With the eddy covariance (EC) technique, net fluxes of carbon dioxide
(
The availability of ecosystem-level observations of net ecosystem exchange
(NEE) of carbon dioxide (
NEE records from periods with low friction velocity (
All these post-processing steps need to be performed routinely for EC data.
Hence, it is desirable to have automated and reproducible post-processing
tools available that can be easily used, extended, and integrated into
researchers' own workflow. For this purpose we have compiled all routines for the
important
The objectives of the paper are to
provide a reference that describes the methodology of the processing
used in the show that the obtained results do not differ systematically from results
obtained with standard post-processing implemented in the FLUXNET community
Appendix
The post-processing relies on half-hourly or hourly measurements of NEE and
ancillary meteorological data of
The post-processing follows a specific workflow:
determination and filtering of periods with low turbulent mixing
( replacement of missing data in the half-hourly/hourly records
(gap filling), and partitioning of NEE into the gross fluxes GPP and R
Usage of the
The workflow starts with importing the half-hourly (or hourly) data,
in this example the year 1998 of the DE-Tha site. Next, a probability
distribution of
Determining periods with low turbulent mixing is a critical step in the EC
data post-processing. Standard steady-state and integral turbulence
characteristics tests in the initial processing exclude the most problematic
records of
Concept of the
There are at least two methods of estimating the
The method of
The nighttime data (default:
Records during the nighttime with
Alternatively, breakpoint detection can be applied to the unbinned data,
which avoids the sensitivity of the moving point method to the specifics of
the binning schemes
Estimates of the
After quality checks and
In the look-up table (LUT) approach, the fluxes are binned by the meteorological conditions within a certain time window. Within the chosen time window and respective bin, each meteorological variable deviates less than a fixed margin to ensure similar meteorological conditions. The missing value of the flux is then calculated as the average value of the binned records and its uncertainty estimated from their standard deviation.
The original LUT of
NEE fluxes have a mean diurnal course (MDC) that follows the course of the
sun with only respiration during nighttime and a combination of respiration
and photosynthesis during daytime. This autocorrelation of the fluxes is
exploited by taking the average value at the same time of day within a moving
time window of adjacent days (Falge et al., 2001). In
Though the MDC method only showed a medium performance in the gap-filling
comparison for net carbon fluxes by
The so-called marginal distribution sampling (MDS) by
The filling of each half-hourly NEE with the MDS algorithm depends on the
availability of the meteorological data of Rg, Tair, and VPD. (1) If all three meteorological
variables are available, LUT will be used with default margins of
50
The MDS algorithm is optimized for carbon dioxide and water fluxes and can
also be used to estimate the uncertainty of the half-hourly fluxes. In the
comparison of gap-filling methods by
The gross fluxes of GPP into the land system and R
The two most widely used methods are the so-called nighttime partitioning and
daytime partitioning
The method of
Next, temperature sensitivity,
Subsequently, the respiration at reference temperature,
Finally, R
The method of
Parameter
Note, that contrary to the nighttime-based flux partitioning, both GPP and
R
The post-processing steps' implementations of
Data of 25 sites of the LaThuile FLUXNET dataset
Description of sites and times used for benchmarking
Estimation of the
The different estimates of the
The biggest difference of
There are further slight differences between
Differently to DP06,
The general relationship in the estimation of the
Strong correspondence in NEE based on the
When propagating the differences in
The agreement between NEE based on
However, the increase in estimated precision, i.e., lower standard deviation,
of the
While the default seasons and their aggregation are in line with previous
approaches,
The gap-filling implementation of
Compared to the BGC16, the new implementation of the MDS algorithm in
There were also slight differences in the sequence of window sizes between
While
In the benchmark,
Predictions of NEE by
The good agreement between NEE based on gap filling by
The nighttime-based flux partitioning was benchmarked to BGC16, which used
pvWave code developed by
The main features of the
Annual aggregated values of R
In order to evaluate the effects of the differences introduced in the code
described above, we also computed R
Predictions of annually aggregated ecosystem respiration,
R
The two implementations agreed very
well for most sites at an annual timescale. Because of no systematic
deviations across sites, the spatial upscaling of fluxes should not be
affected by
The daytime flux partitioning was benchmarked with results of the BGC online
tool (BGC16, Sect.
BGC16 differed from
While for separating nighttime data
For the estimation of temperature sensitivity
For uncertainty estimation, BGC16 relied on the curvature of the LRC fit's
optimum instead of a bootstrap procedure. Hence, it could not take into
account the uncertainty of
For weighting the records in the LRC fit, BGC16 used the raw estimated NEE
uncertainty of each record. It did not check for high leverage of spurious
low NEE uncertainty estimates. Its estimates, therefore, were in some windows
very strongly influenced by a few records, and failed if a NEE uncertainty
estimate of zero was provided. Moreover, when there were missing values or
values below zero in a given NEE uncertainty, it set all uncertainty to 1, while
Treatment of missing values was not considered by BGC16 and assumed to be handled prior to the processing. Hence, it did not handle missing VPD values and did not retry the LRC fit without the VPD effect in order to also use records with missing VPD. Moreover, as described above, when there were missing values of NEE uncertainty, weighting records in the LRC fit were omitted.
For compatibility with BGC16, the above code differences can be disabled in
Annually GPP predictions of both implementations showed no significant bias
across the test sites (Fig.
