def merge(data, attach, standard_times=True, suffix='era', report=None, verbose=0):
"""Merge data and attach to one dataframe
Matching criteria are date and p
return dataframe with merged columns
Parameters
----------
data DataFrame Original Data
attach DataFrame Data that should be merged with Original
standard_times bool Consider only 0 and 12 UTC for merging
suffix str Suffix for Columns
verbose int verboseness
"""
# convert sonde data to standard_times to match better with ERA-Interim data?
if 'p' not in data.columns or 'p' not in attach.columns:
raise ValueError("Requires p in dataframes")
# print_verbose("Matching levels: %d"%(np.sum( np.in1d( data.p.unique(),attach.p.unique())) ) ,verbose)
# Match Index names
data.index.name = 'date'
attach.index.name = 'date'
if standard_times:
data = standard_dates_times(data, report=report, verbose=verbose)
journal('[MERGE] Data with standard times: %s' % str(data.shape), report, verbose)
if '_' not in suffix[0]:
suffix = '_' + suffix
return pd.merge(data.reset_index(),
attach.rename(columns=lambda x: x + suffix).reset_index(),
left_on=['date', 'p'],
right_on=['date', 'p'+suffix],
how='left').drop('p'+suffix, 1).set_index('date', drop=True)
def standard_data(data, interpolate=True, int_vars=None, method='murphy_koop', levels=None, init_replace=True,
replace=False, daynight=True, report=None, verbose=0, **kwargs):
"""Standardize radiosonde data
1. Quality Control
2. Datetime fix (0/12 Soundings)
3. Interpolation / std levels (ERA Levels)
4. Variable conversion (td, vp, dpd)
5. Final Quality Control
Parameters
----------
data pd.DataFrame Profile Data in Database format
interpolate bool Interpolate to levels (default: era_plevels)
method str saturation water vapor pressure formulation
levels list pressure levels for interpolation
init_replace bool First quality check repalce?
replace bool replace flagged values with NAN
daynight bool use only: 00 and 12 UTc
verbose int verboseness
kwargs dict
Returns
-------
pd.DataFrame
Raises
------
"""
from raso.config import era_plevels
from qc import control
from dpd_anomaly import dpd_anomaly
from interpolation import interp_dataframe
from standard_dates_times import standard_dates_times
funcid = "[STD] Data "
if not isinstance(data, pd.DataFrame):
raise ValueError(funcid + "Requires a DataFrame Database style")
start = set_starttime()
############################################################################
#
# Create a private copy and modify it
#
data = data.copy() # active copy
############################################################################
#
# QC ( replace with NAN)
#
data = control(data, replace=init_replace, report=report, verbose=verbose - 1)
#
# DPD 30 Detection
#
if 'dpd' in data.columns:
# for IGRA ?
data = dpd_anomaly(data, var='dpd', num_years=10, before_year='1994', replace=replace,
report=report, verbose=verbose - 1)
#
# Standard dates, times, remove duplicates!!
#
standard_dates_times(data, inplace=True, report=report, verbose=verbose - 1)
journal(funcid + 'Standard Dates-Times (%s)' % str(data.shape), report, verbose)
#
if daynight:
itime = data.index.hour * 100 + data.index.minute
itx = (itime == 0) | (itime == 1200) # including minutes
data = data.ix[itx, :].copy() # Limit to 0 and 12
journal(funcid + "Data [0, 12]: %s" % str(data.shape), report, verbose)
############################################################################
#
# Convert humidity variable to relative humidity for interpolation
# See manual
standard_rel_humidity(data, inplace=True, replace=replace, method=method, report=report, verbose=verbose - 1)
############################################################################
#
# Standard levels / interpolate
#
take_era = False
if levels is None:
levels = era_plevels # a lot of levels ?
