Source code for pyart.io.cfradial

"""
Utilities for reading CF/Radial files.

"""

import datetime
import getpass
import platform
import warnings

import netCDF4
import numpy as np

from ..config import FileMetadata
from .common import stringarray_to_chararray, _test_arguments
from ..core.radar import Radar
from ..lazydict import LazyLoadDict


# Variables and dimensions in the instrument_parameter convention and
# radar_parameters sub-convention that will be read from and written to
# CfRadial files using Py-ART.
# The meta_group attribute cannot be used to identify these parameters as
# it is often set incorrectly.
_INSTRUMENT_PARAMS_DIMS = {
    # instrument_parameters sub-convention
    'frequency': ('frequency'),
    'follow_mode': ('sweep', 'string_length'),
    'pulse_width': ('time', ),
    'prt_mode': ('sweep', 'string_length'),
    'prt': ('time', ),
    'prt_ratio': ('time', ),
    'polarization_mode': ('sweep', 'string_length'),
    'nyquist_velocity': ('time', ),
    'unambiguous_range': ('time', ),
    'n_samples': ('time', ),
    'sampling_ratio': ('time', ),
    # radar_parameters sub-convention
    'radar_antenna_gain_h': (),
    'radar_antenna_gain_v': (),
    'radar_beam_width_h': (),
    'radar_beam_width_v': (),
    'radar_receiver_bandwidth': (),
    'radar_measured_transmit_power_h': ('time', ),
    'radar_measured_transmit_power_v': ('time', ),
    'radar_rx_bandwidth': (),           # non-standard
    'measured_transmit_power_v': ('time', ),    # non-standard
    'measured_transmit_power_h': ('time', ),    # non-standard
}


[docs]def read_cfradial(filename, field_names=None, additional_metadata=None, file_field_names=False, exclude_fields=None, include_fields=None, delay_field_loading=False, **kwargs): """ Read a Cfradial netCDF file. Parameters ---------- filename : str Name of CF/Radial netCDF file to read data from. field_names : dict, optional Dictionary mapping field names in the file names to radar field names. Unlike other read functions, fields not in this dictionary or having a value of None are still included in the radar.fields dictionary, to exclude them use the `exclude_fields` parameter. Fields which are mapped by this dictionary will be renamed from key to value. additional_metadata : dict of dicts, optional This parameter is not used, it is included for uniformity. file_field_names : bool, optional True to force the use of the field names from the file in which case the `field_names` parameter is ignored. False will use to `field_names` parameter to rename fields. exclude_fields : list or None, optional List of fields to exclude from the radar object. This is applied after the `file_field_names` and `field_names` parameters. Set to None to include all fields specified by include_fields. include_fields : list or None, optional List of fields to include from the radar object. This is applied after the `file_field_names` and `field_names` parameters. Set to None to include all fields not specified by exclude_fields. delay_field_loading : bool True to delay loading of field data from the file until the 'data' key in a particular field dictionary is accessed. In this case the field attribute of the returned Radar object will contain LazyLoadDict objects not dict objects. Delayed field loading will not provide any speedup in file where the number of gates vary between rays (ngates_vary=True) and is not recommended. Returns ------- radar : Radar Radar object. Notes ----- This function has not been tested on "stream" Cfradial files. """ # test for non empty kwargs _test_arguments(kwargs) # create metadata retrieval object filemetadata = FileMetadata('cfradial', field_names, additional_metadata, file_field_names, exclude_fields) # read the data ncobj = netCDF4.Dataset(filename) ncvars = ncobj.variables # 4.1 Global attribute -> move to