---
title: "Conformance with the CF Metadata Conventions"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Conformance with the CF Metadata Conventions}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
```

`CFtime` is based on version 1.12 of the CF Metadata Conventions. The text for
the *time coordinate* in the conventions may be consulted [here](https://cfconventions.org/Data/cf-conventions/cf-conventions-1.12/cf-conventions.html#time-coordinate).

> The *time coordinate* is one of four coordinate types that receive "special
treatment" in the conventions. The other three are longitude, latitude and the
vertical. If you require convention-compliant support for any of these three
other coordinate types, please consider using package [`ncdfCF`](https://cran.r-project.org/package=ncdfCF) 
which supports all three coordinate types and links with `CFtime` for support of 
the time coordinate.

This document sets out how the `CFtime` package conforms to the CF Metadata
Conventions, by section of the conventions. This information is mostly useful
for developers and expert users.

If you have issues reading a netCDF file that is due to conformance of package
`CFtime` with the CF Metadata Conventions, please [open an issue on GitHub](https://github.com/pvanlaake/CFtime/issues).

Please note that there are many netCDF files out there that are not claiming 
adherence to the CF Metadata Conventions but whose time coordinate can still be 
successfully handled by `CFtime`: the `netCDF` library itself provides the basic
plumbing.

## 4.4. Time Coordinate

A `CFTime` object is constructed from information that is contained in the 
`units` and `calendar` attributes for a time coordinate read out of a netCDF 
file. The package does not actually access the netCDF file, it only uses the
information that was read out of the file by e.g. `ncdfCF`. Consequently, the
`CFtime` package can also construct a `CFTime` object from suitable character
strings.

This package is agnostic to the orientation of the axes in any data variable
that references the time coordinate. Consequently, the `standard_name` and 
`axis` attributes are not considered by this package (but the `ncdfCF` package
handles both). Identification of a time coordinate is done by the `units`
attribute, alone.

## 4.4.1. Time Coordinate Units

The `CFtime` package fully supports the units `"second"`, `"minute"`, `"hour"` 
and `"day"`, including abbreviated and/or plural forms. Unit `"second"` is the
SI second, a `"minute"` equals 60 seconds, an `"hour"` equals 3,600 seconds,
and a `"day"` equals 86,400 seconds. This is exactly as expected, but refer to
the `utc` calendar, below, for peculiarities of that calendar.

The `units` `"month"` and `"year"` are accepted on input but not using their
definition in UDUNITS. Instead, `"year"` is a calendar year, so either 360, 365 
or 366 days depending on its value and the calendar. A `"month"` is similarly a
calendar month. Use of either of these time units is discouraged by the CF
Metadata Conventions.

Other UDUNITS time units are not supported by this package.

All variants of the glue word `"since"` are accepted, being `"after"`, `"from"`,
`"ref"` and `"per"`.

The *"reference datetime string"* should be formatted using the UDUNITS broken
timestamp format or following [ISO8601](https://en.wikipedia.org/wiki/ISO_8601)
rules, but noting that datetimes valid in specific calendars other than
Gregorian (such as `2023-02-30` in the `360_day` calendar) are acceptable as
well. The UDUNITS "packed" format is not supported.

Timezone information can only use `00`, `15`, `30` and `45` to indicate minutes;
other minute offsets have never been used anywhere. A time zone value of `"UTC"`
is accepted, as an extension to the conventions. Even though the conventions 
don't indicate it, the `tai` and `utc` calendars can carry no time zone 
indication as that does not exist for either of these calendars.

## 4.4.2. Calendar

If a `calendar` attribute is not given, `"standard"` is assumed, being the
default calendar as per the conventions.

* `standard` (or the deprecated `gregorian`): Fully conformant, but leap seconds
are never considered (see below). The combination of a *reference datetime* and
other *datetimes* spanning the gap between 1582-10-05 and 1582-10-15, in either 
direction, is supported.
* `proleptic_gregorian`: Fully conformant, but leap seconds are never considered
(see below).
* `julian`: Fully conformant, but, despite the suggestion in the conventions,
leap seconds do not exist in this calendar and are thus never considered.
* `utc`: Fully conformant. Leap seconds are always accounted so when a leap 
second is included, UTC time progresses like 
`23:59:58 ... 23:59:59 ... 23:59:60 ... 00:00:00 ... 00:00:01`.
This also extends to minutes `23:59:00 ... 23:59:60 ... 00:00:59 ... 00:01:59`, 
always adding 60 seconds. Likewise for `hours` and `days`. Units `"year"` and 
`"month"` are not allowed, and neither is any time zone indication.
* `tai`: Fully conformant. Units `"year"` and `"month"` are not allowed, and 
neither is any time zone indication.
* `no_leap` / `365_day`: Fully conformant.
* `all_leap` / `366_day`: Fully conformant.
* `360_day`: Fully conformant.
* `none`: Fully conformant.

## 4.4.3. Leap seconds

The `utc` calendar fully supports leap seconds.

The `julian` calendar has no concept of leap seconds so these are never
possible or considered. Using a leap second in a `julian` calendar is an error.

In the `standard` and `proleptic_gregorian` calendars only the variant without
leap seconds is considered. The `units_metadata` attribute is not considered, so
assumed to be `"leap seconds: unknown"`. The assumption here is that if second
accuracy for a data producer is essential, then the entire tool chain from
observation equipment, to processing, to file recording will have to be of
known characteristics with regards to UTC time and leap seconds and thus the
`utc` calendar would be used, rather than `standard` or `proleptic_gregorian`
with a caveat communicated through the `units_metdata` attribute.

## 4.4.4. Time Coordinates with no Annual Cycle

Fully conformant.

## 4.4.5. Explicitly Defined Calendar

Not implemented.

## 7.4. Climatological statistics

The time coordinate with climatological bounds is fully supported, with the
exception of using year 0 in the `standard` and `julian` calendars to flag
climatological data (which is a deprecated practice as per the CF Metadata
Conventions.

This package is agnostic to the underlying statistical operation that produced
the *climatological time* so the `cell_methods` attribute is not inspected nor
interpreted.