Impacts of experimentally imposed drought on leaf respiration and morphology in an Amazon rain forest

Authors


*Correspondence author. E-mail: daniel.metcalfe@ouce.ox.ac.uk

Summary

1. The Amazon region may experience increasing moisture limitation over this century. Leaf dark respiration (R) is a key component of the Amazon rain forest carbon (C) cycle, but relatively little is known about its sensitivity to drought.

2. Here, we present measurements of R standardized to 25 °C and leaf morphology from different canopy heights over 5 years at a rain forest subject to a large-scale through-fall reduction (TFR) experiment, and nearby, unmodified Control forest, at the Caxiuanã reserve in the eastern Amazon.

3. In all five post-treatment measurement campaigns, mean R at 25 °C was elevated in the TFR forest compared to the Control forest experiencing normal rainfall. After 5 years of the TFR treatment, R per unit leaf area and mass had increased by 65% and 42%, respectively, relative to pre-treatment means. In contrast, leaf area index (L) in the TFR forest was consistently lower than the Control, falling by 23% compared to the pre-treatment mean, largely because of a decline in specific leaf area (S).

4. The consistent and significant effects of the TFR treatment on R, L and S suggest that severe drought events in the Amazon, of the kind that may occur more frequently in future, could cause a substantial increase in canopy carbon dioxide emissions from this ecosystem to the atmosphere.

Introduction

Leaf dark respiration (R) of carbon dioxide (CO2) is a key component of the Amazon rain forest ecosystem carbon (C) cycle but remains poorly understood and rarely measured, compared to other ecosystem fluxes such as photosynthesis and soil CO2 efflux (Malhi, Baldocchi & Jarvis 1999; Chambers et al. 2004b; Meir et al. 2008). This lack of knowledge impedes attempts to predict the impacts of current and future environmental change upon C cycling in the Amazon rain forest.

In the case of the Amazon, of particular interest is the effect of water availability upon R because the region may experience increasingly frequent and severe drought events associated with global climate change, fire and deforestation over the next 100 years (Werth & Avissar 2002; Christensen et al. 2007; Cox et al. 2008; Harris, Huntingford & Cox 2008; Malhi et al. 2008). Previous periods of drought during El Niño events have appeared to cause a shift in regional scale C exchange across the entire Amazon forest from a net C sink to a source of up to 1·5 × 109 t C year−1 (Roedenbeck et al. 2003; Zeng, Mariotti & Wetzel 2005). Models that simulate the interactions between forest and atmosphere have been able to approximate this inter-annual pattern of regional C fluxes by simulating a simultaneous decline in C uptake via photosynthesis and a rise in ecosystem respiration during drier and warmer years (Tian et al. 1998; Peylin et al. 2005; Zeng, Mariotti & Wetzel 2005). Several studies have attempted to experimentally corroborate these model predictions. While there is reasonable supporting evidence for a drought-induced reduction in photosynthesis (Chaves & Oliveira 2004; Flexas et al. 2004; Fisher et al. 2007), most available data on ecosystem respiration suggest that it decreases under dry conditions (Saleska et al. 2003; Vourlitis et al. 2005; Hutyra et al. 2007; Meir et al. 2008). Much of this drought-induced inhibition may be attributable to a decline in the largest single component of ecosystem respiration-soil CO2 efflux (Davidson et al. 2000; Schwendenmann et al. 2003; Sotta et al. 2004, 2007; Metcalfe et al. 2007).

However, ecosystem respiration is a composite flux derived from not only soil CO2 efflux but also R and other sources, which may each respond to environmental change in different ways. Foliar C emissions per unit ground area are the integrated product of R, leaf area index (L) and the vertical distribution of R and L through the forest canopy. L, in turn, is the product of foliar biomass per unit ground area (M) and specific leaf area (S). All of these parameters are potentially sensitive to changes in water availability (Nepstad et al. 2002; Hanson & Wullschleger 2003; Miranda et al. 2005; Wright et al. 2006; Fisher et al. 2007; Myneni et al. 2007; Brando et al. 2008). There currently exists relatively little information about the individual responses of R, S, L and M to drought in the Amazon and even fewer data on the net effect of drought-induced changes in these parameters upon stand-scale foliar C emissions. Thus, it remains unclear to what extent measured responses of Amazon forest soil CO2 efflux may be offset or accentuated by simultaneous shifts in foliar C emissions.

