def l5qc(main_gui, cf, ds4): ds5 = pfp_io.copy_datastructure(cf, ds4) # ds5 will be empty (logical false) if an error occurs in copy_datastructure # return from this routine if this is the case if not ds5: return ds5 # set some attributes for this level pfp_utils.UpdateGlobalAttributes(cf, ds5, "L5") # parse the control file for information on how the user wants to do the gap filling l5_info = pfp_gf.ParseL5ControlFile(cf, ds5) if ds5.returncodes["value"] != 0: return ds5 # check to see if we have any imports pfp_gf.ImportSeries(cf, ds5) # re-apply the quality control checks (range, diurnal and rules) pfp_ck.do_qcchecks(cf, ds5) pfp_gf.CheckL5Drivers(ds5, l5_info) if ds5.returncodes["value"] != 0: return ds5 # now do the flux gap filling methods # *** start of the section that does the gap filling of the fluxes *** pfp_gf.CheckGapLengths(cf, ds5, l5_info) if ds5.returncodes["value"] != 0: return ds5 # apply the turbulence filter (if requested) pfp_ck.ApplyTurbulenceFilter(cf, ds5, l5_info) # fill short gaps using interpolation pfp_gf.GapFillUsingInterpolation(cf, ds5) # gap fill using marginal distribution sampling if "GapFillUsingMDS" in l5_info: pfp_gfMDS.GapFillUsingMDS(ds5, l5_info, "GapFillUsingMDS") # do the gap filling using SOLO if "GapFillUsingSOLO" in l5_info: pfp_gfSOLO.GapFillUsingSOLO(main_gui, ds5, l5_info, "GapFillUsingSOLO") if ds5.returncodes["value"] != 0: return ds5 # fill long gaps using SOLO if "GapFillLongSOLO" in l5_info: pfp_gfSOLO.GapFillUsingSOLO(main_gui, ds5, l5_info, "GapFillLongSOLO") if ds5.returncodes["value"] != 0: return ds5 # merge the gap filled drivers into a single series pfp_ts.MergeSeriesUsingDict(ds5, l5_info, merge_order="standard") # check that all targets were gap filled pfp_gf.CheckL5Targets(ds5, l5_info) if ds5.returncodes["value"] != 0: return ds5 # calculate Monin-Obukhov length pfp_ts.CalculateMoninObukhovLength(ds5) # write the percentage of good data as a variable attribute pfp_utils.get_coverage_individual(ds5) # write the percentage of good data for groups pfp_utils.get_coverage_groups(ds5) # remove intermediate series from the data structure pfp_ts.RemoveIntermediateSeries(ds5, l5_info) return ds5
def l4qc(main_gui, cf, ds3): ds4 = pfp_io.copy_datastructure(cf, ds3) # ds4 will be empty (logical false) if an error occurs in copy_datastructure # return from this routine if this is the case if not ds4: return ds4 # set some attributes for this level pfp_utils.UpdateGlobalAttributes(cf, ds4, "L4") # check to see if we have any imports pfp_gf.ImportSeries(cf, ds4) # re-apply the quality control checks (range, diurnal and rules) pfp_ck.do_qcchecks(cf, ds4) # now do the meteorological driver gap filling # parse the control file for information on how the user wants to do the gap filling l4_info = pfp_gf.ParseL4ControlFile(cf, ds4) if ds4.returncodes["value"] != 0: return ds4 # *** start of the section that does the gap filling of the drivers *** # read the alternate data files ds_alt = pfp_gf.ReadAlternateFiles(ds4, l4_info) # fill short gaps using interpolation pfp_gf.GapFillUsingInterpolation(cf, ds4) # gap fill using climatology if "GapFillFromClimatology" in l4_info: pfp_gf.GapFillFromClimatology(ds4, l4_info, "GapFillFromClimatology") # do the gap filling using the ACCESS output if "GapFillFromAlternate" in l4_info: pfp_gfALT.GapFillFromAlternate(main_gui, ds4, ds_alt, l4_info, "GapFillFromAlternate") if ds4.returncodes["value"] != 0: return ds4 # merge the first group of gap filled drivers into a single series pfp_ts.MergeSeriesUsingDict(ds4, l4_info, merge_order="prerequisite") # re-calculate the ground heat flux but only if requested in control file opt = pfp_utils.get_keyvaluefromcf(cf,["Options"], "CorrectFgForStorage", default="No", mode="quiet") if