10th International Aerosol Conference September 2 - September 7, 2018 America's Center Convention Complex St. Louis, Missouri, USA
Abstract View
Evaluating the Transferability of Low-Cost Sensor Calibration between Different Regions and Spatial and Temporal Variation in Air Quality in Hartford, CT
Kyle Terracciano, Fatema Parvez, Carl Malings, Rebecca Tanzer, R. Subramanian, KRISTINA WAGSTROM, University of Connecticut
Abstract Number: 1531 Working Group: Low-Cost and Portable Sensors
Abstract One of the biggest challenges facing researchers hoping to deploy low-cost monitors is developing reliable calibration approaches for many of these technologies. Zimmerman et al. (2018) present a novel approach applying machine learning to calibrate the Real-time Affordable Multi-Pollutant (RAMP) sensor package for CO, NO2, O3, and SO2. The RAMP uses electrochemical sensors for each of the gases. This calibration approach relies on data collected when monitors are collocated with reference monitors to develop a calibration model using random forests (RF). Each RAMP uses a PurpleAir or Met-One Neighborhood PM monitor for fine particulate mass (PM2.5).
In this work, we evaluate the transferability of the calibration model developed in Pittsburgh, Pennsylvania for use in a different, but meteorologically similar, region - central Connecticut. It is imperative that we determine how well a calibration model developed in one area performs when moved to another region. For this test, we have selected two locations with fairly similar meteorology: southwestern Pennsylvania and central Connecticut. We will share the results of a three week collocation test in Hartford, CT using three RAMP monitors calibrated in Pittsburgh, PA using a random forest model hybridized with a multiple linear regression (MLR) model for higher concentrations. The RAMP data will be compared with data from a local regulatory monitoring station (maintained by the Connecticut Department of Energy and Environmental Protection) to determine the performance of the Pittsburgh calibration. We will also share the results of an additional three week study with the monitors distributed over three sites in a Central Connecticut, to examine spatial and temporal variability in air quality around Hartford.