The overall objective of this study, therefore, was to assess the sensitivity to drought of R, S, L and M at an eastern Amazon rain forest site. The impact of sustained drought was estimated by comparing measurements made in a 1-ha plot where c. 50% of incident rainfall had been excluded (through-fall reduction or TFR plot) to an adjacent, similar but unmodified, Control plot. While the TFR treatment was not replicated (Hurlbert 1984, 2004) because of logistical and financial constraints, it did provide insights into ecosystem processes that would otherwise have been impossible to capture in smaller scale, more easily replicated experiments (Reviews: Carpenter 1996; Sullivan 1997; Osmond et al. 2004; Stokstad 2005; Field studies: Nepstad et al. 2002; Davidson, Ishida & Nepstad 2004; Fisher et al. 2007; Metcalfe et al. 2007; Sotta et al. 2007; Brando et al. 2008). Specifically, we used a ‘before-after-control-impact’ (Underwood 1997; Rasmussen et al. 2001; Gotelli & Ellison 2004) approach to test for significant shifts in R and S both (i) Over time- before and after the imposition of the TFR treatment and between dry and wet seasons, and (ii) Between the TFR treatment and Control for each individual measurement campaign. Finally, we use existing L data to upscale leaf-level R measurements to derive plot estimates of foliar night-time C effluxes.

Materials and methods

Study site

The experimental site is located in the Caxiuanã National Forest, Pará State, north-eastern Brazil (1°43′3·5″S, 51°27′36″W). The forest is a lowland terra firme rain forest situated on a level plain 10–15 m above river water level, with a high annual rainfall (∼2500 mm) and a pronounced seasonality in leaf fall which peaks during the strong dry season (see Table 1 for additional plot details). Plant species diversity is high at around 100 species per hectare, of which over half are Sapotaceae, Fabaceae, Violaceae and Chrysobalanacae, and less than 1% are lianas. Mean annual air temperature is c. 25 °C and the diurnal variation is typically less than 3 °C. The most widespread soil type is a highly weathered yellow Oxisol (US Department of Agriculture soil taxonomy). In January 2002, a 1-ha area of forest was modified with the installation of plastic panels at 2 m height to exclude c. 50% of incident rainfall (TFR plot). This reduction in rainfall is similar to a key early long-term climate prediction for the region (Cox et al. 2000). The perimeter of the TFR plot was trenched to a mean depth of 1 m and lined with plastic sheeting to minimize lateral flow of water into the site. Intercepted water was channelled away to a run-off area 50 m away from the plot. An adjacent 1-ha Control plot with similar topography, soil type and vegetation structure (Fisher et al. 2007) was used to assess natural patterns of L, S and R, in the absence of any TFR treatment. Supplementary measurements during the first 3 years of the TFR treatment demonstrated that soil water potential, tree stem sapflow, stomatal conductance and photosynthesis were all substantially reduced in the TFR plot compared to the Control, particularly during the dry season (Fisher et al. 2007). At the beginning of the experiment in January 2001, 30 m tall canopy access towers were installed near the centre of both plots. All measurements were taken at least 20 m inside the perimeter of each plot to minimize edge effects.

Table 1.   Key vegetation and soil features for each plot surveyed
Plot characteristicsControlTFR
  1. TFR, through-fall reduction. Tree number and basal area represents all individuals over 10 cm diameter at breast height, measured in January 2005. Soil values are collated from data in Sotta et al. (2007).

Vegetation
 Tree number ha−1434421
 Stem basal area (m2 ha−1)23·924·0
 Tree species ha−1118113
Soil 0–10 cm
 Clay content (%)1813
 Silt content (%)54
 Sand content (%)7783
 pH44
 Carbon content (g kg−1)912
 Nitrogen content (g kg−1)0·40·3
 Phosphorus content (mg dm−3)33
 Carbon : nitrogen ratio2335
 Soil cation exchange (cmol dm−3)0·80·7

Measurement of leaf dark respiration and specific leaf area

R and S from trees on both plots were recorded on six occasions between November 2001 and January 2007: once before and five times after imposition of the TFR treatment. L data are also available from the same periods on both plots (Fisher et al. 2007). All measurement campaigns sampled fully expanded, non-senescent, un-diseased leaves, and recorded additional information about the height and tree species of the sampled leaves. Thus, R measurements from these leaves should primarily reflect ‘maintenance’ respiration rather than ‘growth’ respiration associated with metabolic costs of constructing new plant tissue (McCree 1970). All leaves were sampled during the daytime (08.30–15.00 h) and kept in the dark until CO2 gas exchange had stabilized (usually after 5–10 min) before R at ambient air CO2 concentration (360–380 p.p.m.) and humidity (60–80%) was recorded, thus minimizing biases potentially introduced by light-enhanced dark respiration and the photorespiratory post-illumination burst (Atkin, Evans & Siebke 1998).

