Understanding the Long-term Performance and the Reject Counts for the Alphasense OPC-N3

TRISTALEE MANGIN, Kamaljeet Kaur, Nancy Daher, Kerry Kelly, University of Utah

     Abstract Number: 96
     Working Group: Instrumentation and Methods

Abstract
As climate change leads to more arid landscapes, interest in windblown dust and particulate matter 10 microns in diameter and smaller (PM10) is growing. The drying of the Great Salt Lake (GSL) is a prime example of how water consumption and increasing aridity have led to historically low lake levels and increasing PM10 levels. Unlike PM2.5, PM10 particles settle quickly, making it difficult to obtain representative PM10 concentration measurements; yet, regulatory measurements for PM10 are generally sparser than for PM2.5. To address this gap, we deployed a network of low-cost air quality devices using Alphasense OPC-N3 to measure PM10 in communities near the GSL. In February 2024, we began co-locating three sensors, each containing an OPC-N3, with a Thermo Scientific Model 5030 SHARP w/ VSCC beta attenuation monitor at a Division of Air Quality monitoring station. This study evaluates how relative humidity (RH) affects OPC-N3 performance in a region and strategies for developing effective correction models (including multiple linear regression (MLR), k-Kohler fit, and machine learning approaches). Finally, we discuss how to interpret the reject count parameters provided by the OPC-N3 and how environmental factors (like RH and wind speed) affect these parameters. Preliminary findings show that an MLR model with concentration and RH parameters resulted in correlation, R2, of 0.73, 0.65, and 0.20, Spring, Summer, and Winter, respectively. The k-Kohler model fitting yielded similar R2 values for Summer and Spring (0.80, 0.66) but performed better for Winter (0.57). During Winter, the higher RH complicates the OPC corrections for both models. If we set a threshold on RH at <80%, both the MLR and k-Kohler models resulted in Winter R2 values of 0.67 and 0.66. The machine learning models are still in development, and the analysis of the reject count parameters is still ongoing.