Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

Long-term flux measurements of water vapour and carbon dioxide, as structured within networks like FLUXNET (Baldocchi et al. 2001), are of strong interest for the scientific community for ecosystem level process studies (e.g. Valentini et al. 2000), derivation of regional to global budgets (e.g. Jung et al. 2009) or to evaluation of land surface models (e.g. Bonan et al. 2011).

The Waldstein–Weidenbrunnen site (DE-Bay) has been intensively studied for more than 20 years, starting with the first carbon dioxide flux measurements in 1996. A history of the site including the main activities during the last two decades is given in Chap. 1 The climate at the Waldstein–Weidenbrunnen site is warm-temperate in the sense of FLUXNET conventions, and the climate type is Cfb according to Köppen–Geiger classification (Kottek et al. 2006). It has been included in the EUROFLUX sites (Valentini et al. 2000), with eddy-covariance-measured carbon and water fluxes obeying the EUROFLUX methodology (Aubinet et al. 2000, 2003). The site took part in the development of quality standards, especially with respect to eddy-covariance data quality (see Chap. 12), footprint (Rebmann et al. 2005; Göckede et al. 2002, 2008) and gap-filling (Rebmann et al. 2004; Ruppert et al. 2006) for eddy-covariance-measured carbon and water fluxes of the EUROFLUX project.

The first flux data sets from 1997 to 2001 were published by Rebmann (2004), Rebmann et al. (2004), the data has been used for site-specific process studies of carbon and water cycling (e.g. Valentini et al. 2000; Matteucci et al. 2000; Bernhofer et al. 2003). A long-term data set has now been compiled for the Waldstein–Weidenbrunnen site. The present study adds the analysis of the data from 2002 up to 2014. We show the long-term behaviour of the site with respect to evapotranspiration and net carbon uptake, as well as gross primary production and ecosystem respiration derived by flux partitioning algorithms. Furthermore, we investigate the inter-annual variation of evapotranspiration with respect to the Budyko framework, which is a non-dimensional characterization of a region by the relationships between evapotranspiration, potential evaporation and precipitation, and compare our data with the results of Williams et al. (2012). Finally, we put our measurements into the regional context and compare our results to selected examples from related studies.

Table 4.1 Selection of instruments used for the long-term budgets
Full size table

2 Methods

2.1 Site Description and Measurement Set-up

The long-term observations of carbon dioxide and water vapour fluxes have been conducted above a Norway spruce forest in the Fichtelgebirge, at the Weidenbrunnen site in the Waldstein measurement area. Chapter 2 provides details about this site including information about location, climate and vegetation. The eddy-covariance set-up has been in operation on the 32 m high walk-up scaffold tower, henceforth called Main Tower or MT (see Chap. 2, Figs. 2.2 and 2.7) since 1996, with a change of instrumentation in 2001 and 2007. A second eddy-covariance complex has been established on a 36 m high slim tower, henceforth called Turbulence Tower or TT in 2007 (see Chap. 2, Figs. 2.2 and 2.8). In this chapter we analyse the data from 2002 up to 2014, with the relevant instrumentation summarized in Table 4.1, while additional information about further routine set-up is given in Appendix A.

2.2 Data Processing

2.2.1 Turbulent Flux Processing

Turbulent fluxes of carbon dioxide and latent heat have been calculated with the internationally compared software package TK2/TK3 (Mauder and Foken 2011, 2015), obeying micrometeorological standards with respect to corrections and data quality control (Foken et al. 2012). Coordinate rotation has been carried out with the planar-fit method (Wilczak et al. 2001) for each month separately, based on an analysis by Siebicke et al. (2012). The net ecosystem exchange NEE is then the sum of the eddy-covariance carbon dioxide flux and the change in storage of the air column below the sensor. As CO2 profile measurements were only available during intensive observation periods, the storage flux has been calculated from mean CO2 concentrations at the top of the tower for the whole period. Ruppert et al. (2006) showed for the Waldstein site, that this method captures the canopy storage reasonably well, as long as unrealistic values of the open-path gas analyser were rejected. The evapotranspiration ET in mm is the water equivalent of the eddy-covariance latent heat flux.

