测量海浪周期的变量的“单位”属性是“秒”。这不是日期时间字段，但xarray会自动将此变量提取为timedelta64。由于单位不是“自......以来的秒数”，我认为xarray应该将其视为普通的float32数据数组，但显然不是这样。有没有办法让我告诉xarray将波周期变量摄取为float32，或者在摄取后将它们从timedelta64转换回原始值？我仍然希望它将“time”变量转换为timedelta64，所以我不想关闭整个数据集的翻译，只针对特定变量（Tper，sper，wper）。
以下是我在TDS服务器中进行wave预测时使用的基本OPeNDAP网址：
连连呢？谢谢！
您可以在此处使用OPeNDAP页面查看“ncdump”类似的输出：
或者您可以在OPeNDAP URL上运行ncdump，如下所示：
这导致以下结果：
netcdf WaveWatch_III_Hawaii_Regional_Wave_Model_best {
dimensions:
lat = 101 ;
lon = 141 ;
time = 54453 ;
z = 1 ;
variables:
float lon(lon) ;
lon:units = "degrees_east" ;
lon:long_name = "longitude" ;
lon:standard_name = "longitude" ;
lon:short_name = "lon" ;
lon:axis = "x" ;
lon:_CoordinateAxisType = "Lon" ;
float lat(lat) ;
lat:units = "degrees_north" ;
lat:long_name = "latitude" ;
lat:standard_name = "latitude" ;
lat:short_name = "lat" ;
lat:axis = "y" ;
lat:_CoordinateAxisType = "Lat" ;
float z(z) ;
z:units = "meters" ;
z:long_name = "depth below mean sea level" ;
z:standard_name = "depth" ;
z:short_name = "depth" ;
z:axis = "z" ;
z:_CoordinateAxisType = "Height" ;
double time(time) ;
time:long_name = "Forecast time for ForecastModelRunCollection" ;
time:standard_name = "time" ;
time:calendar = "proleptic_gregorian" ;
time:units = "hours since 2011-06-21 00:00:00.000 UTC" ;
time:missing_value = NaN ;
time:_CoordinateAxisType = "Time" ;
double time_run(time) ;
time_run:long_name = "run times for coordinate = time" ;
time_run:standard_name = "forecast_reference_time" ;
time_run:calendar = "proleptic_gregorian" ;
time_run:units = "hours since 2011-06-21 00:00:00.000 UTC" ;
time_run:missing_value = NaN ;
time_run:_CoordinateAxisType = "RunTime" ;
double time_offset(time) ;
time_offset:long_name = "offset hour from start of run for coordinate = time" ;
time_offset:standard_name = "forecast_period" ;
time_offset:calendar = "proleptic_gregorian" ;
time_offset:units = "hours since 2011-06-21T00:00:00Z" ;
time_offset:missing_value = NaN ;
float Thgt(time, z, lat, lon) ;
Thgt:units = "meters" ;
Thgt:long_name = "significant wave height" ;
Thgt:standard_name = "sea_surface_wave_significant_height" ;
Thgt:short_name = "Thgt" ;
Thgt:valid_range = 0.f, 60.f ;
Thgt:_FillValue = NaNf ;
Thgt:coordinates = "time_run time z lat lon " ;
float Tper(time, z, lat, lon) ;
Tper:units = "seconds" ;
Tper:long_name = "peak wave period" ;
Tper:standard_name = "sea_surface_wave_period_at_variance_spectral_density_maximum" ;
Tper:short_name = "Tper" ;
Tper:valid_range = 0.f, 60.f ;
Tper:_FillValue = NaNf ;
Tper:coordinates = "time_run time z lat lon " ;
float Tdir(time, z, lat, lon) ;
Tdir:units = "degrees" ;
Tdir:long_name = "peak wave direction" ;
Tdir:standard_name = "sea_surface_wave_from_direction" ;
Tdir:short_name = "Tdir" ;
Tdir:valid_range = 0.f, 360.f ;
Tdir:_FillValue = NaNf ;
Tdir:coordinates = "time_run time z lat lon " ;
float shgt(time, z, lat, lon) ;
shgt:units = "meters" ;
shgt:long_name = "swell significant wave height" ;
shgt:standard_name = "sea_surface_swell_wave_significant_height" ;
shgt:short_name = "shgt" ;
shgt:valid_range = 0.f, 60.f ;
shgt:_FillValue = NaNf ;
shgt:coordinates = "time_run time z lat lon " ;