Prediction of annually aggregated GPP from
The largest differences in aggregated fluxes between implementations were due
to the extrapolation of fitted parameters to periods where no parameter fits
were obtained. In many of these cases, there were fits at the boundaries of
these periods, whose validity was questionable. Whether these fits passed the
quality check or not had a large influence on the extrapolation and hence on
the aggregated values. For example, at RU-Cok parameter estimates for valid
periods agreed between implementations. However, no valid parameters could be
obtained for winter months. While
Uncertainty estimates of gross fluxes approximately doubled with
Agreement between aggregated fluxes predicted by the daytime method and
absence of bias for the test sites
(Fig.
The daytime flux partitioning is quite sensitive to the details of the LRC
fit. Small changes in treatment of extreme or missing NEE uncertainty
estimates or changes in pre-processing and treatment of missing values cause
different estimates of LRC parameters and propagate to predicted fluxes of
GPP and R
The estimated uncertainties are even more sensitive. Both implementations
occasionally produce unreasonably high outliers that affect the aggregated
values.
The
The freely available R-package enables researchers to integrate the flux data processing into their own offline environment or work stream without the need of uploading data. This seamless integration allows overall workflow to be improved, processing routines to be sped up, and ultimately cleaner, reproducible scientific results to be generated.
The compatibility of the implemented methods with the available standard tools
provides continuity of the data analysis when adopting
A number of enhancements provide more flexibility to the user in the
processing of their data. For instance, the new processing allows
multi-year data to be treated without breaks at annual boundaries that can significantly
affect sites in the Southern Hemisphere or sites characterized by vegetation
activity in winter. Another new feature of
Sensitivity of the results to subtle details of the implementation, however,
calls for caution when interpreting results. This is especially true for
Continued integration of new methodological developments into the package will support research using EC data. We strive to provide new developments in a basic and extensible manner, while paying attention to compatibility with results of reference implementations.
In summary, research using (half-)hourly eddy covariance data can benefit
from building blocks for standardized and extensible post-processing provided
by
The
The
Some general principles and choices in the design of
This section reports an example R session using
The workflow starts with importing the half-hourly data. The example reads a
text file with data of the year 1998 from the DE-Tha site and converts the
separate decimal columns year, day, and hour to a POSIX timestamp column.
Next, it initializes the
The second step is the estimation of the distribution of
The friction velocity,
The output reports
The subsequent post-processing steps will be repeated using the three
quantiles of the
For this example of an evergreen forest site, the same annually aggregated
The second post-processing step is filling the gaps using information from
the valid data. In this case, the same annual
The screen output (not shown here) already shows that the
For each of the different
The third post-processing step is partitioning the net flux (NEE) into its
gross components GPP and R
Now we are ready to invoke the partitioning, here by the nighttime approach,
for each of the several filled NEE
columns.
The results are stored in columns
The visualizations of the results in a fingerprint plot give a compact
overview.
First, the mean of the GPP across all the years is computed for each
The difference between these aggregated values is a first estimate of the
uncertainty range in GPP due to uncertainty of the
In this run of the example a relative error of about 4.7 % is inferred.
For a better but more time-consuming uncertainty estimate, specify a larger
sample of
The results still reside inside the
With changing surface roughness, e.g., during harvest or leaf fall, the
Moreover, the users can also specify other user-defined seasons, e.g., when
harvest dates are known (see package vignette DEGebExample). They can create
a grouping by specifying exact starting days of the periods by the function
With all methods, there is a required minimum number of 160 records within a
season. If there are too few records, the data of the seasons within a year
are combined and the
ALM designed the R-package REddyProc in consultation with MR, based on his original pvWave algorithm. TW extended the functionality of the package and maintained the code. MM and TW conducted the analysis and JK, KS, LS, and OM contributed to the code and/or analysis. TW took the lead in writing the manuscript with contributions from all authors.
The authors declare that they have no conflict of interest.
This work used eddy covariance data acquired and shared by the FLUXNET community, including these networks: AmeriFlux, AfriFlux, AsiaFlux, CarboAfrica, CarboEurope-IP, CarboItaly, CarboMont, ChinaFlux, FLUXNET Canada, GreenGrass, ICOS, KoFlux, LBA, NECC, TERN OzFlux, TCOS-Siberia, and USCCC. The ERA-Interim reanalysis data are provided by ECMWF and processed by LSCE. The FLUXNET eddy covariance data processing and harmonization was carried out by the European Fluxes Database Cluster, the AmeriFlux Management Project, and the Fluxdata project of FLUXNET, with the support of CDIAC and the ICOS Ecosystem Thematic Centre, and the OzFlux, ChinaFlux, and AsiaFlux offices.
The authors acknowledge Dario Papale, Gilberto Pastorello, and Trevor F. Keenan for the discussions on the benchmarking of REddyProc and pvWave code. Mirco Migliavacca and Markus Reichstein acknowledge the Alexander von Humboldt Foundation that funded part of this research activity through the Max Planck Research Award to Markus Reichstein. Mirco Migliavacca acknowledges the MSCA-ITN project TRUSTEE.
Ladislav Šigut was supported by the Ministry of Education, Youth and Sports of the Czech Republic within the CzeCOS program, grant number LM2015061, and within the National Sustainability Program I (NPU I), grant number LO1415. The article processing charges for this open-access publication were covered by the Max Planck Society.Edited by: Paul Stoy Reviewed by: three anonymous referees