The first five measurement campaigns (conducted between November 2001 and 2003) used the following methodology: 17–26 leaves from nine trees, and 18–26 leaves from eight trees were sampled around the canopy access towers on the Control and TFR plots respectively (Table 2). Measurements were taken from the same trees and from leaves at the same canopy heights in each measurement campaign. R was measured in situ from un-excised leaves with an infra-red gas analyser (IRGA) connected to a leaf measurement cuvette (LI-COR 6400 portable photosynthesis system with 6400-02B leaf cuvette; Lincoln, NE, USA). Leaf discs of a known area were cut from leaves on the same branch as leaves used for R measurement, dried at 70 °C until constant mass and weighed. S was calculated for each leaf disc sampled by dividing dry mass by one-sided area.

Table 2.   Tree species sampled on the plots
ControlTFR
  1. TFR, through-fall reduction. An individual representative of each species (two individuals of Quiina florida on the Control plot) was repeatedly sampled between November 2001 and 2003. In the final measurement campaign (January 2007) trees sampled were not identified to species level.

Duguetia echinophoraDuguetia echinophora
Hasseltia floribundaHirtela bicornis
Licania heteromorphaLecythis confertiflora
Manilkara bidentataLicaria armeniaca
Mezilaurus lindawianaLicania canescens
Pouteria laterifloraManilkara paraensis
Protium heptaphyllumMouriri duckeana
Quiina floridaSwartzia racemosa

The final measurement campaign (in January 2007) sampled a total of 33 and 28 individual leaves from the Control and TFR plots, respectively, from 15 separate trees per plot. Of these trees sampled, 10 were randomly selected as the closest tree to every 10 m intersection point along two 40 m long transects in the centre of the plot. A branch from the outer canopy of each selected tree was excised at between one and three different canopy heights. No attempt was made to cut and re-cut branches under water because this would not have guaranteed that gas exchange remained unaltered (Santiago & Mulkey 2003). Instead, we designed an experiment to quantify and, if necessary, correct for any impacts of branch excision (see text in Methods section). To facilitate sampling of leaves higher up in the emergent canopy and to replicate measurements on individual trees made earlier, an additional five trees per plot were similarly sampled around the canopy access towers on each plot. R was measured for most leaves within 3 h of branch excision, using an IRGA connected to a leaf measurement cuvette (CIRAS-1 IRGA with PLC6 leaf cuvette; PP Systems, Hitchen, UK). The interval of time between branch excision and R measurement was noted for each leaf sampled. There was no significant difference in the mean time between excision and R measurement on the plots.

After measurement, the same leaves were photographed to calculate leaf area with digital image analysis, and then dried at 70 °C until constant mass and weighed. S was calculated for each entire leaf sampled (including petioles) by dividing dry mass by one-sided area.

Measurements made with the LI-COR 6400 IRGA maintained a flow rate of 500 μmol s−1, with a mean ± standard error (SE) difference between Cr and Cs of 0·52 ± 0·07 p.p.m. The CIRAS-1 IRGA was set to a lower flow rate of 200 μmol s−1, and consequently the observed mean ± SE CO2 difference was 0·91 ± 0·06 p.p.m. The inward diffusion of respired CO2 from leaf material clamped under the cuvette gasket (Pons & Welschen 2002) and the diluting effect of water vapour produced by the leaf was corrected for.

Leaf temperatures recorded automatically by the IRGA systems during R measurement varied between 22 and 31 °C. Species-specific R temperature response functions were not available for all of the trees sampled, so measurements were standardized to a reference temperature of 25 °C (R25) with the following formula that describes the average R temperature response across 116 terrestrial plant species (Atkin & Tjoelker 2003; Atkin, Bruhn & Tjoelker 2005):

image

where Ra is R recorded at ambient temperature (Ta).

