Interactions between nocturnal turbulent flux, storage and advection at an ‘ideal’ eucalypt woodland site

Abstract

<p><strong>Abstract.</strong> While the eddy covariance technique has become an important technique for estimating long-term ecosystem carbon balance, under certain conditions the measured turbulent flux of carbon at a given height above an ecosystem does not represent the true surface flux. Profile systems have been deployed to measure periodic storage of carbon below the measurement height, but have not been widely adopted. This is most likely due to the additional expense and complexity, and possibly also the perception – given that net storage over intervals exceeding 24 hours is generally negligible – that these measurements are not particularly important. In this study, we used a three year record of net ecosystem exchange of carbon and simultaneous measurements of carbon storage to ascertain the relative contributions of turbulent carbon flux, storage and advection (calculated as a residual quantity) to the nocturnal carbon balance, and to quantify the effect of neglecting storage. The conditions at the site are in relative terms highly favourable for eddy covariance measurements, yet we found a substantial contribution (~ 40 %) of advection to nocturnal turbulent flux underestimation. The most likely mechanism for advection is cooling-induced drainage flows, the effects of which were observed in the storage measurements. The remaining ~ 60 % of flux underestimation was due to storage of carbon. We also showed that substantial underestimation of carbon uptake (approximately 80 gC m<sup>−2</sup> a<sup>−1</sup>, or 25 % of annual carbon uptake) arose when standard methods of nocturnal flux correction were implemented in the absence of storage estimates. These biases were much larger than quantifiable uncertainties in the data. Neglect of storage also distorted the relationships between the carbon exchange processes (respiration and photosynthesis) and their key controls (light and temperature, respectively). We conclude that addition of storage measurements to eddy covariance sites with all but the lowest measurement heights should be a high priority for the flux measurement community.</p>

OzFlux data: network integration from collection to curation

“…The resultant storage correction terms have been applied to the AU-Tum and AU-Whr data, processing is in progress for AU-Cum and AU-Wom and will be included in future revisions of the OzFlux data set. McHugh et al (2016) provide details of profile systems at AU-Whr and AU-Wom.…”

“…Short gaps of 1 day or shorter in duration are the most common, but the greatest contribution to the total amount of missing data comes from long gaps of 30 days or more. The high proportion of missing data due to gaps longer than 30 days means that gap-filling techniques such as mean diurnal variation and marginal distribution sampling (Moffat et al, 2007) that are based on site data alone will perform poorly. As a substitute for climatologytype approaches, OzFluxQC uses data from three alternative sources to fill time series of radiation, meteorological and soil data from flux towers: one based on observations and two based on model or reanalysis outputs.…”

“…Provision of high-quality EC data requires that "bad" observations be removed, generating gaps in the data set. These data sets are increasingly used for construction of local to global carbon, water or energy budgets and in modelling applications (Moffat et al, 2007), as Falge et al (2001b) had predicted they would. However, gaps in the data have to be filled to produce defensible sums or to incorporate into modelling frameworks.…”

“…In this method, gaps are filled from the measured flux during a period of similar vapour pressure deficit, solar radiation and air temperature within a ±7-day window, which is expanded to a ±14-day window if no other similar periods are present within ±7 days (Reichstein et al, 2005). Gap filling via lookup tables is widely used, and the errors from filling gaps this way are smaller than for most methods, other than artificial neural networks (ANNs), which produce comparably small errors when trained upon a full year (instead of the ±7-to 14-day windowed approach used in the lookup tables; Moffat et al, 2007). Two types of ANNs are in use for filling of gaps in flux data.…”

“…Two types of ANNs are in use for filling of gaps in flux data. The first type are feed-forward ANNs (Papale and Valentini, 2003), which are in widespread use for gap filling of flux data (Beringer et al, 2007;Moffat et al, 2007;Beringer et al, 2016b). Alternatively, a self-organising linear output (SOLO) model based upon a self-organising feature map (SOFM) is a type of ANN that is used in hydrology for its small error and resistance to overtraining (Hsu et al, 2002).…”