The flux data has been quality controlled with a flagging system ranging from 1 to 9, combining tests for stationarity and integral turbulence characteristics (Foken and Wichura 1996; Foken et al. 2004). Therefore a u ∗-filtering on NEE flux measurements has not been applied (Ruppert et al. 2006, see Chap. 12). We kept flux data of best and intermediate data quality (1–6) for the long-term budgets, while the parameterization of the gap-filling routine (this section) was done with data of best quality (1–3) only. In addition, all data during rain or fog events and the following hour have been discarded due to steamed up windows of the open-path gas analyser. No quality checks in terms of advection have been applied, as the relationship between advection and diagnostics of the eddy-covariance flux is still unclear. Especially the night-time u ∗-criteria was shown to be not consistent with advection (Aubinet et al. 2010). More details about advection at Waldstein–Weidenbrunnen is given in Chaps. 6, 12 and 14.

The footprint of the eddy-covariance data has been calculated with a Lagrangian stochastic forward model (method presented by Göckede et al. 2006). The calculations for 2003 (Göckede et al. 2008, before Kyrill) and for the IOPs in 2007 and 2008 (Siebicke 2008, after Kyrill) showed that although this significant wind-throw destroyed large forest areas in the further vicinity of the sensors, the contribution from the forest to the flux is in the range of 80–97% (main tower: nearly all of the data; turbulence tower: approx. 75% of the data) and is therefore even slightly larger than the contribution in 2003. This can be explained by the canopy height of a growing forest (2003: 19 m, 2008: 25 m, see Chap. 2), diminishing the average footprint extent. The differences between the footprints of the main tower and the turbulence tower can be attributed to the fact that the turbulence tower is slightly closer to the Köhlerloh clearing (showing larger contribution), and further from the Pflanzgarten clearing (showing less contribution). Twenty-five percent of the data at TT is still not exceeding 80% target contribution, but within a range of approximately 74–80%. All in all, we judged the contribution of the target area as sufficient for the calculation of long-term budgets and did not exclude data on the basis of footprint issues.

2.2.2 Meteorological Data

The meteorological data, comprised of the radiation balance (four components) and air temperature, has been used as input for the gap-filling routine, while additional measurements of the precipitation and weather code served for quality checks (Table 4.1). After visual plausibility checks of the data, gaps have been filled (where possible) by linear regression from measurements at other heights or at the Waldstein–Pflanzgarten site.

2.2.3 Gap-Filling

As a last quality check, a multi-step error filter as used by Lüers et al. (2014) has been performed on the meteorological data, the mean CO2 and H2O densities as well as on the fluxes. Besides fixed thresholds, adjustable quantile and standard deviation filters have been applied in order to remove physically non-plausible outliers from the data. In addition to the quality checks mentioned earlier in this section, this procedure removed 2.4% of measured NEE and 1.3% evapotranspiration records. After all quality checks, and due to longer periods of sensor malfunction and power failures, remaining measured NEE (ET) in 2007 and 2010 were only 4.1% (4.0%) and 17.4% (17.5%), respectively (Table 4.2). Therefore, these 2 years have been discarded from further analysis. Available measured NEE (ET) for the remaining years range from 26.5% (26.6%) to 58.4% (56.2%).

Table 4.2 Percentages of observations available for the annual budgets of NEE and evapotranspiration after all quality checks (measured), percentages of data gapfilled with non-linear regression (nonlin reg, Ruppert et al. 2006) and percentages of data filled with the half-hourly ensemble average of all years (ens avg); 2007 and 2010 were discarded from further analysis due to low data coverage
Full size table

Gap-filling of the NEE data was then performed with non-linear regression: The Lloyd–Taylor function (Lloyd and Taylor 1994) is used for night-time respiration with the air temperature as the explanatory variable and a Michaelis–Menten type function (Michaelis and Menten 1913) for day-time NEE, binned in temperature classes of 2 K width and using global radiation as the explanatory variable. The procedure is explained in detail by Ruppert et al. (2006), and they showed with a data set from 2003 that the explanatory variables used explain most of the variability of NEE at the Waldstein site.

Gaps in ET measurements have been filled with the help of Priestley–Taylor potential evaporation E p (Priestley and Taylor 1972). Measured net radiation as well as air temperature was used to calculate E p, with the ground heat flux being 5% of the net radiation (Rebmann et al. 2004). For gap-filling, E p has been scaled to measured ET by linear regression.

Large gaps in the data in winter-time lead to an under-representation of training data for these conditions. Therefore, regressions have not been performed on an annual basis but for two periods only: 2002–2006 (measurements at the main tower) and 2008–2014 (measurements at the turbulence tower). Further adjustments include a temperature threshold for winter-time assimilation, preventing the occurrence of unrealistic carbon uptake during winter-time. Reasons for these adjustments, and their impact, are presented in Sect. 4.3.2.