float sper(time, z, lat, lon) ;
sper:units = "seconds" ;
sper:long_name = "swell peak wave period" ;
sper:standard_name = "sea_surface_swell_wave_period" ;
sper:short_name = "sper" ;
sper:valid_range = 0.f, 60.f ;
sper:_FillValue = NaNf ;
sper:coordinates = "time_run time z lat lon " ;
float sdir(time, z, lat, lon) ;
sdir:units = "degrees" ;
sdir:long_name = "swell peak wave direction" ;
sdir:standard_name = "sea_surface_swell_wave_from_direction" ;
sdir:short_name = "sdir" ;
sdir:valid_range = 0.f, 360.f ;
sdir:_FillValue = NaNf ;
sdir:coordinates = "time_run time z lat lon " ;
float whgt(time, z, lat, lon) ;
whgt:units = "meters" ;
whgt:long_name = "wind significant wave height" ;
whgt:standard_name = "sea_surface_wind_wave_significant_height" ;
whgt:short_name = "whgt" ;
whgt:valid_range = 0.f, 60.f ;
whgt:_FillValue = NaNf ;
whgt:coordinates = "time_run time z lat lon " ;
float wper(time, z, lat, lon) ;
wper:units = "seconds" ;
wper:long_name = "wind peak wave period" ;
wper:standard_name = "sea_surface_wind_wave_period" ;
wper:short_name = "wper" ;
wper:valid_range = 0.f, 60.f ;
wper:_FillValue = NaNf ;
wper:coordinates = "time_run time z lat lon " ;
float wdir(time, z, lat, lon) ;
wdir:units = "degrees" ;
wdir:long_name = "wind peak wave direction" ;
wdir:standard_name = "sea_surface_wind_wave_from_direction" ;
wdir:short_name = "wdir" ;
wdir:valid_range = 0.f, 360.f ;
wdir:_FillValue = NaNf ;
wdir:coordinates = "time_run time z lat lon " ;
// global attributes:
:title = "WaveWatch III (WW3) Hawaii Regional Wave Model" ;
:_CoordSysBuilder = "ucar.nc2.dataset.conv.CF1Convention" ;
:Conventions = "CF-1.6, ACDD-1.3" ;
:cdm_data_type = "Grid" ;
:featureType = "GRID" ;
:location = "Proto fmrc:WaveWatch_III_Hawaii_Regional_Wave_Model" ;
:id = "ww3_hawaii" ;
:naming_authority = "org.pacioos" ;
:Metadata_Link = "http://pacioos.org/metadata/ww3_hawaii.html" ;
:ISO_Topic_Categories = "oceans" ;
:summary = "Through a collaborative effort with NOAA/NCEP and NWS Honolulu, the University of Hawaii has implemented a global-scale WaveWatch III (WW3) model, which in turn provides boundary conditions for this Hawaii regional WW3: a 7-day model with a 5-day hourly forecast at approximately 5-km or 0.05-deg resolution. The primary purpose of this regional model is to capture island effects such as island shadowing, refraction, and accurate modeling of local wind waves. Hawaii WW3 is forced with winds from the University of Hawaii Meteorology Department\'s operational mesoscale model, which has a more suitable spatial resolution than the global scale wind field. The Hawaii regional WW3 also provides boundary conditions for nearshore island-scale models via Simulating WAves Nearshore (SWAN). While considerable effort has been made to implement all model components in a thorough, correct, and accurate manner, numerous sources of error are possible. As such, please use these data with the caution appropriate for any ocean related activity." ;
