Extension of the Averaging Time in EddyCovariance Measurements and Its Effect on the Energy Balance Closure
 1.6k Downloads
 40 Citations
Abstract
The modified ogive analysis and the block ensemble average were employed to investigate the impact of the averaging time extension on the energy balance closure over six landuse types. The modified ogive analysis, which requires a steadystate condition, can extend the averaging time up to a few hours and suggests that an averaging time of 30 min is still overall sufficient for eddycovariance measurements over low vegetation. The block ensemble average, which does not require a steadystate condition, can extend the averaging time to several days. However, it can improve the energy balance closure for some sites during specific periods, when secondary circulations exist in the vicinity of the sensor. These nearsurface secondary circulations mainly transport sensible heat, and when nearground warm air is transported upward, the sensible heat flux observed by the block ensemble average will increase at longer averaging times. These findings suggest an alternative energy balance correction for a groundbased eddycovariance measurement, in which the attribution of the residual depends on the ratio of sensible heat flux to the buoyancy flux. The fraction of the residual attributed to the sensible heat flux by this energy balance correction is larger than in the energy balance correction that preserves the Bowen ratio.
Keywords
Energy balance closure Ensemble average LITFASS Ogive analysis1 Introduction
In the past few years, despite improvements in measuring and data processing techniques (see review in Foken 2008a; Foken et al. 2011), the energy balance closure problem still remains. According to several studies using largeeddy simulation (LES), the energy balance is significantly improved with contributions from secondary circulations or turbulent organized structures (Kanda et al. 2004; Inagaki et al. 2006; Steinfeld et al. 2007). These secondary circulations are largescale eddies (several km) and relatively stationary (either static or move very slowly). They are generated by surface heterogeneity (Stoy et al. 2013) and normally move away from the ground . Due to their large size and slow motion, their contributions to the low frequency part of the turbulent spectrum cannot be detected by the eddycovariance (EC) measurement, which is typically averaged over a period of 30 min. This results in the underestimation of \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\), when normally measured by the EC technique.
An extension of the averaging time was suggested and expected to result in a greater contribution from the low frequency parts. Two different approaches were introduced for this task: the ogive analysis (Desjardins et al. 1989; Oncley et al. 1990) and the block ensemble average (Bernstein 1966, 1970; Finnigan et al. 2003). The ogive analysis, which requires a steadystate condition, uses the turbulent spectrum to estimate the turbulent fluxes for different frequency ranges, allowing assessment of the contribution of the low frequencies to the turbulent fluxes measured by the EC method. In Foken et al. (2006), the ogive analysis was applied to the data measured over a maize field of the LITFASS2003 experiment and was focused mainly on data from three selected days, where the averaging time was extended up to 4 h. It was found that the time extension did not significantly increase the turbulent fluxes overall.
For the block ensemble average, which does not require a steadystate condition, low frequency contributions were added to the turbulent fluxes from longterm fluctuations over several hours to days. Detailed discussion of the block ensemble average can be found in Aubinet et al. (2012). In Mauder and Foken (2006), the block ensemble average was also applied to the dataset from the same maize field of the LITFASS2003 experiment. The length of the selected dataset was 15 days, while the block ensemble averaging period varied from 5 min to 5 days. This study showed that the block ensemble average can close the energy balance at longer averaging times. Extensive discussion of the energy balance closure of the LITFASS2003 experiment can be found in Foken et al. (2010).
In our study, we applied both ogive analysis and block ensemble average to the data from the LITFASS2003 experiment, which was collected by multiple EC towers over a large heterogeneous area. This surface heterogeneity may induce secondary circulations, some of which may still exist in the vicinity of the measuring stations and influence the measured fluxes. We examine the impact of averaging time extension over different types of surface that were broadly exposed to the same atmospheric condition. However, the data obtained from different measuring stations do not always have the same sampling rate, thus minor modifications in both ogive analysis and block ensemble average were made. These modifications were validated by repeating the ogive analysis in Foken et al. (2006) and the block ensemble average in Mauder and Foken (2006). Since the energy balance of this dataset was previously analyzed over 30 min (Beyrich et al. 2006; Foken et al. 2010), we could mainly concentrate on the averaging time beyond 30 min, which is more related to the low frequency contributions.
2 Material and Method
2.1 LITFASS2003 Experiment and Data Processing
The LITFASS2003 experiment was performed between 19 May 2003 and 18 June 2003 near the RichardAßmannObservatory of the German Meteorological Service in Lindenberg, Germany. The local time zone in this area is UTC + 1. During the experiment, there were 14 groundbased micrometeorological measuring stations over 13 sites, and two elevated measuring stations on the tower at 50 and 90 m heights. This experiment covered an area of 20 \(\times \) 20 km\(^2\) and made up of six major landuse types: grass, maize, rape, cereals (include rye, barley and triticale), lake and forest. More information about the LITFASS2003 experiment can be found in Beyrich and Mengelkamp (2006).
Summary of selected measuring stations from the LITFASS2003 experiment during 20 May 2003, 1200 UTC–18 June 2003, 0000 UTC
Station  Canopy  \(h_\mathrm{c}\) (m)  \(z_\mathrm{m}\) (m)  \(\theta \) (\(^{\circ }\))  Turbulence sensors  \({\Delta } t\) (min)  \(Res\) (W m\(^{2}\))  \(\%Res\) (%) 

HV  Pine forest  14  30.5  30–330  USA1/LI7500  10  133  24 
A5  Rye  0.75–1.50  2.8  60–30  USA1/KH20  5  144  30 
A6  Maize  0.05–0.70  2.7  90–270  CSAT3/LI7500  5  122  31 
A7  Rape  0.70–0.90  3.4  30–240  CSAT3/KH20  5  87  22 
NV2  Grass  0.05–0.25  2.4  60–180  USA1/LI7500  5  75  22 
NV4  Grass  0.05–0.25  2.4  150–330  USA1/LI7500  5  82  24 
FS  Lake  0  3.9  180–30  USA1/LI7500  10  –  – 
M50  Grass  0.05–0.25  50.7  90–300  USA1/LI7500  5  –  – 
M90  Grass  0.05–0.25  90.7  90–300  USA1/LI7500  5  –  – 
All selected stations were equipped with EC systems as listed in Table 1. Fourcomponent net radiometers and soil heat flux plates were also installed, therefore all the energy balance components in Eq. 1 were measured at each station. Details of these measurements were well described in Mauder et al. (2006) and Liebethal et al. (2005). These measurements allow estimation of the residual, which, on average, reached its maximum at 1000–1200 UTC. For low vegetation, its average value during this time ranged from 75 to 145 W m\(^{2}\) (or 20–30 % of the available energy as shown in Table 1).