metadata dictionary metadata = dict([(k, getattr(ncobj, k)) for k in ncobj.ncattrs()]) if 'n_gates_vary' in metadata: metadata['n_gates_vary'] = 'false' # corrected below # 4.2 Dimensions (do nothing) # 4.3 Global variable -> move to metadata dictionary if 'volume_number' in ncvars: metadata['volume_number'] = int(ncvars['volume_number'][:]) else: metadata['volume_number'] = 0 global_vars = {'platform_type': 'fixed', 'instrument_type': 'radar', 'primary_axis': 'axis_z'} # ignore time_* global variables, these are calculated from the time # variable when the file is written. for var, default_value in global_vars.items(): if var in ncvars: metadata[var] = str(netCDF4.chartostring(ncvars[var][:])) else: metadata[var] = default_value # 4.4 coordinate variables -> create attribute dictionaries time = _ncvar_to_dict(ncvars['time']) _range = _ncvar_to_dict(ncvars['range']) # 4.5 Ray dimension variables # 4.6 Location variables -> create attribute dictionaries latitude = _ncvar_to_dict(ncvars['latitude']) longitude = _ncvar_to_dict(ncvars['longitude']) altitude = _ncvar_to_dict(ncvars['altitude']) if 'altitude_agl' in ncvars: altitude_agl = _ncvar_to_dict(ncvars['altitude_agl']) else: altitude_agl = None # 4.7 Sweep variables -> create atrribute dictionaries sweep_mode = _ncvar_to_dict(ncvars['sweep_mode']) fixed_angle = _ncvar_to_dict(ncvars['fixed_angle']) sweep_start_ray_index = _ncvar_to_dict(ncvars['sweep_start_ray_index']) sweep_end_ray_index = _ncvar_to_dict(ncvars['sweep_end_ray_index']) if 'sweep_number' in ncvars: sweep_number = _ncvar_to_dict(ncvars['sweep_number']) else: nsweeps = len(sweep_start_ray_index['data']) sweep_number = filemetadata('sweep_number') sweep_number['data'] = np.arange(nsweeps, dtype='float32') warnings.warn("Warning: File violates CF/Radial convention. " + "Missing sweep_number variable") if 'target_scan_rate' in ncvars: target_scan_rate = _ncvar_to_dict(ncvars['target_scan_rate']) else: target_scan_rate = None if 'rays_are_indexed' in ncvars: rays_are_indexed = _ncvar_to_dict(ncvars['rays_are_indexed']) else: rays_are_indexed = None if 'ray_angle_res' in ncvars: ray_angle_res = _ncvar_to_dict(ncvars['ray_angle_res']) else: ray_angle_res = None # Uses ARM scan name if present. if hasattr(ncobj, 'scan_name'): mode = ncobj.scan_name else: # first sweep mode determines scan_type try: mode = netCDF4.chartostring( sweep_mode['data'][0])[()].decode('utf-8') except AttributeError: # Python 3, all strings are already unicode. mode = netCDF4.chartostring(sweep_mode['data'][0])[()] mode = mode.strip() # options specified in the CF/Radial standard if mode == 'rhi': scan_type = 'rhi' elif mode == 'vertical_pointing': scan_type = 'vpt' elif mode == 'azimuth_surveillance': scan_type = 'ppi' elif mode == 'elevation_surveillance': scan_type = 'rhi' elif mode == 'manual_ppi': scan_type = 'ppi' elif mode == 'manual_rhi': scan_type = 'rhi' # fallback types elif 'sur' in mode: scan_type = 'ppi' elif 'sec' in mode: scan_type = 'sector' elif 'rhi' in mode: scan_type = 'rhi' elif 'ppi' in mode: scan_type = 'ppi' else: scan_type = 'other' # 4.8 Sensor pointing variables -> create attribute dictionaries azimuth = _ncvar_to_dict(ncvars['azimuth']) elevation = _ncvar_to_dict(ncvars['elevation']) if 'scan_rate' in ncvars: scan_rate = _ncvar_to_dict(ncvars['scan_rate']) else: scan_rate = None if 'antenna_transition' in ncvars: antenna_transition = _ncvar_to_dict(ncvars['antenna_transition']) else: antenna_transition = None # 4.9 Moving platform geo-reference variables # Aircraft specific varaibles if 'rotation' in ncvars: rotation = _ncvar_to_dict(ncvars['rotation']) else: rotation = None if 'tilt' in ncvars: tilt = _ncvar_to_dict(ncvars['tilt']) else: tilt = None if 'roll' in ncvars: roll = _ncvar_to_dict(ncvars['roll']) else: roll = None if 'drift' in ncvars: drift = _ncvar_to_dict(ncvars['drift']) else: drift = None if 'heading' in ncvars: heading = _ncvar_to_dict(ncvars['heading']) else: heading = None if 'pitch' in ncvars: pitch = _ncvar_to_dict(ncvars['pitch']) else: pitch = None if 'georefs_applied' in ncvars: georefs_applied = _ncvar_to_dict(ncvars['georefs_applied']) else: georefs_applied = None # 4.10 Moments field data variables -> field attribute dictionary if 'ray_n_gates' in ncvars: # all variables with dimensions of n_points are fields. keys = [k for k, v in ncvars.items() if v.dimensions == ('n_points', )] else: # all variables with dimensions of 'time', 'range' are fields keys = [k for k, v in ncvars.items() if v.dimensions == ('time', 'range')] fields = {} for key in keys: field_name = filemetadata.get_field_name(key) if field_name is None: if exclude_fields is not None and key in exclude_fields: if key not in include_fields: continue if include_fields is None or key in include_fields: field_name = key else: continue fields[field_name] = _ncvar_to_dict(ncvars[key], delay_field_loading) if 'ray_n_gates' in ncvars: shape = (len(ncvars['time']), len(ncvars['range'])) ray_n_gates = ncvars['ray_n_gates'][:] ray_start_index = ncvars['ray_start_index'][:] for dic in fields.values(): _unpack_variable_gate_field_dic( dic, shape, ray_n_gates, ray_start_index) # 4.5 instrument_parameters sub-convention -> instrument_parameters dict # 4.6 radar_parameters sub-convention -> instrument_parameters dict keys = [k for k in _INSTRUMENT_PARAMS_DIMS.keys() if k in ncvars] instrument_parameters = dict((k, _ncvar_to_dict(ncvars[k])) for k in keys) if instrument_parameters == {}: # if no parameters set to None instrument_parameters = None # 4.7 lidar_parameters sub-convention -> skip # 4.8 radar_calibration sub-convention -> radar_calibration keys = _find_all_meta_group_vars(ncvars, 'radar_calibration') radar_calibration = dict((k, _ncvar_to_dict(ncvars[k])) for k in keys) if radar_calibration == {}: radar_calibration = None # do not close file if field loading is delayed if not delay_field_loading: ncobj.close() return Radar( time, _range, fields, metadata, scan_type, latitude, longitude, altitude, sweep_number, sweep_mode, fixed_angle, sweep_start_ray_index, sweep_end_ray_index, azimuth, elevation, instrument_parameters=instrument_parameters, radar_calibration=radar_calibration, altitude_agl=altitude_agl, scan_rate=scan_rate, antenna_transition=antenna_transition, target_scan_rate=target_scan_rate, rays_are_indexed=rays_are_indexed, ray_angle_res=ray_angle_res, rotation=rotation, tilt=tilt, roll=roll, drift=drift, heading=heading, pitch=pitch, georefs_applied=georefs_applied)
def _find_all_meta_group_vars(ncvars, meta_group_name): """ Return a list of all variables which are in a given meta_group. """ return [k for k, v in ncvars.items() if 'meta_group' in v.ncattrs() and v.meta_group == meta_group_name] def _ncvar_to_dict(ncvar, lazydict=False): """ Convert a NetCDF Dataset variable to a dictionary. """ # copy all attribute except for scaling parameters d = dict((k, getattr(ncvar, k)) for k in ncvar.ncattrs() if k not in ['scale_factor', 'add_offset']) data_extractor = _NetCDFVariableDataExtractor(ncvar) if lazydict: d = LazyLoadDict(d) d.set_lazy('data', data_extractor) else: d['data'] = data_extractor() return d class _NetCDFVariableDataExtractor(object): """ Class facilitating on demand extraction of data from a NetCDF variable. Parameters ---------- ncvar : netCDF4.Variable NetCDF Variable from which data will be extracted. """ def __init__(self, ncvar): """ initialize the object. """ self.ncvar = ncvar def __call__(self): """ Return an array containing data from the stored variable. """ data = self.ncvar[:] if data is np.ma.masked: # If the data is a masked scalar, MaskedConstant is returned by # NetCDF4 version 1.2.3+. This object does not