8

我有一个带有时间序列的 netcdf 文件，时间变量具有以下典型元数据：

    double time(time) ;
            time:standard_name = "time" ;
            time:bounds = "time_bnds" ;
            time:units = "days since 1979-1-1 00:00:00" ;
            time:calendar = "standard" ;
            time:axis = "T" ;

RI 内部想要将时间转换为 R 日期对象。我现在通过读取单位属性并拆分字符串并使用第三个条目作为我的原点以硬连线的方式实现这一点（因此假设间距是“天”并且时间是 00:00 等）：

require("ncdf4")
f1<-nc_open("file.nc")
time<-ncvar_get(f1,"time")
tunits<-ncatt_get(f1,"time",attname="units")
tustr<-strsplit(tunits$value, " ")
dates<-as.Date(time,origin=unlist(tustr)[3])

这个硬连线解决方案适用于我的具体示例，但我希望 R 中可能有一个包可以很好地处理时间单位的 UNIDATA netcdf 日期约定并将它们安全地转换为 R 日期对象？

4

3 回答 3

5

I have just discovered (two years after posting the question!) that there is a package called ncdf.tools which has the function:

convertDateNcdf2R

which

converts a time vector from a netCDF file or a vector of Julian days (or seconds, minutes, hours) since a specified origin into a POSIXct R vector.

Usage:

convertDateNcdf2R(time.source, units = "days", origin = as.POSIXct("1800-01-01", 
    tz = "UTC"), time.format = c("%Y-%m-%d", "%Y-%m-%d %H:%M:%S", 
    "%Y-%m-%d %H:%M", "%Y-%m-%d %Z %H:%M", "%Y-%m-%d %Z %H:%M:%S"))

Arguments:

time.source

numeric vector or netCDF connection: either a number of time units since origin or a netCDF file connection, In the latter case, the time vector is extracted from the netCDF file, This file, and especially the time variable, has to follow the CF netCDF conventions.

units

character string: units of the time source. If the source is a netCDF file, this value is ignored and is read from that file.

origin

POSIXct object: Origin or day/hour zero of the time source. If the source is a netCDF file, this value is ignored and is read from that file.

Thus it is enough to simply pass the netcdf connection as the first argument and the function handles the rest. Caveat: This will only work if the netCDF file follows CF conventions (e.g. if your units are "years since" instead of "seconds since" or "days since" it will fail for example).

More details available here: https://rdrr.io/cran/ncdf.tools/man/convertDateNcdf2R.html

于 2019-12-12T17:11:48.600 回答
3

没有，据我所知。我有这个方便的功能 using lubridate，它与您的基本相同。

getNcTime <- function(nc) {
    require(lubridate)
    ncdims <- names(nc$dim) #get netcdf dimensions
    timevar <- ncdims[which(ncdims %in% c("time", "Time", "datetime", "Datetime", "date", "Date"))[1]] #find time variable
    times <- ncvar_get(nc, timevar)
    if (length(timevar)==0) stop("ERROR! Could not identify the correct time variable")
    timeatt <- ncatt_get(nc, timevar) #get attributes
    timedef <- strsplit(timeatt$units, " ")[[1]]
    timeunit <- timedef[1]
    tz <- timedef[5]
    timestart <- strsplit(timedef[4], ":")[[1]]
    if (length(timestart) != 3 || timestart[1] > 24 || timestart[2] > 60 || timestart[3] > 60 || any(timestart < 0)) {
        cat("Warning:", timestart, "not a valid start time. Assuming 00:00:00\n")
        warning(paste("Warning:", timestart, "not a valid start time. Assuming 00:00:00\n"))
        timedef[4] <- "00:00:00"
    }
    if (! tz %in% OlsonNames()) {
        cat("Warning:", tz, "not a valid timezone. Assuming UTC\n")
        warning(paste("Warning:", timestart, "not a valid start time. Assuming 00:00:00\n"))
        tz <- "UTC"
    }
    timestart <- ymd_hms(paste(timedef[3], timedef[4]), tz=tz)
    f <- switch(tolower(timeunit), #Find the correct lubridate time function based on the unit
        seconds=seconds, second=seconds, sec=seconds,
        minutes=minutes, minute=minutes, min=minutes,
        hours=hours,     hour=hours,     h=hours,
        days=days,       day=days,       d=days,
        months=months,   month=months,   m=months,
        years=years,     year=years,     yr=years,
        NA
    )
    suppressWarnings(if (is.na(f)) stop("Could not understand the time unit format"))
    timestart + f(times)
}