opt.lower() != "no": pfp_ts.CorrectFgForStorage(cf, ds4, Fg_out='Fg', Fg_in='Fg_Av', Ts_in='Ts', Sws_in='Sws') # re-calculate the net radiation pfp_ts.CalculateNetRadiation(cf, ds4, Fn_out='Fn', Fsd_in='Fsd', Fsu_in='Fsu', Fld_in='Fld', Flu_in='Flu') # re-calculate the available energy pfp_ts.CalculateAvailableEnergy(ds4, Fa_out='Fa', Fn_in='Fn', Fg_in='Fg') # merge the second group of gap filled drivers into a single series pfp_ts.MergeSeriesUsingDict(ds4, l4_info, merge_order="standard") # re-calculate the water vapour concentrations pfp_ts.CalculateHumiditiesAfterGapFill(ds4, l4_info) # re-calculate the meteorological variables pfp_ts.CalculateMeteorologicalVariables(ds4, l4_info) # check for any missing data pfp_utils.get_missingingapfilledseries(ds4, l4_info) # write the percentage of good data as a variable attribute pfp_utils.get_coverage_individual(ds4) # write the percentage of good data for groups pfp_utils.get_coverage_groups(ds4) # remove intermediate series from the data structure pfp_ts.RemoveIntermediateSeries(ds4, l4_info) return ds4
def l6qc(main_gui, cf, ds5): ds6 = pfp_io.copy_datastructure(cf, ds5) # ds6 will be empty (logical false) if an error occurs in copy_datastructure # return from this routine if this is the case if not ds6: return ds6 # set some attributes for this level pfp_utils.UpdateGlobalAttributes(cf, ds6, "L6") # parse the control file l6_info = pfp_rp.ParseL6ControlFile(cf, ds6) # check to see if we have any imports pfp_gf.ImportSeries(cf, ds6) # check units of Fc Fc_list = [label for label in ds6.series.keys() if label[0:2] == "Fc"] pfp_utils.CheckUnits(ds6, Fc_list, "umol/m2/s", convert_units=True) # get ER from the observed Fc pfp_rp.GetERFromFc(cf, ds6) # return code will be non-zero if turbulance filter not applied to CO2 flux if ds6.returncodes["value"] != 0: return ds6 # estimate ER using SOLO if "ERUsingSOLO" in l6_info: pfp_rp.ERUsingSOLO(main_gui, ds6, l6_info, "ERUsingSOLO") if ds6.returncodes["value"] != 0: return ds6 # estimate ER using FFNET #pfp_rp.ERUsingFFNET(cf, ds6, l6_info) # estimate ER using Lloyd-Taylor pfp_rp.ERUsingLloydTaylor(cf, ds6, l6_info) # estimate ER using Lasslop et al pfp_rp.ERUsingLasslop(ds6, l6_info) # merge the estimates of ER with the observations pfp_ts.MergeSeriesUsingDict(ds6, l6_info, merge_order="standard") # calculate NEE from Fc and ER pfp_rp.CalculateNEE(cf, ds6, l6_info) # calculate NEP from NEE pfp_rp.CalculateNEP(cf, ds6) # calculate ET from Fe pfp_rp.CalculateET(ds6) # partition NEE into GPP and ER pfp_rp.PartitionNEE(ds6, l6_info) # write the percentage of good data as a variable attribute pfp_utils.get_coverage_individual(ds6) # write the percentage of good data for groups pfp_utils.get_coverage_groups(ds6) # remove intermediate series from the data structure pfp_ts.RemoveIntermediateSeries(ds6, l6_info) # do the L6 summary pfp_rp.L6_summary(cf, ds6) return ds6
def l5qc(cf, ds4): ds5 = pfp_io.copy_datastructure(cf, ds4) # ds4 will be empty (logical false) if an error occurs in copy_datastructure # return from this routine if this is the case if not ds5: return ds5 # set some attributes for this level pfp_utils.UpdateGlobalAttributes(cf, ds5, "L5") ds5.cf = cf # create a dictionary to hold the gap filling data ds_alt = {} # check to see if we have any imports pfp_gf.ImportSeries(cf, ds5) # re-apply the quality control checks (range, diurnal and rules) pfp_ck.do_qcchecks(cf, ds5) # now do the flux gap filling methods label_list = pfp_utils.get_label_list_from_cf(cf) for label in label_list: # parse the control file for information on how the user wants to do the gap filling pfp_gf.GapFillParseControlFile(cf, ds5, label, ds_alt) # *** start of the section that does the gap filling of the fluxes *** # apply the turbulence filter (if