All kinds of plants store energy in their canopies. This canopy heat storage has two main contributions from the plant material (or biomass) and the air between plants. In Oncley et al. (2007), it was shown that over low vegetation, such as a cotton field, both contributions of canopy heat storage are relatively small and negligible. According to the study in maize and soybean (Meyers and Hollinger 2004), the stored energy in biomass is significant when a canopy is fully developed, while \(Q_\mathrm{G}\) is very low. During the LITFASS2003 experiment, the maize field grew from bare soil up to approximately 0.5 m height at the end of the experiment. Therefore, the stored energy in biomass can be neglected in our analyses. However, a forest’s canopy heat storage is significant (Lindroth et al. 2010) and needs to be included in the energy budget equation (Eq. 1). Unfortunately, we did not collect all required biomass properties of the forest during the LITFASS2003 experiment, so a forest’s canopy heat storage could not be precisely estimated. Hence, all analyses of this site were done without a canopy heat storage term. Since a forest’s canopy heat storage during the daytime would release heat back to the atmosphere during the nighttime, it is more important at the subdiurnal scale (Haverd et al. 2007). Therefore, the omission of a forest’s canopy heat storage would have a minimal effect over a longterm basis. For the lake, due to its characteristics (e.g. large heat capacity), its energy budget cannot be described by Eq. 1. Therefore, its residual is not reported in Table 1.
During the LITFASS2003 campaign, the raw data were processed and averaged over 30 min. For this task, all the participating groups agreed to use the software package TK2 (Mauder and Foken 2004), which has been tested and compared internationally (Mauder et al. 2008). During flux calculation, several corrections were applied. Crosscorrelation analysis was used for fixing the time delay between the sonic anemometer and hygrometer, and a correction was used for the spectral loss in the high frequency range (Moore 1986). The planarfit rotation was used to align the sonic anemometer with a longterm mean streamline (Wilczak et al. 2001); a correction was used to convert the sonic temperature, as recorded by the sonic anemometer, to the actual temperature (Schotanus et al. 1983); a correction was used to correct for density fluctuation (Webb et al. 1980). Crosswind correction was used to account for a different type of sonic anemometer (Liu et al. 2001), and a correction was used for the cross sensitivity between \({\hbox {H}_2\hbox {O}}\) and \({\hbox {O}_2}\) molecules (Tanner et al. 1993), which was only applied for the Krypton hygrometer KH20 (deployed in A5 and A7). More details of these corrections can be found in Mauder et al. (2006) and Foken et al. (2012).
After all flux corrections, quality flags were assigned to each 30min period. These quality flags are the steadystate flag, the integral turbulence characteristic (ITC) flag (Foken and Wichura 1996) and combined flag. The steadystate flag is a result of the steadystate test and represents the stationarity of the data. The ITC flag represents the development of turbulent conditions, which is the result of the flux variance similarity test. The combined flag is the combination of the steadystate and ITC flags. All these flags range from 1 to 9 (from best to worst). High quality data, considered suitable for fundamental scientific research, have flag values of 1–3. More details of the data quality analysis can be found in Foken et al. (2004, 2012).
2.2 Data Selection
In most selected measuring stations, ground heat flux and radiation data are only available from 20 May 2003, 1200 UTC, so the period 20 May 2003, 1200 UTC–18 June 2003, 0000 UTC was used in our study. To ensure high data quality, as well as to minimize the irrelevant factors that might influence turbulent fluxes, we imposed sets of data selection criteria to the ogive analysis and block ensemble average separately. For the ogive analysis, we increased the averaging time to up to 4 h. This 4h period consists of eight consecutive subperiods (or blocks) of 30 min. We performed the ogive analysis over any 4h period only if all blocks satisfied the selection criteria.
The next data selection criterion is a steadystate condition of the time series, which is indicated by the steadystate flag (Sect. 2.1). We only accepted data with high quality flags (flag 1–3). In this study, we did the ogive analysis of the energy balance components (\(Q_\mathrm{H}\) and \(Q_\mathrm{E}\)) and CO\(_2\) flux (\(F_\mathrm{c}=\overline{w^{\prime } c^{\prime }}_\mathrm{CO_{2}}\)) separately. For the energy balance components, we only considered the steadystate flags of friction velocity (\(u_{*}\)), \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\). We only performed the ogive analysis on any periods during which these three steadystate flags qualified simultaneously. For \(F_\mathrm{c}\), we considered only steadystate flags of \(u_{*}\) and \(\mathrm{C}O_{2}\) flux, and performed the ogive analysis on any periods where these two steadystate flags were accepted simultaneously.
We avoided the transition period by excluding from our analysis the time period covering one hour before to one hour after both sunrise and sunset. We also specified the threshold value of each turbulent flux as a minimum requirement in our analysis. For \(u_*\), which indicates the intensity of turbulence (Massman and Lee 2002), the threshold value is 0.1 m s\(^{1}\); this was set to rule out very small turbulent fluxes, which can result from instrumentation noise. This limit normally excludes periods with very weak flow as well. For \(Q_\mathrm{H}\), \(Q_\mathrm{E}\) and \(F_\mathrm{c}\), threshold values were formulated to avoid complication with their measurement errors. According to Mauder et al. (2006), based on a 30min averaging time, the measurement errors of \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\) are 10–20 % of the turbulent flux at 30 min, or 10–20 W m\(^{2}\), whichever is larger. For \(u_*\) and \(F_\mathrm{c}\), the measurement errors are 0.02–0.04 m s\(^{1}\) and 0.5–1 \(\upmu \)mol m\(^{2} \)s\(^{1}\), respectively (Meek et al. 2005). Therefore, we set the threshold values of \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\) to be 20 W m\(^{2}\), and the threshold value of \(F_\mathrm{c}\) to be 1 \(\upmu \)mol m\(^{2}\) s\(^{1}\). Unusually large uncertainty of \(F_\mathrm{c}\) during the nighttime was taken into account by using only data periods with \(u_* >\) 0.25 m s\(^{1}\) (Hollinger and Richardson 2005).