To investigate the potential confounding influences of branch excision on the R values recorded in January 2007, the following experiment was devised. An un-excised leaf was placed within the IRGA cuvette and R was measured every minute for 1 h. After this period, the branch attached to the leaf within the cuvette was excised, but R measurement was continued at the same temporal frequency for 5 h, to observe whether there was any change in R with time since branch excision. Over this period, the sensor was regularly automatically calibrated with air passed through a molecular sieve to remove all CO2. Before and during measurements the molecular sieve was frequently checked to ensure that it was not exhausted. This procedure was repeated three times, on consecutive days from three individual leaves each on separate trees of different species. All leaves sampled showed no change in R over the hour prior to excision, but after excision R rose gradually over time, approximately doubling after 5 h compared to the pre-excision mean value (data not shown). A third-order polynomial model was fitted to the mean trend of R over time since branch excision (R2 = 0·77). This model was not chosen as a realistic mechanistic simulation of plant gas exchange, but purely for limited predictive purposes over the duration of the measurements because it provided the best fit to the data. This equation, together with data collected on the interval of time between branch excision and R measurements for each leaf, was used to correct for the confounding effect of excision and storage on January 2007 measurements by calculating R at time since excision = 0 for each leaf sampled. No immediate effect of excision itself on R was apparent.

Measurement of leaf area index and foliage mass

Mean plot L estimated during the first five measurement campaigns (conducted between November 2001 and 2003) is presented in Fisher et al. (2007). These data were derived from canopy images captured at 100 points per plot with LAI-2000 plant canopy analysers (LI-COR Inc.). For this study, additional L data were collected in January 2007 based upon canopy images per plot collected at 25 locations along a with a digital camera and fish-eye lens (Nikon Coolpix 900; Nikon Corporation, Melville,USA) and subsequently analysed with digital image analysis software (Hemiview 2.1 SR1; Delta-T Devices Ltd, Cambridge, UK). All images of the canopy were recorded in the early morning or late afternoon, during periods of fully diffuse incoming radiation along a regular grid within both plots (following the methodology of Aragão et al. (2005). The distribution of L with height above the ground on both plots was estimated once – in November 2001 – by recording L with the LAI-2000 plant canopy analysers (LI-COR Inc.) every 2 m up each of the plot canopy access towers. Plot-level M was estimated for each measurement campaign by multiplying mean plot leaf mass per unit leaf area (1/S) by L.

Estimating stand-scale night-time foliar carbon efflux

To illustrate how R, S and L interact, and to facilitate direct comparison of our leaf-level R measurements with other ecosystem C fluxes, we derived approximate estimates of stand-scale foliar C efflux. To do so, we calculated mean ± 95% confidence intervals of L and R per unit area separately for three canopy height layers (≤10, 11–20, ≥21 m). Night-time foliar C emissions per unit ground area were estimated for each canopy layer as the product of R per unit leaf area multiplied by L. Given the low temporal frequency of our direct measurements and the focus on between-plot (rather than seasonal or annual) differences we opted for the relatively simple, more transparent, up-scaling approach of assuming constant night-time air temperature of 25 °C and 12 h of dark conditions each day throughout the year. For the purposes of this analysis, we also assumed that dark-equilibrated R recorded during the day in this study was representative of night-time leaf respiration (Chambers et al. 2009; but see Hubbard, Ryan & Lukens 1995). Means from each canopy layer were summed to derive total plot estimates. Where necessary, 95% confidence intervals were propagated by quadrature of absolute errors for addition and subtraction, and quadrature of relative errors for multiplication and division (Mood, Graybill & Boes 1974; Cavaleri, Oberhauer & Ryan 2008). This assumes that errors are independent and normally distributed.

Data analysis

To assess the impact of the TFR treatment on S and R the following two statistical analyses were performed. (i) Within-plot change over time since the imposition of the TFR treatment was quantified with a repeated-measures analysis of variance (RM-ANOVA). Data from the final measurement campaign were not included in the RM-ANOVA because a different set of trees were sampled with a different methodology, whereas the previous five campaigns repeatedly sampled leaves from the same trees and same canopy heights. To examine specifically which time periods differed from each other in terms of S and R, pairwise comparisons between measurement campaigns were conducted within the RM-ANOVA analysis. (ii) Between-plot differences in R and S over all measurement campaigns were quantified with a Generalized Linear Model (GLM) with plot as a fixed-effects factor, and leaf height, tree family and sampling time specified as random-effects factors to control for the potentially confounding effect of sampling differences between plots. Using this method, plot differences were examined both for all data at each measurement campaign, and for all data in different canopy height categories (≤10, 11–20, ≥21 m). In addition, the links between leaf height, R and S were assessed with a Spearman’s Rank Correlation. Statistical analyses were carried out with SPSS 14·0 for Windows (SPSS Inc., Chicago, IL, USA). Key outputs of the analyses were an F-statistic (for the RM-ANOVA and GLM), a correlation coefficient (r, for the Spearman’s Rank Correlation analysis) and a significance P-value for all tests. Data were transformed with a natural logarithm, where necessary, to conform to the assumptions of parametric analysis.