After this procedure, some gaps were still left due to missing meteorological input data. Filling these gaps by mean diurnal variation as reviewed by, for example, Falge et al. (2001) seems to be not appropriate, as the gaps occur in larger blocks (due to instrument failures or power outages). An ensemble average annual cycle has therefore been calculated for 2002–2014 on half-hourly basis, and used to fill these gaps. A summary of data availability and gap-filling gives Table 4.2, and a detailed list on a monthly basis is in Appendix B of this book.

3 Results and Discussion

3.1 Energy Balance Closure of EC Measurements

The non-closure of the observed surface-energy balance introduces a systematic error to the long-term flux measurements (Foken 2008). We calculate the energy balance closure ratio EBR as the slope of the geometric mean regression of turbulent fluxes vs. available energy (R n − Q G, with Q G = 0. 05 ⋅ R n). The average closure for the whole period is 81.3 ± 5.2% (mean and standard deviation), while the EBR significantly changes from the first to the second period: EBR at the main tower (2002–2006) was 75.4 ± 1.9%, while EBR at the turbulence tower (2007–2014) increases to 84.9 ± 2.2%. A summary of annual EBR is given in Table 4.3, and exemplary scatterplots are displayed in Fig. 4.1. The energy balance closure for the period 1997–1999 was 72.6% (Aubinet et al. 2000), which is similar to the closure from 2002 to 2006. Recent hypotheses as to the reasons for the unclosed energy balance suggest that the missing flux is mainly attributed to the sensible heat flux (Charuchittipan et al. 2014, see Chap. 12). We therefore decided not to correct the latent heat flux (and NEE) for missing turbulent energy.

Table 4.3 Annual energy balance closure ratios EBR of eddy-covariance data
Full size table
Fig. 4.1
figure 1

Energy balance closure ratio from available energy (R n − Q G) vs. turbulent fluxes for 2002/2003 at main tower and 2011/2012 at turbulence tower

Full size image

3.2 Adaptations of the Gap-Filling Method for NEE

With the given set-up, however, the long-term measurements face specific problems during winter-time: data availability is very low due to frequently steamed windows of the open-path gas analyser, and the heating of the sonic anemometer often failed to prevent icing of the sensor head. Moreover, the remaining winter-time data showed carbon uptake in many cases, which is at least not representative for winter-time, and most likely not true, as the respiration components might be subjected to advection, leaving the measured (Reynolds-) flux as an incomplete representation of NEE. Utilizing this data for the gap-filling routine (Sect. 4.2.2) would significantly underestimate winter-time respiration. The following adjustments have therefore been made:

Multi-annual parameterization :

In order to enlarge data coverage in winter-time, non-linear regressions have been performed only for the two periods with different sensor set-up (instead of a parameterization for each year): 2002–2006 and 2008–2014. In order to reduce scatter and to preserve the relative weight of the lower temperature regimes, the regression has been performed on the median of temperature classes (2 K width). All necessary equations and parameters are listed in Table 4.4

Temperature threshold :

Winter-time assimilation (November–April) is only accepted when the air temperature is above 6 ℃. Otherwise, assimilation is set to zero and only parameterized ecosystem respiration is used. A similar method has been applied at a disturbed spruce forest site in the Bavarian Forest by Lindauer et al. (2014), who set GPP to zero from December to March, as GPP from measurements in this period are zero on average. This is not the case here, as the rare measurements with the open-path gas analyser show, on average, assimilation for November–April. However, these measurements are only available under dry, clear-sky conditions, which are not representative of average winter-time conditions. Excluding these data diminishes annual net carbon uptake ( − NEE) by 74 ± 50 g C m−2 a−1, on average, in the period 2002–2014.