Similar selection criteria cannot apply to the block ensemble average, as this involves averaging times of several hours to days. Therefore, the quality control of this part was done by discarding any periods with more than 10 % of missing raw data. This missing data could result from various factors, e.g. electrical failure.
2.3 Modified Ogive Analysis
Ogive case definition in analogy to Foken et al. (2006)
Case  Criterion 

1  \({\Delta }_{\max }/\overline{F}_{30} \le \eta \) 
2  \({\Delta }_{\max }/\overline{F}_{30} > \eta \) and \({\Delta }_{\max } < 0\) 
3  \({\Delta }_{\max }/\overline{F}_{30} > \eta \) and \({\Delta }_{\max } > 0\) 
To extend the investigation beyond three golden days and cover more landuse types, the modified ogive analysis was applied to a selected dataset (as described in Sect. 2.2) from selected groundbased stations of the LITFASS2003 experiment (Table 1). The modified ogive analysis was applied to all selected sites for the energy balance component, while for \(F_\mathrm{c}\), it was only applied to these sites: maize, grass and forest. This is because \(F_\mathrm{c}\) measurements were not available at the rye (A5) and rape (A7) sites (both equipped with KH20), and \(F_\mathrm{c}\) was very low over the lake. Note that in this study, we did not apply the flux correction as mentioned in Sect. 2.1. Since each point of the cospectrum corresponds to the turbulent flux at a different duration, the choice of suitable duration for the flux corrections would be ambiguous. According to Mauder and Foken (2006), these flux corrections would reduce the residual by 17 %, and we may therefore assume that this reduction would have reflected in an increase of the sensible and latent heat fluxes.
2.4 Block Ensemble Average
When the averaging period \(P\) is extended to be much longer than 30 min, it is very difficult to maintain a steadystate condition. Without a steadystate condition, the Reynolds averaging rules no longer hold, in which case the time average is no longer a good representative statistic. Finnigan et al. (2003) and Bernstein (1966, 1970) proposed using the block ensemble average as it always obeys the Reynolds averaging rules, allowing the formulation to be carried out without a steadystate condition.
 1.
To balance the unsteady horizontal flux divergence and transient changes in source and storage terms.
 2.
To carry the low frequency contribution to the longterm vertical flux.
This approach was applied with the dataset from the Amazonian rain forest in Finnigan et al. (2003), where it was shown that the residual goes to zero at around 4 h. A similar strategy was applied to the 15day dataset from the maize field (A6) of the LITFASS2003 experiment during the period 2 June 2003, 1800 UTC–18 June 2003, 0000 UTC (Mauder and Foken 2006), which we also used in our study as a period \(NP\). Overlapping blocks ensemble averages were used, with the starting point of each consecutive block being shifted by 5 min, where the period \(P\) of the block ensemble average was varied from 5 min to 5 days. Flux corrections were applied as described in Sect. 2.1 in each individual block. It was shown that the energy balance was closed within a day and mainly due to the increase of \(\left\langle Q_\mathrm{H} \right\rangle \).
In the present study, to investigate whether this approach could generally close the energy balance, we applied the block ensemble average to selected groundbased stations of the LITFASS2003 experiment as listed in Table 1 and used an identical period as in Mauder and Foken (2006). We made slight changes compared to Mauder and Foken (2006), because the turbulent data from lake and forest as well as the ground heat fluxes and net radiations from most selected stations are only available every 10 min. Our block ensemble period \(P\) was varied from 10 min to 5 days, with the starting points of each consecutive block being shifted by 10 min. We also used the same flux corrections as in Mauder and Foken (2006).
2.5 Scale Analysis
Since we are interested in how secondary circulations contribute to the low frequency part of the turbulent spectra, we used the wavelet analysis to resolve the underlying scale of motion. The wavelet analysis routine employed in this study is similar to Mauder et al. (2007), which is based on the algorithm described in Torrence and Compo (1998). The Morlet wavelet was used in this routine due to its localization strength in the frequency domain. We can only apply the wavelet analysis to the data from rye, maize and grassland, whose high frequency data are available. The CO\(_2\) flux is not discussed, as it is not related to the energy balance. Other than the high frequency data requirement, the wavelet analysis also consumes large amounts of computing resources, so it is almost impossible to apply over quite large datasets. Therefore, a specific period when largescale structures exist needs to be identified before the wavelet analysis is performed over this specific period.
3 Results and Discussion
3.1 Modified Ogive Analysis
As mentioned in Sect. 2.3, we investigated the impact of averaging time extension with the modified ogive analysis on energy balance components and \(F_\mathrm{c}\) separately. Our data selection criteria (Sect. 2.2) ruled out most nighttime periods in both analyses, because their turbulent fluxes were below thresholds. We expected measuring stations with broader undisturbed wind sectors, namely rye (A5), grass (NV) and forest (HV), to have more qualified periods. This was confirmed by the highest number of qualified periods from grass and rye stations, however, the number of qualified periods of the forest station for the modified ogive analysis of the energy balance components was less than for the other two measuring stations. This is because measurements of \(Q_\mathrm{E}\) at the forest were often rejected due to poor steadystate conditions during daytime. This contrasts with data from the lake (FS), where \(Q_\mathrm{H}\) was often rejected for the same reason. Over low vegetation, steadystate flags of \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\) were normally qualified during 0600–1600 UTC. Some random unsteady periods mostly appeared in the afternoon. For all selected measuring stations, steadystate flags of \(F_\mathrm{c}\) (if measurements were available) were randomly disqualified throughout the day, while steadystate flags of \(u_*\) were mostly qualified. Hence, passing the steadystate criterion is mainly dependent on the stationarity of \(Q_\mathrm{H}\), \(Q_\mathrm{E}\) and \(F_\mathrm{c}\). In the end, at each measuring station, only 5–20 % of available periods were left for the modified ogive analysis, and these periods occurred mainly during 0600–1600 UTC. For the energy balance components, all periods had unstable stratification, while for \(F_\mathrm{c}\), there were a few periods with stable stratification.