take_era = True
if interpolate:
if int_vars is None:
int_vars = ['p', 't', 'r']
data = interp_dataframe(data, variables=int_vars, levels=levels, report=report, verbose=verbose - 1)
data = data.query('orig>0').copy() # .drop('orig',1) # Only Standard Levels
else:
cmd = "|".join(["p==%d" % ip for ip in levels])
data = data.query(cmd)
journal(funcid + 'Level selection (%d), ERA (%s)' % (len(levels), take_era), report, verbose)
############################################################################
#
# Final Quality Control
#
data = control(data, replace=replace, report=report, verbose=verbose - 1)
############################################################################
#
# Convert RH to vp, dpd
#
standard_water_vapor(data, inplace=True, replace=replace, method=method, report=report, verbose=verbose - 1)
#
standard_dewpoint_depression(data, inplace=True, replace=replace, method=method, report=report, verbose=verbose - 1)
#
# DPD 30 Detection
#
if 'dpd' in data.columns:
data = dpd_anomaly(data, var='dpd', num_years=10, before_year='1994', replace=replace, report=report,
verbose=verbose - 1)
############################################################################
if verbose > 0:
print_time(start)
return data
def merge_interpolate(data1, data2, variables=None, dropna=True, min_count=5, standard_time=True, d2name='era',
report=None, verbose=0):
"""
Interpolate and Merge two datasets:
1. convert to standard times (0,6,12,18)
2. Merge Datasets
3. Vertical logp interpolation
Interpolation levels are the union of both datasets
Parameters
----------
data1 DataFrame ['p', vars ... ]
data2 DataFrame ['p', vars ... ]
variables list list of variables to use
dropna bool remove missing values before interpolation
min_count int minimum values per profile
standard_times bool convert data to standard times
Returns
-------
newdata DataFrame ['p', vars ... ]
Raises
------
ValueError not a pandas DataFrame
RuntimeError missing p column
"""
from interpolation import interp_profile
funcid = "[M] "
if not isinstance(data1, pd.DataFrame) or not isinstance(data2, pd.DataFrame):
raise ValueError(funcid + "Requires a pandas DataFrame as input")
if 'p' not in data1.columns or 'p' not in data2.columns:
raise RuntimeError(funcid + "Missing pressure column: p")
# check variables ...
if variables is not None:
# check if p,t,r are in variables?
if 'p' not in variables:
if isinstance(variables, str):
variables = list(variables)
variables.append('p')
data1 = data1[variables].copy()
data2 = data2[variables].copy()
journal(funcid + "Subsetting ... %s" % ",".join(variables), report, verbose)
data1.index.name = 'date'
data2.index.name = 'date'
# match indices & merge (outer -> add both sides
if standard_time:
data1 = standard_dates_times(data1) # 0,6,12,18
else:
data1 = data1.copy()
# rename era variables and make sure we have floats!!
data2 = data2.astype('float64').rename(columns=lambda x: x + "_%s" % d2name).rename(
columns={'p_%s' % d2name: 'p'}).reset_index()
# add marker columns !?
data2['orig_%s' % d2name] = 1
data1['orig_raso'] = 1
journal(funcid + " Merging %s with %s on date and p" %(str(data1.shape), str(data2.shape)), report, verbose)
# Merge
# CHECK ? no data is lost ?
alldata = pd.merge(data1.reset_index(),
data2,
left_on=['date', 'p'],
right_on=['date', 'p'],
how='outer').set_index('date', drop=True)
# 0 raw, 1 both, 2 era
alldata['source'] = np.where(np.isfinite(alldata['orig_%s' % d2name]), 2, 0) # set ERA to 2
alldata['source'] = np.where((np.isfinite(alldata['orig_raso'])) & (alldata['source'] == 2), alldata['source'] - 1,
alldata['source']) # set RASO+ERA to 1
alldata.drop(['orig_era', 'orig_raso'], 1, inplace=True)
alldata['source'] = alldata['source'].astype(str)
if 'orig' in alldata.columns:
alldata['orig'] = np.int_(alldata['orig'].values)
alldata['orig'] = alldata['orig'].astype(str)
# every profile / define wrapper function
# pout -> raso p-levels
# pin -> era p-levels
# data -> only era data
mod_interp = lambda x: interp_profile(x, pout=x['p'].values, pcolumn='p', dropna=dropna, min_values=min_count)
journal(funcid+ "Variables: %s" % ",".join(alldata.columns), report, verbose)
journal(funcid + " Interpolating %s (NA: %s, Min: %d)" % (str(alldata.shape), dropna, min_count), report, verbose)
newdata = alldata.groupby(alldata.index).apply(mod_interp)
# drop ... ?
newdata = newdata.reset_index().drop('level_1', axis=1)
# Multi-index to long / sort
newdata = newdata.sort_values(by=['date', 'p']).set_index('date', drop=True)
alldata['source'] = alldata['source'].astype(int)
if 'orig' in alldata.columns:
alldata['orig'] = alldata['orig'].astype(int)
journal(funcid + " finished: %s > %s" % (str(data1.shape), str(newdata.shape)), report, verbose)
return newdata