Results

Plot trends in stand-scale night-time foliar carbon efflux

R per unit area together with L data, apportioned into canopy height categories (≤10, 11–20, ≥21 m), were used to estimate total ± 95% confidence intervals foliar night-time C emissions prior to the TFR treatment of 3·4 ±0·1 and 3·4 ±0·3 t ha−1 year−1 on the Control and TFR plots respectively (Table 3). According to these estimates, elevated R on the TFR plot relative to the Control was not fully offset by the drought-associated decline in L, such that night-time foliar C emissions on the TFR plot were 0·7 ± 0·4 and 1·8 ± 0·9 t C ha−1 year−1 greater than the Control plot based upon measurements in November 2003 and January 2007 respectively (Table 3).

Table 3.   Plot estimates of night-time stand-scale foliar C efflux (t C ha−1 year−1)
 November 2001May 2002November 2002May 2003November 2003January 2007
  1. TFR, through-fall reduction. Values are estimated from instantaneous measurements at each date extrapolated over a year assuming constant air temperature of 25 °C and 12 h of dark conditions every night.

Control3·4 ± 0·14·0 ± 0·44·2 ± 1·93·5 ± 0·64·6 ± 0·24·1 ± 0·4
TFR3·4 ± 0·34·2 ± 0·43·1 ± 0·24·8 ± 1·15·3 ± 0·35·9 ± 0·8
Difference TFR − Control0·0 ± 0·40·2 ± 0·6−1·1 ± 1·91·4 ± 1·20·7 ± 0·41·8 ± 0·9

Plot trends in leaf dark respiration

Plot mean R values were comparable to existing data from other studies in the Amazon forest (Table 4). In the Control plot, there was no significant overall change in R per unit area (RM-ANOVA, d.f. = 4, = 0·77, = 0·56) and mass (RM-ANOVA, d.f. = 4, = 1·78, = 0·17) between November 2001 and 2003 (Table 5, Fig. 1c,d). In contrast, over the same period on the TFR plot, R per unit area (RM-ANOVA, d.f. = 4, = 5·24, = 0·004) and mass (RM-ANOVA, d.f. = 4, = 5·14, = 0·004) increased significantly (Table 5, Fig. 1c,d). Specifically, while R per unit area on the TFR plot began to rise immediately after imposition of the treatment (Table 5, Fig. 1d), it only became significantly higher compared to pre-treatment values during the peaks of the dry season in November 2002 (RM-ANOVA pairwise comparison, = 0·011) and 2003 (RM-ANOVA pairwise comparison, = 0·003). R per unit mass showed a slightly different temporal trend following the TFR treatment (Table 5, Fig. 1c) becoming significantly higher compared to the pre-treatment mean only by November 2003 (RM-ANOVA pairwise comparison, = 0·027).

Table 4.   A summary of available data on respiration per unit area at 25 °C from leaves at primary lowland terra firme rain forest sites in the Amazon
SourceCoordinatesRespiration (μmol m−2 s−1)Notes
  1. TFR, through-fall reduction. Respiration values represent mean ± standard deviation. Respiration data collected in this study in November 2001 and January 2007 used different methods and sampling strategies.

  2. *Measured at 25 °C.

  3. †Corrected to 25 °C using site-specific temperature response curves.

  4. ‡Corrected to 25 °C using a generic temperature response curve (Atkin & Tjoelker 2003; Atkin, Bruhn and Tjoelker 2005).

Reich et al. (1998)1°56′N, 67°03′W0·91 ± 0·23* 
Meir, Grace & Miranda (2001)10°05′S, 61°55′W0·36 ± 0·20† 
Domingues et al. (2005)3°33′S, 55°83′W0·43 ± 0·36‡
0·57 ± 0·39‡
Wet season
Dry season
Miranda et al. (2005)11°25′S, 55°20′W0·33 ± 0·17‡
0·66 ± 0·37‡
Wet season
Dry season
Cavaleri, Oberhauer & Ryan (2008)10°20′N, 83°50′W0·59 ± 0·44† 
This study1°43′3·5″S, 51°27′36″W0·32 ± 0·13‡
0·33 ± 0·17‡
0·41 ± 0·20‡
0·55 ± 0·27‡
Control plot, November 2001
TFR plot, November 2001
Control plot, January 2007
TFR plot, January 2007
Table 5.   Significance P-values for differences in respiration and specific area from branches re-sampled over five successive measurement campaigns in both plots
November 2001May 2002November 2002May 2003November 2003
  1. TFR, through-fall reduction. Values are derived from an RM-ANOVA pairwise comparison. Significant (< 0·05) and very significant (< 0·01) differences are marked with single and double asterisks respectively. Values in bold highlight differences compared to data collected prior to imposition of the TFR treatment (November 2001).

