Instructions

Upload a CSV file with columns (id, lat, lon) using the Browse button on the sidebar. Below is an example of the contents of the file:

id,lat,lon
0,47.5,-122.5
1,47.5,-122.25
2,47.5,-122.0
3,47.5,-121.75
4,47.5,-121.5

The id column can be any identifier, or the column can be ommited, in which case the row number will be used as the id. Make sure that the latitude is before the longitude column in the CSV file. The valid range for latitude is 20.0 to 52.0 and longitude is -135.0 to -60.0, which cover the contiguous United States.

The resolution corresponds to how much neighboring information is captured by the embedding. If local is selected, the original weather covariates will be returned. Currently, all the embeddings correspond to the variables:

• air.2m.mon.mean.nc: temperature at 2m
• apcp.mon.mean.nc: total precipitation
• rhum.2m.mon.mean.nc: relative humidity
• vwnd.10m.mon.mean.nc: (north-south) wind component
• uwnd.10m.mon.mean.nc: (east-west) wind component

The radius corresponds to the number of neighboring raster cells to include in weather2vec representation. A resolution of 96km means that the embeddings encodes informations from all nearby raster cells whose centers are less than 96km. All embeddings have 10 hidden dimensions.

The embeddings also record information of the 12-month moving average. For this reason, the 'local' embeddings also have dimension 10, the first 5 dimensions correspond to the 5 meteorological variables in a given month, and the last 5 dimensions correspond to their 12-month moving average. For the non-local embeddings, the order of the variables is not interpretable.

Download

Citation

Tec, M., Scott, J.G. and Zigler, C.M., 2023. "Weather2vec: Representation learning for causal inference with non-local confounding in air pollution and climate studies". In: Proceedings of the AAAI Conference on Artificial Intelligence.

@inproceedings{tec2023weather2vec,
  title={Weather2vec: Representation learning for causal inference with non-local confounding in air pollution and climate studies},
  author={Tec, Mauricio and Scott, James G and Zigler, Corwin M},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={37},
  number={12},
  pages={14504--14513},
  year={2023}
}