:keywords = "Earth Science Services > Models > Ocean General Circulation Models (OGCM)/Regional Ocean Models, Earth Science Services > Models > Weather Research/Forecast Models, Earth Science > Oceans > Ocean Waves > Significant Wave Height, Earth Science > Oceans > Ocean Waves > Wave Period, Earth Science > Oceans > Ocean Waves > Wave Speed/Direction" ;
:keywords_vocabulary = "GCMD Science Keywords" ;
:platform = "Models/Analyses > > Operational Models" ;
:platform_vocabulary = "GCMD Platform Keywords" ;
:instrument = "Not Applicable > Not Applicable" ;
:instrument_vocabulary = "GCMD Instrument Keywords" ;
:locations = "Continent > North America > United States Of America > Hawaii, Ocean > Pacific Ocean > Central Pacific Ocean > Hawaiian Islands" ;
:locations_vocabulary = "GCMD Location Keywords" ;
:standard_name_vocabulary = "CF Standard Name Table v39" ;
:comment = "Model runs produced by Dr. Kwok Fai Cheung (cheung@hawaii.edu)." ;
:geospatial_lat_min = 18. ;
:geospatial_lat_max = 23. ;
:geospatial_lon_min = 199. ;
:geospatial_lon_max = 206. ;
:geospatial_vertical_min = 0. ;
:geospatial_vertical_max = 0. ;
:geospatial_bounds = "POLYGON ((18 -161.0, 23 -161.0, 23 -154.0, 18 -154.0, 18 -161.0))" ;
:geospatial_bounds_crs = "EPSG:4326" ;
:time_coverage_start = "2011-06-21T21:00:00Z" ;
:geospatial_lat_units = "degrees_north" ;
:geospatial_lat_resolution = 0.05 ;
:geospatial_lon_units = "degrees_east" ;
:geospatial_lon_resolution = 0.05 ;
:geospatial_vertical_units = "meters" ;
:geospatial_vertical_positive = "up" ;
:geospatial_vertical_resolution = 0. ;
:time_coverage_resolution = "PT1H" ;
:creator_email = "cheung@hawaii.edu" ;
:creator_name = "Kwok Fai Cheung" ;
:creator_type = "person" ;
:creator_url = "http://www.ore.hawaii.edu/OE/cheung_research.htm" ;
:creator_institution = "University of Hawaii" ;
:date_created = "2011-06-22" ;
:date_issued = "2011-06-22" ;
:date_modified = "2014-06-23" ;
:date_metadata_modified = "2017-01-30" ;
:institution = "University of Hawaii" ;
:project = "Pacific Islands Ocean Observing System (PacIOOS)" ;
:program = "Pacific Islands Ocean Observing System (PacIOOS)" ;
:contributor_name = "Jim Potemra" ;
:contributor_role = "distributor" ;
:publisher_email = "info@pacioos.org" ;
:publisher_name = "Pacific Islands Ocean Observing System (PacIOOS)" ;
:publisher_url = "http://pacioos.org" ;
:publisher_institution = "University of Hawaii" ;
:publisher_type = "group" ;
:license = "The data may be used and redistributed for free but is not intended for legal use, since it may contain inaccuracies. Neither the data Contributor, University of Hawaii, PacIOOS, NOAA, State of Hawaii nor the United States Government, nor any of their employees or contractors, makes any warranty, express or implied, including warranties of merchantability and fitness for a particular purpose, or assumes any legal liability for the accuracy, completeness, or usefulness, of this information." ;
:acknowledgement = "The Pacific Islands Ocean Observing System (PacIOOS) is funded through the National Oceanic and Atmospheric Administration (NOAA) as a Regional Association within the U.S. Integrated Ocean Observing System (IOOS). PacIOOS is coordinated by the University of Hawaii School of Ocean and Earth Science and Technology (SOEST)." ;
