Search
CH-AWS FP2022 (2015-2021)
PI dataset of CH-AWS fluxes 2015-2021.
STATUS: Released.
Contents
Flux Products
For an explanation of variables in output files, please see Variable Abbreviations and ReddyProc Data Output Format
FP2022.1
- Date: 8 Feb 2022
- Identifier:
ID20220202233037
- File:
CH-AWS_FP2022.1_2015-2021_ID20220202233037_30MIN.csv
- Description: Initial release using a variable USTAR threshold for all years, MDS gap-filling and partitioning done in ReddyProc
- Included variables: NEE, GPP, Reco, TA, SW_IN, VPD and auxiliary output
- Recommended flux variables:
- gap-filled: NEE_U50_f, GPP_DT_U50, Reco_DT_U50, LE_f, ET_f
- It is recommended to use GPP and Reco from the daytime partitioning method (_DT), because daytime fluxes are generally more reliable across years. However, partitioned fluxes from the nighttime method are also included in this dataset.
- measured only (highest quality): NEE_CUT_QCF0, LE_QC0
- Release candidates:
- None
Plots
Description
- Ecosystem fluxes over 7 years from two different IRGAs and sonic anemometers
- This is the first version of the CH-AWS PI dataset.
Key Stats
- Best estimate cumulative carbon uptake (2015-2021): 1.8 kg C m-2 (
NEE_U50_f
), avg. 254 g C m-2 yr-1 - Available directly measured, highest-quality fluxes:
- directly measured fluxes with quality flag
QCF
= 0 (no gap-filling) NEE_U50_QCF0
: 37680 half-hours (30.7% of potential values) [1], [2], [3]LE_QCF0
: 32938 half-hours (26.8% of potential values) [1], [3]
- directly measured fluxes with quality flag
- Data basis for budget calculations:
NEE_U50_f
: 46.5% measured / 53.5% gap-filled [1], [2], [4]LE_f
andET_f
: 45.7% measured / 54.3% gap-filled [1], [4]
[1] … after all quality checks, including outlier removal
[2] … after USTAR
threshold application
[3] … QCF
= 0 (daytime and nighttime)
[4] … measured data include quality flag QCF
= 0 and QCF
= 1 (OK quality) for daytime data, and QCF
= 0 for nighttime data
Dataset Production
- For info about the flux processing chain and flux levels see: Flux Processing Chain
- Google Sheet that was used to document processing progress: CH-AWS / FP2022 (2015-2021)
Software
- bico v1.2.1 for the conversion of binary raw data files to ASCII
- fluxrun v1.0.3 for the flux calculation using EddyPro v7.0.8
- diive v0.21.0 (legacy version) for file merging, quality control, storage correction, outlier removal
- ReddyProc v1.2.2 for USTAR threshold detection, MDS gap-filliing and partitioning, in R Studio v1.3.959
Processing Steps
For an overview of the general processing chain see here: Flux Processing Chain
Steps to create a flux dataset comprising CO2 flux (NEE), LE and H. Note that H2O flux and ET are not prepared, only LE is used from the water fluxes. ET is later calculated (converted from LE) in ReddyProc.
Level-1 / Flux calculations 2015-2021
-
- for IRGA75 (2015) and IRGA72 (2015-2021) fluxes
- Level-1 / Merging
- Separate for each IRGA
- Fluxes for each IRGA were merged, one separate file for each IRGA, yielding two datasets: IRGA75 and IRGA72
- Level-1.1 / Self-heating correction for IRGA75 not applied
- Was not done because a winter comparison of parallel measurements IRGA75 / IRGA72 showed only small differences.
- However, a longer time period of parallel measurements would be necessary for final conclusions.
- Level-2 / Quality Flag Expansion
- Separate for each IRGA
- Quality flags were expanded for CO2 flux, LE and H
- No flux outliers were flagged in this step, no absolute flux limits applied. This is done later after the storage correction.