Table 4.4 Parameterization of the non-linear regression functions for gap-filling of NEE and latent heat Q E, upper part: Michaelis–Menten type function (Michaelis and Menten 1913) for day-time NEE; middle part: Lloyd–Taylor function (Lloyd and Taylor 1994) for ecosystem respiration R eco; lower part: coefficient for Priestley–Taylor function (Priestley and Taylor 1972); grey values displayed in italic type were discarded from the analysis due to low data coverage
Full size table

3.3 Carbon and Water Vapour Fluxes

The ecosystem fluxes of net ecosystem exchange NEE, gross primary production GPP and ecosystem respiration R eco as well as evapotranspiration ET and potential evaporation E p for the calculated period from 2002 up to 2014, as well as the results by Rebmann et al. (2004) from 1997 to 2001, are summarized as annual budgets in Table 4.5. We analyse these fluxes in more detail in the next sections. Monthly budgets were provided in Appendix B. In the following the data is partly analysed in its whole length, but due to changes in instrument set-up (Table 4.1) and availability of measurements (Table 4.2), three periods were discriminated as well:

1997–2001 :

MT, Gill R2, LiCor Li6262, data analysis: Rebmann et al. (2004).

2002–2006 :

MT, Gill R2/R3, LiCor Li7500, data analysis: this chapter.

2008–2014 :

TT, Metek USA1, LiCor Li7500, data analysis: this chapter.

Table 4.5 Annual budgets of net ecosystem exchange NEE, gross primary production GPP and ecosystem respiration R eco as well as evapotranspiration ET and potential evaporation E p (Priestley and Taylor 1972); italic values were discarded from the analysis due to low data coverage
Full size table

3.3.1 Carbon Exchange

The annual sums of NEE show a net uptake of carbon throughout the whole time series and a clear trend towards larger uptake (Table 4.5, Fig. 4.2). The three periods differ significantly from each other in both flux magnitudes and variance: 1997–2001 shows low values and variation with 40 ± 12 g C m−2 a−1 on average, except for the year 2000 with an uptake of 146 g C m−2 a−1. The net carbon sink in 2002–2006 ranges from 238 to 377 g C m−2 a−1, and finally reaches a range of 491–692 g C m−2 a−1 in 2008–2014. The highest uptakes were observed for 2011–2014, with 615 ± 79 g C m−2 a−1 on average. In comparison with a mean biome flux of 398 ± 42 g C m−2 a−1 for evergreen humid temperate forests from a global database (Luyssaert et al. 2007), our numbers seem to be in the same range on average, but low for 1997–2001 and very high in 2008–2014. Nevertheless, all values are within the range for temperate forest sites shown by Luyssaert et al. (2010).

Fig. 4.2
figure 2

Annual sums of carbon flux components, displayed as absolute deviations from the ensemble average for the whole period; positive values for NEE (GPP) indicate less net (gross) uptake than on average, positive R eco indicate higher respiration than on average

Full size image

While the trend is significant over the whole period and specifically for 2002–2014 (Mann–Kendall trend test, p < 0. 05), it is reflected in the latter two sub-periods, but is not significant for 2002–2006 (p = 0. 46) and is only a tendency for 2008–2014 (p = 0. 13). This implies a superposition of a real trend and site-specific effects related to the sub-periods, which will be discussed later. The flux partitioning estimates suggest that this increase in carbon uptake can be mainly attributed to an enhancement in GPP, while R eco is increasing as well but to a lesser extent than does GPP: a linear regression of R eco vs. GPP gives a slope of 0.21, indicating that a rising GPP causes an increase in NEE by 79% of this growth, while only 21% is consumed by R eco (Fig. 4.3). The coherence between GPP and R eco seems to be robust, and is reasonable in an ecophysiological sense as observed for many sites (Fernández-Martínez et al. 2014; Kutsch and Kolari 2015), although to some extent this coherence must be attributed to a spurious correlation resulting from the calculation for GPP = NEE − R eco as a balance residual, see Vickers et al. (2009) and the follow-up discussion (Lasslop et al. 2010; Vickers et al. 2010).

Fig. 4.3
figure 3

Annual R eco vs. − GPP

Full size image
Fig. 4.4
figure 4

Seasonal carbon exchange at Waldstein–Weidenbrunnen, left panels: Cumulative sums of daily net ecosystem exchange NEE, gross primary production GPP and ecosystem respiration R eco for the period 2002–2006 (left) and 2008–2014 (middle); the thick grey line shows the cumulative sum of the ensemble average for the whole period 2002–2014. Right panel: Monthly sums of NEE, GPP and R eco for 1997–2001 (light grey), 2002–2006 (black) and 2008–2014 (darkgrey); the symbols represent the month with the highest uptake/respiration within the respective year

Full size image

These changes in NEE, GPP and R eco are reflected in the seasonal patterns as presented in cumulative time series and monthly sums (Fig. 4.4). The ensemble average NEE shows net respiration during winter-time from the beginning of November until mid-March, while net uptake takes place in the other months. All years are similar in the winter-time respiration, but the latter periods 2002–2006 and 2008–2014 exhibit subsequently stronger uptake from April to October, especially June, July and August for 2008–2014. The extraordinarily large carbon uptake in 2011, 2012 and 2014 are mainly characterized by an earlier change from net respiration to uptake in March and even stronger uptake from July to October. The time of maximum net uptake shifts from May (1997–2001) to June (2008–2014). Such a shift is reflected in GPP (maximum gross uptake in 2002–2006 mainly in June to mainly in July for the period 2008–2014) as well as in R eco (maximum respiration from mainly July to July/August).

