OzFlux Data Release
TERN-OzFlux is a network of observation sites across Australia and New Zealand providing
measurements of the exchange of heat, moisture and CO2 between Australian ecosystems and the atmosphere. TERN-OzFlux is a TERN facility and is also part of the global FluxNet initiative.
OzFlux wishes to announce the release of approximately 150 site-years of data from 22 sites across the network. This announcement introduces this release of the data set and provides some details about the type of data available and how to access the data. If you have questions or comments about the data set, please feel free to contact the OzFlux Data Manager (Peter Isaac, firstname.lastname@example.org). We would also like to hear from end users of the data set so that we can learn from their experiences and improve the utility of the OzFlux data. To this end, a brief email stating how you intend to use the data and any suggestions for improvements would be greatly appreciated.
1) The Data
OzFlux data is divided in to 6 levels:
- L1 – as received from the flux tower.
- L2 – QA/QC data.
- L3 – post-processing and corrections.
- L4 – gap filled meteorology.
- L5 – gap filled fluxes.
- L6 – NEE partitioned in to GPP and ER.
The data in this release are for levels L1 to L6 and have been prepared using the OzFluxQC suite of Python scripts.
OzFlux makes a distinction between data processed by a central facility and data processed by the site PI. In general, data processed by the site PIs is preferred because it is expected that the choice of processing options has been based on local knowledge of the site. This release consists primarily of data processed by the central facility but is notintended to be a substitute for L6 data produced by site PIs themselves. OzFlux is committed to helping site PIs process their own data by providing the tools to do so (eg OzFluxQC), by providing training in the use of the tools (eg the annual data workshops) and by providing one-on-one help sessions as required.
To maintain the distinction between data produced centrally and data produced by the site PI, the data are made available in separate directories: “default” for data produced centrally and “site_pi” for data produced by site PIs.
The OzFlux data in this release are predominantly in netCDF files. Details of the netCDF files provided by OzFlux are given later in this email.
2) Availability of the Data
The OzFlux data is available through 4 avenues.
The OzFlux OPeNDAP/THREDDS server contains L3 to L6 data for 22 sites in one file per level with each file containing multiple years. Data on the server is organised into directories based on site name. Each site directory is further divided into sub-directories for each level; L4 to L6 inclusive. Each level sub-directory is further divided into “default” and “site-pi” sub-directories; “default” contains data processed by the central facility and “site-pi” contains data processed by the site PIs.
The OzFlux OPeNDAP/THREDDS server can be viewed using a web browser and can be accessed by any DAP-aware program such as Panoply. Example code for accessing the OPeNDAP server in IDL and Python is also available from OzFlux, contact Peter Isaac for a copy.
2.2 The OzFlux Data Portal
The OzFlux Data Portal contains L1 to L3 data in yearly files for 34 sites. All L1 to L3 data files on the portal have been produced by the site PIs. Some sites also have L4 to L6 data produced by the site PIs available on the portal. Most of the data is freely available, a small quantity is restricted access to protect graduate student IP. If you would like to use restricted access data please contact the data owner, a link for this is provided on the collection page.
2.3 The OzFlux Share on CloudStor
OzFlux also maintains a shared area on CloudStor. The shared area contains the netCDF files available from the OPeNDAP/THREDDS server as well as summary Excel files (daily, monthly, annual and cumulative statistics) and plots. CloudStor can be accessed using a web browser or through the ownCloud synchronisation client.
Contact Peter Isaac if you would like access to the CloudStor shared area.
OzFlux has recently contributed about 150 site-years of data to the current FluxNet synthesis initiative. The data submitted to FluxNet is L3 data (ie not gap filled or partitioned) and is in the FluxNet CSV format. This data is now available from the FluxNet collection on the OzFlux Data Portal.
The OzFlux data is currently covered by the TERN BY-SA-NC license. This license requires end users of the data to attribute the source of the data (“BY”), to share any products derived from the data under the same license (“SA”) and to contact the data owner if the data is used commercially (“NC”). Attribution text is available from the site page on the OzFlux Data Portal. OzFlux will be reviewing it’s choice of license in early 2016 and is likely to adopt the CC V4 BY license as recommended by the Terrestrial Ecosystem Research Network (TERN).
The OzFlux Fair Use Policy is available here.
