The spatial variability of CO2 storage and the ...

Agricultural and Forest Meteorology 150 (2010) 226–237

Contents lists available at ScienceDirect

Agricultural and Forest Meteorology journal homepage: www.elsevier.com/locate/agrformet

The spatial variability of CO2 storage and the interpretation of eddy covariance fluxes in central Amazonia A.C. de Arau´jo a,*, A.J. Dolman a, M.J. Waterloo a, J.H.C. Gash a,c, B. Kruijt b, F.B. Zanchi a, J.M.E. de Lange a, R. Stoevelaar a, A.O. Manzi d,e, A.D. Nobre d,e, R.N. Lootens a, J. Backer a a

VU University Amsterdam, Department of Hydrology and Geo-Environmental Sciences, De Boelelaan 1085, 1081 HV Amsterdam, The Netherlands ESS-CC (Earth System Science-Climate Change), Wageningen-UR, PO Box 47, 6700 AA Wageningen, The Netherlands Centre for Ecology and Hydrology, Wallingford OX10 8BB, UK d Instituto Nacional de Pesquisas da Amazonia (INPA), Av Andre Arau´jo, 2936 – Aleixo, Manaus, AM, CEP 69060-001, Brazil e Instituto Nacional de Pesquisas Espaciais (INPE), Av. dos Astronautas, 1758 – Jd. Granja, Saõ Jose´ dos Campos, SP, CEP 12227-010, Brazil b c

A R T I C L E I N F O

A B S T R A C T

Article history: Received 16 January 2009 Received in revised form 11 September 2009 Accepted 7 November 2009

The landscape of central Amazonia is composed of plateaus and valleys. Previous observations have shown preferential pooling of CO2 in the valleys, suggesting that the change in CO2 storage in the canopy air space (S) will be spatially variable at the scale of the topography. This may affect the interpretation of the net ecosystem CO2 exchange (NEE) rates measured on the plateaus if they have used one single atmospheric CO2 concentration ([CO2]) vertical profile measurement system. We have measured the diel, spatial and seasonal variation of S along the topography by using a set of automated [CO2] vertical profile measurement systems. In addition, NEE, the above-canopy turbulent exchange of CO2 (Fc) and meteorological variables were also measured on a micrometeorological tower located on the plateau. The nocturnal accumulation of CO2 was larger on the slopes and in the valleys than on the plateau and was larger in the dry period than in the wet period. In addition, the release of this CO2 occurred later in the day on the slopes and in the valleys than on the plateau. Differences in the flow regime above the canopy along the topographical gradient, lateral drainage of respired CO2 downslope, and temporal, spatial, and seasonal variation of soil CO2 efflux (Rsoil) are thought to have contributed to this. These conditions cause S to be higher in magnitude on the slopes and in the valleys than on the plateau during midmorning hours. We demonstrate that there is a larger underestimation of Reco by nighttime eddy covariance (EC) measurements in the dry period than in the wet period. In addition, Reco – as derived from measurements only on the plateau (Fc + SP) – does not agree with that derived by an independent method. Yet S fluxes peaked at about 18:00–20:00 on the slopes and in the valleys, following a continuous decrease after this period until reaching a minimum just after dawn. NEE derived from Fc measured on the plateau and S measured on the plateau, slope and valley increased the estimates of Reco on the plateau by about 30% and 70% in the wet and dry periods, respectively. Particularly for flux-tower sites over complex terrain, we recommend measuring the spatial variability of CO2 at, at least two, more points along the topography to determine to what extent horizontal gradients and storage changes may contribute to tower fluxes. Finally, for sites that present topographical characteristics similar to that described in this study, care must be taken with the use of single in-canopy profiles of [CO2] to correct EC fluxes. ß 2009 Elsevier B.V. All rights reserved.

Keywords: Amazonia Lateral drainage CO2 vertical profile NEE Complex terrain Carbon balance Storage Ecosystem respiration Topography

1. Introduction Estimates based on eddy covariance (EC) measurements have suggested that the Amazon forest might be either a carbon sink (of

* Corresponding author. Current address: Instituto Nacional de Pesquisas da Amazonia – INPA, Av Andre´ Arau´jo, 2936, Aleixo, Programa LBA, Campus-II, Manaus, AM, CEP 69060-001, Brazil. Tel.: +55 92 3643 3638; fax: +55 92 3643 3238. E-mail addresses: [email protected], [email protected] (A.C. de Arau´jo). 0168-1923/$ – see front matter ß 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.agrformet.2009.11.005

about 1.2–7.2 ton C ha1 y1), a carbon source (of about 0.3– 1.4 ton C ha1 y1) or even carbon-neutral (Grace et al., 1995a,b; Malhi et al., 1998; de Arau´jo et al., 2002; Carswell et al., 2002; Saleska et al., 2003; Miller et al., 2004; Kruijt et al., 2004; Hutyra et al., 2007, 2008). In addition to the underlying heterogeneity across the Amazon forest that gives rise to significant variability among the EC estimates of the carbon balance at individual sites, there is a potential bias in EC carbon balance observations if inadequate measurement of ecosystem respiration (Reco) on calm nights is not properly accounted for (Goulden et al., 1996; de Arau´jo et al., 2002;