Respiration per unit area
Control plot
 November 2001    
 May 20020·21   
 November 20020·940·08  
 May 20030·280·880·40 
 November 20030·890·440·890·22
TFR plot
 November 2001    
 May 20020·31   
 November 20020·011*0·19  
 May 20030·210·750·28 
 November 20030·003**0·01**0·270·06
Respiration per unit mass
Control plot
 November 2001    
 May 20020·16   
 November 20020·450·13  
 May 20030·310·980·64 
 November 20030·280·130·300·042*
TFR plot
 November 2001    
 May 20020·25   
 November 20020·480·89  
 May 20030·520·038*0·17 
 November 20030·027*0·045*0·080·004**
Specific leaf area
Control plot
 November 2001    
 May 20020·60   
 November 20020·049*0·25  
 May 20030·420·740·63 
 November 20030·230·120·047*0·014*
TFR plot
 November 2001    
 May 20020·51   
 November 20020·240·016*  
 May 20030·070·002**0·039* 
 November 20030·930·310·08 
Figure 1.

 Seasonal trends in site rainfall (a), plot surface soil water content (b), respiration per unit mass (c) and area (d), specific leaf area (e), leaf area index (f) and foliage mass (g) over the study period. Values denote plot means ± 95% confidence intervals. TFR, through-fall reduction. Rainfall is presented both on a daily basis (grey) and as a 15-day running average (black). The vertical dashed arrows in (a) mark leaf sampling times, and in (b)–(g) mark the date of TFR treatment establishment. L data for the measurement campaigns between November 2001 and 2003 are reproduced from Fisher et al. (2007).

Directly comparing plot R estimates from each measurement campaign and controlling for plot sampling differences in tree family, leaf height and sampling time showed that there was no significant plot difference in R per unit area (GLM, d.f. = 1, = 0·30, = 0·68) and mass (GLM, d.f. = 1, = 4·34, = 0·28) prior to imposition of the TFR treatment (Fig. 1c,d). However, after the TFR treatment, measured R on the TFR plot increased (Table 5, Fig. 1c,d) until it became significantly higher than the Control in January 2007 (GLM, d.f. = 1, Per unit mass: = 6·12, = 0·035; Per unit area: = 16·91, = 0·002). This plot difference in R was mainly attributable to a significantly lower rate of R per unit mass (GLM, d.f. = 1, = 28·38, < 0·001) and area (GLM, d.f. = 1, = 45·77, < 0·001) in lower canopy leaves (≤10 m above the ground surface) on the TFR plot compared to the Control (Fig. 2a,b). No significant plot difference was found in upper canopy leaves (≥21 m above the ground surface).

Figure 2.

 Plot differences in respiration per unit mass (a) and area (b), specific leaf area (c) and leaf area index (d) amongst canopy height categories (≤10, 11–20, ≥21 m). TFR, through-fall reduction. Values denote plot means ± 95% confidence intervals from data pooled across all post-treatment sampling sessions, with the exception of the L canopy profile which was measured only once, before the drought treatment in November 2001. ***Significant plot differences (< 0·001).

There was a highly significant positive correlation between leaf height above the ground and R per unit area in the Control plot (Fig. 2b, Spearman’s Rank Correlation, = 147, = 0·48, < 0·001), while on the TFR plot the trend was weaker but still present (Fig. 2b, Spearman’s Rank Correlation, = 141, = 0·16, = 0·054). However, when quantified on a per unit mass basis, R in the Control plot ceased to show any significant pattern of change with leaf height above the ground (Fig. 2a).