3.3.2 Water Vapour Fluxes

The annual sums of ET show a distinct increase for the whole time series (Table 4.5, Fig. 4.5), with values ranging from 211 to 301 mm a−1 (1997–2001), from 352 to 415 mm a−1 (2002–2006) and from 435 to 570 mm a−1 (2008–2014). The average fluxes increase from 277 ± 44 mm a−1 for 1997–2001 to 493 ± 60 mm a−1 for 2008–2014. Similar to NEE, the trend is significant for the whole period, with no trend visible for 2002–2006 and a tendency apparent for 2008–2014 (p = 0. 13). Thus the patterns for annual ET and NEE for all years are consistent, with high correlation (R 2 = 0. 9). Such increase, however, is not reflected in the potential evaporation E p, nor in precipitation, which both show variability but no trend, suggesting that ET is (still) not strongly limited by these factors.

Fig. 4.5
figure 5

Annual sums of ET, E p and precipitation P

Full size image

Budyko (1974) provides an appropriate framework to examine relationships between ET, E p and precipitation with non-dimensional numbers: evaporative index (ET, normalized with precipitation) vs. dryness index (E p, normalized with precipitation). In Fig. 4.6, dashed straight lines denote the supply limit (ET = P) and the demand or energy limit (ET = E p). Within this framework, annual values at the Waldstein site suggest not only a trend towards the energy limit (driven by the ET trend, while E p shows no trend), but also a large variability (caused mainly by the variability in P, which is not reflected by ET or E p). Although a trend to drier spring seasons has been detected (see Chap. 3), annual ET is still not limited by precipitation. Williams et al. (2012) conducted such an analysis for a global set of flux stations. We compared our values with a corresponding subset of numbers from warm-temperate evergreen coniferous forest sites and found a similar range, and the inter-annual variability at the Waldstein site also nearly covers the variation of the station subset chosen from Williams et al. (2012).

Fig. 4.6
figure 6

Annual potential evaporation, normalized with annual precipitation vs. normalized evapotranspiration according to the Budyko framework. The grey dashed line is the original parameterizations according to Budyko (1974), the red line according to Choudhury (1999) with an exponent n = 1 and the black line also according to Choudhury (1999), but with an exponent n = 1. 49 as proposed for the sites examined by Williams et al. (2012) (Adapted and completed with kind permission of Ⓒ John Wiley and Sons 2012, All rights reserved)

Full size image
Fig. 4.7
figure 7

Seasonal water vapour exchange at Waldstein–Weidenbrunnen, left panels: Cumulative sums of daily evapotranspiration ET for the period 2002–2006 (left) and 2008–2014 (middle); the thick grey line shows the cumulative sum of the ensemble average for the whole period 2002–2014. Right panel: Monthly sums of ET for 1997–2001 (light grey), 2002–2006 (black) and 2008–2014 (darkgrey); the symbols represent the month with the highest ET within the respective year

Full size image

The inter-annual seasonal variations of ET follow the patterns found for NEE (Fig. 4.7): a subsequent increase of ET from April to October, and only minor changes in winter-time. For ET, only 2012 and 2014 were extraordinarily large.

3.4 Factors Influencing the Carbon and Water Vapour Exchange

In Sect. 4.3.3 we showed remarkable trends of carbon and water vapour exchange with a large range in flux magnitude. There are many possible reasons, which may be attributed either to a gradual shift, or to step changes after breakpoints. Some of them, like instrument set-up and location, are rather local effects, and variations due to these factors are not related to the flux variability of the ecosystem that we wish to characterize and therefore add to the uncertainty of the measurements. The history of the forest’s state is site-specific as well, but is certainly relevant for the ecosystem flux. Other factors may have regional to global implication, as, for example, climate change.

