tasks.md

Tasks

Preprocess AQ

• write download functions
• write function to filter by quality flags
• write function to filter for rural/urban/suburban & background/traffic
• write function to join pollutant tables by country
• download station data for all countries (2015-2023)
• preprocess all station data
• gapfill all PM2.5 with predictors according to EEA
  • PM10
  • coordinates
  • population density
  • sun shine duration
  • ERA5 alternative: surface_net_solar_radiation

Temporal aggregates (annual, monthly, daily)

• NO2 mean
• SO2 mean
• PM2.5 mean
• PM10
  • mean
  • 90.4 percentile of daily mean
• O3
  • mean
  • 93.2 percentile of max. daily 8h rolling mean

Supplementary data

Station-level (static)

• Elevation: COP-DEM
• Corine Land Cover
  • reclassify to 8 classes (Horalek 2019, section 3.4)
  • aggregate to single-class 1 km fractional cover
  • aggregate to single-class 10 km fractional cover
  • aggregate to single-class 1 km fractional cover within 5 km radius
• Population Density

Measurement-level (hourly)

• CAMS data (atmospheric transport model outputs for each pollutant, hourly)
  • reanalysis for 2015-2022
    • validated reanalysis: 2015-2020
    • interim reanalysis: 2021-2022
  • forecasts for 2023 (3-year rolling archive)
  • download
    • 2015-2022
    • 2023
• ECWMF ERA5 Land data (hourly):
  • wind speed (from u and v)
  • surface net solar radiation
  • temperature
  • relative humidity (from temp. and dew point temp.)
  • download
  • calculate wind speed/direction & humidity
  • temporal aggregates (daily, monthly, annual)
    • mean
    • percentiles
• Sentinel-5P TROPOMI (daily)
  • annual
  • monthly
  • daily

Interpolation

• function to read aq data
• function to read and warp required covariates to a common grid
• function wrapping lm
• function for residual kriging in parallel
• functions for (LOO-) cross-validation
• function to combine lm and kriging prediction
• functions for plotting prediction and standard error

Map merging

• Weights: Traffic Exposure
  • buffer and rasterize GRIP vector data for road classes 1-3
  • distance to nearest road (by type)
• Weights: Urban Character
  • scale and reclassify population density grid
• function to weight and merge map layers in parallel
  • RB: rural background stations
  • UB: urban/suburban background stations
  • JB: joint rural/urbal background stations
  • UT: urban/suburban traffic station (not for O3)
  • adjust RB and UB where necessary using JB
  • adjust UT where necessary using UB (not O3)
• write final maps (prediction and se) to COG
  • gdal_translate in out -of "COG" -co "COMPRESS=DEFLATE" -co "PREDICTOR=3" -co "BIGTIFF=YES"

Results

• write as cloud-optimized GeoTiffs
• convert and publish as Zarr stores

Potential improvements to the current method

• interpolation using
  • standard Random Forest
  • Random Forest with awareness for spatial correlation (RF-GLS)