- Step 1:
- Expanded quality flags (
QCF
) were created for each IRGA, based on the merged dataset of the respective IRGA. QCF for fluxes: CO2, LE and H. For example, the quality flags for IRGA72 were created for all years from the merged IRGA72 dataset. - Used quality flags: SSITC, spectral correction factor, spikes (raw data), drop-out (raw data), absolute limits (raw data), AGC (>90% is bad, only for IRGA75 CO2 and LE), SIGNAL_STRENGTH (<60% is bad, only for IRGA72 CO2 and LE).
QCF
: 0=best quality data, 1=OK quality for long-term budgets, 2=bad data, do not use
- Expanded quality flags (
- Step 2:
- Flux values where
QCF = 0
orQCF = 1
were kept in the dataset. This removes all flux values whereQCF = 2
or where QCF is missing. - The same was done with the
QCF
flags for CO2 flux, LE and H. Flags whereQCF = 0
orQCF = 1
were kept in the dataset, making sure that flag values were available for time periods when the corresponding half-hourly flux was also available. This removes all flag values whereQCF = 2
or where QCF was missing. - These steps make sure that the time series for each flux and for each flag comprises values of good and OK values only and other values are NaN.
- The variable name suffix
_QC01
marks that only fluxes of quality 0 and 1 are available in the respective data columns.
- Flux values where
- Level-3.1 / Storage Correction
- Separate for each IRGA
- The storage term was added to CO2 flux, LE and H
- Level-3.1 / Merging of IRGA datasets
- Combine the IRGA datasets (IRGA75 and IRGA72) into one dataset
- Data from the IRGA75 dataset have the suffix
_IRGA75
, and data from the IRGA72 have the suffix_IRGA72
. - Fluxes (NEE, LE, H) from the 2 IRGAs were then collected in one single column, marked by the suffix
_IRGA7572
. - For the merging, the IRGA72 fluxes were used as the starting dataset. Then, the IRGA75 fluxes were added, filling data gaps in the time series with IRGA75 values (only the case in 2015).
- This was done not only for the fluxes, but also for the
QCF
andUSTAR
data. This way all data points should have their respective correct quality flag andUSTAR
value, even though there is this mixing of 2 sonics and 2 IRGAs with sometimes differentQCF
criteria. - Add additional meteo data:
- SW_IN, VPD, TA and RH. Necessary to divide the data into daytime and nighttime data based on SW_IN, for separate outlier removal.The other meteo variables are needed later during gapfilling and partitioning, therefore they are also added already here.
- Used data: 2015-2020 from the FLUXNET Warm Winter 2020 dataset (
VPD_F
,SW_IN_F
,TA_F
andRH
) and 2021 from our recent measurements
- Level-3.2 / Outlier Removal: Absolute Flux Limits and Despiking
- For all IRGAs combined
- Proceeding with NEE and LE (no H)
- NEE:
- Step 1: Application of absolute flux limits
- Note that this step was done before the application of the Hampel filter in the next step, so that the Hampel filter has a similar data basis (data range) for all years. The IRGA75 fluxes show generally considerably more noise than IRGA72 fluxes.
- Outliers were removed separately for daytime and nighttime data.
- daytime: values outside a physically plausible range of ±50 µmol CO2 m-2 s-1 were considered outliers
- nighttime: values outside a range between +20 and -5 µmol CO2 m-2 s-1 were considered outliers. The upper limit was chosen based on the range of highest quality nighttime NEE, which is mostly below +20 µmol m-2 s-1 during nighttime. The lower limit was chosen to remove nighttime carbon uptake spikes which are deemed implausible due to the absence of radiation.
- Step 2: Despiking filter
- daytime and nighttime (in combination): using a Hampel filter using the median absolute deviation (MAD) in a running time window of 432 records (9 days). The limit above which a NEE data point was defined as an outlier was set at 10 sigmas (NEE). The filter was applied multiple times until all outliers were removed from the dataset.