Results from the modified ogive analysis of the energy balance components (\(Q_\mathrm{H}\) and \(Q_\mathrm{E}\)) from selected stations of the LITFASS2003 experiment between 20 May 2003, 1200 UTC–18 June 2003, 0000 UTC
Station  Flux  \(\eta \)  Case 1  Case 2  Case 3  

(Tot no.)  (%)  \(\left\langle \overline{F}_{30} \right\rangle \)  no. (%)  \(\left\langle \overline{F}_{30} \right\rangle \)  \(\left\langle {\Delta }_{\max } \right\rangle \)  no. (%)  \(\left\langle \overline{F}_{30} \right\rangle \)  \(\left\langle {\Delta }_{\max } \right\rangle \)  no. (%)  
Forest  \(Q_\mathrm{H}\)  10  261  74.8  205  \(\)33  3.3  224  33  22.0 
HV  20  252  96.7  237  \(\)56  0.8  217  70  2.4  
(123)  \(Q_\mathrm{E}\)  10  107  43.1  128  \(\)33  9.8  119  27  47.2 
20  112  75.6  126  \(\)45  4.9  125  40  19.5  
Rye  \(Q_\mathrm{H}\)  10  148  88.1  99  \(\)15  2.8  85  19  9.2 
A5  20  143  97.7  –  –  0.0  61  36  2.3  
(218)  \(Q_\mathrm{E}\)  10  145  89.9  118  \(\)20  4.6  131  23  5.5 
20  143  97.2  116  \(\)26  0.9  132  30  1.8  
Maize  \(Q_\mathrm{H}\)  10  106  84.6  98  \(\)12  2.6  116  28  12.8 
A6  20  108  94.9  –  –  0.0  92  39  5.1  
(117)  \(Q_\mathrm{E}\)  10  134  82.9  77  \(\)20  12.0  80  18  5.1 
20  127  95.7  91  \(\)37  2.6  57  22  1.7  
Rape  \(Q_\mathrm{H}\)  10  127  90.4  83  \(\)13  8.5  94  12  1.1 
A7  20  123  100.0  –  –  0.0    –  0.0  
(94)  \(Q_\mathrm{E}\)  10  181  98.9  –  –  0.0  141  16  1.1 
20  181  100.0  –  –  0.0    –  0.0  
Grass  \(Q_\mathrm{H}\)  10  117  93.0  101  \(\)15  6.0  132  23  1.0 
NV  20  116  99.5  99  \(\)27  0.5    –  0.0  
(201)  \(Q_\mathrm{E}\)  10  131  86.1  95  \(\)19  2.0  118  19  11.9 
20  140  97.5  94  \(\)31  0.5  114  27  2.0  
Lake  \(Q_\mathrm{H}\)  10  40  95.8  –  –  0.0  31  14  4.2 
FS  20  40  100.0  –  –  0.0    –  0.0  
(72)  \(Q_\mathrm{E}\)  10  197  95.8  93  \(\)15  1.4  121  14  2.8 
20  193  100.0  –  –  0.0    –  0.0 
Results from the modified ogive analysis of friction velocity (\(u_*\)) and CO\(_2\) flux (\(F_\mathrm{c}\))
Station  Flux  \(\eta \)  Case 1  Case 2  Case 3  

(Tot no.)  (%)  \(\left\langle \overline{F}_{30} \right\rangle \)  no. (%)  \(\left\langle \overline{F}_{30} \right\rangle \)  \(\left\langle {\Delta }_{\max } \right\rangle \)  no. (%)  \(\left\langle \overline{F}_{30} \right\rangle \)  \(\left\langle {\Delta }_{\max } \right\rangle \)  no. (%)  
Forest  \(u_*\)  10  0.64  99.5  –  –  0.0  0.38  0.06  0.5 
HV  \((\hbox {m s}^{1})\)  20  0.64  100.0  –  \(\)2.48  0.0  –  –  0.0 
(192)  \(F_\mathrm{c}\)  10  8.68  58.3  8.25  \(\)1.57  12.5  7.43  1.54  29.2 
(\(\upmu \)mol m\(^{2}\) s\(^{1}\))  20  8.29  89.1  7.73  \(\)2.48  4.2  8.21  3.23  6.8  
Maize  \(u_*\)  10  0.31  97.4  0.26  \(\)0.03  0.9  0.15  0.03  1.8 
A6  \((\hbox {m s}^{1})\)  20  0.31  100.0  –  \(\)1.70  0.0  –  –  0.0 
(114)  \(F_\mathrm{c}\)  10  9.09  62.3  7.13  \(\)1.56  14.0  7.34  2.40  23.7 
(\(\upmu \)mol m\(^{2}\) s\(^{1}\))  20  8.69  78.9  7.10  \(\)1.70  10.5  7.52  4.09  10.5  
Grass  \(u_*\)  10  0.33  88.8  –  –  0.0  0.27  0.04  11.2 
NV  \((\hbox {m s}^{1})\)  20  0.33  96.6  –  –  0.0  0.22  0.05  3.4 
(206)  \(F_\mathrm{c}\)  10  9.95  74.3  8.80  \(\)1.67  14.1  7.65  1.27  11.7 
(\(\upmu \)mol m\(^{2}\) s\(^{1}\))  20  9.57  94.7  8.47  \(\)2.74  3.9  9.41  2.82  1.5 
\(\left\langle \overline{F}_{30} \right\rangle \) of \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\) were closely grouped over low vegetation. \(\left\langle \overline{F}_{30} \right\rangle \) of \(Q_\mathrm{H}\) was largest over the forest and smallest over the lake, and vice versa for \(Q_\mathrm{E}\). Over lake and low vegetation, the modified ogive analysis classified most qualified periods of both \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\) as case 1. This suggests that a 30min averaging time is generally sufficient to capture most turbulent fluxes. However, there were significant numbers of cases 2 and 3 of both \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\) from rye, grass, maize and, remarkably, forest stations. These periods of cases 2 and 3 of rye, grass and maize sites were closely related to the stationarity of \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\) over a 4h period. For these three sites, periods of case 1 usually had a 4h steadystate flag of 1 for \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\), while cases 2 and 3 usually had steadystate flags of 2 or greater. This relation was not readily apparent in the forest site, implying that when the atmosphere becomes less stationary at longer averaging time, the measured fluxes over low vegetation can be either increased or decreased. If we restrict our consideration to rye, grass, maize and forest sites, we found that the number of case 3 was normally greater than the number of case 2 in both \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\). This would tell us that the averaging time extension most likely increases \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\). The average maximum flux difference (\(\left\langle {\Delta }_{\max } \right\rangle \)) for \(Q_\mathrm{H}\) was mostly higher than for \(Q_\mathrm{E}\). \(\left\langle {\Delta }_{\max } \right\rangle \) is increased with larger size of an error band (\(\eta \)), while lower numbers of cases 2 and 3 were observed. This would indicate that the fewer periods left had a larger \(\left\langle {\Delta }_{\max } \right\rangle \). However, even with the greatest \(\left\langle {\Delta }_{\max } \right\rangle \) added on top of flux corrections, the energy increase is still not sufficient to close the energy balance. Furthermore, from scalar similarity of \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\), we expected these fluxes to increase or decrease together. This means we should see case 2 or case 3 in both \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\) simultaneously, which was rarely observed.
