datasets

Open datasets from my research, free to download and reuse.

High-Resolution Daily PM2.5 for the Contiguous US (2005–2021)

DOI 10.5281/zenodo.20402757 License: CC BY 4.0

Daily surface PM2.5 estimates at 1 km across the contiguous United States, from August 2005 to December 2021. The estimates come from a bidirectional LSTM with attention that combines EPA ground monitors, satellite remote sensing, and wildfire smoke density observations. It does especially well on high-pollution days, such as wildfire smoke events, where many existing products underestimate.

Model predictions during the 2020 California August Complex Fire
Predictions during the 2020 August Complex Fire in California: (1) CONUS-wide predictions, (2) zoomed-in predictions, (3) NOAA smoke density polygons, and (4) bias at EPA monitors.
   
Coverage Contiguous United States
Resolution 1 km, daily
Period Aug 25, 2005 – Dec 31, 2021
Variable Surface PM2.5 concentration (µg/m³)
Format One ZIP per year (monthly files inside), about 85 GB in total
License CC BY 4.0

Download (Zenodo) Paper Code

Citation. If you use this dataset, please cite the paper:

Wang, Z., Crooks, J. L., Regan, E. A., & Karimzadeh, M. (2025). High-Resolution Estimation of Daily PM2.5 Levels in the Contiguous US Using Bi-LSTM with Attention. Remote Sensing, 17(1), 126. https://doi.org/10.3390/rs17010126

@article{wang2025pm25,
  author  = {Wang, Zhongying and Crooks, James L. and Regan, Elizabeth A. and Karimzadeh, Morteza},
  title   = {High-Resolution Estimation of Daily {PM2.5} Levels in the Contiguous {US} Using Bi-{LSTM} with Attention},
  journal = {Remote Sensing},
  year    = {2025},
  volume  = {17},
  number  = {1},
  pages   = {126},
  doi     = {10.3390/rs17010126}
}

Questions or ideas for using the data? Email me at zywang16@mit.edu.