System/Sensor Breakdown

Raw data (off-scale, out of ensemble range)

Spikes

Anemometer/Rotation Angle

Missing Correction Terms (Pressure, Temperature, Profile measurements for CO₂ or heat storage)

Stationarity Tests

Integral Turbulence Characteristics

Foot-print/Source Weight

Biological Constraints (Growing season - Leafless period, Bud Break - Leaf Senescence)

Physical Constraints (Available Energy, Energy Balance Closure)

About 36 % gaps and rejected data are found in the seasonal and daily courses of NEE at Walker Branch 1997 (Fig.1). The rejection or gap probability decreases using the CO₂ flux measured above the canopy and not corrected for storage in the canopy air layer or drift. Those terms are determined independently and can add additional gap percentages.

Applied rejection criteria for flux (CO₂ Flux, LE, H, Momentum Flux), storage (CO₂, LE, H in air layer, H in biomass), and main meteorology variables should be documented. Documentation would be useful in interpreting gap periods, decision which gap filling strategy is most applicable, and investigating and possibly reducing rejection causes. A binary flag system of rejection criteria for each variable would allow tracking multiple causes of data rejection. Each bin of the binary code represents one particular reason for data rejection. Instead of the binary value the corresponding decimal number could be stored as in the flag column of each variable:

Linear interpolation is often being used for small gaps (2-3 half-hourly means missing), and especially useful to complete meteorological variables as temperature or relative humidity. Applied to NEE data the results are conflicting: depending on the size of the gaps more or less reasonable results are obtained (Fig.2).

Figure 2: Linear interpolation used to fill gaps in seasonal and daily courses of NEE at Walker Branch 1997.

This gap filling method uses mean daily courses to fill gaps in half-hourly or hourly data. Differences in methods are based on the length of the time interval, which is used to calculate the mean daily course (usually 4 to 15 days). Further, a so-called running mean can be applied (i.e. the 15 days previous to the gap are used to calculate the daily course). On the other hand, the entire time period can be cut into short periods, where gaps within one period are filled with the representative half-hourly value from the mean daily course for that period.

Whereas monthly intervals - especially during changing weather, bud-break or leaf-fall periods - are supposed to be too large (see standard deviations in Fig. 3), 15 day or shorter periods are often applied and on an average seem to show smaller standard deviations (Fig.4). But interpretation of this figure remains somewhat difficult, as the standard deviations in shorter time periods are strongly dependent on weather changes during the investigated period.

Figure 3: July mean day course of NEE at Walker Branch 1997.

Gap filling for NEE:

Carbon fluxes from each site observed during daytime, low atmospheric stability, and periods of similar leaf area were sorted into 2 degree C air temperature classes at the time of observation and plotted against incident photosynthetic photon flux density (PPFD) measured above the canopy. As shown in the examples of Figure 7, the light response curves apparently show high scatter, indicated by the standard deviations. High scatter in the data is expected and depends on additional factors not yet included in the regression analysis, as effects of daytime, seasonality of leaf physiology, water availability, or inhomogeneity in fetch/footprint area of the tower site. Nevertheless, the data provide good estimates of the estimated respiration rate as light decreases to zero (Rday), of the quantum yield a' as the primary slope, and of the maximum carbon uptake (Fcsat) at light saturation. The light dependence for each 2 degree temperature class is described via a modified Michaelis-Menten type function with three parameters: Rday, the respiration term combining leaf, bole and soil respiration, a', the quantum yield of the canopy, and Fcsat, the saturation value in the absence of respiration.

(Eq. 1)

Parameter estimates were obtained from nonlinear regression analyses applied to the data of each temperature class. Responses studied in this manner provide a large data set describing the parameters Rday, Fcsat, and a' as a function of temperature.

The temperature dependency of the respiration term, Rday, may be described by an exponential function of air temperature, although Rday show slightly decreased values at higher temperatures.

(Eq. 2)

where fair and Eair are scaling constant and activation energy, respectively. TK is air temperature in Kelvin, and R is the gas constant (8.31 J K-1 mol-1). A concern in the analysis of the respiration data is that a clear separation of the different respiration sources (leaves, bole, and soil) is not possible. The contribution of these sources will change over time, and in response to different developmental factors (e.g. air, soil, and bole temperature, soil water potential, etc.). Nevertheless, we decided to include only the response to air temperature.

The temperature dependence of the potential rate of carbon uptake at saturating light, Fcsat, is described by:

(Eq. 3)

where Tair is air temperature (K), R is the gas constant (8.31 J K-1 mol-1), D Ha(Fcsat) is the activation energy, D Hd(Fcsat) the energy of deactivation, D S(Fcsat) is an entropy term and c(Fcsat) is a scaling constant.

Values of a', the quantum yield of the canopy, may be directly estimated from the slope of the light-response-curve of canopy carbon dioxide exchange. In this analysis a' were determined together with Rday and Fcsat for each temperature class using Eq. 1 and non-linear regression methods, and showed an optimum curve with temperature. Whereas at leaf level the light use efficiency (a ) at saturating CO2 shows no temperature response, Ehleringer and Björkman (1977) showed leaf quantum yield as temperature dependent. The quantum yield of a canopy, a', is an operationally determined parameter based on stand gas exchange data. Therefore light distribution effects are included in the estimate of the quantum yield, and a’ may be expected to decrease for inhomogeneous, clumped canopy architecture.

Describing the optimum temperature response of a', an equation analogous to the response of Fcsat is used:

(Eq. 4)

where Tair is air temperature (K), R is the gas constant (8.31 J K-1 mol-1), D Ha(a') is the activation energy, D Hd(a') the energy of deactivation, D S(a') is an entropy term and c(a') is a scaling constant.

Semi-empirical gap filling provides preservation of flux responses to meteorological drivers (as temperature, incident light, or vapor pressure deficit), but also shows the need of complete meteorological data sets (compare white bars in Fig. 8).

Figure 8: Semi-empirical fill of gaps in seasonal and daily courses of NEE at Walker Branch 1997 on a half-hourly data basis. Gaps in meteorological drivers are shown in white. The annual sum of NEE is calculated based on semi-empirical filled data, but using running bi-weekly day courses for periods, where air temperature or PAR was missing.