Values of \(\left\langle \overline{F}_{30} \right\rangle \) of \(u_*\) were highest over the forest and smallest over the lake, and were closely grouped together over low vegetation. Our modified ogive analysis classified most periods from all sites as case 1. This suggests that the time extension has almost no impact on \(u_*\), regardless of canopy types.
For \(F_\mathrm{c}\), all sites gave comparable values of \(\left\langle \overline{F}_{30} \right\rangle \). Case 1 was still in the majority, with a larger fraction of cases 2 and 3 than the energy balance components. Forest also had larger fraction of cases 2 and 3 than did low vegetation. Overall, there was a greater number of case 3 than number of case 2, and \(\left\langle {\Delta }_{\max } \right\rangle \) also increased with \(\eta \). The 4h steadystate flags were normally 1 for case 1 and higher for cases 2 and 3. However, case 2 generally had higher steadystate flags than case 3. This suggests that when the atmosphere becomes less stationary at longer averaging times, the measured values of \(F_\mathrm{c}\) tend to increase. However, when the degree of unsteadiness becomes stronger, the measured values of \(F_\mathrm{c}\) start to decrease.
3.2 Block Ensemble Average
The outcome of the block ensemble averaging was quite unexpected to us. It could close the energy balance over only maize, rye and rape sites. For maize and rye sites, the closures were at around 15–30 h, which is close to the results obtained in Mauder and Foken (2006). These closures were mainly caused by an increase of \(\left\langle Q_\mathrm{H} \right\rangle \) with a longer block ensemble averaging period \(P\). For the rape site, both \(\left\langle Q_\mathrm{H} \right\rangle \) and \(\left\langle Q_\mathrm{E} \right\rangle \) were approximately constant at all values of \(P\). During the observation period, this site was also influenced by rain events in the southern part of the LITFASS area. Therefore, its closure at very large values of \(P\) was not enhanced by the block ensemble average. For grassland and lake, \(\left\langle Q_\mathrm{H} \right\rangle \) was decreased with greater \(P\), which was canceled by the increase in \(\left\langle Q_\mathrm{E} \right\rangle \), and caused the residual to be approximately constant at all values of \(P\). For lake and forest sites, we must interpret the results carefully, because the lake has different characteristics from other terrain sites and we cannot precisely estimate the canopy heat storage (Sect. 2.1) of the forest from our data.
At all sites, both \(\left\langle Q_\mathrm{H} \right\rangle \) and \(\left\langle Q_\mathrm{E} \right\rangle \) were approximately constant within the first few hours. Over longer \(P\), \(\left\langle Q_\mathrm{E} \right\rangle \) was more steady than \(\left\langle Q_\mathrm{H} \right\rangle \). The inflection at the diurnal scale was found at all sites of both \(\left\langle Q_\mathrm{H} \right\rangle \) and \(\left\langle Q_\mathrm{E} \right\rangle \). As all these selected sites are practically in the same \(20 \times 20\) km\(^2\) area, the diurnal effects should not be very different and the degree of inflection should be comparable. Therefore, the stronger inflection over some sites and fluxes may not be entirely caused by the diurnal effects.
As the block ensemble average could not close the energy balance for all selected sites from 2 June 2003, 1800 UTC to 18 June 2003, 0000 UTC, we need to determine the reason for this and whether it would be the same in a different observation period \(NP\). We know that the \(\tilde{w}\tilde{c}\) term of the block ensemble average is related to the low frequency flux contribution. In principle, \(\tilde{w}\tilde{c}\) represents the flux contributions beyond the averaging period \(P\). If we set \(P\) to be 30 min, \(\tilde{w}\tilde{c}\) would represent additional flux after the 30min averaging time. Hence, longterm observation of \(\tilde{w}\tilde{c}\) would show variation of additional fluxes from low frequency contributions, which may be related to observed block ensemble average fluxes. These variations can be observed more clearly when the observation period \(NP\) is sufficiently long to suppress any transient effects in the block ensemble average fluxes.
Our observation period \(NP\), which covered an entire period of the LITFASS2003 experiment, was 20 May 2003, 1200 UTC–18 June 2003, 0000 UTC. We used \(\tilde{w}\tilde{c}\) from all 30min nonoverlapping blocks (\(P=30\) min) within this period \(NP\) to construct the Hovmøller diagrams of \(\tilde{Q}_\mathrm{H}\) (mesoscale flux of sensible heat or \(\tilde{w}\tilde{T}\) in units of energy, \(T\) is temperature) and \(\tilde{Q}_\mathrm{E}\) (mesoscale flux of latent heat or \(\tilde{w}\tilde{a}\) in units of energy, \(a\) is absolute humidity). These diagrams show the variation of additional fluxes beyond the 30min averaging interval. According to Sect. 2.4, \(\tilde{w}\tilde{c}\) can be very large in any arbitrary blocks. We therefore expected to observe some randomly occurring large \(\tilde{Q}_\mathrm{H}\) and \(\tilde{Q}_\mathrm{E}\) in these diagrams.
