Evaluating Data Product Exposure Metrics for Use in Epidemiologic Studies of Dust Storms.

Dust storms are increasing in frequency and correlate with adverse health outcomes but remain understudied in the United States (U.S.), partially due to the limited spatio-temporal coverage, resolution, and accuracy of current data sets. In this work, dust-related metrics from four public areal data products were compared to a monitor-based "gold standard" dust data set. The data products included the National Weather Service (NWS) storm event database, the Modern-Era Retrospective analysis for Research and Applications-Version 2, the EPA's Air QUAlity TimE Series (EQUATES) Project using the Community Multiscale Air Quality Modeling System (CMAQ), and the Copernicus Atmosphere Monitoring Service global reanalysis product. California, Nevada, Utah, and Arizona, which account for most dust storms reported in the U.S., were examined. Dichotomous and continuous metrics based on reported dust storms, particulate matter concentrations (PM10 and PM2.5), and aerosol-type variables were extracted or derived from the data products. Associations between these metrics and a validated dust storm detection method utilizing Interagency Monitoring of Protected Visual Environments monitors were estimated via quasi-binomial regression. In general, metrics from CAMS yielded the strongest associations with the "gold standard," followed by the NWS storm database metric. Dust aerosol (0.9-20 µm) mixing ratio, vertically integrated mass of dust aerosol (9-20 µm), and dust aerosol optical depth at 550 nm from CAMS generated the highest standardized odds ratios among all metrics. Future work will apply machine-learning methods to the best-performing metrics to create a public dust storm database suitable for long-term epidemiologic studies.

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.