preserve the dtype # and fill_value of the original NetCDF variable and causes issues # in Py-ART. # Rather we create a masked array with a single masked value # with the correct dtype and fill_value. self.ncvar.set_auto_mask(False) data = np.ma.array(self.ncvar[:], mask=True) # Use atleast_1d to force the array to be at minimum one dimensional, # some version of netCDF return scalar or scalar arrays for scalar # NetCDF variables. return np.atleast_1d(data) def _unpack_variable_gate_field_dic( dic, shape, ray_n_gates, ray_start_index): """ Create a 2D array from a 1D field data, dic update in place. """ fdata = dic['data'] data = np.ma.masked_all(shape, dtype=fdata.dtype) for i, (gates, idx) in enumerate(zip(ray_n_gates, ray_start_index)): data[i, :gates] = fdata[idx:idx+gates] dic['data'] = data return
[docs]def write_cfradial(filename, radar, format='NETCDF4', time_reference=None, arm_time_variables=False): """ Write a Radar object to a CF/Radial compliant netCDF file. The files produced by this routine follow the `CF/Radial standard`_. Attempts are also made to to meet many of the standards outlined in the `ARM Data File Standards`_. .. _CF/Radial standard: http://www.ral.ucar.edu/projects/titan/docs/radial_formats/cfradial.html .. _ARM Data File Standards: https://docs.google.com/document/d/1gBMw4Kje6v8LBlsrjaGFfSLoU0jRx-07TIazpthZGt0/edit?pli=1 To control how the netCDF variables are created, set any of the following keys in the radar attribute dictionaries. * _Zlib * _DeflateLevel * _Shuffle * _Fletcher32 * _Continguous * _ChunkSizes * _Endianness * _Least_significant_digit * _FillValue See the netCDF4 documentation for details on these settings. Parameters ---------- filename : str Filename to create. radar : Radar Radar object. format : str, optional NetCDF format, one of 'NETCDF4', 'NETCDF4_CLASSIC', 'NETCDF3_CLASSIC' or 'NETCDF3_64BIT'. See netCDF4 documentation for details. time_reference : bool True to include a time_reference variable, False will not include this variable. The default, None, will include the time_reference variable when the first time value is non-zero. arm_time_variables : bool True to create the ARM standard time variables base_time and time_offset, False will not create these variables. """ dataset = netCDF4.Dataset(filename, 'w', format=format) # determine the maximum string length max_str_len = len(radar.sweep_mode['data'][0]) for k in ['follow_mode', 'prt_mode', 'polarization_mode']: if ((radar.instrument_parameters is not None) and (k in radar.instrument_parameters)): sdim_length = len(radar.instrument_parameters[k]['data'][0]) max_str_len = max(max_str_len, sdim_length) str_len = max(max_str_len, 32) # minimum string legth of 32 # create time, range and sweep dimensions dataset.createDimension('time', None) dataset.createDimension('range', radar.ngates) dataset.createDimension('sweep', radar.nsweeps) dataset.createDimension('string_length', str_len) # global attributes # remove global variables from copy of metadata metadata_copy = dict(radar.metadata) global_variables = ['volume_number', 'platform_type', 'instrument_type', 'primary_axis', 'time_coverage_start', 'time_coverage_end', 'time_reference'] for var in global_variables: if var in metadata_copy: metadata_copy.pop(var) # determine the history attribute if it doesn't exist, save for # the last attribute. if 'history' in metadata_copy: history = metadata_copy.pop('history') else: user = getpass.getuser() node = platform.node() time_str = datetime.datetime.now().isoformat() t = (user, node, time_str) history = 'created by %s on %s at %s using Py-ART' % (t) dataset.setncatts(metadata_copy) if 'Conventions' not in dataset.ncattrs(): dataset.setncattr('Conventions', "CF/Radial") if 'field_names' not in dataset.ncattrs(): dataset.setncattr('field_names', ', '.join(radar.fields.keys())) # history