requested) pfp_ck.ApplyTurbulenceFilter(cf, ds5) # fill short gaps using interpolation pfp_gf.GapFillUsingInterpolation(cf, ds5) # do the gap filling using SOLO pfp_gfSOLO.GapFillUsingSOLO(cf, ds4, ds5) if ds5.returncodes["solo"] == "quit": return ds5 # gap fill using marginal distribution sampling pfp_gfMDS.GapFillFluxUsingMDS(cf, ds5) # gap fill using climatology pfp_gf.GapFillFromClimatology(ds5) # merge the gap filled drivers into a single series pfp_ts.MergeSeriesUsingDict(ds5, merge_order="standard") # calculate Monin-Obukhov length pfp_ts.CalculateMoninObukhovLength(ds5) # write the percentage of good data as a variable attribute pfp_utils.get_coverage_individual(ds5) # write the percentage of good data for groups pfp_utils.get_coverage_groups(ds5) return ds5
def l4qc(cf, ds3): # !!! code here to use existing L4 file # logic # if the L4 doesn't exist # - create ds4 by using copy.deepcopy(ds3) # if the L4 does exist and the "UseExistingL4File" option is False # - create ds4 by using copy.deepcopy(ds3) # if the L4 does exist and the "UseExistingL4File" option is True # - read the contents of the L4 netCDF file # - check the start and end dates of the L3 and L4 data # - if these are the same then tell the user there is nothing to do # - copy the L3 data to the L4 data structure # - replace the L3 data with the L4 data #ds4 = copy.deepcopy(ds3) ds4 = pfp_io.copy_datastructure(cf, ds3) # ds4 will be empty (logical false) if an error occurs in copy_datastructure # return from this routine if this is the case if not ds4: return ds4 # set some attributes for this level pfp_utils.UpdateGlobalAttributes(cf, ds4, "L4") ds4.cf = cf ## calculate the available energy #if "Fa" not in ds4.series.keys(): #pfp_ts.CalculateAvailableEnergy(ds4,Fa_out='Fa',Fn_in='Fn',Fg_in='Fg') # create a dictionary to hold the gap filling data ds_alt = {} # check to see if we have any imports pfp_gf.ImportSeries(cf, ds4) # re-apply the quality control checks (range, diurnal and rules) pfp_ck.do_qcchecks(cf, ds4) # now do the meteorological driver gap filling for ThisOne in cf["Drivers"].keys(): if ThisOne not in ds4.series.keys(): logger.warning("Series " + ThisOne + " not in data structure") continue # parse the control file for information on how the user wants to do the gap filling pfp_gf.GapFillParseControlFile(cf, ds4, ThisOne, ds_alt) # *** start of the section that does the gap filling of the drivers *** # fill short gaps using interpolation pfp_gf.GapFillUsingInterpolation(cf, ds4) # gap fill using climatology pfp_gf.GapFillFromClimatology(ds4) # do the gap filling using the ACCESS output pfp_gfALT.GapFillFromAlternate(cf, ds4, ds_alt) if ds4.returncodes["alternate"] == "quit": return ds4 # gap fill using SOLO pfp_gfSOLO.GapFillUsingSOLO(cf, ds3, ds4) if ds4.returncodes["solo"] == "quit": return ds4 # merge the first group of gap filled drivers into a single series pfp_ts.MergeSeriesUsingDict(ds4, merge_order="prerequisite") # re-calculate the ground heat flux but only if requested in control file opt = pfp_utils.get_keyvaluefromcf(cf, ["Options"], "CorrectFgForStorage", default="No", mode="quiet") if opt.lower() != "no": pfp_ts.CorrectFgForStorage(cf, ds4, Fg_out='Fg', Fg_in='Fg_Av', Ts_in='Ts', Sws_in='Sws') # re-calculate the net radiation pfp_ts.CalculateNetRadiation(cf, ds4, Fn_out='Fn', Fsd_in='Fsd', Fsu_in='Fsu', Fld_in='Fld', Flu_in='Flu') # re-calculate the available energy pfp_ts.CalculateAvailableEnergy(ds4, Fa_out='Fa', Fn_in='Fn', Fg_in='Fg') # merge the second group of gap filled drivers into a single series pfp_ts.MergeSeriesUsingDict(ds4, merge_order="standard") # re-calculate the water vapour concentrations pfp_ts.CalculateHumiditiesAfterGapFill(ds4) # re-calculate the meteorological variables pfp_ts.CalculateMeteorologicalVariables(ds4) # the Tumba rhumba pfp_ts.CalculateComponentsFromWsWd(ds4) # check for any missing data pfp_utils.get_missingingapfilledseries(ds4) # write the percentage of good data as a variable attribute pfp_utils.get_coverage_individual(ds4) # write the percentage of good data for groups pfp_utils.get_coverage_groups(ds4) return ds4