A.C. de Arau´jo et al. / Agricultural and Forest Meteorology 150 (2010) 226–237

Saleska et al., 2003; Miller et al., 2004; Goulden et al., 2006; Hutyra et al., 2007, 2008; Lloyd et al., 2007; de Arau´jo et al., 2008a). For example, on these nights, even though changes in CO2 stored in the air below the EC measurements reference height are taken into account, there is an underestimation of Reco. Furthermore, lateral drainage of CO2 away from the EC measurement tower and high spatiotemporal variability of CO2 fluxes above and beneath the canopy also cause EC measurements to underestimate Reco. The temporal and spatially integrated net ecosystem CO2 exchange rate (NEE) between an ecosystem and the atmosphere is defined as NEE ¼ F c þ S

(1)

where Fc is the above-canopy turbulent exchange of CO2 (mmol m2 s1), and S is the change in CO2 storage in the canopy air space (mmol m2 s1) (Wofsy et al., 1993). Usually, Fc and S are measured on a single tower because it is assumed that surface characteristics in the EC measurement-tower footprint are horizontally homogeneous (Loescher et al., 2006; Finnigan, 2006; Dolman et al., 2008). Unfortunately, the weak vertical mixing during nighttime conditions in Amazonian forest leads to a violation of the assumption of horizontal homogeneity (Hutyra et al., 2007). Under such conditions, when thermal stratification is strong and turbulence is low, S assumes greater importance and often constitutes the dominant term in Eq. (1) (Finnigan, 2006). In addition, it is well known that changes in S around sunrise and sunset can be significant relative to Fc, especially over periods of an hour or less (Xu et al., 1999; Finnigan, 2006). If S were accurately determined across the EC measurement-tower footprint this would not be a problem, however, spatial heterogeneity makes it problematic to determine S accurately. Furthermore, significant horizontal advection may occur, which violates assumptions that neglect horizontal terms in Eq. (1). The central Amazonia landscape is composed of plateaus and valleys with a maximum difference in height of about 60 m. de Arau´jo et al. (2008a,b) observed preferential pooling of CO2 in the valleys during nighttime in the dry season. They also observed that the CO2 stored in the valleys took longer to be released than that on the plateaus. However, empirical data were lacking on synchronous measurements of diel variability of [CO2] along the topography in both wet and dry seasons. Temperature inversion layers that develop above and within the canopy along the topography, spatial variability of soil CO2 efflux (Rsoil), weak vertical mixing in the valley and lateral drainage of respired CO2 downslope were considered to influence the observed variability in atmospheric CO2 concentration ([CO2]) across the topography (de Arau´jo et al., 2008a,b). It is very likely that this variability in [CO2] causes S to be spatially very heterogeneous within the footprint of the EC tower-based measurements. This may affect the interpretation of NEE on the plateau (de Arau´jo et al., 2002, 2008a). Indeed, Sun et al. (2007) have suggested that the conventional method of estimating NEE using isolated towers may underestimate respiration at night and photosynthesis during daytime, even over relatively flat terrain. This study investigates the spatial and temporal variability of S along a topographical gradient at a site in central Amazonia using a set of automated [CO2] vertical profile measurement systems, and assesses how the spatiotemporal variability in S may affect the NEE measured on the plateau of this landscape.

227

drained Oxisols and Ultisols on the plateaus and slopes, respectively, and poorly drained Spodosols in the valleys (Chauvel et al., 1987; Ferraz et al., 1998; Luizaõ et al., 2004). On the plateau, the forest is taller with higher biomass than on the slope and in the valley (Guillaumet, 1987; Higuchi et al., 1998). The diversity of species per unit area is above 200 species ha1 (Jardim and Hosokawa, 1987; de Oliveira and do Amaral, 2004, 2005). Rsoil is generally higher in the valley than on the plateau in the dry season and vice versa in the wet season (de Souza, 2004; Chambers et al., 2004; van Diepen, 2006). Leaf nitrogen concentration is higher on the plateau than in the valley, whereas the leaf C:N ratio is higher in the valley than on the plateau (Luizaõ et al., 2004; Nardoto, 2005). Annual rainfall is about 2400 mm and average air temperature ranges between 26 8C (in April) and 28 8C (in September). The wetter period extends from December to April, and the drier period from June to September, when rainfall is less than 100 mm per month (de Arau´jo et al., 2002). For more detailed information about the meteorology and hydrology of this site see de Arau´jo et al. (2002), Waterloo et al. (2006), Cuartas et al. (2007), Tomasella et al. (2008), and Hodnett et al. (submitted). 3. Material and methods 3.1. [CO2] vertical profile measurement systems The [CO2] vertical profiles were measured with a system developed using the guidelines given by Xu et al. (1999) and Molder et al. (2000). This system consisted of a cord supporting the solenoid valves, and tube and a box that contained the infrared gas analyzer (IRGA), air pumps, data logger and interface. Details are shown in Fig. 1. Each system consisted of a high-density polyethyl-

2. Site description Measurements were made in a pristine tropical rain forest in central Amazonia, about 60 km northwest of Manaus, Brazil. The site is located in the Asu catchment within the Reserva Biolo´gica do Cuieiras that has an area of about 230 km2. The forest belongs to the Instituto Nacional de Pesquisas da Amazoˆnia (INPA). Soils are well

Fig. 1. Diagram of the [CO2] vertical profile measurement system used to measure the [CO2] along the topography in a site in central Amazonia. The vertical part refers to a string with six boxes, and the excerpt from the second box (from top to bottom) is intended to show the set up of each box (i.e., screen, solenoid valve, wiring, tube and air filter). The horizontal box contains the IRGA, air pumps, wiring, pressure sensor, filters, data logger and interface.