Plot trends in specific leaf area

Overall, S showed significant change over time in both the Control (Table 5, Fig. 1e, RM-ANOVA, d.f. = 4, = 3·16, = 0·036) and TFR plots (Table 5, Fig. 1e, RM-ANOVA, d.f. = 4, = 4·34, = 0·009). However, this change reflected bidirectional seasonal and annual variation (Fig. 1e) rather than progressive unidirectional change over time: in the Control plot only mean S measured in November 2002 was significantly different from the pre-treatment mean (Table 5, RM-ANOVA pairwise comparison, = 0·049) while none of the S measurements in the TFR plot diverged significantly from initial values recorded in November 2001 (Table 5). Once variation derived from sampling differences was excluded, no significant difference in S between plots during any individual measurement campaign was found. Neither was there any significant plot difference in mean S in the different canopy height categories (Fig. 2), although on both plots, leaves in the upper canopy tended to have lower S compared to understorey leaves (Fig. 3c, Control plot: Spearman’s Rank Correlation, = 135, = −0·56, < 0·001; TFR plot: Spearman’s Rank Correlation, = 129, = −0·54, < 0·001).

Plot trends in leaf area index

Before imposition of the TFR treatment both plots had similar L (∼5·5 m2 m−2), but after the treatment L on the TFR plot declined, reaching a value of c. 4·5 m2 m−2 (∼1 m2 m−2 or 20% lower than the Control) after almost 2 years of the TFR treatment (Fig. 1f). On both plots, there was an abrupt and substantial decline in L measured during November 2002 which was gradually recovered over subsequent years (Fig. 1f).

According to tower L height profile measurements in November 2001, slightly less L was located in the lower canopy on the TFR plot compared to the Control, such that 65% and 71% of total L occurred above 20 m on the Control and TFR plots respectively (Fig. 3d).

Plot trend in foliage mass

On both plots over all measurement campaigns, M varied between 4 and 5·5 t ha−1 (Fig. 1g). There was no clear difference in M between the plots over the first few years of the TFR treatment, because lower TFR plot L (Fig. 1f) was offset by higher S (Fig. 1e) compared to the Control. This changed by the last measurement campaign (January 2007) when the TFR plot S fell back to levels similar to that of the Control, but L remained relatively low, with the consequence that estimated M was 1·2 ± 0·7 lower on the TFR plot compared to the Control (Fig. 1e–g).

Discussion

Estimating stand-scale night-time foliar carbon efflux

Combining leaf-level R and S data with L, to produce estimates of stand-scale foliar respiration, illustrates how their individual responses interact to determine ecosystem foliar C emissions, and how this differs between plots. The results indicate that the effect of the decline in the amount of respiring leaf tissue in the TFR plot, measured as L, was outweighed by the simultaneous increase in R per unit area. The estimated net consequence of these opposing changes was that night-time foliar C emissions from the TFR plot increased by 1·4 ± 1·2, 0·7 ± 0·4 and 1·8 ± 0·9 t ha−1 year−1 compared to the Control in the last three measurement campaigns 1·4, 1·9 and 5·1 years after imposition of the TFR treatment respectively (Table 3). The three previous measurements (before, 0·4 and 0·9 years after the TFR treatment) showed no clear difference in night-time foliar C emissions between plots (Table 3). By comparison, the largest single ecosystem respiration term – soil CO2 efflux – was estimated to be 4·2 and 0·7 t C ha−1 year−1 lower on the TFR plot relative to the Control in 2003 (Sotta et al. 2007) and 2006 (Metcalfe et al. 2007) respectively. Thus, incorporating the plot differences in night-time foliar C emissions estimated in this study offsets a large portion of the estimated drought-induced reduction in soil CO2 efflux. Given that atmospheric CO2 fertilization of Amazon vegetation is unlikely to stimulate net ecosystem uptake far in excess of 0·5 t C ha−1 year−1 (Phillips et al. 1998; Chambers et al. 2001, 2004a; Baker et al. 2004) we suggest that inclusion, and improved quantification, of night-time foliar C emissions in atmosphere-biosphere models could be vital for accurate prediction of changes in Amazon forest C exchange in response to climate change.

The approach taken in this study to up-scaling leaf-level R measurements makes a number of assumptions which deserve examination. First, it is unclear how confidently R measured during the day (albeit equilibrated to dark conditions) may be extrapolated to night-time conditions (Hubbard, Ryan & Lukens 1995), although at a similar Amazon rain forest type to our study site, R measured from 27 trees showed no clear diurnal variation in R, and no significant difference between night- and day-time values (Chambers et al. 2009). Secondly, our assumption of constant temperature (25 °C) on both plots and 12 h of darkness each day throughout the year is clearly simplistic though any error introduced is minor compared to natural intra- and inter-species variation, and affects both plots similarly. Thus, this method facilitates direct comparison of plots and, in the absence of more detailed process-level data, provides a transparent basis for up-scaling instantaneous measurements.

