plastic_AQ_causal

Associated paper: Considine, EM and Nethery, RC. 2026. "A Spatiotemporal, Quasi-experimental Causal Inference Approach to Characterize the Effects of Global Plastic Waste Export and Burning on Air Quality Using Remotely Sensed Data." Journal of the Royal Statistical Society - Series C: Applied Statistics

Our final analytic dataset is available on Harvard DataVerse.

Statistical Methodology

See the READMEs in the EIF_method folder and the EIF_simulations folder for the case study methodology and simulations methodology, respectively.

Use the following to install the necessary packages:

conda env create -f env_plastic-AQ.yaml

Note that the locpol_with_lolocv folder is based closely on source code from the locpol package, but has the added functionality of selecting a bandwidth (for kernel regression) based on leave-location-out (LOLO) -- or any other manually-specified folds -- cross-validation. By contrast, CAST_CreateSpaceTimeFolds_copy.R is unchanged from its definition in the CAST package, copied here because it is not available on conda-forge (which is used to build our conda environment).

Exploratory Data Analysis

Scripts are in the EDA/ folder.

1. VIIRS_GPW_preliminary.Rmd
2. Jakarta_map.R
3. EDA_time-varying.Rmd

Data Processing

Scripts are in the Process_data/ folder.

Global Plastic Watch

1. Request an API key using the form linked in their FAQs
2. Download data: GPW_API.ipynb [Most recently downloaded on 9/5/24]
3. Obtain maximum footprint of each site: Create_GPW_shp.R

VIIRS Active Fire Points

1. Manually download VIIRS S-NPP data in CSV format, 2012-2019, from NASA FIRMS, into a folder called Data/VIIRS. [Downloaded on 6/18/24]
2. Spatial join with GPW footprints: VIIRS_overlay.R
3. Later, after obtaining province boundaries, obtain the total number of fires (not overlapping with GPW sites) within each province: Wildfires.Rmd

Global Maritime Traffic Density

1. Manually download GMTD loitering data from the interactive map. Detailed metadata can be found here. [Most recently downloaded on 9/5/24]

• Select "Loitering" and "WGS84"
• Click the rectangle next to "Draw bounding box on map". You will need to do this several times, as the total number of pixels to cover Indonesia exceeds the maximum download size from this webpage. I drew four slightly-overlapping bounding boxes to achieve this.
• In each bounding box's pop-up, select the "Multiple Months" tab, beginning January 2018 and ending December 2018. Export as GeoTIFF.
• Place each (unzipped) folder into a folder called Data/GMTD/

2. Calculate the annual average, mosaic the rasters together, and create the port proximity index: Process_gmtd.R

Province Boundaries

Manually download the Level-1 shapefile for Indonesia from https://gadm.org/download_country.html, save in a folder called Data/Province_boundaries

Meteorology (Temperature, Relative Humidity)

1. Run Download_rasters_GEE.iynb on Google Drive (Colaboratory)

• If you don't already have a project on Google Earth Engine, you will need to make one. If you run into trouble, try creating a project on Google Cloud Console, then adding yourself to that project on Google Earth Engine.

2. Move the resulting files to a local folder called Data/GEE_meteorology
3. Extract the values at each GPW site: Extract_meteo.R

PM_2.5

1. Manually download the folders 2012-2019 here -- these should be the monthly Asia subsets. Place these folders in a folder called Data/PM25
2. Process these data: Process_PM25.R

Finally, merge everything with the script Merge_time-varying.R