3.4.1 Development of the Spruce Forest at Waldstein Site

The Norway spruce forest at the Waldstein site has undergone tremendous changes, partly described in Chap. 2, with obvious implications for ecosystem performance. The most important steps are:

  • forest visibly suffering from forest decline, with nearly no growth and only slight increase in canopy height from 17.8 m (1995) up to 19 m (2003).

  • reconvalescence after a liming application in 2001: re-greening of needles observed and increase in canopy height up to 25 m (2008) and 28 m (2011).

  • Increase in heterogeneity of the forest after the wind-throw in 2007 due to the storm “Kyrill”, with a clearing south of the site and a reduced tree density west of the site (see Chap. 7), although the percentage of forest in the footprint estimates were not changed significantly. The trees from the blowdown have been removed.

It is obvious that regeneration is accompanied by a large carbon uptake as well as enhanced ET. The latter is in agreement with findings in sap-flow measurements (see Chap. 5). Their measurements show that both stand-scale sapwood area and sap flow increased at the Waldstein site, which might be attributed to the effect of liming. The sap-flow measurements made in 1995 at the Waldstein–Weidenbrunnen and five other sites in the area (Alsheimer et al. 1998) always had fluxes below 3 mm d−1, while the measurements in 2007 and 2008 (Chap. 5) also had fluxes larger than 3 mm d−1. The implications of heterogeneity are discussed in a broader perspective in Chap. 19

3.4.2 Instrumental and Methodological Issues

The annual trends of NEE and ET show distinct jumps in magnitude between the periods 1997–2001, 2002–2006 and 2008–2014. The breakpoints coincide with a change in measurement set-up as summarized at the beginning of Sect. 4.3.3. A potential problem related to the low fluxes in 1997–2001 might arise from the 12-bit digitalization done within the Li6262 in contrast to the 16-bit digitalization within the Li7500. Such digitalization problems also existed for the Gill R2 and R3 sonic anemometer up to 2003 (Foken et al. 2004).

Fig. 4.8
figure 8

Integral turbulence characteristics for vertical wind velocity σ w u ∗ −1 from ultrasonic anemometer measurements, left: June 2006, Main Tower, Gill solent R3; right: June 2008, Main Tower, Metek USA1

Full size image

Furthermore, we investigated the impact of the change from R3 to the Metek USA1 sonic anemometer through a comparison of the integral turbulence characteristics for the vertical wind velocity σ w u ∗ −1, as its stability dependence is quite weak in the near-neutral range (Thomas and Foken 2002). Monthly averaged values (March–October, 2002–2014) yield an average σ w u ∗ −1 = 1. 32 ± 0. 026 for 2002–2006 (R2/R3) and 1. 20 ± 0. 026 for 2008–2014 (USA1). A larger number indicates a lower momentum flux for the same observed variance σ w 2 and therefore a possible underestimation of fluxes with the R2/R3. Or, from another perspective, a larger fraction of the variance in R2/R3 observation must be attributed to random noise, which does not contribute to the momentum flux (which likely has an impact on \(\overline{w^{{\prime}}q^{{\prime}}}\) and \(\overline{w^{{\prime}}c^{{\prime}}}\) as well). An exemplary time series is shown for R3, June 2006 and USA1, June 2008 (Fig. 4.8). There is no remarkable difference between R2 and R3 or between the USA1 at the Main Tower and the USA1 at the Turbulence Tower (not shown). The latter indicates that the move of the instrumentation to the Turbulence Tower in 2007 has, at least, no effect with respect to flow distortion by tower elements. On the other hand, the energy imbalance decreased since measurements have been performed on the turbulence tower (Sect. 4.3.1). This suggests that the increase in fluxes is only weakly connected to differences in the sonic anemometer, but might be related to the increased heterogeneity since 2007, which is discussed in Chap. 19

Another problem arises from the methodological differences between the calculation and gap-filling of the fluxes for 1997–2001 (Rebmann 2004; Rebmann et al. 2004) and the procedure conducted in this chapter. In flux quality assessment, for instance, Rebmann et al. (2004) used the u ∗-criterion, while in this chapter the integral turbulence characteristics were utilized. Although similar equations have been used for flux partitioning, parameterization differs in the selection of data subsets for individual regressions. We therefore compared both methods with the data of 2002 (very large fraction of parameterized fluxes, see Appendix B), yielding NEE = −238 g C m−2 a−1 for the methods used here and NEE = −326 g C m−2 a−1 with the methods by Rebmann et al. (2004). While this difference is surely not negligible, it shows that the distinct jump in flux magnitude is reproduced and therefore not a matter of methodological differences. Flux partitioning, however, is affected much more, with differences of 490 g C m−2 a−1 for ecosystem respiration. This can be partly explained with a potential overestimation of night-time respiration (and therefore also extrapolated day-time respiration) due to the usage of the u ∗-criterion (see Chap. 12), and underlines the huge variability among different approaches.