228


ene (HDPE) tube (Synflex 1300-0440x, 6.35 mm OD, with aluminum layer and non-buffering ethylene copolymer coating, Saint-Gobain Inc., MA, USA) with six 2/2 impulse-bi-stable solenoid valves (04241-106-20, 24 VDC, 10 W, Fluid Automation Systems, Geneva, Switzerland) attached to it at fixed distances, which allowed switching to any of the inlets (Fig. 1). Stainless steel filters (SS-4F-K460, 60 mm, Swagelok, OH, USA) were installed at the air intakes to avoid small particles entering the tube. A second filter (NL 1, G 1/8, 5 mm, Rexroth, Bosch group, Germany) to prevent water droplets entering the air pump was installed downstream of the last solenoid valve and it was placed inside a box (40678, Zarges, Weilheim, Germany) (Fig. 1). Tube fittings (SS-400-61, 6.35 mm OD, Swagelok, OH, USA) were used at every connection for sealing the tubing system. An air pump (N 89 KNDC, 24 VDC, KNF Neuberger, Freiburg, Germany) was used to draw the air from any sampling height through the tube at a flow rate of about 9 L min1 (the purge pump in Fig. 1). The air sampling tube had an internal volume of about 0.4 L that corresponds to a maximum residence time of about 3 s. The tubing material used inside the box was polyurethane (1100U06 01, 6 mm OD, Legris, Rennes, France). A single ‘‘Y’’ type instant fitting (3140 04 06, 6 mm OD, Legris, Rennes, France) was installed before the inlet of the purge pump to enable air to be drawn by a second air pump (NMP 830 KNDC B, 24 VDC, KNF Neuberger, Freiburg, Germany) at a flow rate of about 2.5 L min1 (the sample pump in Fig. 1). This pump pushed the air through a filter (4226, Acrodisc CR PTFE, 1 mm, 25 mm OD, Pall Life Sciences, Ann Arbor, MI, USA) and an IRGA (Gascard II, 0–1000 mmol mol1 CO2 measuring range, 24 VDC, Edinburgh Sensors, Division of Edinburgh Instruments Ltd., Livingston, Scotland, UK). This IRGA measures the CO2 mole density (mol m3) and converts it into mole fraction mixing ratio of CO2 in air (mmol mol1). It has a stated precision and repeatability of 2% of the reading, and is equipped with a temperature sensor at the optical bench that is used to automatically compensate the output for variations in temperature (Edinburgh Sensors, 2003). To circumvent the potential problem of leaks in the tube, every system was tested in both laboratory and field conditions with all solenoid valves closed. A pressure sensor (24PCEFA1D, Honeywell Sensing and Control, Golden Valley, USA) was installed permanently before the inlets of both air pumps to measure the vacuum level inside the tubing (Fig. 1). The pressure sensor readings (V) were measured by a homemade interface (CO2 VU-Interface, Vrije Universiteit, Amsterdam, The Netherlands), which in combination with a homemade data logger (VU 32-Channel-Logger, Vrije Universiteit, Amsterdam, the Netherlands) controlled the opening and closing of the solenoid valves (Fig. 1). These sensors were used to perform a valve test before the start of measurements at any sampling level. The time elapsed for closing all valves, checking the vacuum level and opening the valve of the sampling level to be measured was about 0.6 s. Measurements of [CO2] were made at fixed heights (0.5, 3, 7, 11, 20 and 30 m agl—above-ground level). Each height was sampled for 2.5 min, and data were recorded over the last 0.5 min. These readings were logged every 5 s. The system cycled through the entire profile every 15 min (the continuous mode set up of the system). When the main power supply (explained below) was kept on for 24 h, the cycles were measured continuously. In addition, the system could operate in four different modes to save power. For example, when the main power supply was switched off the cycles were measured continuously from 18:00 until 10:00 in the morning of the next day (when the destruction of the nocturnal boundary layer is likely to be completed in the valley), then from 10:00 to 18:00, the system measured only one cycle at fixed times (12:00, 14:00, and 16:00). After 18:00, the cycles were again measured continuously. The [CO2] at each height was calculated as the average of the six stored data points. The time attributed to all heights sampled within one complete cycle corresponds to the time that the cycle

Fig. 2. Composite of satellite images from a tographical gradient in central Amazonia along which [CO2] vertical profiles, S, Fc, NEE, and meteorological variables were measured. An IKONOS image was overlaid on a SRTM image to generate the composite. The open circles denote the places where the measurements were made. The letters P, S and V refer to plateau, slope and valley, respectively. The names K34 and B34 refer to the micrometeorological towers on the plateau and in the valley, respectively.

ended. The system had a low power consumption (about 22 W) and was powered either by solar panels or by a diesel generator located about 2 km northwest of the K34 tower. In addition, each system had two 100 Ah batteries as backup. 3.1.1. Positioning of [CO2] vertical profile measurement systems [CO2] vertical profiles were measured at five different positions along the topographic gradient (Fig. 2). One system on the plateau (P) was attached to the main tower (known as K34), 118 m asl— above sea level. Two systems were positioned on representative slopes: the first (S1) was attached to a tower 90 m asl and about 550 m from the K34 tower; the second (S2) was suspended from the highest branch of a tall tree located about midway down the slope at 89.2 m asl and about 260 m from the K34 tower. Finally, the last two systems were installed at representative sites in the valleys. The first system (V1) was suspended in the same way as S2, but in the valley at 77.3 m asl and about 790 m from the K34 tower, while the second (V2) was attached to a 42 m triangular, steel mast, the B34 tower (de Arau´jo, 2009) at about 75 m asl and 750 m from K34 tower. These systems were installed in early April 2006. Due to several technical problems, such as lightning strikes and oxidation of solenoid valves, the data reported in this study were collected between late April and early October 2006, a period during which all five [CO2] vertical profile measurement systems worked simultaneously. 3.2. Turbulent variables The fluxes of CO2, sensible and latent heat, and momentum transfer were measured from a tower on a medium sized plateau, the K34 tower (de Arau´jo et al., 2002). The raw data were processed using the software Alteddy (version 3.1), based on Aubinet et al. (2000). Detailed information about this software is available on the internet (http://www.climatexchange.nl/projects/alteddy/). Fluxes, means and variances were calculated every half-hour. 3.3. Meteorological variables The meteorological variables measured on K34 tower were described by de Arau´jo et al. (2002). However, data acquisition and


