Timestamps can be provided as vectors of character strings, POSIXct or
Date.
Matching also returns index values for timestamps that fall between two
elements of the time series - this can lead to surprising results when time
series elements are positioned in the middle of an interval (as the CF
Metadata Conventions instruct us to "reasonably assume"): a time series of
days in January would be encoded in a netCDF file as
c("2024-01-01 12:00:00", "2024-01-02 12:00:00", "2024-01-03 12:00:00", ...)
so x <- c("2024-01-01", "2024-01-02", "2024-01-03") would result in
(NA, 1, 2) (or (NA, 1.5, 2.5) with method = "linear") because the date
values in x are at midnight. This situation is easily avoided by ensuring
that y has bounds set (use bounds(y) <- TRUE as a proximate solution if
bounds are not stored in the netCDF file). See the Examples.
If bounds are set, the indices are taken from those bounds. Returned indices
may fall in between bounds if the latter are not contiguous, with the
exception of the extreme values in x.
Values of x that are not valid timestamps according to the calendar of y
will be returned as NA.
x can also be a numeric vector of index values, in which case the valid
values in x are returned. If negative values are passed, the positive
counterparts will be excluded and then the remainder returned. Positive and
negative values may not be mixed. Using a numeric vector has the side effect
that the result has the attribute "CFTime" describing the temporal dimension
of the slice. If index values outside of the range of y (1:length(y)) are
provided, an error will be thrown.