def l3qc(cf, ds2): """ """ # make a copy of the L2 data ds3 = copy.deepcopy(ds2) # set some attributes for this level pfp_utils.UpdateGlobalAttributes(cf, ds3, "L3") # put the control file name into the global attributes ds3.globalattributes['controlfile_name'] = cf['controlfile_name'] # check to see if we have any imports pfp_gf.ImportSeries(cf, ds3) # apply linear corrections to the data pfp_ck.do_linear(cf, ds3) # ************************ # *** Merge humidities *** # ************************ # merge whatever humidities are available pfp_ts.MergeHumidities(cf, ds3, convert_units=True) # ************************** # *** Merge temperatures *** # ************************** # get the air temperature from the CSAT virtual temperature pfp_ts.TaFromTv(cf, ds3) # merge the HMP and corrected CSAT data pfp_ts.MergeSeries(cf, ds3, "Ta", convert_units=True) pfp_utils.CheckUnits(ds3, "Ta", "C", convert_units=True) # *************************** # *** Calcuate humidities *** # *************************** # calculate humidities (absolute, specific and relative) from whatever is available pfp_ts.CalculateHumidities(ds3) # ******************************** # *** Merge CO2 concentrations *** # ******************************** # merge the 7500 CO2 concentration # PRI 09/08/2017 possibly the ugliest thing I have done yet # This needs to be abstracted to a general alias checking routine at the # start of the L3 processing so that possible aliases are mapped to a single # set of variable names. if "CO2" in cf["Variables"]: CO2 = "CO2" elif "Cc" in cf["Variables"]: CO2 = "Cc" else: msg = "Label for CO2 ('CO2','Cc') not found in control file" logger.error(msg) return pfp_ts.MergeSeries(cf, ds3, CO2, convert_units=True) # ****************************************** # *** Calculate meteorological variables *** # ****************************************** # Update meteorological variables pfp_ts.CalculateMeteorologicalVariables(ds3) # ************************************************* # *** Calculate fluxes from covariances section *** # ************************************************* # check to see if the user wants to use the fluxes in the L2 file if not pfp_utils.cfoptionskeylogical(cf, Key="UseL2Fluxes", default=False): # check the covariance units and change if necessary pfp_ts.CheckCovarianceUnits(ds3) # do the 2D coordinate rotation pfp_ts.CoordRotation2D(cf, ds3) # do the Massman frequency attenuation correction pfp_ts.MassmanStandard(cf, ds3) # calculate the fluxes pfp_ts.CalculateFluxes(cf, ds3) # approximate wT from virtual wT using wA (ref: Campbell OPECSystem manual) pfp_ts.FhvtoFh(cf, ds3) # correct the H2O & CO2 flux due to effects of flux on density measurements pfp_ts.Fe_WPL(cf, ds3) pfp_ts.Fc_WPL(cf, ds3) # ************************************** # *** Calculate Monin-Obukhov length *** # ************************************** pfp_ts.CalculateMoninObukhovLength(ds3) # ************************** # *** CO2 and Fc section *** # ************************** # convert CO2 units if required pfp_utils.ConvertCO2Units(cf, ds3, CO2=CO2) # calculate Fc storage term - single height only at present pfp_ts.CalculateFcStorageSinglePoint(cf, ds3, Fc_out='Fc_single', CO2_in=CO2) # convert Fc and Fc_storage units if required pfp_utils.ConvertFcUnits(cf, ds3) # merge Fc and Fc_storage series if required merge_list = [ label for label in cf["Variables"].keys() if label[0:2] == "Fc" and "MergeSeries" in cf["Variables"][label].keys() ] for