should be the last attribute, ARM standard dataset.setncattr('history', history) # arm time variables base_time and time_offset if requested if arm_time_variables: dt = netCDF4.num2date(radar.time['data'][0], radar.time['units'], only_use_cftime_datetimes=False, only_use_python_datetimes=True) td = dt - datetime.datetime.utcfromtimestamp(0) base_time = { 'data': np.array([td.seconds + td.days * 24 * 3600], 'int32'), 'string': dt.strftime('%d-%b-%Y,%H:%M:%S GMT'), 'units': 'seconds since 1970-1-1 0:00:00 0:00', 'ancillary_variables': 'time_offset', 'long_name': 'Base time in Epoch', } _create_ncvar(base_time, dataset, 'base_time', ()) time_offset = { 'data': radar.time['data'], 'long_name': 'Time offset from base_time', 'units': radar.time['units'].replace('T', ' ').replace('Z', ''), 'ancillary_variables': 'time_offset', 'calendar': 'gregorian', } _create_ncvar(time_offset, dataset, 'time_offset', ('time', )) # standard variables _create_ncvar(radar.time, dataset, 'time', ('time', )) _create_ncvar(radar.range, dataset, 'range', ('range', )) _create_ncvar(radar.azimuth, dataset, 'azimuth', ('time', )) _create_ncvar(radar.elevation, dataset, 'elevation', ('time', )) # optional sensor pointing variables if radar.scan_rate is not None: _create_ncvar(radar.scan_rate, dataset, 'scan_rate', ('time', )) if radar.antenna_transition is not None: _create_ncvar(radar.antenna_transition, dataset, 'antenna_transition', ('time', )) # fields for field, dic in radar.fields.items(): _create_ncvar(dic, dataset, field, ('time', 'range')) # sweep parameters _create_ncvar(radar.sweep_number, dataset, 'sweep_number', ('sweep', )) _create_ncvar(radar.fixed_angle, dataset, 'fixed_angle', ('sweep', )) _create_ncvar(radar.sweep_start_ray_index, dataset, 'sweep_start_ray_index', ('sweep', )) _create_ncvar(radar.sweep_end_ray_index, dataset, 'sweep_end_ray_index', ('sweep', )) _create_ncvar(radar.sweep_mode, dataset, 'sweep_mode', ('sweep', 'string_length')) if radar.target_scan_rate is not None: _create_ncvar(radar.target_scan_rate, dataset, 'target_scan_rate', ('sweep', )) if radar.rays_are_indexed is not None: _create_ncvar(radar.rays_are_indexed, dataset, 'rays_are_indexed', ('sweep', 'string_length')) if radar.ray_angle_res is not None: _create_ncvar(radar.ray_angle_res, dataset, 'ray_angle_res', ('sweep', )) # instrument_parameters if ((radar.instrument_parameters is not None) and ('frequency' in radar.instrument_parameters.keys())): size = len(radar.instrument_parameters['frequency']['data']) dataset.createDimension('frequency', size) if radar.instrument_parameters is not None: for k in radar.instrument_parameters.keys(): if k in _INSTRUMENT_PARAMS_DIMS: dim = _INSTRUMENT_PARAMS_DIMS[k] _create_ncvar(radar.instrument_parameters[k], dataset, k, dim) else: # Do not try to write instrument parameter whose dimensions are # not known, rather issue a warning and skip the parameter message = ("Unknown instrument parameter: %s, " % (k) + "not written to file.") warnings.warn(message) # radar_calibration variables if radar.radar_calibration is not None and radar.radar_calibration != {}: size = [len(d['data']) for k, d in radar.radar_calibration.items() if k not in ['r_calib_index', 'r_calib_time']][0] dataset.createDimension('r_calib', size) for key, dic in radar.radar_calibration.items(): if key == 'r_calib_index': dims = ('time', ) elif key == 'r_calib_time': dims = ('r_calib', 'string_length') else: dims = ('r_calib', ) _create_ncvar(dic, dataset, key, dims) # latitude, longitude, altitude, altitude_agl if radar.latitude['data'].size == 1: # stationary platform _create_ncvar(radar.latitude, dataset, 'latitude', ()) _create_ncvar(radar.longitude, dataset, 'longitude', ()) _create_ncvar(radar.altitude, dataset, 'altitude', ()) if radar.altitude_agl is not None: _create_ncvar(radar.altitude_agl, dataset, 