Disaggregate daily meteorology to hourly or 3-hourly

Distributes daily meteorology across the day using the diurnal structure of an hourly reanalysis record for the same location, so that a daily series - a bias-corrected daily record, or a climate-scenario series from a delta-change step - can drive a sub-daily lake model. Daily means (and daily rainfall totals) are conserved exactly, with one deliberate exception: if the wind vector components MET_wnduvu / MET_wnduvv are present they are shaped directly and MET_wndspd is recomputed from them, so the disaggregated speed re-aggregates to the magnitude of the daily vector mean rather than to a supplied daily scalar mean speed (the former is the smaller of the two).

Usage

disaggregate_met_to_hourly(
  daily,
  donor,
  method = c("fragments", "diurnal"),
  timestep = c("hour", "3hour"),
  swr = c("clearsky", "shape"),
  interval = c("ending", "beginning", "instant"),
  lat = NULL,
  lon = NULL,
  elev = 0,
  tz = NULL,
  analogue_window = 15,
  match_on_value = TRUE,
  sample_top_k = 1,
  blend_hours = 2,
  seed = NULL,
  expand = FALSE,
  verbose = TRUE
)

Arguments

daily

data frame with Date (class Date, or daily POSIXct) and MET_* columns. Rainfall and snowfall are treated as daily totals, every other variable as a daily mean.

donor

hourly data frame with Date (POSIXct) and MET_* columns for the same location, e.g. from extract_era5_lake_met(). A longer donor record gives a better analogue pool.

method

"fragments" (default) or "diurnal".

timestep

"hour" (default) or "3hour".

swr

"clearsky" (default) to rebuild shortwave from solar geometry, or "shape" to treat it like any other multiplicative variable. "clearsky" needs lat and lon.

interval

what a timestamp denotes, passed to estimate_hourly_swr(). Default "ending", matching the accumulated flux convention of ERA5 and of extract_era5_hourly_met().

lat, lon

location in decimal degrees; taken from the lat / lon attributes of daily or donor when NULL.

elev

elevation, m, used only when expand = TRUE.

tz

timezone of daily and of the output. Defaults to the tz attribute of daily, then of donor, else "UTC".

analogue_window

half-width, in days of the year, of the donor pool for method = "fragments" (default 15).

match_on_value

rank candidate donor days by similarity of their daily values, not just day-of-year (default TRUE).

sample_top_k

sample the donor uniformly from the k best candidates instead of always taking the closest. 1 (default) is deterministic; larger values add realistic variety.

blend_hours

smooth the borrowed shape across midnight over this many steps either side, to avoid a sawtooth discontinuity in temperature and humidity (default 2; 0 disables). Daily aggregates are restored afterwards, so conservation is unaffected. Never applied to rainfall or snowfall.

seed

optional RNG seed, for reproducibility when sample_top_k > 1.

expand

run expand_met() on the result to regenerate the dependent variables at the new timestep.

verbose

print progress.

Value

data frame with Date (POSIXct in tz, regular sub-daily steps) and the same MET_* columns as daily. Attributes tz, method, timestep, and donor_days (the analogue chosen for each target day, method = "fragments" only).

Details

Two ways of borrowing the structure:

"fragments"

(default) the method of fragments: for each target day an analogue day is chosen from the donor record - close in day-of-year and, when match_on_value, close in daily values - and its within-day shape is applied to every variable at once. Because a single donor day supplies all variables, the temperature, humidity and wind sub-daily patterns stay mutually consistent, and day-to-day variability in the shape is retained.

"diurnal"

a deterministic month-by-hour mean diurnal cycle from build_diurnal_climatology(). Smoother and reproducible, but every day of a given month gets the same shape.

Shortwave radiation is handled separately and by default comes from solar geometry (estimate_hourly_swr()), which guarantees a single-peaked curve, true night-time zeros and a peak at local solar noon. Rainfall in "fragments" mode borrows its shape from a wet donor day, chosen by closeness of daily total, so that wet-hour intermittency is preserved; a dry target day stays dry throughout.

Variables present in daily but not in donor are held constant through each day. Wind direction (MET_wnddir) is never shaped about a daily mean - it is recomputed from the disaggregated MET_wnduvu / MET_wnduvv when those are present, otherwise held constant per day.

See also

build_diurnal_climatology(), estimate_hourly_swr(), met_to_daily() for the reverse operation.

Examples

set.seed(1)
## donor: two years of synthetic hourly data with a diurnal cycle
t <- seq(as.POSIXct("2022-01-01", tz = "UTC"), by = "hour",
         length.out = 24 * 730)
hr <- as.integer(format(t, "%H"))
donor <- data.frame(Date = t,
                    MET_tmpair = 15 + 5 * sin((hr - 9) / 24 * 2 * pi) + rnorm(length(t)),
                    MET_wndspd = pmax(0.1, 4 + 2 * sin((hr - 12) / 24 * 2 * pi)))
## daily series to disaggregate
daily <- data.frame(Date = seq(as.Date("2024-01-01"), by = "day", length.out = 10),
                    MET_tmpair = rnorm(10, 16, 2),
                    MET_wndspd = runif(10, 2, 7))
out <- disaggregate_met_to_hourly(daily, donor, swr = "shape", verbose = FALSE)
head(out)
#>                  Date MET_tmpair MET_wndspd
#> 1 2024-01-01 00:00:00   13.15318   4.683362
#> 2 2024-01-01 01:00:00   12.49455   4.077291
#> 3 2024-01-01 02:00:00   13.52026   3.512522
#> 4 2024-01-01 03:00:00   12.11978   3.027544
#> 5 2024-01-01 04:00:00   12.80467   2.655407
#> 6 2024-01-01 05:00:00   14.34843   2.421472