label in merge_list: pfp_ts.MergeSeries(cf, ds3, label, save_originals=True) # correct Fc for storage term - only recommended if storage calculated from profile available pfp_ts.CorrectFcForStorage(cf, ds3) # ************************* # *** Radiation section *** # ************************* # merge the incoming shortwave radiation pfp_ts.MergeSeries(cf, ds3, 'Fsd') # calculate the net radiation from the Kipp and Zonen CNR1 pfp_ts.CalculateNetRadiation(cf, ds3, Fn_out='Fn_KZ', Fsd_in='Fsd', Fsu_in='Fsu', Fld_in='Fld', Flu_in='Flu') pfp_ts.MergeSeries(cf, ds3, 'Fn') # **************************************** # *** Wind speed and direction section *** # **************************************** # combine wind speed from the Wind Sentry and the SONIC pfp_ts.MergeSeries(cf, ds3, 'Ws') # combine wind direction from the Wind Sentry and the SONIC pfp_ts.MergeSeries(cf, ds3, 'Wd') # ******************** # *** Soil section *** # ******************** # correct soil heat flux for storage # ... either average the raw ground heat flux, soil temperature and moisture # and then do the correction (OzFlux "standard") pfp_ts.AverageSeriesByElements(cf, ds3, 'Ts') pfp_ts.AverageSeriesByElements(cf, ds3, 'Sws') if pfp_utils.cfoptionskeylogical(cf, Key='CorrectIndividualFg'): # ... or correct the individual ground heat flux measurements (James' method) pfp_ts.CorrectIndividualFgForStorage(cf, ds3) pfp_ts.AverageSeriesByElements(cf, ds3, 'Fg') else: pfp_ts.AverageSeriesByElements(cf, ds3, 'Fg') pfp_ts.CorrectFgForStorage(cf, ds3, Fg_out='Fg', Fg_in='Fg', Ts_in='Ts', Sws_in='Sws') # calculate the available energy pfp_ts.CalculateAvailableEnergy(ds3, Fa_out='Fa', Fn_in='Fn', Fg_in='Fg') # create new series using MergeSeries or AverageSeries pfp_ck.CreateNewSeries(cf, ds3) # re-apply the quality control checks (range, diurnal and rules) pfp_ck.do_qcchecks(cf, ds3) # coordinate gaps in the three main fluxes pfp_ck.CoordinateFluxGaps(cf, ds3) # coordinate gaps in Ah_7500_Av with Fc pfp_ck.CoordinateAh7500AndFcGaps(cf, ds3) # check missing data and QC flags are consistent pfp_utils.CheckQCFlags(ds3) # get the statistics for the QC flags and write these to an Excel spreadsheet pfp_io.get_seriesstats(cf, ds3) # write the percentage of good data as a variable attribute pfp_utils.get_coverage_individual(ds3) # write the percentage of good data for groups pfp_utils.get_coverage_groups(ds3) return ds3
def l3qc(cf, ds2): """ """ # make a copy of the L2 data ds3 = copy.deepcopy(ds2) # set some attributes for this level pfp_utils.UpdateGlobalAttributes(cf, ds3, "L3") # check to see if we have any imports pfp_gf.ImportSeries(cf,ds3) # apply linear corrections to the data pfp_ck.do_linear(cf,ds3) # parse the control file for information on how the user wants to do the gap filling l3_info = pfp_compliance.ParseL3ControlFile(cf, ds3) if l3_info["status"]["value"] != 0: logger.error(l3_info["status"]["message"]) return ds3 # ************************ # *** Merge humidities *** # ************************ # merge whatever humidities are available pfp_ts.MergeHumidities(cf, ds3, convert_units=True) # ************************** # *** Merge temperatures *** # ************************** # get the air temperature from the CSAT virtual temperature pfp_ts.TaFromTv(cf, ds3) # merge the HMP and corrected CSAT data pfp_ts.CombineSeries(cf, ds3, "Ta", convert_units=True) pfp_utils.CheckUnits(ds3, "Ta", "degC", convert_units=True) # *************************** # *** Calcuate humidities *** # *************************** # calculate humidities (absolute, specific and relative) from whatever is available pfp_ts.CalculateHumidities(ds3) # ******************************** # *** Merge CO2 concentrations *** # ******************************** # merge the CO2 concentration