control systems were upgraded with new data loggers (CR-10X, or before December 2003, CR-10, Campbell Scientific, Logan, UT, USA), and a new IRGA for measuring the [CO2] and water vapor concentration ([H2O]) vertical profiles (LI-840, Li-Cor, Lincoln, Nebraska, USA, or before July 2004, CIRAS-SC, PP Systems, UK). 3.4. Calculation of S S was calculated from [CO2] vertical profiles according to Aubinet et al. (2001) as Z hm Pa @c @z (2) S¼ RT a 0 @t where Pa is the atmospheric pressure (N m2), R is the molar gas constant (N m mol1 K1), Ta is the air temperature (K), hm is the maximum measurement height agl (m), c is the [CO2] (mmol mol1), t is time (s), and z is height agl (m). For more detailed and updated information about the calculation of S see Finnigan (2006, 2009) and Kowalski (2008). 3.5. Rsoil measurements Measurements of Rsoil were made on the plateau, on the slope and in the valley using automated systems. On the plateau, the system (LI-8100, Lincoln, Nebraska, USA) was installed about 25 m southwest of K34 tower, from 28 August to 14 September 2006. On the slope, the same system was installed about 15 m east of the mid-slope tower, from 28 June to 18 July 2006. In the valley, a second system was installed permanently about 35 m south of B34 tower. For this position, we have selected the data collected from 10 September to 10 October 2006. More details of these latter measurements are given by Zanchi et al. (in preparation). 3.6. Calibration Each IRGA was calibrated at approximately fortnightly intervals. Standard gases were primary certified CO2 standard (Praxair, Osasco, Saõ Paulo, Brazil, 512 mmol mol1 2%, or before July 2006, 488 mmol mol1 2%), and were used to perform the ‘‘span’’ (gain). Ultra-pure nitrogen gas (5.0 analytical, Praxair, Osasco, Saõ Paulo, Brazil) containing neither CO2 nor H2O was used to perform the ‘‘zero’’ (offset). During the first calibrations of each IRGA used for measuring the [CO2] vertical profiles the absolute difference of the ‘‘zeros’’ presented a variability of about 40 mmol mol1, whereas the relative difference of the ‘‘span’’ values was less than 1%. We therefore decided to maintain the potentiometers of both ‘‘zero’’ and ‘‘span’’ of each IRGA at their original positions, recording the readings in a logbook. Calibration readings were applied a posteriori to correct the raw data. Further calibrations have shown a decrease in the absolute difference of the ‘‘zeros’’ to about 5–8 mmol mol1. As will be shown later, the averaged [CO2] difference among plateau, slopes and valleys was minimum during daytime conditions and about 25 mmol mol1, sufficiently large for making comparisons among the different positions. The IRGAs used on the EC system and in the original [CO2] vertical profile system at K34 showed almost no variability in either ‘‘zero’’ or ‘‘span’’. The overall precision and accuracy of the readings from the IRGAs were about 2 mmol mol1 and 2% of the reading, respectively. The accuracy of S and Fc terms is insensitive to variability of the ‘‘zero’’ because they are derived from differences between successive measurements or have subtracted the mean values, but do depend on the ‘‘span’’ of the IRGA. 4. Results According to [CO2] vertical profile data availability, we have classified the periods from late April to late May 2006 (doy 114–

