ck_spei() with the default log-logistic distribution returned NA for
every value in a calendar month whose climatic water balance had zero or
negative L-skewness, with the warning "L-skewness out of range". The
three-parameter log-logistic of Vicente-Serrano et al. (2010) is the
generalised logistic distribution (Hosking 1997) with shape equal to
minus the L-skewness, but the package used the parameterisation
beta = 1 / tau3, which only exists for positive skewness. The fit now
uses the generalised logistic form, which covers any L-skewness between
-1 and 1, as the SPEI package does. Values for months with positive
skewness are unchanged (they agree to within 1e-13); months that were
NA now have values.
This release corrects three defects that produced wrong numbers in default code paths. All three were found by an audit that checked each index against its published definition rather than against the package's own tests.
ck_spei() with the default distribution = "log-logistic" did not
return a standardised index. On an 80-year synthetic series it gave
mean 1.61 and standard deviation 3.10, where a standardised index is
standard normal by construction. Two errors in the L-moment fit were
responsible: the shape parameter was computed as
tau3 * pi / (3 * sin(tau3 * pi / 3)) instead of 1 / tau3, and the
scale parameter omitted a factor of the shape parameter
(Vicente-Serrano et al. 2010, appendix). Fitting to data simulated
from a known log-logistic now recovers its parameters; output is mean
0.00 and standard deviation 0.99. ck_spi() and
distribution = "gev" were unaffected.(0, 1),
the range in which a three-parameter log-logistic exists. Calendar
months whose water balance is symmetric or left-skewed return NA
with a warning instead of a value drawn from a distribution that does
not describe them. Use distribution = "gev" for such series.ck_ehf() had the two Excess Heat Factor terms the wrong way round.
It computed max(EHIsig, 1) * EHIaccl; Nairn and Fawcett (2013)
define EHIsig * max(EHIaccl, 1). The consequence was that the sign
of the EHF was set by acclimatisation rather than by heat, so any
spell warmer than the preceding 30 days counted as a heatwave whether
or not it was hot. On a 30-year synthetic series 5,455 days were
flagged where 540 should have been, and 4,915 of those were below the
reference 95th percentile.ck_pet_pm() returned 4,619,067 mm/day. The
hour-angle argument is now clamped to [-1, 1], which resolves both
poles correctly and stops acos() emitting "NaNs produced" warnings.
This affected ck_pet(), ck_pet_pm(), and any ck_spei() built on
their output, at latitudes beyond about 66.5 degrees.ck_max_5day_precip() returned the sum of a short period as though it
were a five-day maximum. A period of fewer than five days now returns
NA.Physically impossible input was previously accepted and computed on. This mattered most for missing-data sentinels such as -999, which networks including GHCN use and which were being counted as frost days and subtracted from rainfall totals.
tmin above tmax is now an error; it means the two columns have
been swapped or the data are corrupt. Previously ck_diurnal_range()
returned a negative range.ck_available() and ck_metadata() now know about every index the
package computes. The index table had been maintained as three
separate hand-written lists, and the twelve indices added in 0.2.0
(ck_hwn(), ck_hwf(), ck_hwd(), ck_hwm(), ck_hwa(),
ck_cwn(), ck_cwf(), ck_cwd(), ck_cwm(), ck_cwa(),
ck_ehf() and ck_pet_pm()) reached only the ck_compute() list.
ck_available() returned 42 rows instead of 54, and
ck_metadata("ehf") failed with "Unknown index". All four views
(ck_available(), ck_metadata(), ck_catalogue() and the
ck_compute() dispatch) are now derived from one internal table, so
they cannot drift apart again. ck_catalogue() and ck_etccdi_27()
are unchanged.ck_warm_spell() estimated its series quantile with R's default
Hyndman-Fan type 7. Every other percentile in the package uses type
8, the 'ETCCDI' convention. It now uses type 8 as well. The two
estimators differ by order 1/n, so on a decade of daily data the
threshold moves by around 0.003 degrees C and counts are typically
unchanged; the gap is larger on short series.clear_cache() is deprecated and will be removed in 0.4.0. It warns
and returns FALSE invisibly. climatekit performs no I/O and has
never written a cache, so the function never had anything to clear.