'altitude_agl', ()) else: # moving platform _create_ncvar(radar.latitude, dataset, 'latitude', ('time', )) _create_ncvar(radar.longitude, dataset, 'longitude', ('time', )) _create_ncvar(radar.altitude, dataset, 'altitude', ('time', )) if radar.altitude_agl is not None: _create_ncvar(radar.altitude_agl, dataset, 'altitude_agl', ('time', )) # time_coverage_start and time_coverage_end variables time_dim = ('string_length', ) units = radar.time['units'] start_dt = netCDF4.num2date( radar.time['data'][0], units, only_use_cftime_datetimes=False, only_use_python_datetimes=True) if start_dt.microsecond != 0: # truncate to nearest second start_dt -= datetime.timedelta(microseconds=start_dt.microsecond) end_dt = netCDF4.num2date( radar.time['data'][-1], units, only_use_cftime_datetimes=False, only_use_python_datetimes=True) if end_dt.microsecond != 0: # round up to next second end_dt += (datetime.timedelta(seconds=1) - datetime.timedelta(microseconds=end_dt.microsecond)) start_dic = {'data': np.array(start_dt.isoformat() + 'Z', dtype='S'), 'long_name': 'UTC time of first ray in the file', 'units': 'unitless'} end_dic = {'data': np.array(end_dt.isoformat() + 'Z', dtype='S'), 'long_name': 'UTC time of last ray in the file', 'units': 'unitless'} _create_ncvar(start_dic, dataset, 'time_coverage_start', time_dim) _create_ncvar(end_dic, dataset, 'time_coverage_end', time_dim) # time_reference is required or requested. if time_reference is None: if radar.time['data'][0] == 0: time_reference = False else: time_reference = True if time_reference: ref_dic = {'data': np.array(radar.time['units'][-20:], dtype='S'), 'long_name': 'UTC time reference', 'units': 'unitless'} _create_ncvar(ref_dic, dataset, 'time_reference', time_dim) # global variables # volume_number, required vol_dic = {'long_name': 'Volume number', 'units': 'unitless'} if 'volume_number' in radar.metadata: vol_dic['data'] = np.array([radar.metadata['volume_number']], dtype='int32') else: vol_dic['data'] = np.array([0], dtype='int32') _create_ncvar(vol_dic, dataset, 'volume_number', ()) # platform_type, optional if 'platform_type' in radar.metadata: dic = {'long_name': 'Platform type', 'data': np.array(radar.metadata['platform_type'], dtype='S')} _create_ncvar(dic, dataset, 'platform_type', ('string_length', )) # instrument_type, optional if 'instrument_type' in radar.metadata: dic = {'long_name': 'Instrument type', 'data': np.array(radar.metadata['instrument_type'], dtype='S')} _create_ncvar(dic, dataset, 'instrument_type', ('string_length', )) # primary_axis, optional if 'primary_axis' in radar.metadata: dic = {'long_name': 'Primary axis', 'data': np.array(radar.metadata['primary_axis'], dtype='S')} _create_ncvar(dic, dataset, 'primary_axis', ('string_length', )) # moving platform geo-reference variables if radar.rotation is not None: _create_ncvar(radar.rotation, dataset, 'rotation', ('time', )) if radar.tilt is not None: _create_ncvar(radar.tilt, dataset, 'tilt', ('time', )) if radar.roll is not None: _create_ncvar(radar.roll, dataset, 'roll', ('time', )) if radar.drift is not None: _create_ncvar(radar.drift, dataset, 'drift', ('time', )) if radar.heading is not None: _create_ncvar(radar.heading, dataset, 'heading', ('time', )) if radar.pitch is not None: _create_ncvar(radar.pitch, dataset, 'pitch', ('time', )) if radar.georefs_applied is not None: _create_ncvar(radar.georefs_applied, dataset, 'georefs_applied', ('time', )) dataset.close()