229

151) and from early June to late September 2006 (doy 152–273) as representative of the wet and dry season, respectively. We will refer to them hereafter as the wet and dry periods. During the wet period, the average daily rainfall was about 11 mm (data not shown), which is as high as the long-term monthly averages for the months of January, February, March and April in this region (de Arau´jo et al., 2002; Chambers et al., 2004). The average daily rainfall during the dry period was about 4 mm (data not shown), which is similar to the long-term average for the dry season in this region (de Arau´jo et al., 2002; Chambers et al., 2004). 4.1. [CO2] vertical profiles Diel averaged curves of [CO2] from all vertical profiles showed an increase during nighttime with maximum after dawn (between 06:00 and 08:00) and minimum late in the afternoon (about 16:00) for every position along the topographical gradient in both wet and dry periods (data not shown). However, [CO2] vertical profiles were different both among the different positions and between the wet and dry periods. The [CO2] was lower on the plateau than on the slopes and in the valleys during both daytime and nighttime periods (Fig. 3). During daytime, except for the measurements made at 0.5 m agl, [CO2] was about 25 mmol mol1 higher on the slopes (S1 and S2) and in the valleys (V1 and V2) than on the plateau (P). During nighttime, the buildup of CO2 was larger on the slopes and in the valleys than on the plateau. Nighttime values of [CO2] were lower in the wet period than in the dry period at all topographical positions (Fig. 3). In the levels near to and above the canopy layer top, daytime values of [CO2] were also lower in the wet period than in the dry period, markedly on the slopes and in the valleys (Fig. 3). In addition, the [CO2] drawdown (depression of the near-surface CO2 concentration below the freeatmosphere value) was larger in the wet period than in the dry period. 4.2. S Positive values of S (i.e., indicating accumulation of CO2 in the canopy air space) were observed from about 14:00–16:00 until 06:00–07:00 (Fig. 4). Particularly on the slopes (S1 and S2) and in the valleys (V1 and V2), nighttime average curves of S showed a maximum about 18:00–20:00, and after this period a continuous decrease until reaching a minimum at about 06:00–06:30 (Fig. 4). During daytime period, the negative values of S (i.e., indicating removal of CO2 out of the canopy air space) were much lower on the slopes and in the valleys than on the plateau (P) (Fig. 4). The nighttime average curves of S were more variable on the plateau than on the slopes and in the valleys in both wet and dry periods (Fig. 4). On the plateau and during nighttime, it was hard to see any difference in S measured in both wet and dry periods (Fig. 4). In contrast, on the slopes and in the valleys, S was higher in the dry period than in the wet period. During daytime, specifically between 06:00 and 10:00, S fluxes were much lower (more negative) in the dry period than in the wet period at all topographical positions (Fig. 4). 4.3. Fc and NEE Fig. 5 shows the diel average curves of Fc, S and NEE on the plateau of K34 tower, in the wet and dry periods. Positive values of Fc and NEE (i.e., indicating that CO2 is leaving the ecosystem as it passes the EC measurement reference height) were observed from about 17:00 until 08:00, when they usually reached a maximum.

230


Fig. 3. Composite of averaged vertical profiles of [CO2] measured on the plateau area (P) along the K34 tower, on the slope areas (S1, S2) along a tower and a tall tree, and in the valley areas (V1, V2) along a tall tree and the B34 tower, during nighttime and daytime conditions in the wet and dry periods of the year 2006. Each point corresponds to halfhour average (SE). The shaded boxes are intended to show the points whose [CO2] were below the tropospheric background [CO2] measured in the marine boundary layer at Ascension Island, UK (7.928S 14.428W; 54 m asl), on October 2006 (represented by the vertical solid line and about 380.77 mmol mol1). The tropospheric background [CO2] was determined by adding the annual [CO2] growth rate for the year 2006 (about 1.73 mmol mol1 y1) to the [CO2] on October 2005 (about 379.04 mmol mol1) (Conway et al., 2007). Time is presented as local time.

During this period, the curves of NEE were more variable than those of Fc, specifically between 18:00 and 24:00, and were in phase with that of S. As to the negative values of Fc and NEE (i.e., indicating a net uptake of CO2 by the ecosystem), they were observed from about 09:00 until 17:00 and from about 07:30 until 17:00, respectively (Fig. 5a and c). In addition, the curves of Fc and NEE were in phase from about 08:00 onwards, and quite similar from about 12:00 until 17:00. During nighttime, Fc and NEE were larger in the wet period than in the dry period (Fig. 5a and c). The average values of Fc were about 3.34 (0.18) and 1.96 (0.18) mmol m2 s1 for wet and dry periods, respectively, whereas that of NEE were about 5.37 (0.26) and 4.54 (0.27) mmol m2 s1, respectively. Daytime values of Fc reached a minimum (more negative) at about noon in the wet and dry periods and they were of the same magnitude (about 16.11 1.21 and 17.07 2.17 mmol m2 s1, respectively) (Fig. 5a). As to NEE, they reached a minimum at about noon and 10:00 in the wet and dry periods, respectively and they were more negative in the dry period than in the wet period (about 20.88 and 17.84 mmol m2 s1, respectively) (Fig. 5c). In addition, the time in which NEE started to become strongly less negative varied between the wet and dry periods. In the wet period, it was observed at about 13:30 and in the dry period at about 11:00 (Fig. 5c).

5. Discussion 5.1. Diel, spatial and seasonal variation 5.1.1. [CO2] vertical profiles The [CO2] was lower on the plateau than on the slopes and in the valleys during both daytime and nighttime periods (Fig. 3). This agrees with the findings of de Arau´jo et al. (2008a,b). Less vertical mixing in the valley than on the plateau and higher Rsoil on the slope and in the valley than on the plateau in the dry season (discussed further) were considered driving the variability in [CO2] along this topography (de Arau´jo et al., 2008a). During nighttime, the buildup of CO2 was larger on the slopes and in the valleys than on the plateau. Nocturnal cold air drainage (i.e., katabatic flow), thermal stratification above and within the canopy layer, and again differences in Rsoil among plateau, slope and valley (explained below) were considered as possible causes for the observed variability in [CO2] along this topography (de Arau´jo et al., 2008a). Daytime and nighttime values of [CO2] were lower in the wet period than in the dry period at all topographic positions (Fig. 3). de Arau´jo et al. (2008a) observed a similar pattern on the plateau of K34 tower. They argued that during nighttime there was less vertical mixing in the dry season than in the wet season. As


