Introduction

Understanding responses of tropical trees to climate change and increasing levels of atmospheric [CO2] is important as tropical forests process large amounts or carbon and water through photosynthesis and transpiration (Malhi and Grace 2000). Thus, small changes in growth rates or water use efficiencies of tropical trees affect the carbon and water cycles, and the rate of climate change itself (Betts et al. 2004; Henderson-Sellers et al. 1995). A powerful way of obtaining insight into the response of trees to climate change and [CO2] is by use of tree rings (Fritts 1976). Tree rings not only record historical growth rates, but also provide an archive of stable isotope ratios in tree ring cellulose over the life-time of the tree (McCarroll and Loader 2004). Stable carbon isotope ratios (δ13C) in tree rings are the result of discrimination against the heavier 13CO2 during carboxylation and diffusion through the stomata, which are linearly related to the ratio of intercellular and atmospheric [CO2] (c i/c a) (Francey and Farquhar 1982). This ratio is driven by the demand of CO2 for photosynthesis (A) and supply through stomatal conductance (g s). The carbon isotope signal of plant matter relative to atmospheric δ13CO2 (Δ13C) is therefore interpreted to reflect the balance between photosynthesis and stomatal conductance (Francey and Farquhar 1982).

Water use efficiency is the ratio of A to transpiration (E), and is a measure of the amount of water used per carbon gain. Δ13C is related to the ratio of A/g s and termed intrinsic water use efficiency (W i), intrinsic because it assumes constant evaporative demand (Ehleringer et al. 1993; Osmond et al. 1980). By taking into account changes in the leaf-to-air water vapour pressure over time, we may adjust for losses through changes in the evaporative demand and obtain a measure of true changes in water use efficiency (A/E) (Ehleringer et al. 1993; Seibt et al. 2008). Changes in water use efficiency are very important as changes in the rate of assimilation affect the uptake of [CO2] by trees from the atmosphere, while changes in stomatal conductance and transpiration rates may have large consequences for the hydrological cycle (Betts et al. 2004; Henderson-Sellers et al. 1995).

In general, plants tend to reduce stomatal conductance (g s) and increase their assimilation (A) in response to increased [CO2]. Thus, their intrinsic water use efficiency (A/g s) increases (Ehleringer et al. 1993; Farquhar et al. 1989), enabling them to absorb the same amount of carbon for less water loss. Such increases in W i have been observed in short-term experiments of tree responses to elevated [CO2] (Norby et al. 1999 and references therein), and over long-time periods using records of δ13C in tree rings that reflect the global increase in atmospheric [CO2] (Feng 1999; Waterhouse et al. 2004). An increase in W i in response to increasing [CO2] since the industrial revolution has been found in nearly all temperate trees studied (Feng 1999; Saurer et al. 2004; Waterhouse et al. 2004), while for the tropics very few records of long-term W i exist (Hietz et al. 2005; Nock et al. 2010). Studies on carbon isotope ratios over longer time scales in tropical trees are particularly important as the degree to which tropical forests have responded to increasing [CO2] and acted as carbon sinks over the last century is still a topic of heated scientific debate (Körner 2003; Lewis et al. 2009; Lloyd and Farquhar 2008). Decadal scale inventory studies show an increase in biomass of tropical forests, thought to be due to CO2 fertilisation (Lewis et al. 2009; Lloyd and Farquhar 2008). However, others do not find such growth increases, and argue that CO2 fertilisation does not affect the growth of tropical forest (Clark et al. 2010; Körner 2003). Predicted changes in the hydrological cycle include increases in soil humidity and runoff, and reductions in rainfall (Betts et al. 2004; Gedney et al. 2006; Henderson-Sellers et al. 1995). Assessing the magnitude of physiological responses of trees to increasing [CO2] is important to evaluate the potential impact of increasing [CO2] on carbon and hydrological cycles of tropical forests.