编辑：人们可能还想看看ncdf4.helpers::nc.get.time.series

EDIT2：请注意，新提出的且目前正在开发中的超棒stars软件包将自动处理日期，请参阅第一篇博文以获取示例。

EDIT3：另一种方法是units直接使用包，这就是stars使用。可以做这样的事情：（仍然不能正确处理日历，我不确定是否units可以）

getNcTime <- function(nc) { ##NEW VERSION, with the units package
    require(units)
    require(ncdf4)
    options(warn=1) #show warnings by default
    if (is.character(nc)) nc <- nc_open(nc)
    ncdims <- names(nc$dim) #get netcdf dimensions
    timevar <- ncdims[which(ncdims %in% c("time", "Time", "datetime", "Datetime", "date", "Date"))] #find (first) time variable
    if (length(timevar) > 1) {
        warning(paste("Found more than one time var. Using the first:", timevar[1]))
        timevar <- timevar[1]
    }
    if (length(timevar)!=1) stop("ERROR! Could not identify the correct time variable")
    times <- ncvar_get(nc, timevar) #get time data
    timeatt <- ncatt_get(nc, timevar) #get attributes
    timeunit <- timeatt$units
    units(times) <- make_unit(timeunit)
    as.POSIXct(time)
}
于 2017-09-02T07:53:51.237 回答
3

我无法让@AF7 的函数处理我的文件，所以我自己编写了。下面的函数创建一个 POSIXct 日期向量，从 nc 文件中读取开始日期、时间间隔、单位和长度。它适用于许多（但可能不是每个......）形状或形式的 nc 文件。

 ncdate <- function(nc) {
    ncdims <- names(nc$dim) #Extract dimension names
    timevar <- ncdims[which(ncdims %in% c("time", "Time", "datetime", "Datetime",
                                          "date", "Date"))[1]] # Pick the time dimension
    ntstep <-nc$dim[[timevar]]$len
    tm <- ncvar_get(nc, timevar) # Extract the timestep count
    tunits <- ncatt_get(nc, timevar, "units") # Extract the long name of units
    tspace <- tm[2] - tm[1] # Calculate time period between two timesteps, for the "by" argument 
    tstr <- strsplit(tunits$value, " ") # Extract string components of the time unit
    a<-unlist(tstr[1]) # Isolate the unit .i.e. seconds, hours, days etc.
    uname <- a[which(a %in% c("seconds","hours","days"))[1]] # Check unit
    startd <- as.POSIXct(gsub(paste(uname,'since '),'',tunits$value),format="%Y-%m-%d %H:%M:%S") ## Extract the start / origin date
    tmulti <- 3600 # Declare hourly multiplier for date
    if (uname == "days") tmulti =86400 # Declare daily multiplier for date
    ## Rename "seconds" to "secs" for "by" argument and change the multiplier.
    if (uname == "seconds") {
        uname <- "secs"
        tmulti <- 1 }
    byt <- paste(tspace,uname) # Define the "by" argument
    if (byt == "0.0416666679084301 days") { ## If the unit is "days" but the "by" interval is in hours
    byt= "1 hour"                       ## R won't understand "by < 1" so change by and unit to hour.
    uname = "hours"}
    datev <- seq(from=as.POSIXct(startd+tm[1]*tmulti),by= byt, units=uname,length=ntstep)
}

编辑

To address the flaw highlighted by @AF7's comment that the above code would only work for regularly spaced files, datev could be calculated as

 datev <- as.POSIXct(tm*tmulti,origin=startd)
于 2017-09-20T14:59:33.757 回答