More interestingly, large \(\tilde{Q}_\mathrm{H}\) were observed consecutively for a few days during 1 June 2003–5 June 2003, over rye, maize, grass and lake. This period was the dry period between the rain events and was not influenced by any significant synoptic events. These large \(\tilde{Q}_\mathrm{H}\) were positive for rye and maize, and mainly negative for grass and lake. Large \(\tilde{Q}_\mathrm{E}\) was not found in this same period. As described in Sect. 2.4, large \(\tilde{Q}_\mathrm{H}\) (or large \(\tilde{w}\tilde{T}\)) could compensate for a strong horizontal divergence in an individual block. However, consecutive occurrences indicate that they were certainly not transient effects. A strong horizontal divergence would imply a strong horizontal advection, which is related to secondary circulations. Hence, we believe that these patterns of large \(\tilde{Q}_\mathrm{H}\) were caused by nearsurface secondary circulations. To support this statement, we inspected the Hovmøller diagram of \(\tilde{Q}_\mathrm{H}\) obtained from the measurement at 90 m height (M90). At this height, there always exist secondary circulations, which means that we should observe series of large \(\tilde{Q}_\mathrm{H}\) more often than in ground measurements. We did actually observe series of large positive and negative \(\tilde{Q}_\mathrm{H}\) throughout the entire period of the LITFASS2003 experiment (Fig. 3c).
Accepted models state that secondary circulations can only reach down to levels near the earth’s surface under the free convection condition, which occurs when the buoyancy term dominates the shear production term, as \(z/L \le 1\). This situation is also accompanied by low friction velocity (Eigenmann et al. 2009). As we did not observe any free convection during 1 June 2003, 1500 UTC–5 June 2003, 1500 UTC, we believe that these nearsurface secondary circulations were caused (either thermally or mechanically) by the surface heterogeneity between different landuse types (Stoy et al. 2013).
3.3 Scale Analysis
We used the wavelet analysis to resolve the scales of motion during 1 June 2003, 1500 UTC–5 June, 2003 1500 UTC with data from the rye, maize and grassland stations. The wavelet analyses of rye and grassland stations are shown in Figs. 5 and 6, respectively. Results for the maize field and the rye field are very similar. From these wavelet crossscalograms, we found small and large scales of motion. The duration of the small scales is around a few min, which should be captured by the eddycovariance measurement over a 30min averaging period. They are present during the daytime at all sites and transport both \(Q_\mathrm{H}\) and \(Q_\mathrm{E}\). The size of the larger scale is approximately one day, and mainly transports \(Q_\mathrm{H}\). It tends to increase \(Q_\mathrm{H}\) in the maize and rye sites, while decreasing \(Q_\mathrm{H}\) over grass. This is consistent with the patterns of \(\tilde{Q}_\mathrm{H}\) and the blockensemble average fluxes. This scale of motion would not be captured by the eddycovariance measurement over the 30min averaging period.
 \({Q_1\!:}\)

\(\tilde{w}>0\) and \(\tilde{T} >0\) or \(\tilde{a} >0\) warm air rising or moist air rising,
 \({Q_2\!:}\)

\(\tilde{w}>0\) and \(\tilde{T} <0\) or \(\tilde{a} <0\) cold air rising or dry air rising,
 \({Q_3\!:}\)

\(\tilde{w}<0\) and \(\tilde{T} <0\) or \(\tilde{a} <0\) cold air sinking or dry air sinking,
 \({Q_4\!:}\)

\(\tilde{w}<0\) and \(\tilde{T} >0\) or \(\tilde{a} >0\) warm air sinking or moist air sinking.
4 Conclusions
The modified ogive analysis, which requires steadystate conditions, reveals that extension of the averaging time by a few hours does not significantly improve the energy balance. The time extension has a greater impact over tall vegetation. Therefore, the 30min averaging time is still, overall, sufficient for the eddycovariance calculation. Sensible heat flux, latent heat flux and CO\(_2\) flux are more sensitive to the time extension than is friction velocity. Over low vegetation, when the atmosphere becomes less stationary with greater averaging times, these three turbulent fluxes tend to increase. However, an increase in the degree of unsteadiness tends to decrease the CO\(_2\) flux. The increase of sensible heat flux is generally greater than the increase of latent heat flux. Over a longer period, the increases or decreases of sensible and latent heat fluxes do not always behave according to the scalar similarity as expected. And lastly, the sizes of the increases in both sensible and latent heat fluxes are not sufficient to close the energy balance at all sites.
Without assuming steadystate conditions, the block ensemble average can extend the averaging time to several days through the inclusion of the periodtoperiod fluctuations (\(\tilde{w}\tilde{c}\), \(c\) is temperature or humidity) in the mean vertical flux. However, this does not usually close the energy balance. The Hovmøller diagram, which shows variations of \(\tilde{w}\tilde{c}\) over a long period, can help to identify when secondary circulations exist in the vicinity of the sensor by exhibiting consecutive large \(\tilde{w}\tilde{T}\). From our findings, secondary circulations that exist near the earth’s surface mainly transport sensible heat. This result also supports the poor scalar similarity between the sensible and latent heat fluxes in the low frequency region (Ruppert et al. 2006; Foken et al. 2011).
Since secondary circulations move very slowly and are relatively large in size, a singletower EC measurement averaging over 30 min is unable to detect them. If the sensor is, coincidentally, at the right time and location when secondary circulations transport nearground warm air upwards, the block ensemble average at a longer period yields higher sensible heat flux, which improves the energy balance closure. However, when these nearsurface secondary circulations transport warm air aloft downwards, the block ensemble average yields a lower sensible heat flux at a longer averaging time. This suggests that nearsurface secondary circulations do transport significant amounts of energy, and these are responsible for the energy balance closure problem rather than sensor deficiencies.
To account for low frequency contributions to turbulent fluxes caused by nearsurface secondary circulations, we must accept that the scalar similarity between the sensible and latent heat fluxes is no longer valid at all scales. Therefore, the widely used energy balance correction in Twine et al. (2000), EBCBo, which assumes the scalar similarity between sensible and latent heat fluxes by preserving the Bowen ratio, would not generally hold.