Stable carbon isotopes have been applied widely in temperate trees (McCarroll and Loader 2004 and references therein), whereas applications in tropical trees are scarce. Some have used carbon isotopes to detect annual cycles in ringless species (Pons and Helle 2010 and references therein), but very few studies analysed stable isotope ratios in annual rings over longer timescales (Cullen et al. 2008; Fichtler et al. 2010; Gebrekirstos et al. 2009). Here, we analysed carbon isotope ratios (δ13C) over the last 40 years in tree rings of Mimosa acantholoba, a tropical dry forest pioneer species from southern Mexico. This species forms annual rings and showed strong growth responses to inter-annual variation in rainfall (Brienen et al. 2010). Annual records of δ13C in tree rings are used to calculate intrinsic water use efficiency (W i,) and intercellular [CO2], c i. Specific questions addressed in this study are: (1) to what degree did c i, and thus W i, change in response to increasing atmospheric [CO2] over the last four decades? (2) Are long-term changes in W i reflected in long-term changes in growth rates? (3) Is inter-annual variation in c i reflected in variation in rainfall, temperature, cloud cover, and incoming solar radiation?

Methods

Study area and climate

The study area is located on the Pacific slope of the Isthmus of Tehuantepec, close to the village of Nizanda in the state of Oaxaca, South Mexico (16°39′N, 95°00′W). The natural vegetation consists of tropical deciduous dry forest (Brienen et al. 2010). Mean annual temperature in the study region is 26°C and total annual rainfall is ~930 mm. Rainfall is highly seasonal with a pronounced dry season from November until May (<50 mm/month), and a wet season from June until October that accounts for 90% of the annual rainfall.

Variation in annual rainfall is high, varying fivefold over the period 1968–2007, from 380–1,850 mm. The principal drivers behind this variation in rainfall are sea surface temperature anomalies in the Pacific and Atlantic Oceans. Effects of El Niño–Southern Oscillation (ENSO) on climate in the region are particularly pronounced with reduced rainfall during El Niño years and higher temperatures and solar radiation during the wet season (Brienen et al. 2010).

Study species

Mimosa acantholoba (Willd.) Poir. (Fabaceae) is a common dry-forest pioneer tree that reaches up to 7 m in height, ca. 20 cm in diameter and has a maximum age of ca. 40 years. In the study area, M. acantholoba is one of the first pioneers to colonize abandoned agricultural field and often forms mono-dominant stands (Lebrija-Trejos et al. 2008). M. acantholoba is strictly deciduous, shedding leaves at the end of the wet season (November–December) and forming new leaves usually after the first rains (May–June).

Mimosa acantholoba forms distinct annual rings characterized by a higher density of vessels and larger vessel size at the beginning of each growth zone (i.e., semi-ring porous growth zones; Brienen et al. 2009).

Isotope analysis

We selected five stem discs of trees between 28 and 40 years old. These discs were collected in February 2008 from three different forest stands that were abandoned between 40 and 64 years ago, and used previously to study secondary forest succession (Brienen et al. 2009) and climate–growth relations of this species (Brienen et al. 2010). Wood samples were isolated from exactly dated annual rings for each of the five trees. This was done manually along a small and thin section of each disc with a sharp knife. To avoid loss of material and possible cross-contamination during grinding, samples were cut into fine pieces by hand with a sharp blade. Approximately 100 mg of each sample was placed in a 2-ml sealable plastic vial and extracted first with 1 ml diglyme + 0.25 ml 10 M HCl and subsequently with 1.5 ml NaClO2–acetic acid (5 g NaClO2 dissolved in 500 ml of distilled water and 0.7 ml of glacial acetic acid) as detailed elsewhere (Hietz et al. 2005). The resulting cellulose was homogenized with a UP200S ultrasonic homogenizer (Hielscher Ultrasonics, Teltow, Germany) and lyophilized (Laumer et al. 2009).

About 1 mg of purified cellulose was weighed into tin capsules and carbon isotope composition (δ13Ccell) measured by gas isotope ratio mass spectrometry (IRMS). The IRMS system consisted of an elemental analyzer (EA 1110, CE Instruments) interfaced by a ConFlo II to the IRMS (DeltaPLUS, Finnigan MAT, Thermo Electron). Reference gas (CO2, Air Liquide) was injected before and after each sample CO2 peak to correct for drift. Laboratory standards were run in between samples and were calibrated against international reference materials (IAEA-CH-6, USGS-40, USGS-41, IAEA-601 and -602). The long-term standard deviation of repeated δ13C measurements of the laboratory standards was 0.1‰. The carbon isotope composition (δ13Ccell) was calculated as follows:

$$ \delta^{ 1 3} {\text{C}}_{\text{cell}} \, \left[ {\permille{\text{ vs}}. \, V {\text{-}}{\text{PDB}}} \right] = \left( {R_{\text{sample}} /R_{\text{standard}} -{ 1}} \right) \times 1 ,000, $$
(1)

where R represents the ratio of 13C/12C of samples and standards.