pfp_ts.CombineSeries(cf, ds3, l3_info["CO2"]["label"], convert_units=True) # ****************************************** # *** Calculate meteorological variables *** # ****************************************** # Update meteorological variables pfp_ts.CalculateMeteorologicalVariables(ds3, l3_info) # ************************************************* # *** Calculate fluxes from covariances section *** # ************************************************* # check to see if the user wants to use the fluxes in the L2 file if not pfp_utils.get_optionskeyaslogical(cf, "UseL2Fluxes", default=False): # check the covariance units and change if necessary pfp_ts.CheckCovarianceUnits(ds3) # do the 2D coordinate rotation pfp_ts.CoordRotation2D(cf, ds3) # do the Massman frequency attenuation correction pfp_ts.MassmanStandard(cf, ds3) # calculate the fluxes pfp_ts.CalculateFluxes(cf, ds3) # approximate wT from virtual wT using wA (ref: Campbell OPECSystem manual) pfp_ts.FhvtoFh(cf, ds3) # correct the H2O & CO2 flux due to effects of flux on density measurements if pfp_ts.Fe_WPL(cf, ds3): return ds3 if pfp_ts.Fco2_WPL(cf, ds3): return ds3 # ************************** # *** CO2 and Fc section *** # ************************** # convert CO2 units if required pfp_utils.ConvertCO2Units(cf, ds3) # calculate Fco2 storage term - single height only at present pfp_ts.CalculateFco2StorageSinglePoint(cf, ds3, l3_info["CO2"]["label"]) # convert Fco2 units if required pfp_utils.ConvertFco2Units(cf, ds3) # merge Fco2 and Fco2_storage series if required pfp_ts.CombineSeries(cf, ds3, l3_info["Fco2"]["combine_list"], save_originals=True) # correct Fco2 for storage term - only recommended if storage calculated from profile available pfp_ts.CorrectFco2ForStorage(cf, ds3) # ************************* # *** Radiation section *** # ************************* # merge the incoming shortwave radiation pfp_ts.CombineSeries(cf, ds3, "Fsd") # calculate the net radiation from the Kipp and Zonen CNR1 pfp_ts.CalculateNetRadiation(cf, ds3) pfp_ts.CombineSeries(cf, ds3, "Fn") # **************************************** # *** Wind speed and direction section *** # **************************************** # combine wind speed from the Wind Sentry and the SONIC pfp_ts.CombineSeries(cf,ds3, "Ws") # combine wind direction from the Wind Sentry and the SONIC pfp_ts.CombineSeries(cf,ds3, "Wd") # ******************** # *** Soil section *** # ******************** # correct soil heat flux for storage # ... either average the raw ground heat flux, soil temperature and moisture # and then do the correction (OzFlux "standard") pfp_ts.CombineSeries(cf, ds3, "Ts") pfp_ts.CombineSeries(cf, ds3, "Sws") if pfp_utils.get_optionskeyaslogical(cf, "CorrectIndividualFg"): # ... or correct the individual ground heat flux measurements (James' method) pfp_ts.CorrectIndividualFgForStorage(cf, ds3) pfp_ts.CombineSeries(cf, ds3, "Fg") else: pfp_ts.CombineSeries(cf, ds3, "Fg") pfp_ts.CorrectFgForStorage(cf, ds3) # calculate the available energy pfp_ts.CalculateAvailableEnergy(ds3) # create new series using MergeSeries or AverageSeries pfp_ck.CreateNewSeries(cf, ds3) # Calculate Monin-Obukhov length pfp_ts.CalculateMoninObukhovLength(ds3) # re-apply the quality control checks (range, diurnal and rules) pfp_ck.do_qcchecks(cf, ds3) # check missing data and QC flags are consistent pfp_utils.CheckQCFlags(ds3) # get the statistics for the QC flags and write these to an Excel spreadsheet pfp_io.get_seriesstats(cf, ds3) # write the percentage of good data as a variable attribute pfp_utils.get_coverage_individual(ds3) # write the percentage of good data for groups pfp_utils.get_coverage_groups(ds3) # remove intermediate series from the data structure pfp_ts.RemoveIntermediateSeries(ds3, l3_info) return ds3