Skip to content

PurpleAirSF: datasets for realistic air quality forecasting

Notifications You must be signed in to change notification settings

JingweiZuo/PurpleAirSF

Repository files navigation

Unleashing Realistic Air Quality Forecasting: Introducing the Ready-to-Use PurpleAirSF Dataset

This is the companion repository for the PurpleAirSF Dataset. The paper can be found here.

The preprocessed datasets can be downloaded in Google Drive.

Datasets with three temporal granularities are provided:

  • PurpleAirSF-10M: 10-minute sampling frequency/granularity
  • PurpleAirSF-1H: one-hour sampling frequency/granularity
  • PurpleAirSF-6H: six-hour sampling frequency/granularity

Figure 1 (Left): 10-min granular data with 316 stations; (Middle): 1-hour granular data with 232 stations; (Right): 6-hour granular data with 112 stations

The statistical summary of the datasets are shown in the below table:

How to use PurpleAirSF?

In each archived file, we provide

- IDS.json:         JSON file including the IDs of the sensor stations 
- sensor-loc.csv:   GPS locations (longitude, latitude) of each sensor station 
- map.html:         Sensor stations visualized in a Map
- data.npy:         Preprocessed meteorological and air quality measures 

The preprocessed data has a shape of (N, L, F)

- N: the number of sensor stations
- L: the entire sequence length
- F: the number of features/measures in each station. 

Users are free to split the dataset with different window size.

Here are a list of ordered measures that we considered during data collection:

    'humidity', 'temperature', 'pressure',
    'pm2.5_alt', 'scattering_coefficient', 'deciviews', 'visual_range',
    '0.3_um_count', '0.5_um_count', '1.0_um_count', '2.5_um_count',
    '5.0_um_count', '10.0_um_count', 'pm1.0_cf_1', 'pm1.0_atm', 'pm2.5_atm',
    'pm2.5_cf_1', 'pm10.0_atm', 'pm10.0_cf_1'

The detailed descriptions of the measures are shown in the below table:

How to obtain the raw data from PurpleAir?

Users can also use our provided scripts to fetch raw data from PurpleAir via PurpleAir API.

Step 1: Private key application

For some feature you need the private keys for the APIs.

  1. Create an PurpleAir account

  2. Write a email to [email protected] with subject "API keys for PurpleAirAPI". They will send you your API private key. Once you have it, just create a file in keys/PurpleAir_API_key.conf with the following structure:

[purpleair.com]
API_readKey = YOUR-PRIVATE-READ-KEY

Step 2: Use the provided scripts for data acquisition and pre-processings

  • 'main_purpleair_to_csv.py': fetch raw data and save to '.csv' files
  • 'csv_data_load.py': pre-process the '.csv' files and save to dataframe with target format, i.e., with shape of (N, L, F)

Citation

If you find this data useful in your research, please consider citing the following paper:

@inproceedings{zuo2023unleashing,
      title  = {Unleashing Realistic Air Quality Forecasting: Introducing the Ready-to-Use PurpleAirSF Dataset}, 
      author  ={Jingwei Zuo and Wenbin Li and Michele Baldo and Hakim Hacid},
      year  ={2023},
      booktitle = {ACM SIGSPATIAL'23}
}