Isotope discrimination theory: calculation of c i and W i

The magnitude of carbon discrimination, Δ13C, by plants can be calculated from stable carbon isotopes in tree ring cellulose, δ13Ccell, as

$$ \Updelta^{ 1 3} {\text{C }}\left( \permille \right) = \left( {\delta^{ 1 3} {\text{C}}_{\text{a}} -\delta^{ 1 3} {\text{C}}_{\text{cell}} } \right)/\left( { 1+ \delta^{ 1 3} {\text{C}}_{\text{cell}} / 1 ,000} \right), $$
(2)

where δ13Ca is the isotopic value of atmospheric CO2, the input signal for the plant. Variation in δ13Ca needs to be taken into account, as the atmosphere has become depleted in heavier 13CO2 over the last two centuries due to combustion of isotopically light fossil fuels. We used published values of δ13Ca for the period 1969–2003 from McCarroll and Loader (2004), and extrapolated the near-linear decline of δ13Ca over the last decades to estimate the values for 2004–2007.

Following Farquhar et al. (1982), carbon discrimination during CO2 fixation of C3 plants is linearly related to the ratio of intercellular to atmospheric [CO2] (c i/c a) by the equation:

$$ \Updelta^{ 1 3} {\text{C }}\left( \permille \right) = a + \left( {b - a} \right)c_{\text{i}} /c_{{{\text{a}},}} $$
(3)

where a (4.4‰) refers to the slower diffusion of 13CO2 relative to 12CO2, and b (27‰) is the fractionation by Rubisco against 13CO2. By combining Eqs. 2 and 3, c i can be calculated using c a, obtained from direct measurements of atmospheric [CO2] from Keeling et al. (2009) (http://cdiac.ornl.gov/trends/co2/sio-mlo.html).

Intrinsic water use efficiency (W i), is defined as the ratio of assimilation rate (A) to stomatal conductance for water vapour (g w) (Ehleringer et al. 1993; Osmond et al. 1980),

$$ W_{\text{i}} = \, A/g_{\text{w}} $$
(4)

Since g w = 1.6g c (g c is the conductance for CO2), and given that the net carbon uptake by diffusion through the stomata (A) follows Fick’s law,

$$ A = g_{\text{c}} \left( {c_{\text{a}} -c_{\text{i}} } \right), $$
(5)

we can calculate W i, using Eqs. 3, 4 and 5,

$$ W_{\text{i}} = A/\left( {g_{\text{c }} 1. 6} \right) = \left( {c_{\text{a}} -c_{\text{i}} } \right)/ 1. 6= c_{\text{a}} \left( {b - \, \Updelta^{ 1 3} {\text{C}}} \right)/ 1. 6\left( {b - \, a} \right). $$
(6)

Although, W i reflects stomatal control of water use, it does not provide an actual measure of true water losses per assimilated unit of carbon as it does not account for evaporative demand. Higher water vapour pressure deficit (vpd) across the stomata increases evaporative demand and will result in increased water losses even if g w and W i remained constant (Ehleringer et al. 1993; Seibt et al. 2008). The instantaneous water use efficiency is defined as the ratio of assimilation and transpiration, A/E, and thus provides a true measure of plant water losses (Ehleringer et al. 1993; Seibt et al. 2008). E can be calculated as,

$$ E \, = \nu {\text{g}}_{\text{w}} $$
(7)

The term ν is the water vapour pressure deficit between leaf and atmosphere, divided by P, the atmospheric pressure,

$$ \nu = \left( {e_{\text{i}} -e_{\text{a}} } \right)/ \, P, $$
(8)

e i and e a are the vapour pressures inside the stomata and in the atmosphere, respectively.

