NetCDF 文件中的写入次数

Question

使用 NETCDF4 包时，在 netCDF 文件中写入次数。

dates = []
for iday in range(84):
    dates.append(datetime.datetime(2016, 10, 1) + atetime.timedelta(hours = iday))
times[:] = date2num(dates, units=times.units, calendar = imes.calendar)
# print times[:]
for ii, i in enumerate(times[:]):
    print i, num2date(i, units=times.units), dates[ii]

时机是正确的：

dates = []
for iday in range(84):
    dates.append(datetime.datetime(2016, 10, 1) + atetime.timedelta(hours = iday))
times[:] = date2num(dates, units=times.units, calendar = imes.calendar)
# print times[:]
for ii, i in enumerate(times[:]):
    print i, num2date(i, units=times.units), dates[ii]

但是在读取 netcdf 文件时：

input_file = '/home/lovechang/test.nc'
data = Dataset(input_file)

times = data.variables['time']
# print times[:]
# print num2date(times[:], units=times.units)
for i in times[:]:
    print i, num2date(i, units=times.units)

结果：

17669815.0 2016-10-04 07:00:00 2016-10-04 07:00:00
17669816.0 2016-10-04 08:00:00.000006 2016-10-04 08:00:00
17669817.0 2016-10-04 09:00:00 2016-10-04 09:00:00
17669818.0 2016-10-04 10:00:00 2016-10-04 10:00:00
17669819.0 2016-10-04 11:00:00.000006 2016-10-04 11:00:00

Ncview 显示时间不准时。

那么时代到底发生了什么？如何在 netcdf 文件中准时写入小时？

Answer 1

根据您选择的时间单位和数据类型，您可能会遇到浮点精度问题。例如，如果您在 中指定时间days since 1970-01-01 00:00，则 32 位浮点是不够的，您应该使用 64 位浮点：

import datetime
import netCDF4

times = [datetime.datetime(2016, 10, 1) + datetime.timedelta(hours=hour)
         for hour in range(84)]

# Create netCDF file
calendar = 'standard'
units = 'days since 1970-01-01 00:00'
ds = netCDF4.Dataset('test.nc', 'w')
timedim = ds.createDimension(dimname='time', size=len(times))

# Write timestamps to netCDF file using 32bit float
timevar32 = ds.createVariable(varname='time32', dimensions=('time',),
                              datatype='float32')
timevar32[:] = netCDF4.date2num(times, units=units, calendar=calendar)

# Write timestamps to netCDF file using 64bit float
timevar64 = ds.createVariable(varname='time64', dimensions=('time',),
                              datatype='float64')
timevar64[:] = netCDF4.date2num(times, units=units, calendar=calendar)

# Read timestamps from netCDF file
times32 = netCDF4.num2date(timevar32[:], units=units, calendar=calendar)
times64 = netCDF4.num2date(timevar64[:], units=units, calendar=calendar)
for time, time32, time64 in zip(times, times32, times64):
    print "original  ", time
    print "  32 bit  ", time32
    print "  64 bit  ", time64
    print

如果您在 中指定时间hours since 2016-10-01 00:00，则即使是整数也足够了（在本示例中）。