Clear cached weather data with the package that downloaded it.tools is no longer a dependency; it was needed only by
clear_cache().ck_dry_days() and ck_wet_days() now state their 'ETCCDI' codes
and warn about the two acronym collisions in the literature: CWD is
both consecutive wet days and cold-wave duration, and CDD is both
consecutive dry days and cooling degree days. ck_compute() keeps
the meaning each ck_* function already carried, and additionally
accepts the unambiguous aliases "consecutive_wet_days",
"consecutive_dry_days", "cold_wave_duration", "cdd" and the
matching cold-wave names.ck_total_precip() now applies the canonical 'ETCCDI' 'PRCPTOT'
wet-day filter (precipitation >= 1 mm) by default. Pass
wet_day_threshold = 0 to recover the previous raw-sum behaviour.ck_first_frost() and ck_last_frost() gain a lat argument and
use hemisphere-appropriate cutoffs. Southern Hemisphere users no
longer get silent NA results.ck_branas() gains a lat argument; Southern Hemisphere growing
season is now October-February of the following year.ck_precip_intensity() (SDII) returns NA for periods with no valid
observations rather than NaN.ck_tx10p(), ck_tn10p(), ck_tx90p(), ck_tn90p() gain a
bootstrap = FALSE argument. When TRUE, the leave-one-out
resampling of Zhang et al. (2005) is applied to remove
self-inclusion bias for analysis years inside the reference
period. This is the canonical 'climdex.pcic' / 'climpact' behaviour
and is required for climate-change attribution work spanning the
base period.ck_hwm(), ck_hwa(), ck_cwm(), ck_cwa() gain a mode = c( "excess", "absolute") argument. "excess" (default) preserves
the existing 'ET-SCI' / 'climpact' convention. "absolute"
returns mean / peak raw temperature on event days, matching
Perkins-Alexander (2013).ck_ehf() implements the Excess Heat Factor of
Nairn and Fawcett (2013), the Australian Bureau of Meteorology
operational heatwave metric. Three annual statistics are exposed
via stat = c("max", "n_positive", "sum_positive").ck_spi() gains distribution = c("gamma", "pearsonIII"). Pearson
III is preferred in arid regions where the wet-day distribution is
highly skewed (Stagge et al. 2015).ck_spei() gains distribution = c("log-logistic", "gev"). GEV is
fitted via Hosking (1985) L-moments.ck_pet_pm() implements the FAO-56 Penman-Monteith
reference evapotranspiration (Allen et al. 1998), the
international standard. Optional inputs include relative humidity,
wind speed, incoming solar radiation, and elevation; FAO-56
fallbacks are used where these are unavailable. ck_pet()
remains as the simpler temperature-only Hargreaves estimator.ck_txx(), ck_tnx(), ck_txn(), ck_tnn() (extreme values);
ck_tx10p(), ck_tn10p(), ck_tx90p(), ck_tn90p(),
ck_r95p(), ck_r99p() (calendar-day percentile base, default
reference period 1961-1990); ck_wsdi(), ck_csdi() (proper
calendar-day spell duration). ck_growing_season() was already
ETCCDI-compliant.ck_warm_spell() is retained as a quick series-quantile
approximation; its documentation now points to ck_wsdi() for the
canonical ETCCDI definition.ck_hwn(), ck_hwf(), ck_hwd(), ck_hwm(), ck_hwa()
(number, frequency, duration, magnitude, amplitude). Cold-wave
duals: ck_cwn(), ck_cwf(), ck_cwd(), ck_cwm(), ck_cwa().ck_cwd() here is the
ET-SCI cold-wave version; ck_wet_days() is the ETCCDI
precipitation index. Both function-level documentation pages
cross-reference the other.ck_etccdi_27() returns the canonical 27 'ETCCDI' indices as a
data frame with code, name, variable, unit, definition,
ck_function, and status columns. Use it to audit coverage or
to locate the function for a given short code.ck_catalogue() returns the full implementation catalogue (51
rows). ck_browse(sector, standard, search) filters by sector
('agriculture', 'health', 'water', 'energy'), standard ('ETCCDI',
'ET-SCI', 'agroclimatic', etc.), or free-text search.ck_apply_grid(x, fun, dates, ...) applies any ck_* function
over the cells of a 'terra' 'SpatRaster' and returns a SpatRaster
with one layer per output period.ck_from_netcdf(path, var) is a thin convenience wrapper around
terra::rast() for 'netCDF' inputs with file-existence and
argument validation.inst/CITATION provides 'bibentry' records for the package and
for Alexander et al. (2006) and Zhang et al. (2011).climdex-migration documents the function-name
crosswalk from 'climdex.pcic' to climatekit, plus interface-shift
notes (numeric-vector inputs, tidy data-frame outputs, default
reference period).ck_compute() for programmatic index selection.ck_available() and ck_metadata() for index discovery.ck_convert_temp() for temperature unit conversion.