:source = "WaveWatch III (WW3) numerical wave model" ;
:references = "http://pacioos.org/waves/model-hawaii/, http://polar.ncep.noaa.gov/waves/wavewatch/" ;
:history = "FMRC Best Dataset" ;
答案 0 :(得分：2)
默认情况下，xarray将具有“{”形式的units属性的变量转换为np.timedelta64，将具有“自......以来的秒数”单位的变量转换为np.datetime64。
这对某些应用程序来说很方便。为了使其更加具体，让我在netCDF文件中突出显示ncdump -h部分：
netcdf WaveWatch_III_Hawaii_Regional_Wave_Model_best {
variables:
double time(time) ;
time:long_name = "Forecast time for ForecastModelRunCollection" ;
time:standard_name = "time" ;
time:calendar = "proleptic_gregorian" ;
time:units = "hours since 2011-06-21 00:00:00.000 UTC" ;
time:missing_value = NaN ;
time:_CoordinateAxisType = "Time" ;
double time_run(time) ;
time_run:long_name = "run times for coordinate = time" ;
time_run:standard_name = "forecast_reference_time" ;
time_run:calendar = "proleptic_gregorian" ;
time_run:units = "hours since 2011-06-21 00:00:00.000 UTC" ;
time_run:missing_value = NaN ;
time_run:_CoordinateAxisType = "RunTime" ;
double time_offset(time) ;
time_offset:long_name = "offset hour from start of run for coordinate = time" ;
time_offset:standard_name = "forecast_period" ;
time_offset:calendar = "proleptic_gregorian" ;
time_offset:units = "hours since 2011-06-21T00:00:00Z" ;
time_offset:missing_value = NaN ;
float Tper(time, z, lat, lon) ;
Tper:units = "seconds" ;
Tper:long_name = "peak wave period" ;
Tper:standard_name = "sea_surface_wave_period_at_variance_spectral_density_maximum" ;
Tper:short_name = "Tper" ;
Tper:valid_range = 0.f, 60.f ;
Tper:_FillValue = NaNf ;
Tper:coordinates = "time_run time z lat lon " ;
我对CF standard names的理解是forecast_period应该等于time和forecast_reference_time之间的差异，即forecast_period = time - forecast_reference_time。如果您使用“小时”形式的单位指定了time_offset变量，那么它将被解码为timedelta64，以及datetime64 time和time_run ，所以xarray的算术实际上会满足这个身份。如果您只想包含其中两个变量并想要动态计算第三个变量，您可能会觉得这很有用。
另一方面，您可能不希望将Tper变量转换为timedelta64。从技术上讲，它也是一个时间段，但与time进行比较并不是一个有意义的变量。
不幸的是，xarray目前无法区分这两种情况而不是查看变量元数据，因此它将所有内容转换为timedelta64（尽管我很乐意讨论您可能在GitHub上提出的想法）。目前，最好的解决方法是在使用xarray解码之前从这些变量中删除units属性，例如，
raw = xr.open_dataset(url, decode_cf=False)
del raw.Tper.attrs['units']
ds = xr.decode_cf(raw)
答案 1 :(得分：1)
如果可能，您是否介意显示ncdump -h file.nc的输出？
我快速检查了一下我运行的WaveWatchIII实验的一些海浪时期数据，一切看起来都很好。
ncdump -h ww3.20140101_20141231.nc
netcdf ww3.20140101_20141231 {
dimensions:
longitude = 46 ;
latitude = 17 ;
time = UNLIMITED ; // (365 currently)
variables:
float longitude(longitude) ;
longitude:units = "degree_east" ;
longitude:long_name = "longitude" ;
longitude:standard_name = "longitude" ;
longitude:valid_min = -180.f ;
longitude:valid_max = 180.f ;
longitude:axis = "X" ;
float latitude(latitude) ;
latitude:units = "degree_north" ;
latitude:long_name = "latitude" ;
latitude:standard_name = "latitude" ;
latitude:valid_min = -90.f ;
latitude:valid_max = 90.f ;
latitude:axis = "Y" ;
double time(time) ;
time:long_name = "julian day (UT)" ;
time:standard_name = "time" ;
time:units = "days since 1850-01-01T00:00:00Z" ;
time:conventions = "relative julian days with decimal part (as parts of the day )" ;
time:axis = "T" ;
...