Using Eq. 6, 7 and 8, we can calculate the instantaneous water use efficiency as,

$$ A/E = W_{\text{i}} /\nu = W_{\text{i}} \, P/\left( {e_{\text{i}} -e_{\text{a}} } \right) $$
(9)

As there are no long-term relative humidity records of the study site to calculate leaf-to-air vapour pressure deficit, we used long-term air temperature and vapour pressure data (e a) from the gridded dataset, CRUTS3.0 [University of East Anglia Climate Research Unit (2009)] to calculate vpd. We calculate e i according to Allen et al. (1998); e i = 0.6108 Exp (17.27 T l)/(T l + 237.3), assuming saturated vapour pressure inside stomata. We also assumed leaf temperature, T l to be equal to air temperature, because of the small leaflet size of Mimosa (width = 5 mm) and high wind speed in the area (mean 9.3 m s−1), resulting in very high convective energy exchange between leaf and air (Nobel 1991). We checked this assumption using detailed climate data for 2007 and 2008 and the energy balance equations of Nobel (1991), and found that T l rarely exceeded T air by more than 0.5°C and never by more than 1°C (data not shown), even if leaves were not transpiring. We corrected for the offset between monthly vpd based on long-term gridded dataset, and vpd during daylight hours (radiation > 20 μmol m−2 s−1) of the wet season (June–October), the period when carbon uptake can take place. To this end, we correlated daytime with gridded vpd for 2007, and used the regression to calculate long-term vpd trends over the wet season during daytime.

Data analysis

We assessed the degree of correspondence of inter-annual variation in δ13Ccell among the five trees by calculating the mean Pearson correlation coefficient of all pair-wise combinations of trees. To study how c i and W i related to atmospheric [CO2], climate, and growth, we calculated yearly means of c i and W i for the five trees. Long-term trends in yearly means of c i and W i over the study period are evaluated using linear regressions. To study physiological responses of trees to inter-annual variation in climate we correlated c i with different climate variables. We choose c i for studying physiological response to climate, instead of W i, as c i was apparently unaffected by increasing atmospheric [CO2], whereas W i showed strong increases over time. To provide detailed insight into the influence of rainfall during different months on c i, we correlated c i with monthly rainfall, running from July of the previous year to December of the current year. We also correlated c i with annual (from November until October) and seasonal (dry and wet season) averages of rainfall, temperature, cloud cover and solar radiation. The previous wet season was included as carbohydrates from previous years may be used for formation of tree ring cellulose in the current year (Fritts 1976; Helle and Schleser 2004). To correct for possible correlations between climate variables, we also calculated partial correlations, which allowed us to study the effect of each climate variable on c i, while controlling for the effects of other climate variables.

Historical local climate data were obtained from several sources. Rainfall (1969–2006) and cloud cover data (1969–2003) from the nearest weather station of Ixtepec (16°33′N, 95°06′W, 14 km from research site) were obtained from the Comisión Nacional del Agua (CONAGUA). As the Ixtepec temperature records showed irregularities, we used monthly gridded temperature anomaly data (1969–2007) from GISSTEMP (Hansen et al. 1999). Average daily solar radiation data at earth surface (kWh m−2 day−1; 1983–2005) were also obtained from a gridded dataset (NASA; http://eosweb.larc.nasa.gov).

We also correlated c i with large-scale climate drivers. As a proxy for large-scale, inter-annual drivers of climate, we used the southern oscillation index (SOI), and sea surface temperature anomalies (SSTA) from the east and west Pacific and from the North tropical Atlantic, the principal regions that showed influences on growth of M. acantholoba in a previous study (Brienen et al. 2010) and affect local climate (Taylor et al. 2002). SSTA data are from the extended SSTa-dataset until 2003 of Kaplan et al. (1998). Data of the SOI, a meteorological index based on air pressure difference between Tahiti and Darwin that is often used to characterize the strength of El Niño events (Trenberth and Caron 2000), were obtained from the National Centre for Atmospheric Research (http://www.cgd.ucar.edu/cas/catalog/climind/soi.html). We define El Niño years as those with 3-monthly means of Niño3.4 SSTA exceeding +0.5 for at least 5 consecutive months (sensu National Oceanic and Atmospheric Administration, http://www.cgd.ucar.edu/cas/ENSO/enso.html).