Although different locations and instrumentations may be one reason for the behaviour of the fluxes in the three periods, modelling approaches in 1998, 2003, 2007, 2008 and 2011 with the ACASA model (see Chap. 16) show similar results of comparisons, at least for day-time data of NEE between modelled and measured fluxes in all years. This suggests that the fluxes were reasonable and the differences, attributed either to changes in forest structure or climate, can be reproduced with a process-based model. Also R eco seems to be realistic, as Rebmann et al. (2004) compared the respiration measured using the eddy-covariance method with the sum of the soil efflux (chamber measurements, Subke and Tenhunen 2004) and the modelled wood and foliage respiration for the years 1997–1999, and found good agreement.

3.4.3 Influential Factors of Regional Relevance

Another reason for the trends found that has regional relevance is, of course, climate change. It is shown in Chap. 3 that the site is affected by climate change, mainly through an increase of the temperature in all months and drought periods occurring mainly in spring, while the annual precipitation sum is nearly constant. We have observed the largest increase in fluxes in summer, where no significant trend in precipitation could be found. It is reasonable to assume that higher temperatures in summer could increase carbon uptake in this well-watered region, where the mean temperature for the period 1971–2000 is 5.3 ℃. On the other hand, the parameterization coefficient α, which we used to scale E p to ET in order to gap-fill the measured ET series, was 0.396 for 2002–2006 and 0.481 for 2007–2014. This, together with the fact that E p shows no trend (which is also true for summer only), suggests that the increase in ET seems to be unrelated to environmental conditions.

Fig. 4.9
figure 9

Long-term increase in net ecosystem exchange NEE at five natural forest sites in the north-eastern USA according to Keenan et al. (2013) and at the Waldstein–Weidenbrunnen site (DE-Bay). Remark: only the station (US-Ho1, Howland Forest) is an evergreen coniferous forest, the others are deciduous broad-leaf forests (Published with kind permission of Ⓒ Nature Publishing Group 2013, All rights reserved)

Full size image

Keenan et al. (2013) attribute an observed positive trend of NEE to an increasing carbon dioxide concentration. They were able to show an increase in water use efficiency as leaf intercellular CO2 concentrations rise, which may cause a higher carbon uptake. The authors showed a significant increase of NEE for five North American stations, and the Waldstein–Weidenbrunnen data also fits with their data (Fig. 4.9). Therefore, an increase of the atmospheric carbon dioxide concentration at Waldstein–Weidenbrunnen from 360 to 400 ppm (1997–2015) could potentially contribute to rising fluxes. We assume that the growth at Waldstein–Weidenbrunnen is not limited by nitrogen availability, as this was the case at the beginning of the time series (Matzner 2004), and atmospheric concentrations decreased only slightly (see Chap. 3). We calculated the ecosystem-scale water use efficiency WUE = GPP∕ET, using only data from April to September, using only measured ET (and GPP was inferred from measured NEE only), and rainy days as well as the days after were excluded to ensure that measured ET mainly accounts for transpiration. WUE for 2002–2014, however, ranges from 3.5 to 4.2 g C (kg H2O)−1 with no visible trend, and even a vague decreasing tendency. Similar to the method used by Keenan et al. (2013), we also calculated the inherent water use efficiency by multiplying WUE with E p in order to eliminate the influence of atmospheric demand, but this does not change the situation. This means that reasons other than rising CO2 levels play an important role.

A very conservative parameter is the carbon use efficiency (CUE)—the ratio of net ecosystem exchange and gross primary production. Kutsch and Kolari (2015) made a re-analysis of the investigations by Fernández-Martínez et al. (2014) and found that nutrient availability has some influence on CUE (not as large as postulated by Fernández-Martínez et al. 2014), and a strong dependence on the heterogeneity (in altitude) of the area in the vicinity of the station. They further conclude that a reasonable range of CUE is between 0 and 0.3 and a strong relationship exists between GPP and R eco. In Fig. 4.10 we add the Waldstein–Weidenbrunnen data to the figure of R eco vs. GPP by Kutsch and Kolari (2015). The data from 2002 to 2006 fit well in the relationship found by Kutsch and Kolari (2015), and the data from 2008 to 2014 are still within the observed range, and closer to the data from sites which are rich in nutrients. CUE increases significantly at the Waldstein site, with CUE = 0. 23 ± 0. 04 (2002–2006) and CUE = 0. 34 ± 0. 03 (2008–2014). This implies that an increase of the heterogeneity in 2007 caused by the wind-throw, and perhaps subsequently enhanced nutrient availability could be responsible for the rising NEE and CUE at the Waldstein site.