231

Fig. 4. Average diel courses of the change in CO2 storage in the canopy air space (S) measured on the plateau area (P) along the K34 tower, on the slope areas (S1, S2) along a tower and a tall tree, and in the valley areas (V1, V2) along a tall tree and the B34 tower: (a, c, e) in the wet period of the year 2006; (b, d, f) in the dry period of the year 2006. In the dry period, continuous measurements of S in the areas S2 and V1 between 10:00 and 18:00 h were hampered by energy supply. The shaded boxes indicate the nighttime periods. Each point corresponds to half-hour average. The error bars were omitted for clarity reasons. Points above the dashed line denote accumulation of CO2 in the canopy air space, and below the line removal of CO2 out of the canopy air space. Time is presented as local time.

mentioned before, it is very likely that nocturnal cold air drainage and the differences in Rsoil among plateau, slope and valley are driving the differences in [CO2] between wet and dry periods. For example, katabatic flow was measured underneath the canopy of an Amazonian forest at about 700 km east of our site during both wet and dry periods (To´ta et al., 2008). In addition, using an isotopic tracer (e.g., carbon isotope ratio of atmospheric CO2 d13Ca), it was observed that in the dry season, the carbon isotope 13 ratio of ecosystem respired CO2 ðd CReco Þ was more negative on the plateau of K34 tower than in the valleys during nights with a relatively less stable atmosphere above the canopy on the plateau (de Arau´jo et al., 2008b). The most likely explanation was katabatic flow that might have transported less negative d13Ca from the plateau to the valleys. Furthermore, near to and at this site, Rsoil varies in a temporal, spatial and seasonal scale (de Souza, 2004; Chambers et al., 2004; Sotta et al., 2004; do Carmo et al., 2006; van Diepen, 2006; Zanchi

et al., in preparation). For example, in the wet season, Rsoil is higher on the plateaus and along the slopes than in the valleys (de Souza, 2004; van Diepen, 2006; Zanchi et al., in preparation). In contrast, in the dry season, Rsoil is higher on the slopes and in the valleys than on the plateaus (de Souza, 2004; Zanchi et al., in preparation). These conditions suggest that the contribution of katabatic flow to the nocturnal buildup of CO2 on the slopes and in the valleys may be more pronounced in the wet season than in the dry season. In the wet season, the nights are less stable on the plateaus and Rsoil is lower in the valleys (de Arau´jo et al., 2008a; de Souza, 2004; van Diepen, 2006; Zanchi et al., in preparation). Higher values of [CO2], on the slopes and in the valleys during nighttime, in the dry period than in the wet period very likely result from a combination of biotic (Rsoil) and physical processes (katabatic flow and stable layer above the canopy on the plateau). Lower values of [CO2] in the wet period than in the dry period during daytime hours may be related to differences in CO2 uptake

232


period at all topographical positions (Section 4.1). As to the lower values (more negative) of S on the slopes (S1 and S2) and in the valleys (V1 and V2) than on the plateau (P) during daytime hours (Fig. 4), they result from larger nocturnal buildup of CO2 on the slopes and in the valleys than on the plateau (Fig. 3). In addition, also from the longer time needed to release the CO2 stored on the slopes and in the valleys than on the plateau (Section 5.1.1; de Arau´jo et al., 2008a; de Arau´jo, 2009). The higher variability of nighttime curves of S on the plateau than on the slopes and in the valleys (Fig. 4), in both wet and dry periods, suggests that differences in the flow regimes above the canopy layer on the plateau and in the valley may exist (de Arau´jo, 2009). Larger S on the slopes and in the valleys in the dry period than in the wet period is an effect of the very likely combination of biotic and physical processes as discussed in the previous section. Particularly in the dry period, nighttime average curves of S were in phase with those of Rsoil on the slope and in the valley, which have shown a peak about 18:00–20:00, followed by a decrease throughout the night until reaching a minimum about 06:00– 06:30 (Fig. 6). As Rsoil is smaller on the slopes and in the valleys than on the plateau in the wet season, it is likely that in the dry season Rsoil may contribute more to the nocturnal buildup of CO2 on the slopes and in the valleys than katabatic flow. Lower S fluxes, between 06:00 and 10:00 in the dry period than in the wet period at all topographical positions (Fig. 4), are an effect of the larger nocturnal buildup of CO2 in the dry period than in the wet period (Fig. 3).

Fig. 5. Average diel courses of (a) above-canopy turbulent exchange of CO2 (Fc), (b) change in the CO2 storage in the canopy air space (S) and (c) net ecosystem CO2 exchange rates between the ecosystem and the atmosphere (NEE) measured on the plateau area of K34 tower in the wet and dry periods of the year 2006. Each point corresponds to half-hour average. The error bars were omitted for clarity reasons. Points above the horizontal dotted line denote CO2 release (for NEE and Fc) or accumulation of CO2 in the canopy air space (for S), and below the line CO2 uptake (for NEE and Fc) or removal of CO2 out of the canopy air space (for S). The dash-dotted line refers to Reco (about 7.8 mmol m2 s1) derived by an independent method. The shaded boxes indicate the nighttime periods. Time is presented as local time.