Notes
Acknowledgments
This research was funded by the German Research Foundation (DFG) within the projects FO 226/201 and BE 2044/41. The authors wish to thank all participants of the LITFASS2003 experiment who produced data used in this analysis. All participants of the LITFASS2003 experiment are listed in Beyrich and Mengelkamp (2006).
References
 Aubinet M, Vesala T, Papale D (eds) (2012) Eddy covariance: a practical guide to measurement and data analysis. Springer, Dordrecht, 438 pp.Google Scholar
 Bernstein AB (1966) Examination of certain terms appearing in Reynolds’ equations under unsteady conditions and their implications for micrometeorology. Q J R Meteorol Soc 92:533–542CrossRefGoogle Scholar
 Bernstein AB (1970) The calculation of turbulent fluxes in unsteady conditions. Q J R Meteorol Soc 96:762–762CrossRefGoogle Scholar
 Beyrich F, Mengelkamp HT (2006) Evaporation over a heterogeneous land surface: EVA\_GRIPS and the LITFASS2003 experiment–an overview. BoundaryLayer Meteorol 121:5–32CrossRefGoogle Scholar
 Beyrich F, Leps JP, Mauder M, Bange J, Foken T, Huneke S, Lohse H, Lüdi A, Meijninger WML, Mironov D, Weisensee U, Zittel P (2006) Areaaveraged surface fluxes over the LITFASS region based on eddycovariance measurements. BoundaryLayer Meteorol 121:33–66CrossRefGoogle Scholar
 de Feriet JK (1951) Average processes and Reynolds equations in atmospheric turbulence. J Meteorol 8:358–361CrossRefGoogle Scholar
 Desjardins RL, MacPherson JI, Schuepp PH, Karanja F (1989) An evaluation of aircraft flux measurements of CO\(_2\), water vapor and sensible heat. BoundaryLayer Meteorol 47:55–69CrossRefGoogle Scholar
 Eigenmann R, Metzger S, Foken T (2009) Generation of free convection due to changes of the local circulation system. Atmos Chem Phys 9:8587–8600Google Scholar
 Finnigan JJ, Clement R, Malhi Y, Leuning R, Cleugh H (2003) A reevaluation of longterm flux measurement techniques part I: averaging and coordinate rotation. BoundaryLayer Meteorol 107:1–48CrossRefGoogle Scholar
 Foken T (2008a) The energy balance closure problem: an overview. Ecol Appl 18:1351–1367Google Scholar
 Foken T (2008b) Micrometeorology. Springer, Berlin 308 ppGoogle Scholar
 Foken T, Wichura B (1996) Tools for quality assessment of surfacebased flux measurements. Agric For Meteorol 78:83–105CrossRefGoogle Scholar
 Foken T, Göckede M, Mauder M, Mahrt L, Amiro B, Munger W (2004) Postfield data quality control. In: Lee X, Massman W, Law B (eds) Handbook of micrometeorology: a guide for surface flux measurement and analysis. Kluwer, Dordrecht, pp 181–208Google Scholar
 Foken T, Wimmer F, Mauder M, Thomas C, Liebethal C (2006) Some aspects of the energy balance closure problem. Atmos Chem Phys 6:4395–4402Google Scholar
 Foken T, Mauder M, Liebethal C, Wimmer F, Beyrich F, Leps JP, Raasch S, DeBruin H, Meijninger W, Bange J (2010) Energy balance closure for the LITFASS2003 experiment. Theor Appl Climatol 101:149–160CrossRefGoogle Scholar
 Foken T, Aubinet M, Finnigan JJ, Leclerc MY, Mauder M, Paw UKT (2011) Results of a panel discussion about the energy balance closure correction for trace gases. Bull Am Meteorol Soc ES92:13ES18.Google Scholar
 Foken T, Leuning R, Oncley SR, Mauder M, Aubinet M (2012) Corrections and data quality control. In: Aubinet M, Vesala T, Papale D (eds) Eddy covariance. Springer, Berlin, pp 85–131CrossRefGoogle Scholar
 Haverd V, Cuntz M, Leuning R, Keith H (2007) Air and biomass heat storage fluxes in a forest canopy: calculation within a soil vegetation atmosphere transfer model. Agric For Meteorol 147:125–139CrossRefGoogle Scholar
 Hollinger DY, Richardson AD (2005) Uncertainty in eddy covariance measurements and its application to physiological models. Tree Physiol 25:873–885CrossRefGoogle Scholar
 Inagaki A, Letzel MO, Raasch S, Kanda M (2006) Impact of surface heterogeneity on energy imbalance: a study using LES. J Meteorol Soc Jpn 84:187–198CrossRefGoogle Scholar
 Kaimal JC, Finnigan JJ (1994) Atmospheric boundary layer flows: their structure and measurement. Oxford University Press, Oxford 289 ppGoogle Scholar
 Kaimal JC, Gaynor JE (1991) Another look at sonic thermometry. BoundaryLayer Meteorol 56:401–410CrossRefGoogle Scholar
 Kaimal JC, Wyngaard JC, Izumi Y, Cote OR (1972) Spectral characteristics of surfacelayer turbulence. Q J R Meteorol Soc 98:563–589CrossRefGoogle Scholar
 Kanda M, Inagaki A, Letzel MO, Raasch S, Watanabe T (2004) LES study of the energy imbalance problem with eddy covariance fluxes. BoundaryLayer Meteorol 110:381–404CrossRefGoogle Scholar
 Leuning R, van Gorsel E, Massman WJ, Isaac PR (2012) Reflections on the surface energy imbalance problem. Agric For Meteorol 156:65–74CrossRefGoogle Scholar
 Liebethal C, Huwe B, Foken T (2005) Sensitivity analysis for two ground heat flux calculation approaches. Agric For Meteorol 132:253–262CrossRefGoogle Scholar
 Lindroth A, Mölder M, Lagergren F (2010) Heat storage in forest biomass improves energy balance closure. Biogeosciences 7:301–313CrossRefGoogle Scholar
 Liu H, Peters G, Foken T (2001) New equations for sonic temperature variance and buoyancy heat flux with an omnidirectional sonic anemometer. BoundaryLayer Meteorol 100:459–468CrossRefGoogle Scholar