To study how physiological tree responses affected tree growth, we correlated c i and W i with annual diameter growth of the five individuals included in this study. There was no age or size-related growth trend present in our data (Brienen et al. 2010), and there was thus no need for detrending our data. Diameter growth was calculated from averaged ring width measurements along two to three radii (Brienen et al. 2010).

Results

Long-term trends in δ13C, c i, and water use efficiencies

The inter-annual pattern of stable carbon isotope ratios in tree ring cellulose (δ13Ccell) between the five trees in this study was well-synchronized with a mean inter-tree correlation of 0.55 (Fig. 1a). This is of a similar magnitude as the inter-tree correlation in growth of the same five individuals (r = 0.50). δ13Ccell remained rather constant between 1968 and 2007 without evident increases or decreases. Because atmospheric δ13C decreased over the same period from −7.1 to −8.2‰ (Fig. 1b), Δ13C decreased. Year-to-year variation in δ13Ccell between 1968 and 2007 was relative high varying from a minimum of −28.3 to −23.0‰. The strong increases in atmospheric [CO2] over the last decades did not result in significant increases in intercellular [CO2] (c i) (P = 0.14). Instead, c i remained relatively constant over time, but with a large year-to-year variation (Fig. 2a). Constant c i and increasing c a implies that intrinsic water use efficiency (W i) increased over time (cf. Eq. 6); we find that W i increased from about 80 to 110 μmol mol−1, an increase of nearly 40% over the past four decades (Fig. 2b), while c a increased from 323 to 384 ppm, an increase of 19%. Average increase in W i was 0.52 μmol mol−1 per ppm increase in atmospheric [CO2].

Fig. 1
figure 1

a Synchronous patterns of δ13Ccel records of five Mimosa acantholoba trees in a Mexican dry forest (tree ages indicated between brackets), and b mean δ13Ccell and δ13Catm between 1968 and 2007 (data source δ13Catm: McCarroll and Loader 2004)

Full size image
Fig. 2
figure 2

Temporal variation in a atmospheric (c a) and intercellular [CO2] (c i), b intrinsic water use efficiency (W i) and instantaneous water use efficiency (A/E), c mean diameter growth rates of five trees included in this study, and d annual rainfall between 1968 and 2007. c i, W i and A/E were inferred from δ13C in tree ring cellulose and growth from ring widths of Mimosa acantholoba in a Mexican dry forest

Full size image

Water vapour pressure deficit increased by about 4% over the last four decades. This weak trend had little effect on instantaneous water use efficiency (A/E) which increased over time by 40%, parallel to W i (Fig. 2b).

Long-term growth and correlations with c i and W i

There was no trend in long-term mean growth rates within the five individuals included in this study (Fig. 2c). Inter-annual variation in growth was synchronized with inter-annual variation in c i (Fig. 2a, c), with years of high mean growth having a higher c i (Fig. 3a). Mean growth of the five trees included in this study correlated positively and strongly with mean c i (r = 0.72, P < 0.001), and was negatively and less strongly correlated to mean W i (r = −0.49, P < 0.01).

Fig. 3
figure 3

a Relationship between mean diameter growth and intercellular [CO2], ci, and b the relationship between annual rainfall and ci of five trees of Mimosa acantholoba in a Mexican dry forest

Full size image

Responses of c i to variation in climate

Intercellular [CO2] (c i) was positively correlated with rainfall during July, August and September of the current rainy season and with November of the previous rainy season (Fig. 4). We also found positive correlations between c i and cloud cover and negative correlations with temperature and solar radiation during the wet season (Table 1). Among all climate variables that were considered in this study, total annual rainfall exhibited by far the strongest influence on c i, explaining about 50% of inter-annual variation in c i (Fig. 3b). This is also illustrated by the strongly synchronous pattern between c i and total annual rainfall (cf. Fig. 2a, c).