short t01(time, latitude, longitude) ;
t01:long_name = "mean period T01" ;
t01:standard_name = "sea_surface_wind_wave_mean_period_from_variance_spectral_density_first_frequency_moment" ;
t01:globwave_name = "mean_period_t01" ;
t01:units = "s" ;
t01:_FillValue = -32767s ;
t01:scale_factor = 0.01f ;
t01:add_offset = 0.f ;
t01:valid_min = 0 ;
t01:valid_max = 5000 ;
...
>>> import xarray as xr
>>> ds = xr.open_dataset('ww3.20140101_20141231.nc')
>>> da = ds['t01']
>>> da
<xarray.DataArray 't01' (time: 365, latitude: 17, longitude: 46)>
[285430 values with dtype=float64]
Coordinates:
* longitude (longitude) float32 -5.0 -4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 ...
* latitude (latitude) float32 30.0 31.0 32.0 33.0 34.0 35.0 36.0 37.0 ...
* time (time) datetime64[ns] 2014-01-01 2014-01-02 2014-01-03
...
Attributes:
long_name: mean period T01
standard_name:
sea_surface_wind_wave_mean_period_from_variance_spectral_...
globwave_name: mean_period_t01
units: s
valid_min: 0
valid_max: 5000
>>> da[10,10,10]
<xarray.DataArray 't01' ()>
array(2.9799999333918095)
Coordinates:
longitude float32 5.0
latitude float32 40.0
time datetime64[ns] 2014-01-11
Attributes:
long_name: mean period T01
standard_name: sea_surface_wind_wave_mean_period_from_variance_spectral_...
globwave_name: mean_period_t01
units: s
valid_min: 0
valid_max: 5000
我记得唯一不同的是我将WW3v4.18 / ftn / ww3_ounf.ftn中的L2374从Jday0=julday(1,1,1990)更改为Jday0=julday(1,1,1850)，并将WW3v4.18 / ftn / ww3_ounp.ftn中的类似行更改为我在1990年之前做过跑步，并不想得到奇怪的负面时间。
我还使用OPeNDAP快速检查了一些（气候模型）数据：
>>> import xarray as xr
>>> remote_data = xr.open_dataset('http://iridl.ldeo.columbia.edu/SOURCES/.Models/.SubX/.RSMAS/.CCSM4/.forecast/.ts/dods')
<xarray.Dataset>
Dimensions: (L: 45, M: 9, S: 15, X: 360, Y: 181)
Coordinates:
* S (S) datetime64[ns] 2017-06-25 2017-07-02 2017-07-09 2017-07-16 ...
* M (M) float32 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0
* X (X) float32 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 ...
* L (L) timedelta64[ns] 0 days 12:00:00 1 days 12:00:00 ...
* Y (Y) float32 -90.0 -89.0 -88.0 -87.0 -86.0 -85.0 -84.0 -83.0 ...
Data variables:
ts (M, L, S, Y, X) float64 ...
Attributes:
Conventions: IRIDL
S：开始日期; M：合奏成员; X：lon; Y：lat; L：交货时间
>>> ts = remote_data['ts'][0, 0, -1, 45, 80]
>>> ts
<xarray.DataArray 'ts' ()>
array(281.383544921875)
Coordinates:
S datetime64[ns] 2017-10-01
M float32 1.0
X float32 80.0
L timedelta64[ns] 12:00:00
Y float32 -45.0
Attributes:
pointwidth: 0.0
standard_name: surface_temperature
long_name: Surface Temperature
level_type: surface
units: Kelvin_scale
cell_method: time: mean
在我的情况下，我应该结合S和L来制作T（时间）坐标。然后将T coord附加到DataArray并删除S和L坐标。不幸的是，我不知道如何做到这一点：
>>> T = ts.coords['S'].values + ts.coords['L'].values
>>> T
numpy.datetime64('2017-10-01T12:00:00.000000000')