% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@MISC{Graf:1053071,
      author       = {Graf, Alexander and Marcon, Lediane and Schmidt, Marius and
                      Kummer, Sirgit and Peichl, Matthias and Larmanou, Eric and
                      Boschetti, Fabio},
      title        = {{H}igh-frequency based {VTT} estimates for {F}ea{V}i{T}a
                      2025},
      publisher    = {Zenodo},
      reportid     = {FZJ-2026-01411},
      year         = {2025},
      abstract     = {Title: High-frequency VTT estimates for FeaViTa 2025  
                      Authors: Graf, Alexander (Project manager), Marcon-Henge,
                      Lediane (Project member), Schmidt, Marius (Data collector),
                      Kummer, Sirgit (Data collector), Peichl, Matthias (Project
                      partner), Larmanou, Eric (Data collector), Boschetti, Fabio
                      (Data collector)   CO2 mixing ratios estimates using the
                      first implementation of the suggested new Virtual Tall Tower
                      (VTT) approach, including diagnostic data (??) for the ITMS
                      Module B2 project FeaViTa. Greenhouse gas measurements, such
                      as CO₂, from existing eddy-covariance (EC) flux stations
                      (typically 2–50 m a.g.l.) can be used to estimate gas
                      concentrations at tall tower (TT) heights (approximately 100
                      m a.g.l. and higher). In this framework, an EC station
                      effectively becomes a virtual tall tower (VTT). This dataset
                      contains the 3rd data deliverable of the ITMS
                      (https://www.itms-germany.de/) project FeaViTa, following up
                      the deliverables: FeaViTa 2024 measurements (DOI
                      10.5281/zenodo.14561379) Classic VTT estimates for FeaViTa
                      2025 (DOI 10.5281/zenodo.16899569) A minor inconsistency
                      towards the latter dataset is that through an update of the
                      external ICOS L2 product for the Svartberget tower, CO2
                      reference data at 35 m and 150 m published with this dataset
                      have slightly changed (for the period of interest for our
                      project by less than 0.03 ppm). We decided to include the
                      newer Svartberget L2 product version in this dataset, since
                      it is more likely to be the long-term future reference for
                      this station. Error statistics of the classic VTT approach
                      changed for less than 0.015 ppm (bias), 0.06 ppm (RMSE) and
                      0.003 (R2) when comparing the new vs. old reference data
                      version. Two EC–TT pairs were used for the VTT
                      calculations. One pair is located in Jülich, Germany (ICOS
                      TT site JUE + ICOS-associated ecosystem EC site DE-RuS; the
                      ICOSclass 1 site RuS is not yet included because of the
                      delay in official final flux data production). The second
                      pair is located in Svartberget, Sweden (ICOS TT site SVB +
                      ICOS ecosystem EC site SE-Svb), where both measurements are
                      taken at the same location. EC (ecosystem) and TT
                      (atmospheric) station data are available for download from
                      the ICOS portal (https://data.icos-cp.eu/portal/) or from
                      our own dataset (https://zenodo.org/records/14561380). VTT
                      calculations were performed using all 2024 data for the
                      Svartberget station, and periods of calibrated measurements
                      in 2024– July 2025 for the Jülich Associate station. The
                      high-frequency approach requires raw (in our case, 20 s-1)
                      fast-response EC station data of CO2 concentration (as dry
                      mixing ratio or together with the thermodynamic state
                      variables needed to convert it), temperature, and optionally
                      vertical wind and humidity. In its most basic form, building
                      on the rationale explored in Graf et al. (2010), the instant
                      minimum fast-response (sonic or corrected) temperature
                      within an averaging interval (e.g. 30 minutes or 1 hour) is
                      used to identify the most likely occurrence of air from the
                      well-mixed part of the (convective) planetary boundary layer
                      (BL) being seen by the EC instruments. In alternative
                      versions, maximum temperature can be used to generate
                      tentative additional estimates for stable (downward sensible
                      heat flux) periods; the minimum or maximum can be replaced
                      by quantiles in an attempt to increase statistical
                      robustness; and quantiles or averages of vertical wind,
                      humidity and CO2 (the latter two depending again on flux
                      direction) can be used to pre-filter for expected BL air.
                      The (instantaneous, average or median) CO2 concentration of
                      these instants is then the VTT (BL concentration) estimate
                      on the calibration scale of the EC gas analyzer. Since
                      typical EC gas analyzers do not have the long-term stability
                      needed to make CO2 concentrations traceable to the standards
                      of atmospheric concentration measurement networks, an
                      additional (typically slow-response and ideally frequently
                      auto-calibrated) CO2 measurement at the EC location is
                      needed. In case of Svartberget, this reference is provided
                      by the atmospheric network measurement at the 35 m EC level
                      and available full-year, in case of Jülich dedicated
                      campaign measurements with an Li810 or Li8100 analyzer were
                      performed (all reference requirements and steps also apply
                      to the classic VTT dataset). Four High-frequency VTT
                      formulations were implemented in this dataset, all after
                      exclusion of outliers and spikes and correction for time
                      lags in the raw data and all on one-hour periods matching
                      the averaging interval of the tall tower data: 1. 