The OzFlux netCDF files conform to V1.6 of the CF Metadata Conventions. The standard_name variable attribute is used where this is defined for variables in the OzFlux data set. All variables have a text description in the long_name variable attribute. The files have 3 dimensions, time, latitude and longitude but latitude and longitude have only one element.
Estimates of ecosystem respiration (ER) are available from 3 methods at present. All 3 methods are based on the use of a u* threshold to identify periods of low turbulence. The u* threshold is determined using the Change Point Detection (CPD) method described in Barr et al 2013 (AgForMet, 171-172, 31-45).
Night time conditions have been defined as downwelling shortwave radiation (Fsd) less than 10 W/m2. Night time, u*-filtered CO2 fluxes (Fc) are assumed to be ER. This stage is common to all 3 methods of estimating a continuous time series of ER.
The 3 methods fall into 2 classes: artificial neural networks (ANN) and Lloyd-Taylor (LT).
The ANN methods use 2 different neural network: SOLO described in Hsu et al 2002 (Water Resources Research, Vol 38, No 12, 1302-1319) and FFNET. Comparisons show that annual totals of ER predicted by both are very similar however at shorter time scales the SOLO values are often more stable than those from FFNET. For this data set, the drivers for both ANN methods were soil temperature (Ts), air temperature (Ta) and soil moisture (Sws). Once the ANNs have been trained using the nocturnal, u*-filtered Fc as the target, the ANNs are applied to the whole, gap filled data set to predict ER during the day time and for low turbulence conditions.
The Lloyd-Taylor method uses the Arrhenius equation as described in Lloyd and Taylor 1994 (Functional Ecology, 8, 315-323). Values of the activation energy (E0) are found by fitting the Arrhenius equation to nocturnal, u*-filtered Fc for each year with baseline respiration (rb) estimated for 15 day windows with a 10 day overlap. ER for day time and low turbulence conditions is then calculated using the gap filled Ta data. The method is described in Lasslop et al 2010 (GCB, 16, 187-208).
The L6 data set contains the following time series for ER:
- ER – nocturnal, u*-filtered Fc (with gaps).
- ER_SOLO – ER with gaps (day time and low turbulence) filled with predictions of ER from the SOLO ANN.
- ER_FFNET – ER with gaps (day time and low turbulence) filled with predictions of ER from the FFNET ANN.
- ER_LT – ER with gaps (day time and low turbulence) filled with predictions of ER using the Lloyd-Taylor method.
The units for all ER variables are umol/m2/s (see the “units” variable attribute).
Net ecosystem exchange (NEE) is calculated using CO2 flux observations (Fc) during the day and at night when u* is above the threshold. When u* is below the threshold NEE is set to ER predicted from one of the 3 methods described above. This leads to 3 estimates of NEE, one for each estimate of ER:
- NEE_SOLO – NEE calculated using ER predicted using the SOLO ANN.
- NEE_FFNET – NEE calculated using ER predicted using the FFNET ANN.
- NEE_LT – NEE calculated using ER predicted using the Lloyd-Taylor method.
The units for all NEE variables are umol/m2/s (see the “units” variable attribute). The OzFlux data set also contains Net Ecosystem Productivity defined as -1*NEE.
Gross Primary Productivity (GPP) is calculated following Chapin et al 2006 (Ecosystems, 9, 1041-1050): GPP = -1*NEE + ER. This leads to 3 estimates of GPP:
- GPP_SOLO – GPP using ER_SOLO and NEE_SOLO.
- GPP_FFNET – GPP using ER_FNET and NEE_FFNET.
- GPP_LT – GPP using ER_LT and NEE_LT.
The units for all GPP variables are umol/m2/s (see the “units” variable attribute).
OzFlux has come a long way in the last 5 years. Many things and many people have contributed to this but it is worth acknowledging and thanking a few:
- Eva van Gorsel and James Cleverly in their roles as Director and Associate-Director, both have made large and largely unrecognised contributions.
- TERN Central for their continued support, both monetary and just as importantly, non-monetary.
- And last but really first, the OzFlux site PIs and community at large who have kept the facility operating despite the challenges with funding. If it were not for the dedication, hard work and straight out creative accounting being practised, we would not be here.
If you have any questions about the OzFlux data set, please feel free to contact Peter Isaac.
The OzFlux Community