High-resolution daily PM₂․₅ maps (1 km) for the contiguous United States (2005-08-25 ➜ 2021-12-31),
generated with a Bi-LSTM-Attention model described in Wang et al., Remote Sensing 17(1):126, 2025.
- Dataset overview
- Quick start (under construction)
- Install & environment
- Data sources
- Citation
- License
- Spatial coverage: CONUS (lower 48 + DC)
- Spatial resolution: 0.01° (~1 km) MODIS sinusoidal grid
- Temporal coverage: 2005-08-25 → 2021-12-31 (2 392 days)
# 1. Clone & install
git clone https://github.com/air-lab/pm25-1km.git
cd pm25-1km
mamba env create -f environment.yml
conda activate pm25-1km
# 2. Download a 3-day sample (~25 MB)
python scripts/download_sample.py --dest data/sample
# 3. Run inference for 2020-07-05
python scripts/infer.py --date 2020-07-05 --input data/sample \
--checkpoint checkpoints/bi_lstm_attention.pth \
--output outputs/pm25_2020-07-05.nc
# 4. Visualise
python notebooks/plot_example.py outputs/pm25_2020-07-05.nc- Conda ≥ 4.12 (or the drop-in replacement Mamba for faster solves).
Miniconda download: https://docs.conda.io/en/latest/miniconda.html - Python 3.8 (pinned inside the env file).
- Git to clone this repository.
- GPU (optional)
Inference works on CPU; training the full 17-year model requires an NVIDIA card with ≥ 10 GB VRAM and CUDA 11.4.
# with Conda
conda env create -f environment.yml
# …or with Mamba (recommended: ~5× faster)
mamba env create -f environment.yml
conda activate pm25-1km # activate the new env| Data Source | Variable | Description | Temporal Resolution | Spatial Resolution |
|---|---|---|---|---|
| MODIS MCD19A2.061 | AOD 047 and AOD 055 | MODIS Terra and Aqua combined Multi-angle Implementation of Atmospheric Correction (MAIAC) Land Aerosol Optical Depth (AOD) gridded Level 2 product produced daily at 1 km resolution. | Daily | 1-km |
| MODIS MCD43A4_006_NDVI | NDVI | NDVI generated from the MODIS/006/MCD43A4 surface reflectance composites. | 16-Day | 0.5-km |
| Daymet | Day Length, Pricipitation, Shortwave Radiation, Max and Min Temperature, and Vapor Pressue | Daily Surface Weather and Climatological Summaries | Daily | 1-km |
| gridMet | Wind direction and velocity | A dataset of daily high-spatial resolution (~4-km, 1/24th degree) surface meteorological data covering the contiguous US from 1979-yesterday. | Daily | 1/24th degree ~4-km |
| GMTED | DEM | The Global Multi-resolution Terrain Elevation Data 2010 (GMTED2010). | Static (2010) | 1-km |
| NOAA Hazard Mapping System | Wildfire Smoke Density | NOAA's Hazard Mapping System (HMS) Smoke Product | Daily | 1-km |
If you use this repository or the derived 1-km PM₂․₅ dataset in your work, please cite:
@article{wang2025high,
title = {High-Resolution Estimation of Daily PM2.5 Levels in the Contiguous US Using Bi-LSTM with Attention},
author = {Wang, Zhongying and Crooks, James L. and Regan, Elizabeth Anne and Karimzadeh, Morteza},
journal = {Remote Sensing},
volume = {17},
number = {1},
pages = {126},
year = {2025},
publisher = {MDPI},
doi = {10.3390/rs17010126}
}Distributed under the Apache License 2.0.