5.1.3. Fc and NEE The nocturnal Reco is inferred from measurements of NEE, which have shown positive values throughout the night (Fig. 5c). Regarding vertical mixing, the nights in Amazon forest are extremely calm and the occurrence of turbulence is uneven (de Arau´jo et al., 2002; Saleska et al., 2003; Goulden et al., 2004; Miller et al., 2004; Hutyra et al., 2007, 2008). As mentioned before, under such conditions, S assumes greater importance and often constitutes the dominant term in Eq. (1), and therefore modulates NEE more than Fc. This is the reason why nighttime Fc curves presented lower variability than that of NEE and were not in phase with that of S. During daytime, the earlier net uptake of CO2 by NEE values in relation to that of Fc, is again related to the greater importance of S in the period just after dawn (Fig. 5). For example, while Fc values

and release. For example, Fc measured near to and on the plateau of this K34 tower has shown larger carbon uptake in the wet season than in the dry season (Malhi et al., 1998; de Arau´jo et al., 2002). Yet, de Arau´jo et al. (2008a) have observed larger [CO2] drawdown in the wet season than in the dry season on the same plateau, similar to what we have observed in this study along the topographical gradient. As to CO2 release, particularly in the dry period, Rsoil is higher on the slope and in the valley than on the plateau and there is less vertical mixing in the valley than on the plateau during daytime (de Souza, 2004; Zanchi et al., in preparation; de Arau´jo, 2009). 5.1.2. S Positive values of S from about mid-afternoon until early in the morning (Fig. 4) denote the increase of [CO2] during the same

Fig. 6. Average diel course of soil CO2 efflux (Rsoil) measured on the plateau, slope and valley of a site in central Amazonia in the dry period of the year 2006. Each point corresponds to half-hour average (SE). The shaded boxes indicate the nighttime periods. Time is presented as local time.


denote a high net CO2 release (i.e., positive values), S values denote an even higher removal of CO2 out of the canopy air space (i.e., negative values). The arithmetic solution of these terms in Eq. (1) causes NEE to become negative earlier than Fc. The nighttime values of Fc and NEE were larger in the wet period than in the dry period (Fig. 5). This may be explained by differences in nocturnal vertical mixing between wet and dry seasons as mentioned above, i.e., nights are relatively less stable in the wet season than in the dry season (Section 5.1.1; de Arau´jo et al., 2008a). Nevertheless, this difference suggests that Reco is larger in the wet period than in the dry period. Indeed, there are remarkable seasonal differences in Rsoil, which amounts to about 40% of Reco for this forest (Chambers et al., 2004). For example, Rsoil on the plateau is larger in the wet season than in the dry season (de Souza, 2004). Independent measurements of Reco made near to our site, reported an average value of about 7.8 mmol m2 s1 (Chambers et al., 2004). It is clear that the values reported in this study, for example, nighttime average NEE of about 5.37 (0.26) and 4.54

233

(0.27) mmol m2 s1 for wet and dry periods, respectively, are not in agreement with that of independent measurement (discussed further). This implies that there is a larger underestimation of Reco by EC measurements in the dry period than in the wet period. During daytime, Fc minimum occurred at noon in the wet and dry periods. This time corresponds to the time that the maximum wind speed, friction velocity (u*) and solar incoming radiation above the K34 canopy occurs (de Arau´jo et al., 2002). This suggests that the time in which Fc minimum occurs is determined by physical drivers rather than by biotical factors. As regards to NEE minimum occurring at different times (at about noon and 10:00 in the wet and dry periods, respectively) and being more negative in the dry period that in the wet period, there appears to be a combination of biotic and physical factors. For example, Fc values from 08:00 until noon were more negative in the dry period than in the wet period (Fig. 5a). It thus suggests an ecophysiological response of vegetation to higher availability of incoming radiation in the dry period (da Rocha et al., 2004, 2009). In addition, S values were also more negative in the dry

Fig. 7. Nighttime averages of the change in CO2 storage in the canopy air space (S) on the plateau (P), slope (S) and valley (V) for different time intervals and u* classes in the wet (a–c) and dry (d–f) periods of the year 2006. The S and V values correspond to the average (SE) of slope areas (S1, S2) and valley areas (V1, V2), respectively. u* is presented in m s1 and time as local time.


234

period than in the wet period (Fig. 5b). However, this difference in S is very likely the result of larger nocturnal accumulation of CO2 in the dry period than in the wet period (Section 5.1.1; de Arau´jo et al., 2008a). Either it takes longer to release the CO2 stored below the canopy on the plateau in the dry period than in the wet period, or there is some CO2 entering the storage on the plateau laterally from the slopes and valleys during this period (de Arau´jo et al., 2002, 2008a; de Arau´jo, 2009). Therefore, the combination of larger Fc and S values in the dry period than in the wet period caused NEE minimum to both occur earlier and be larger in magnitude in the dry period than in the wet period. This was not the case in the wet period, when NEE minimum was influenced solely by Fc (as S was about 0.8 mmol m2 s1). The difference in times in which NEE started to become strongly less negative (i.e., less carbon assimilation) varied between the wet (at about 13:30) and dry (at about 11:00) periods (Fig. 5c). We could say that this difference suggests an ecophysiological response of vegetation to dryness.

Fig. 7 shows that on the plateau and under different classes of u*, nighttime average values of S do not differ significantly from that of the first and second halves of the night both in the wet and dry periods. This was not the case on the slopes and in the valleys, where S was larger in the first half of the night than both in the second half of the night and the nighttime average values of S in the wet and dry periods. In addition, the absolute differences between the first and second halves of the night were smaller in the dry period than in the wet period, most likely due to larger Rsoil in the dry period. Nevertheless, as Rsoil is lower in the wet period on the slopes and valleys than on the plateaus, the observed larger S fluxes in the first half of the night on the slopes and valleys than on the plateaus in this period provides strong evidence that katabatic flow occurs at this site. Yet S also increased at lower u* values as shown elsewhere (Goulden et al., 1996; de Arau´jo et al., 2002; Saleska et al., 2003; Miller et al., 2004; Hutyra et al., 2008; van Gorsel et al., 2007). 5.3. Interpretation of nighttime NEE