Fig. 4.10
figure 10

Ecosystem respiration R eco plotted vs. GPP for the remaining 82 sites according to Kutsch and Kolari (2015) and for the Waldstein–Weidenbrunnen site (DE-Bay). (a ) Red: sites with high nutrient availability. Blue: sites with low nutrient availability. Grey: sites with medium nutrient availability. Open squares: sites removed owing to bad data quality and unclosed carbon balance that could not be corrected. Open circles: removed sites younger than 15 years. Grey stars: removed sites with complex terrain. (b ) Average CUE for sites with low and high nutrient availability with a GPP between 1200 and 2300 g C m−2 a−1 (Published with kind permission of Ⓒ Nature Publishing Group 2015, All rights reserved)

Full size image

4 Conclusion

A long-term data set of eddy-covariance measurements of carbon dioxide and water vapour exchange has been compiled for the Waldstein–Weidenbrunnen site (DE-Bay). While measurements from 1997 to 2001 have already been published, we analysed the years from 2002 until 2014 in a uniform manner with respect to data selection and quality control, processing and gap-filling. Within the latter, gross primary production GPP as well as ecosystem respiration R eco has been estimated with standard flux partitioning algorithms.

The Waldstein–Weidenbrunnen site was a carbon sink in all years, while magnitude and variance of net uptake ( − NEE) increased significantly from values around 40 g C m−2 a−1 for 1997–1999 up to 615 ± 79 g C m−2 a−1 for 2011–2014. This is related to a strong increase in GPP, while R eco is slightly enhanced. Evapotranspiration ET follows the NEE trend coherently, with average fluxes ranging from 277 ± 44 mm a−1 (1997–2001) up to 493 ± 60 mm a−1 (2008–2014), while atmospheric demand does not drive the change.

We discussed various potential drivers for this development, namely instrumentation issues, forest stand history and climate change. Instrumental and methodological problems seem to play a minor role and could not explain the huge flux variability. Climate variability and change do indeed play a role at the site, as warming and rising CO2-concentrations are consistent with the observed trend. The effects, however, cannot be disentangled from site-specific changes such as the recovery from forest decline after liming and an increase in heterogeneity after a wind-throw, as well as structural change within the under-storey, which are likely primarily responsible for the harsh changes in the observed dynamics. This attempt has been made here using only the flux data as the information source, with a more general discussion given in Chap. 19 Although “know thy site” is already a commonplace aphorism regarding long-term flux stations, there is a more general problem behind it, as a transition from an “ideal” to a disturbed or heterogeneous site is surely not a singular occurrence and therefore a systematic bias in regional studies using multiple sites is likely.

The presented data set suffers from problems in the measurements, creating large and systematic gaps in the winter-time and raising the need for assumptions about a temperature threshold for winter-time assimilation, which cannot be proved with the current data set. The uncertainty in NEE measurements therefore by far exceeds the 50 g C m−2 a−1 error proposed by Baldocchi (2003) for ideal sites. Nevertheless, the visible variation and trends should be robust due to consistent data processing, and the comparisons with other sites show that our estimates are realistic.

The Waldstein–Weidenbrunnen site has been intensively studied for a long time and large, detailed data sets exist from intensive observation campaigns. A long-term flux data set as presented in this chapter offers the opportunity to put these experiments in a temporal context and therefore create a better connection among those campaigns. The data set provides a comprehensive basis for water balance studies (see Chap. 15). Despite the trend found in the series there is a high inter-annual variability in the fluxes of NEE and ET, presumably following climatic drivers, which should be investigated in more detail. Seasonal trends of precipitation as detected in Chap. 3, however, do not influence the annual ET budgets. Furthermore, model studies can be deployed with varying climate and stand structure to quantify the influence of the different drivers at the Waldstein site. Such investigations could bring more generality into a case study at a unique location, having a special history.