methods for data cleaning and analysis to ensure valuable data recorded during wildfire events can be utilized to the greatest extent.
Description:
Air sensors can provide continuous PM measurements advantageous for rapidly changing concentrations due to wildfires. However, there may be increased uncertainty in sensor data due to interruptions in sensor connectivity or power. This work simulates the impact of data gaps in a smoke impacted PurpleAir PM2.5 dataset. Data were collected during August 2018 at the Natchez wildfire in Northern California where a PurpleAir was deployed alongside an E-BAM. The PurpleAir logged data to an SD card every 80 seconds while the E-BAM recorded 1-hr averages. To simulate lowered completeness versions of the PurpleAir dataset, subsets within each 1-hr interval (e.g. 50% completeness = 22 80s points/hour) were selected either at random or through an iterative process. Each hourly subset was averaged and corrected with a USEPA correction equation and compared to both the full PurpleAir and E-BAM hourly averages. The NowCast AQI, a rolling 12-hr weighted average for PM2.5 developed by USEPA, was calculated for datasets simulating ‘worst-case’ completeness (1 point/hour) alongside NowCast AQI values for the full PurpleAir dataset and E-BAM hourly averages. Estimations of NowCast AQI category for the ‘worst-case’ scenario with one data point sampled each hour indicate that the PurpleAir reports within one category of the E-BAM ~90% of the time. These results are important as sensor users develop methods for sensor data cleaning and analysis since applying overly strict completeness criteria could exclude limited data available during smoke events, while applying insufficient completeness criteria could provide misleading results.