def _create_ncvar(dic, dataset, name, dimensions): """ Create and fill a Variable in a netCDF Dataset object. Parameters ---------- dic : dict Radar dictionary to containing variable data and meta-data. dataset : Dataset NetCDF dataset to create variable in. name : str Name of variable to create. dimension : tuple of str Dimension of variable. """ # create array from list, etc. data = dic['data'] if isinstance(data, np.ndarray) is not True: warnings.warn("Warning, converting non-array to array:%s" % name) data = np.array(data) # convert string/unicode arrays to character arrays if data.dtype.char == 'U': # cast unicode arrays to char arrays data = data.astype('S') if data.dtype.char == 'S' and data.dtype != 'S1': data = stringarray_to_chararray(data) # determine netCDF variable arguments special_keys = { # dictionary keys which can be used to change the default values of # createVariable arguments, some of these map to netCDF special # attributes, other are Py-ART conventions. '_Zlib': 'zlib', '_DeflateLevel': 'complevel', '_Shuffle': 'shuffle', '_Fletcher32': 'fletcher32', '_Continguous': 'contiguous', '_ChunkSizes': 'chunksizes', '_Endianness': 'endian', '_Least_significant_digit': 'least_significant_digit', '_FillValue': 'fill_value', } kwargs = {'zlib': True} # default is to use compression for dic_key, kwargs_key in special_keys.items(): if dic_key in dic: kwargs[kwargs_key] = dic[dic_key] # the _Write_as_dtype key can be used to specify the netCDF dtype if '_Write_as_dtype' in dic: dtype = np.dtype(dic['_Write_as_dtype']) if np.issubdtype(dtype, np.integer): if 'scale_factor' not in dic and 'add_offset' not in dic: # calculate scale and offset scale, offset, fill = _calculate_scale_and_offset(dic, dtype) dic['scale_factor'] = scale dic['add_offset'] = offset dic['_FillValue'] = fill kwargs['fill_value'] = fill else: dtype = data.dtype # create the dataset variable ncvar = dataset.createVariable(name, dtype, dimensions, **kwargs) # long_name attribute first if present, ARM standard if 'long_name' in dic.keys(): ncvar.setncattr('long_name', dic['long_name']) # units attribute second if present, ARM standard if 'units' in dic.keys(): ncvar.setncattr('units', dic['units']) # set all attributes for key, value in dic.items(): if key in special_keys.keys(): continue if key in ['data', 'long_name', 'units']: continue ncvar.setncattr(key, value) # set the data if data.shape == (): data.shape = (1,) if data.dtype == 'S1': # string/char arrays # KLUDGE netCDF4 version around 1.1.6 do not expand an ellipsis # to zero dimensions (Issue #371 of netcdf4-python). # Solution is so we treat 1 dimensional string dimensions explicitly. if ncvar.ndim == 1: ncvar[:data.shape[-1]] = data[:] else: ncvar[..., :data.shape[-1]] = data[:] else: ncvar[:] = data[:] def _calculate_scale_and_offset(dic, dtype, minimum=None, maximum=None): """ Calculate appropriated 'scale_factor' and 'add_offset' for nc variable in dic in order to scaling to fit dtype range. Parameters ---------- dic : dict Radar dictionary containing variable data and meta-data. dtype : Numpy Dtype Integer numpy dtype to map to. minimum, maximum : float Greatest and smallest values in the data, those values will be mapped to the smallest+1 and greates values that dtype can hold. If equal to None, numpy.amin and numpy.amax will be used on the data contained in dic to determine these values. """ if "_FillValue" in dic: fillvalue = dic["_FillValue"] else: fillvalue = np.NaN data = dic['data'].copy() data = np.ma.array(data, mask=(~np.isfinite(data) | (data == fillvalue))) if minimum is None: minimum = np.amin(data) if maximum is None: maximum = np.amax(data) if maximum < minimum: raise ValueError( 'Error calculating variable scaling: ' 'maximum: %f is smaller than minimum: %f' % (maximum, minimum)) elif maximum == minimum: warnings.warn( 'While calculating variable scaling: ' 'maximum: %f is equal to minimum: %f' % (maximum, minimum)) maximum = minimum + 1 # get max and min scaled, maxi = np.iinfo(dtype).max mini = np.iinfo(dtype).min + 1 # +1 since min will serve as the fillvalue scale = float(maximum - minimum) / float(maxi - mini) offset = minimum - mini * scale return scale, offset, np.iinfo(dtype).min