Drought effects on leaf respiration and morphology

Estimates of R from this study are consistent with previous estimates from other lowland Amazon rain forest ecosystems (Table 4). While there were significant differences in R between the plots after several years of the TFR treatment, these differences were relatively minor compared to global variation amongst studies, sites and biomes (Wright et al. 2006). In this study, evidence for enhancement of R by drought comes from (i) the increase in R over time since imposition of the TFR treatment, (ii) the higher mean R compared to the Control during each individual measurement campaign and (iii) the slight but consistent increase in R on both plots during the dry seasons (Fig. 1, Table 5). While only some of these differences were statistically significant individually, the effect was consistent across all post-treatment measurement campaigns. Leaf nitrogen content did not differ between plots, neither was there a clear correlation between leaf nitrogen content and R (R. Lobo-do-Vale, unpublished data). Other studies in the Amazon have also recorded an increase in R in the dry season compared to the wet season (Table 4). In addition, a comprehensive survey across 208 woody plant species from 20, mainly temperate, sites showed that, for a given S, R was higher at low-rainfall sites compared to higher rainfall sites (Wright et al. 2006; but see Flexas et al. 2005; Atkin & Macherel 2008). The change in R in all these studies could either reflect a shift in R for a given S, or altered S with little concomitant change in R, or some combination of these two processes. Potential mechanisms for enhanced R at a given S under moisture stress include: increased energy demand for the maintenance of vacuolar solute gradients, repair of water-stress-induced cell damage and/or increased wastage respiration via futile cycles (Hue 1982; Lambers 1997; Lambers, Chapin & Pons 1998; Cannell & Thornley 2000; Flexas et al. 2005; Würth et al. 2005; Wright et al. 2006; Atkin & Macherel 2009). Determining which of these processes dominate will be important for modelling the pattern and magnitude of change in R across the Amazon in the face of future climate changes.

The reduction in L on the TFR plot (Fig. 1f) was consistent with existing data on Amazon forest responses to soil moisture deficit (Nepstad et al. 2002; Fisher et al. 2007; Myneni et al. 2007) and over the first 3 years of the TFR treatment was largely caused by declining S, because total plot M remained remarkably similar between the plots despite substantial seasonal variation (Fig. 1g). In January 2007, lower L on the TFR plot could not be attributed to S and was therefore most likely a product of an imbalance between leaf growth and shedding (Fig. 1g). Changes in S may reflect adaptation to drought on the part of TFR plot trees by developing thicker and/or denser leaves (Witkowski & Lamont 1991). The abrupt drop in L during November 2002 followed by a sharp rise in M (Fig. 1f,g) on both plots was also attributable to a decline in S. The reason for this change is not known. The meteorological conditions during this period do not appear to have been anomalous while the equipment and sampling strategy used to quantify L and S remained the same over this period. Taken together this suggests that the decline in L and S over this period was a real biological pattern perhaps linked to seasonal phenology rather than an artefact of methodology.

The pattern of change in R through the canopy was strikingly different between plots: with significantly higher rates of R (Fig. 2a,b) in foliage below 20 m on the TFR plot compared to the Control. It is unlikely that the infrastructure of the TFR plot itself could account for these differences because the panels diverting rainfall were installed to a maximum height of 2 m, while all R measurements were recorded in tree canopies above this height. We suggest that the plot differences in the pattern of change in R through the canopy more likely reflect the fact that most of the lower leaves sampled came from smaller stature trees with shallower root systems, which were likely to suffer more from surface soil moisture limitation.

Conclusion

This study evaluated the drought sensitivity of R at an eastern Amazon rain forest site. Partial rainfall exclusion of a 1-ha area of rain forest was associated with an estimated increase in night-time foliar C emissions of 1·4, 0·7 and 1·8 t ha−1 year−1 compared to forest on a nearby Control plot 1·4, 1·9 and 5·1 years after rainfall exclusion respectively. This drought-induced physiological shift, if shown to occur more widely, might be sufficient to offset current estimates of the Amazon forest C sink, and alter model predictions of future changes in net C emissions from the Amazon basin. To build upon the key conclusions of this study more measurements are required to improve our understanding of the spatial and temporal variation in R, and of leaf respiration under light conditions.

Acknowledgements

This research contributes to the Brazil-led Large Scale Biosphere – Atmosphere Experiment in Amazonia. Fieldwork was supported by an Edinburgh University Small Project Grant. The authors thank Leonardo Sá and Ima Vieira for their scientific support and collaboration, and the Museu Paraense Emilio Goeldi for the use of its field station and laboratory facilities.

Ancillary

Advertisement