Fig. 4
figure 4

Correlations between monthly rainfall and c i of Mimosa acantholoba in a Mexican dry forest for 1970–2006 (bars). Correlations run from previous growing season (j-1 July) to the middle of following dry season (d December). Significance levels are indicated above or below the bars, ***P < 0.001; **P < 0.01. The shaded area in the background is the mean total rainfall for each month

Full size image
Table 1 Correlations (and partial correlations) between c i (intercellular [CO2]) and rain, temperature, cloud cover and solar radiation averaged over the wet (June–October) or dry season (November–May)
Full size table

When correcting for the strong effect of wet season rainfall on c i, using partial correlations, the effects of wet season temperature, cloud cover and solar insolation on c i disappeared (cf. partial correlations, Table 1). Thus, the negative correlations between c i on one hand and temperature, cloud cover and solar radiation on the other, were completely due to correlations of these climate variables with wet season rainfall. When correcting for the influence of wet season rainfall, we found also a positive effect of dry season rainfall.

ci also correlated with large-scale climate drivers, like Pacific and Atlantic sea surface temperature anomalies (SSTA) and the southern oscillation index (SOI) (Table 2). In the first half of the year North tropical Atlantic SSTA exhibited the strongest influence on ci, while in the second half of the year ci was mainly affected by west and central equatorial Pacific SSTA (Table 2). The correlation between ci and SOI and central equatorial Pacific SSTA, two measures of the strength of the EL Niño–southern oscillation disappeared when correcting for the effect of wet season rainfall on ci. ci was significantly reduced during warm and dry El Niño years between 1968 and 2007 (average of 194 ppm during El Niño years vs. 207 ppm during non-El Niño years, t = 3.55, P < 0.001, df = 38).

Table 2 Correlations (and partial correlations) between c i (intercellular [CO2]) and sea surface temperature anomalies (SSTA) in the Pacific and Atlantic Ocean and the southern oscillation index (SOI), averaged over the wet (June–October) or dry season (November–May)
Full size table

Discussion

Long-term trends

Calculations of c i and W i based on δ13C in tree ring cellulose are sensitive to several assumptions. First, we did not correct for a likely offset between wood and leaf δ13C. Wood δ13C is usually enriched compared to leaves due to downstream fractionation of carbohydrates from leaf to stem (McCarroll and Loader 2004). This enrichment is generally not accounted for in tree ring studies and does not affect the general trends. Another simplification of our calculations is that we did not account for mesophyll conductance of CO2 from intercellular space to the chloroplast. Seibt et al. (2008) showed that ignoring mesophyll conductance by using the linear (or reduced) form of the isotopes discrimination model of Farquhar et al. (1982) may underestimate the response of W i and c i to increases in [CO2], and may affect variation between species in δ13C. However, we had no information on mesophyll conductance for our species and using mean values of mesophyll conductance would not improve insights. We therefore preferred the use of the simpler, linear model of Farquhar et al. (1982, cf. Eq. 3), which also allowed comparison with other tree rings studies.

We observed strong increases of nearly 40% in W i over the last four decades. Besides increases in atmospheric [CO2], other mechanisms that may have lead to the improved W i are long-term climate trends. Two climate trends were observed; an increase in wet season temperature of 0.15°C per decade (P < 0.01), and an increase in cloud cover (P < 0.001). Increasing cloud cover probably results in lower, not higher W i, because of lower A with lower irradiance and/or higher g s as a reaction to lower evaporative demand. Temperature increases may indirectly result in higher W i as higher temperatures increase vapour pressure deficits (vpd), to which plants may respond by closing their stomata (reducing g s) (Lloyd and Farquhar 2008). However, we found only a slight increase in vpd over time, and thus trends in true (or instantaneous) water use efficiency (A/E) did not deviate from the trend in W i (Fig. 2b).

