15  Aggregate ERA5 Across Healthsheds

Summarize hourly climate fields into daily boundary-level outputs

This page converts raw ERA5 data into daily healthshed-level summaries, including optional population weighting, timezone adjustment, unit conversion, and derived variables such as relative humidity and wind speed.

Before beginning aggregation, confirm that the required boundary geometry, downloaded ERA5 rasters, elevation data, and optional population rasters are already available. These are files that may have been staged in a previous workflow step.

shp_file <- "path/to/healthsheds.shp"
data_dir <- "path/to/era5_raw"
elevation_file <- "path/to/elevation_healthshed.csv"
population_dir <- "path/to/worldpop"
output_dir <- "path/to/era5_healthshed_outputs"

list(
  shp_file = shp_file,
  data_dir = data_dir,
  elevation_file = elevation_file,
  population_dir = population_dir,
  output_dir = output_dir
)

The notebook begins by defining whether the run targets ERA5 or ERA5-Land, which years to process, the local time shift, the healthshed identifier, and whether values should be population-weighted or area-averaged.

# Pseudocode adapted from the original notebook
is_era5land <- TRUE
id_name <- "fs_uid"
data_years <- 2010:2025
local_time_to_utc <- +3
pop_weighted <- FALSE

source_name <- ifelse(is_era5land, "era5land", "era5")

The core of the notebook does three things for each variable and year:

  1. read the raster and align it to the healthshed geometry;
  2. extract hourly values per healthshed, optionally weighting by population;
  3. clean the hourly series by shifting accumulated variables, adjusting for local time, converting units, and applying elevation-based lapse-rate corrections where needed.
# Pseudocode adapted from the original notebook
# shp <- read_sf(shp_file)
# r <- rast(file.path(data_dir, "<source>_<variable>_<year>.grib"))
# hourly <- get_hourly_var_per_healthshed(r, year, shp, id_name, pop_weighted = pop_weighted)
# cleaned <- clean_raw_hourly_data(
#   hourly,
#   agg_previous = previous_hours,
#   year = year,
#   variable = variable,
#   is_era5land = is_era5land,
#   local_time_to_utc = local_time_to_utc,
#   elevation_file = elevation_file
# )

For the standard ERA5 variables, the notebook loops over all years, aggregates cleaned hourly values to daily mean, min, and max, and saves one output file per variable and year.

# Pseudocode adapted from the original notebook
base_variables <- c(
  "2m_temperature",
  "2m_dewpoint_temperature",
  "surface_pressure",
  "soil_temperature_level_1",
  "volumetric_soil_water_layer_1",
  "volumetric_soil_water_layer_2",
  "total_precipitation",
  "surface_solar_radiation_downwards"
)

# for (variable in base_variables) {
#   for (year in data_years) {
#     daily <- get_daily_avg(cleaned$data_year_clean, id_name = id_name)
#     write.csv(daily, file.path(output_dir, "<source>_<variable>_<year>.csv"))
#   }
# }

The notebook also computes additional products that are not downloaded directly. Relative humidity is derived from temperature and dewpoint; wind speed and direction are derived from the u and v wind components.

# Pseudocode adapted from the original notebook
# relative_humidity <- derive_from_temperature_and_dewpoint(hourly_temp, hourly_dewpoint)
# relative_humidity_daily <- get_daily_avg(relative_humidity, round_digits = 1)

# wind_speed <- sqrt(u^2 + v^2)
# wind_speed_daily <- get_daily_avg(wind_speed, round_digits = 2)

# wind_direction <- (180 + atan2(u, v) * 180 / pi) %% 360
# wind_direction_daily <- get_daily_avg(wind_direction, round_digits = 2)

A final check section can assert the success of the pipeline.

# This page is complete when the operator can explain:
# - which source files feed the aggregation,
# - when population weighting is used instead of area averaging,
# - why hourly data must be cleaned before daily summaries are computed,
# - and which outputs are direct daily summaries versus derived products
#   such as relative humidity and wind metrics.