 Massman WJ, Lee X (2002) Eddy covariance flux corrections and uncertainties in longterm studies of carbon and energy exchanges. Agric For Meteorol 113:121–144CrossRefGoogle Scholar
 Mauder M, Foken T (2004) Documentation and instruction manual of the eddy covariance software package TK2. Work Report University of Bayreuth, Department of Micrometeorology, ISSN 1614–8916, 26, 42 pp.Google Scholar
 Mauder M, Foken T (2006) Impact of postfield data processing on eddy covariance flux estimates and energy balance closure. Meteorol Z 15:597–609CrossRefGoogle Scholar
 Mauder M, Liebethal C, Göckede M, Leps JP, Beyrich F, Foken T (2006) Processing and quality control of flux data during LITFASS2003. BoundaryLayer Meteorol 121:67–88CrossRefGoogle Scholar
 Mauder M, Desjardins RL, Oncley SP, MacPherson I (2007) Atmospheric response to a partial solar eclipse over a cotton field in central california. J Appl Meteorol Climatol 46:1792–1803CrossRefGoogle Scholar
 Mauder M, Foken T, Clement R, Elbers JA, Eugster W, Grünwald T, Heusinkveld B, Kolle O (2008) Quality control of CarboEurope flux data–part 2: intercomparison of eddycovariance software. Biogeosciences 5:451–462CrossRefGoogle Scholar
 Meek DW, Prueger JH, Kustas WP, Hatfield JL (2005) Determining meaningful differences for SMACEX eddy covariance measurements. J Hydrometeorol 6:805–811CrossRefGoogle Scholar
 Meyers TP, Hollinger SE (2004) An assessment of storage terms in the surface energy balance of maize and soybean. Agric For Meteorol 125:105–115CrossRefGoogle Scholar
 Moore CJ (1986) Frequency response corrections for eddy correlation systems. BoundaryLayer Meteorol 37:17–35CrossRefGoogle Scholar
 Nakamura R, Mahrt L (2006) Vertically integrated sensible heat budgets for stable nocturnal boundary layers. Q J R Meteorol Soc 132:383–403CrossRefGoogle Scholar
 Oncley SP, Businger JA, Itsweire EC, Friehe CA, C LJ, Chang SS (1990) Surface layer profiles and turbulence measurements over uniform land under nearneutral conditions. In: 9th symposium on boundary layer and turbulence. American Meteorological Society, Roskilde pp 237–240.Google Scholar
 Oncley SP, Foken T, Vogt R, Kohsiek W, DeBruin HAR, Bernhofer C, Christen A, van Gorsel E, Grantz D, Feigenwinter C, Lehner I, Liebethal C, Liu H, Mauder M, Pitacco A, Ribeiro L, Weidinger T (2007) The energy balance experiment EBEX2000. Part I: overview and energy balance. BoundaryLayer Meteorol 123:1–28CrossRefGoogle Scholar
 Paw UKT, Baldocchi DD, Meyers TP, Wilson KB (2000) Correction of eddycovariance measurements incorporating both advective effects and density fluxes. BoundaryLayer Meteorol 97:487–511CrossRefGoogle Scholar
 Raabe A (1983) On the relation between the drag coefficient and fetch above the sea in the case of offshore wind in the nearshore zone. Z Meteorol 33:363–367Google Scholar
 Ruppert J, Thomas C, Foken T (2006) Scalar similarity for relaxed eddy accumulation methods. BoundaryLayer Meteorol 120:39–63CrossRefGoogle Scholar
 Schotanus P, Nieuwstadt F, De Bruin H (1983) Temperature measurement with a sonic anemometer and its application to heat and moisture fluxes. BoundaryLayer Meteorol 26:81–93CrossRefGoogle Scholar
 Shaw RH (1985) On diffusive and dispersive fluxes in forest canopies. In: Hutchinson BA, Hicks BB (eds) The forestatmosphere interaction. Reidel Publishing Company, Dordrecht, pp 407–419CrossRefGoogle Scholar
 Steinfeld G, Letzel M, Raasch S, Kanda M, Inagaki A (2007) Spatial representativeness of single tower measurements and the imbalance problem with eddycovariance fluxes: results of a largeeddy simulation study. BoundaryLayer Meteorol 123:77–98CrossRefGoogle Scholar
 Stoy PC, Mauder M, Foken T, Marcolla B, Boegh E, Ibrom A, Arain MA, Arneth A, Aurela M, Bernhofer C, Cescatti A, Dellwik E, Duce P, Gianelle D, van Gorsel E, Kiely G, Knohl A, Margolis H, McCaughey H, Merbold L, Montagnani L, Papale D, Reichstein M, Saunders M, SerranoOrtiz P, Sottocornola M, Spano D, Vaccari F, Varlagin A (2013) A datadriven analysis of energy balance closure across fluxnet research sites: the role of landscape scale heterogeneity. Agric For Meteorol 171–172:137–152CrossRefGoogle Scholar
 Tanner BD, Swiatek E, Greene JP (1993) Density fluctuations and use of the krypton hygrometer in surface flux measurements. In: Allen RG (ed) Management of irrigation and drainage systems: integrated perspectives. American Society of Civil Engineers, New York, pp 945–952Google Scholar
 Torrence C, Compo GP (1998) A practical guide to wavelet analysis. Bull Am Meteorol Soc 79:61–78CrossRefGoogle Scholar
 Twine TE, Kustas WP, Norman JM, Cook DR, Houser PR, Meyers TP, Prueger JH, Starks PJ, Wesely ML (2000) Correcting eddycovariance flux underestimates over a grassland. Agric For Meteorol 103:279–300CrossRefGoogle Scholar
 Webb EK, Pearman GI, Leuning R (1980) Correction of flux measurements for density effects due to heat and water vapour transfer. Q J R Meteorol Soc 106:85–100CrossRefGoogle Scholar
 Wilczak J, Oncley S, Stage S (2001) Sonic anemometer tilt correction algorithms. BoundaryLayer Meteorol 99:127–150CrossRefGoogle Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.