                      “W50T0”: After filtering for vertical wind values lower
                      than or equal to their hourly median (approximately focusing
                      on downdrafts), identifying the minimum (positive covariance
                      of temperature and vertical wind) or maximum (negative)
                      sonic-derived air temperature and applying the difference
                      between this instant’s dry air CO2 mixing ratio and the
                      period average (both from the EC) to the reference CO2
                      measurement at EC level. Results are almost identical to
                      those of focusing on negative (centered) vertical wind
                      speeds; the percentile-/median-based approach is preferred
                      here because the below alternative formulations can be
                      implemented by simply changing parameters (percentile
                      limits) of the same algorithm. 2.  “T50W50”: As above
                      but using all values lower than the median of both
                      temperature and vertical wind speed, and then the median CO2
                      mixing ratio of all remaining records as the hourly VTT
                      estimate. For negative heat flux (e.g. typical nighttime)
                      situations, the upper instead of the lower half of
                      temperatures is used. 3.  “W50T50C50H50”: The same
                      approach is applied to all variables potentially informing
                      about the vertical origin of the air, i.e. also CO2 mixing
                      ratio itself and that of H2O, always taking the lower half
                      of values for an upwards net flux of the respective
                      quantity, and the upper one for a downward flux. 4.
                      “W75C75H75T5”: Similar to the above, but the criterion
                      is relaxed $(75\%)$ for all other variables and restricted
                      $(5\%$ of the data) for temperature. The percentiles are
                      always computed from the whole original dataset if possible,
                      but if after filtering for one quantity the desired
                      percentile of the next one is not contained in the remaining
                      data anymore, it is newly computed from them. The filtering
                      is performed starting with the most relaxed toward the most
                      restricted criterion, and if percentiles are identical for
                      several quantities, their order in the short code of the
                      formulation is the filtering order.   CSV files: o  
                      $VTThifreq_Juelich_Associate.csv$ and o  
                      $VTThifreq_Svartberget.csv:$ Hourly time series of CO2
                      averages (EC, calibration reference and TT), diagnostic data
                      and VTT estimates o   $variable_names.csv:$ List of
                      variable names, units, descriptions, and data sources used
                      in the files.   Python codes Below is a short description
                      of the provided Python codes. Scripts should be run in
                      numerical order. More detailed explanations are included as
                      comments within the code. o  
                      $FeaViTa_1a_ReadHighResData.py:$ Read in EC raw data from
                      their original format (half-hourly files in ICOS standard
                      format for class-1 sites, daily custom format files for
                      Jülich associate sites) o  
                      $FeaViTa_2_ConvertHighResData.py:$ Quality filter, determine
                      and shift time lag, Convert units, compute dependent
                      variables and diagnostic files to store the number of
                      outliers, crosscorrelations and shifts, while still
                      remaining in the 20 s-1 domain. o  
                      $FeaViTa_3b_VTTHhiFreqPercentileVersion.py:$ The actual VTT
                      estimation described above, turning 20 s-1 data into hourly
                      timeseries o   $FeaViTa_4_compareVTTtoTT.py:$ Calculates
                      the offset between EC and calibration measurements, applies
                      the offset to the calculated VTT to obtain final CO₂
                      concentrations at tall tower height, and computes comparison
                      statistics. Outputs a CSV file containing the VTT-corrected
                      CO₂ time series, and optionally, another CSV file with
                      main statistics (bias, root mean square error, and Pearson
                      R²) comparing estimated and measured CO₂ concentrations
                      can be exported. This script was already part of the classic
                      VTT data upload but is here stored in an updated version.
                      Also needed but not included, since identical to the classic
                      VTT upload version, are the scripts
                      $FeaViTa_1b3c_ReadTallTowerData.py,$
                      $FeaViTa_1d3_ReadCalAnalyzer.py,$ and
                      $FeaViTa_0a_maxcorrshift.$   Additional files o  
                      $Svartberget_EC_HourlyHiFreqFiles_qc_2004a.zip,$ o  
                      $Svartberget_EC_HourlyHiFreqFiles_qc_2004b.zip$ and o  
                      $Juelich_AssociateEC_HourlyHiFreqFiles_qc.zip:$ 20 s-1 raw
                      data after quality filtering and lag correction, one
                      (in-zip) csv file per station and hour, named by the center
                      of the hour in UTC   References: Graf, A. et al., 2010.
                      Boundedness of Turbulent Temperature Probability
                      Distributions, and their Relation to the Vertical Profile in
                      the Convective Boundary Layer. Bound.-Layer Meteor., 134(3):
                      459-486. https://doi.org/10.1007/s10546-009-9444-9},
      cin          = {IBG-3},
      cid          = {I:(DE-Juel1)IBG-3-20101118},
      pnm          = {2173 - Agro-biogeosystems: controls, feedbacks and impact
                      (POF4-217)},
      pid          = {G:(DE-HGF)POF4-2173},
      typ          = {PUB:(DE-HGF)32},
      doi          = {10.5281/ZENODO.17209765},
      url          = {https://juser.fz-juelich.de/record/1053071},
}