- Step 1: Application of absolute flux limits
- LE:
- Step 1: Application of absolute flux limits
- Note that this step was done before the application of the Hampel filter in the next step, so that the Hampel filter has a similar data basis (data range) for all years. The IRGA75 fluxes show generally considerably more noise than IRGA62 and IRGA72 fluxes.
- daytime: values outside a range of +800 and -40 W m-2 were considered outliers. The limit was chosen because the vast majority of highest quality LE fluxes falls into this range: 95th percentile = +397 W m-2, 5th percentile = +4 W m-2.
- nighttime: values outside a range of +100 and -40 W m-2 were considered outliers.The upper limit was chosen based on the range of highest quality nighttime LE, which is mostly found in this range during nighttime: 95th percentile = +27 W m-2, 5th percentile = -3 W m-2.
- Step 2: Despiking filter
- daytime and nighttime (in combination): no additional filter was applied.
- Step 1: Application of absolute flux limits
- This step produced the following fluxes for further processins:
NEE_QCF01
: despiked NEE, whereby daytime data comprises high-quality and OK data and nighttime data comprises high-quality data only. This variable will be gap-filled after USTAR filtering.NEE_QCF0
: despiked NEE, whereby daytime and nighttime data comprise high-quality data only. UnlikeLE_QCF0
(see below), this variable will not go directly into the final dataset at this point, because these values also need to be USTAR filtered before sharing.LE_QCF01
: despiked NEE, whereby daytime data comprises high-quality and OK data and nighttime data comprises high-quality data only. This variable will be directly gap-filled (no USTAR filtering).LE_QCF0
: despiked LE, whereby daytime and nighttime data comprise high-quality data only. Will be provided in the dataset as highest-quality, directly measured LE.
- Level-3.3 / USTAR threshold
- Application of a variable, seasonal USTAR threshold between
0.08
and0.18
m s-1 to NEE data, values below this threshold were rejected. - We applied a seasonal threshold (different threshold for each of the four seasons: DJF, MAM, JJA, SON).
- The threshold detection was done in
ReddyProc
. Results are similar to FLUXNET Warm Winter 2020 thresholds.
- Application of a variable, seasonal USTAR threshold between
- Level-4.1 / Gap-filling
- Important:
- Data basis for gap-filling:
- Data from Level-3.3
- For NEE daytime, data of highest quality (
QCF = 0
) and data of OK quality (QCF = 1
) were kept in the dataset to maximize availability of directly measured data (following CarboEurope recommendations). - For NEE nighttime, only data of highest quality (
QCF = 0
) were retained. - For LE daytime and nighttime, data of highest quality and OK quality were used.
- Lowest quality data (
QCF = 2
) were rejected in all cases.
- Data basis for gap-filling:
- Gap-filling for NEE and LE was done using the MDS method described in Reichstein et al. (2005).
- ET was calculated from gap-filled LE.
- Important:
- Level-4.2 / Partitioning
- Recommended: NEE partitioning was done using the daytime approach described in Lasslop et al. (2010).
- In addition, NEE partitioning using the nighttime method described in Reichstein et al. (2005) was also done. However, we currently recommend to use the daytime partitioning.
Source Data
- Using data from the Adaptive Setup since 2015 (2015-2021) and Mobile Setup 2006-2016 (2015), for an overview of raw data see here: EC Raw Binary Format (CH-AWS)
Flux Data (Level-1)
2015-2021: HS50-IRGA72
- 2021 Level-1_FR-20220127-164245
- 2020 Level-1_FR-20220111-201927
- 2019 Level-1_XXX: FR-20220126-164653 + FR-20220111-202059
- 2018 Level-1_FR-20220111-202307
- 2017 Level-1_FR-20220126-133051
- 2016 Level-1_FR-20220111-202533
- 2015 Level-1_FR-20220126-131843
2015: R2-IRGA75
- 2015 Level-1_FR-20220126-170707
Meteo Data
TA (Tair), SW_IN (Rg), RH, VPD meteo data (2021) were merged with meteo data from the FLUXNET Warm Winter 2020 dataset (2015-2020)