Joh*_*rer 4 python netcdf opendap python-xarray
测量海浪周期的变量的“单位”属性以“秒”为单位。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，但显然并非如此。有没有办法让我告诉 xarray 将波浪周期变量作为 float32 摄取，或者在摄取后将它们从 timedelta64 转换回原始值？我仍然希望它将“时间”变量转换为 timedelta64，所以我不想关闭整个数据集的翻译，而只是针对特定变量（Tper、sper、wper）。
以下是我在 TDS 服务器中的波浪预测中使用的基本 OPeNDAP URL：
建议？谢谢！
您可以使用此处的 OPeNDAP 页面查看类似“ncdump”的输出：
或者您可以在 OPeNDAP URL 上运行 ncdump，如下所示：
结果如下：
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" ;
默认情况下，xarray 将具有units“秒”形式的属性的变量转换为np.timedelta64，并将具有“自...”形式的单位的变量转换为np.datetime64。
这对于某些应用程序来说很方便。为了使这一点更具体，让我突出显示ncdump -hnetCDF 文件中的一个部分：
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标准名称的理解是forecast_period应该等于time和之间的差异forecast_reference_time，即forecast_period = time - forecast_reference_time。如果您使用time_offset“小时”形式的单位指定变量，那么它将timedelta64与datetime64fortime和一起解码为time_run，因此 xarray 的算术实际上会满足此恒等式。如果您只想包含其中两个变量并想动态计算第三个变量，您可能会发现这很有用。
另一方面，您可能不想将Tper变量转换为timedelta64. 从技术上讲，它也是一个时间段，但它不是一个与 进行比较有意义的变量time。
不幸的是，xarray 目前无法通过查看变量元数据来区分这两种情况，因此它将所有内容转换为timedelta64（尽管我很乐意在 GitHub 上讨论您可能有的想法）。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单位属性，例如，
raw = xr.open_dataset(url, decode_cf=False)
del raw.Tper.attrs['units']
ds = xr.decode_cf(raw)