Annual PM2.5 Component Concentrations for the Contiguous United States, aggregated to common geographies (2000-2019)

This dataset contains many files; use the Tree view below or Dataverse API directory index to get a condensed overview of what is available. See the README-PM-components-aggregated.pdf file for more information. Overview This dataset includes annual predictions of PM2.5 components across the contiguous U.S from 2000 to 2019, aggregated to common geographic levels. These aggregates are derived from predictions originally made on a 50-meter grid within urban areas and on a 1-kilometer grid outside of those areas. The original predictions are available here: Amini, Heresh; Castro, Edgar; Danesh-Yazdi, Mahdieh; Di, Quan; Requia, Weeberb J; Wei, Yaguang; Abu Awad, Yara; Shi, Liuhua; Franklin, Meredith; Kang, Choong-Min; Wolfson, Jack M; James, Peter; Habre, Rima; Zhu, Qiao; Apte, Joshua S; Andersen, Zorana J; Xing, Xiaoshi; Hultquist, Carolynne; Kloog, Itai; Dominici, Francesca; Koutrakis, Petros; Schwartz, Joel, 2025, "Annual PM2.5 Component Concentrations for the Contiguous United States, 50-m Urban Grid and 1-km Non-Urban Grid (2000-2019)", https://doi.org/10.7910/DVN/3H7DNP, Harvard Dataverse, V2 From that link: These predictions were produced by a set of super-learning ensemble models in which the predictions from a variety of machine learners were then used to train a second stage of machine learners. For each component-urbanicity combination, the second-stage machine learner with the highest R2 was chosen as the final model. Overall cross-validated R2 values ranged from 0.856-0.952 for the major components (EC, NH4+, NO3-, OC, and SO42-) in urban areas and 0.878-0.957 outside urban areas, and 0.797-0.878 for the trace elements (Br, Ca, Cu, Fe, K, Ni, Pb, Si, V, and Zn) in urban areas and 0.787-0.881 outside urban areas. Please refer to the README file for the exact values for each element and location. Predictions and supplementary information about how aggregations were performed are available as comma-separated values (CSV) and RDS files. When using this dataset, please also cite the Related Publications in the Metadata tab below (also available in the README file), which further detail the data sources and processes used to produce these predictions. Bulk downloads wget and other download programs can be used to bulk download files using the Dataverse API directory index: wget --recursive --execute robots=off --no-host-directories --span-hosts --content-disposition 'https://dataverse.harvard.edu/api/datasets/:persistentId/dirindex?persistentId=doi:10.7910/DVN/F6PQAA'