Data Information

General Notes

  • Please contact Dr. Syndonia Bret-Harte or Dr. Gaius Shaver if you need guidance regarding appropriate use and citation.
     
  • Data are in comma-delimited format, please contact us if you need another format.  Metadata are web-viewable to quickly vizualize the data file contents.
     
  • Current-year datasets will be updated on a frequent basis (semi-monthly).
     
  • Each current-year file has a date stamp in the name or header reflecting when it was last updated.  Please use this date to determine if you have the most current data set.
     
  • All infomation about sensor types, heights and other details can be found in the 'Sensor Details' links below.
     

About Data Processing

  • Missing and flagged data are replaced with '-9999' for current year and  'NaN' for older data sets
  • Time is expessed as local time, AKST (UTC-9hrs) for Imnavait and MAGST (UTC+12hrs) for Pleistocene Park
  • Gas and energy fluxes are calculated from the high frequency time series using EdiRe or EddyPro software with these basic corrections:  despiking, coordinate rotation, spectral correction, the 'WPL' correction, and the 'Burba' correction (where appropriate, to account for instrument heating effects).
  • Some data have been filtered (_f) and gap-filled (_gf).
  • 2014 data and onward will present gas fluxes in umol sec-1 m-2 (older data appear in mmol sec-1 m-2)

     

  • Data prior to 2014 were processed and gapfilled using different tools than data starting in 2014.  The two approaches are described below.

 

Data from 2014 to current:

  • Fluxes were computed from the high frequency time series using Licor EddyPro software, filtered using a custom script package, then gapfilled using the REddyProc R-package
  • In EddyPro, the advanced processing option was used, which included extended statistical tests, custom QA flagging and use of biomet data imported from the datalogger.  The standard EddyPro choices for stationarity flagging and footprint modeling were used.  When appropriate, the so-called Burba correction was calculated using a Matlab tool provided by Licor. 
  • Post-processing, the flux and met. data were then filtered using custom Matlab scripts to provide basic QA/QC, tower rejection angle, u* and signal strength thresholds and other filters such as seasonal limits on NEE. 
  • In ReddyProc, gapfilling was performed using the NEE partitioning option.  The partitioned ER and GPP were then filtered to reflect site seasonality (e.g. GPP outside of snow-free period forced to zero). 
  • Please contact the PI's or website admin for further details on data processing.

 

Data PRIOR to 2014:

Fluxes were computed from the high frequency time series using EdiRe software, then filtered and gapfilled using a custom script package.

In EdiRe, the following basic corrections were used:

  • The 'WPL' correction
  • A coordinate rotation
  • A spectral correction
  • The 'Burba' correction (where appropriate to account for instrument heating effect)
*Correct instrument lag adjustment occurs during datalogging
 

The following QA/QC variables are applied to the flux data:

  • Stationarity tests
  • Footprint analysis
  • Gas analyzer diagnostics are used as a QA/QC variable for both flux and radiation data
  • Rejection angles of 10° at Imnavait and 45° at Pleistocene Park are used when EC instruments were downwind of tower to remove flow distortions 
 
The flux and meteorological data is further post-processed to accomplish the following:
  • Data are within engineering specifications of each instrument. 
  • Removal of impossible measurements (e.g. negative precipitation)
  • Standardize previously flagged data including '-9999'  to 'NaN'
  • Removal of outliers via a three-standard deviation filter
  • Removal of unrealistic changes in a time series with a step change filter
  • A similarity filter to remove errors from instruments that generate a string of identical values when not working

  Data are gap-filled with the following methods: 

  • P-th order autoregressive model.  Two values for each missing element are predicted with both a forward-and backward-looking algorythm.  The mean of these two values is used to fill the gap.  Currently the model is applied 168 elements in both directions into missing data.  This model can produce impossible values, these are filtered out.
     
  • MDV method.   Missing values from a time series are estimated using binned half-hourly averages of filtered values from the following or the previous seven days.