A nested machine learning approach to short-term PM2.5 prediction in metropolitan areas using PM2.5 data from different sensor networks.

Many predictive models for ambient PM2.5 concentrations rely on ground observations from a single monitoring network consisting of sparsely distributed sensors. Integrating data from multiple sensor networks for short-term PM2.5 prediction remains largely unexplored. This paper presents a machine learning approach to predict ambient PM2.5 concentration levels at any unmonitored location several hours ahead using PM2.5 observations from nearby monitoring sites from two sensor networks and the location's social and environmental properties. Specifically, this approach first applies a Graph Neural Network and Long Short-Term Memory (GNN-LSTM) network to time series of daily observations from a regulatory monitoring network to make predictions of PM2.5. This network produces feature vectors to store aggregated daily observations as well as dependency characteristics to predict daily PM2.5. The daily feature vectors are then set as the precondition of the hourly level learning process. The hourly level learning again uses a GNN-LSTM network based on daily dependency information and hourly observations from a low-cost sensor network to produce spatiotemporal feature vectors capturing the combined dependency described by daily and hourly observations. Finally, the spatiotemporal feature vectors from the hourly learning process and social-environmental data are merged and used as the input to a single-layer Fully Connected (FC) network to output the predicted hourly PM2.5 concentrations. To demonstrate the benefits of this novel prediction approach, we have conducted a case study using data collected from two sensor networks in Denver, CO, during 2021. Results show that the utilization of data from two sensor networks improves the overall performance of predicting fine-level, short-term PM2.5 concentrations compared to other baseline models.

Characterization of aerosol plumes from singing and playing wind instruments associated with the risk of airborne virus transmission.

The exhalation of aerosols during musical performances or rehearsals posed a risk of airborne virus transmission in the COVID-19 pandemic. Previous research studied aerosol plumes by only focusing on one risk factor, either the source strength or convective transport capability. Furthermore, the source strength was characterized by the aerosol concentration and ignored the airflow rate needed for risk analysis in actual musical performances. This study characterizes aerosol plumes that account for both the source strength and convective transport capability by conducting experiments with 18 human subjects. The source strength was characterized by the source aerosol emission rate, defined as the source aerosol concentration multiplied by the source airflow rate (brass 383 particle/s, singing 408 particle/s, and woodwind 480 particle/s). The convective transport capability was characterized by the plume influence distance, defined as the sum of the horizontal jet length and horizontal instrument length (brass 0.6 m, singing 0.6 m and woodwind 0.8 m). Results indicate that woodwind instruments produced the highest risk with approximately 20% higher source aerosol emission rates and 30% higher plume influence distances compared with the average of the same risk indicators for singing and brass instruments. Interestingly, the clarinet performance produced moderate source aerosol concentrations at the instrument's bell, but had the highest source aerosol emission rates due to high source airflow rates. Flute performance generated plumes with the lowest source aerosol emission rates but the highest plume influence distances due to the highest source airflow rate. Notably, these comprehensive results show that the source airflow is a critical component of the risk of airborne disease transmission. The effectiveness of masking and bell covering in reducing aerosol transmission is due to the mitigation of both source aerosol concentrations and plume influence distances. This study also found a musician who generated approximately five times more source aerosol concentrations than those of the other musicians who played the same instrument. Despite voice and brass instruments producing measurably lower average risk, it is possible to have an individual musician produce aerosol plumes with high source strength, resulting in enhanced transmission risk; however, our sample size was too small to make generalizable conclusions regarding the broad musician population.

Measurements and Simulations of Aerosol Released while Singing and Playing Wind Instruments.

Outbreaks from choir performances, such as the Skagit Valley Choir, showed that singing brings potential risk of COVID-19 infection. There is less known about the risks of airborne infection from other musical performances, such as playing wind instruments or performing theater. In addition, it is important to understand methods that can be used to reduce infection risk. In this study, we used a variety of methods, including flow visualization, aerosol and CO2 measurements, and computational fluid dynamics (CFD) modeling to understand the different components that can lead to transmission risk from musical performance and risk mitigation. This study was possible because of a partnership across academic departments and institutions and collaboration with the National Federation of State High School Associations and the College Band Directors National Association. The interdisciplinary team enabled us to understand the various aspects of aerosol transmission risk from musical performance and to quickly implement strategies in music classrooms during the COVID-19 pandemic. We found that plumes from musical performance were highly directional, unsteady and varied considerably in time and space. Aerosol number concentration measured at the bell of the clarinet was comparable to that of singing. Face and bell masks attenuated plume velocities and lengths and decreased aerosol concentrations measured in front of the masks. CFD modeling showed differences between indoor and outdoor environments and that the lowest risk of airborne COVID-19 infection occurred at less than 30 min of exposure indoors and less than 60 min outdoors.