5.2. Nighttime S and vertical mixing In Sections 4.2 and 5.1.2, it was observed that nighttime average curves of S peaked at about 18:00–20:00 on the slopes and in the valleys, following a continuous decrease after this period until reaching a minimum at about 06:00–06:30 (Fig. 4). This was not observed on the plateau during the present study. However, a close inspection of the data presented by de Arau´jo et al. (2002, Fig. 16a) reveals a similar pattern to that on the slopes and in the valleys for this plateau during windy nights. This has also been observed at other plateau forests in central Amazonia (Malhi et al., 1998, Fig. 5a; Hutyra et al., 2008, Fig. 6). In addition, van Gorsel et al. (2007) have also observed the same pattern in a Eucalyptus forest over complex terrain in southeast Australia. They demonstrated that differences in S between the first and second halves of the night were related to suppression of turbulence, stability of the flow regime within the canopy and gravity currents that drain CO2 enriched air away.

As S fluxes were measured within typical ‘flux footprint’ area (de Arau´jo et al., 2002), it is quite likely that the high S fluxes measured on the slopes and in the valleys during nighttime hours have had contribution of CO2 that drains laterally from the plateau. Hence, the use of S fluxes measured on the slopes and in the valleys may increase the estimated nighttime NEE on the plateau. In this and previous studies (Malhi et al., 1998; de Arau´jo et al., 2002; Kruijt et al., 2004), storage corrections were only calculated from local profiles, disregarding the variability of S between plateau, slope and valley. To account for a variable S, as derived from the data in this study, it should be weighted with a footprint model and then spatially averaged. However, applying a footprint model during nighttime conditions in the Amazon forest provides unrealistic values of the source area, e.g., ‘the infinite fetch’ (unpublished data). In Section 5.1.3, it was observed that nighttime average values of NEE underestimate independent measurements of Reco at this

Table 1 Average values of CO2 fluxes from plateau (P), slope (S) and valley (V) in central Amazonia for wet and dry periods in 2006. The average values of CO2 fluxes are presented in mmol m2 s1 and in three different classes of time: full nighttime length (from 17:30 to 06:30), first half of the night (from 17:30 to 24:00), and second half of the night (from 00:30 to 06:30). The u* is presented in m s1. Period

Wet

Dry

Threshold

Time interval

Fc

S

P

P

S

V

Fc + SP

Fc + SP,S,V

NEE Difference

u* > 0

17:30–06:30 17:30–24:00 00:30–06:30

3.34 2.29 4.50

1.74 1.97 1.48

1.85 2.80 0.91

2.40 3.77 1.07

5.08 4.26 5.98

6.58 7.02 6.28

29.59 64.73 5.09

u* 0.2

17:30–06:30 17:30–24:00 00:30–06:30

2.52 1.60 3.50

2.01 2.06 1.95

2.28 3.17 1.42

2.84 4.19 1.58

4.53 3.66 5.45

6.39 6.84 6.08

41.18 86.66 11.57

u* 0.15

17:30–06:30 17:30–24:00 00:30–06:30

2.09 1.15 3.06

2.15 2.02 2.29

2.51 3.43 1.66

3.08 4.39 1.90

4.24 3.17 5.35

6.30 6.65 6.12

48.56 109.55 14.24

u* > 0

17:30–06:30 17:30–24:00 00:30–06:30

1.96 1.69 2.28

2.60 2.60 2.60

3.38 4.00 2.67

3.12 3.91 2.37

4.56 4.29 4.88

6.90 7.47 6.34

51.40 74.34 30.00

u* 0.2

17:30–06:30 17:30–24:00 00:30–06:30

1.06 1.03 1.10

2.82 2.65 3.03

3.80 4.01 3.58

3.97 4.92 3.12

3.88 3.68 4.13

6.86 7.44 6.33

76.51 102.47 53.51

u* 0.15

17:30–06:30 17:30–24:00 00:30–06:30

0.78 0.79 0.77

2.77 2.53 3.03

3.68 4.29 3.09

4.05 4.97 3.30

3.55 3.32 3.80

6.53 7.36 5.82

83.71 121.70 53.14

S for slope (S) and valley (V) represent the average for slopes (S1 and S2) and valleys (V1 and V2). S for plateau, slope and valley (SP,S,V) were weighted according to the proportion of plateaus (about 40%) and valleys (60%) in a radius of about 1 km within the K34 tower footprint as in de Arau´jo et al. (2002). The absolute difference between the terms (Fc + SP) and (Fc + SP,S,V) in relation to (Fc + SP) was used to calculate the difference (%). Values in bold highlight the best estimates of micrometeorological measurements against independent measurements of Reco (about 7.8 mmol m2 s1).


site. In addition, it was also observed that S fluxes were larger in the first half of the night on the slopes and valleys than on the plateaus (Section 5.2). We decided therefore to compare nighttime NEE derived from Fc and S measured only on the plateau with that derived from Fc measured on the plateau and S measured on the plateau, slope and valley for the same periods as in Fig. 7. The NEE (Fc + SP) derived from measurements only on the plateau clearly do not match that of independent measurements even in the first half of the night (Table 1), as one would expect according to van Gorsel et al. (2007, 2009). This was also observed by van Gorsel et al. (2009) for a tropical forest site in French Guyana. However, when S measured on the plateau, slope and valley were included, NEE (Fc + SP,S,V) was much closer to independent estimates of Reco, mainly in the first half of the night (bold values in Table 1). Specifically for low u* values (e.g.,