Another mechanism that could affect trends in W i is related to the effects of tree age or ontogeny on isotope composition in tree rings, also called the “juvenile effect” (Bert et al. 1997; Francey and Farquhar 1982; Marshall and Monserud 1996; McCarroll and Loader 2004). Although the causes behind this juvenile effect are not entirely known, one possible mechanism behind observed higher δ13C in juvenile trees (or lower apparent W i) is the use of recycled air that is depleted in 13C by young trees growing close to the forest floor (Schleser and Jayasekera 1985). This effect does not appear to extend higher than the lowest forest strata even in a dense rainforest (Buchmann et al. 1997), and it is very implausible to have affected δ13C signals in our species. Another cause behind the juvenile effect, shading, can also be ruled out, as M. acantholoba is a pioneer species and individuals receive full sunlight throughout their entire life. A third possible cause for the juvenile effect is changes in hydraulic conductance from soil to leaves when trees are get taller (Ryan and Yoder 1997). As our trees reach maximum heights of 7 m, this is also unlikely to cause strong changes in hydraulic limitation. We therefore think it unlikely that a putative juvenile effect played a significant role in the W i trend observed, although we could not statistically test for an age effect as we only included five trees of similar ages in this study. Even if we assumed a relatively strong age-related trend of 1‰ over the entire trajectory (cf. Duquesnay et al. 1998), we would still find an increase in W i of 26% over the last decades. We therefore conclude that response to increasing [CO2] is the dominant cause for the improved W i observed in this study. This is in line with many other studies that showed that increases in W i coincided with increases in atmospheric [CO2] since ca. 1850 (Bert et al. 1997; Feng 1999; Hietz et al. 2005; Saurer et al. 2004; Waterhouse et al. 2004).

Reported trends in W i in temperate trees are mostly in the range of 0.20–0.45 μmol mol−1 per ppm increase in atmospheric [CO2] (Bert et al. 1997; Feng 1999; Saurer et al. 2004), with maximum increases of up to 0.54 (Waterhouse et al. 2004). For tropical trees, few studies on long-term trends in W i exist so far. Hietz et al. (2005) and Nock et al. (2010) reported increases in W i for tropical moist forest tree species from Brasil and Thailand within the range of temperate species: 0.26 and 0.34 μmol mol−1 per ppm. For semi-arid woodlands in Ethiopia, Gebrekirstos et al. (2009) report large differences in W i trends in four tree species, varying from constant W i to increases of up to 0.44 μmol mol−1 per ppm (based on δ13C records presented in Gebrekirstos et al. (2009)). The increase in W i observed in our study (0.52) is thus higher than in most previously reported increases. It is still uncertain what determines the observed differences in W i responses to increases in [CO2]. Substantial differences between sites within the same species (Arneth et al. 2002; Saurer et al. 2004; Waterhouse et al. 2004) indicate that differences in soil water availability, air humidity, and temperature play an important role in the physiological responses of plants to increasing atmospheric [CO2].

To gain a better understanding of physiological reactions of trees to increasing [CO2], it may help to use the three scenarios of gas exchange responses of Saurer et al. (2004). Under increasing c a , we may observe; (1) a constant c a–c i with no improvement in W i and no active stomatal responses, (2) a constant c i /c a indicating that the linkage between stomatal conductance and assimilation (Wong et al. 1979) is largely maintained under increased c a (Medlyn et al. 2001; Sage 1994), or (3) the maintenance of a constant c i. The most common response inferred from tree rings is maintenance of a constant c i/c a (Feng 1998; Hietz et al. 2005; Leavitt and Lara 1994; Nock et al. 2010; Saurer et al., 2004), but constant c i has also been reported (Francey and Farquhar 1982; Linares et al. 2009; Waterhouse et al. 2004). Short-term experiments also show constant c i/c a for a variety of species, except under conditions of drought stress where c i showed less increase (Sage 1994). This indicates that stomata may become more conservative during drought stress, and is in accordance with observations that drought-stressed plants show stronger stomatal responses to [CO2] (Field et al. 1997; Medlyn et al. 2001; Wullschleger et al. 2002). A stronger stomatal response of drought-stressed trees may explain the constant c i and relative high increases in W i at our site, which is dry compared to other studies.

Higher W i can result from decreasing stomatal conductance, increased assimilation or a combination of both (cf. Eq. 4). Although, it is not possible to determine the contribution of each factor by δ13C alone, trends in c i over time may give us some insight into changes of assimilation rates over time. Assuming that the relation between c i and photosynthetic rate of leaves (A) did not change due to increased atmospheric [CO2], we may cautiously conclude that instantaneous assimilation rates did not change as c i did not change. However, plants raised at higher [CO2] often show lower rates of photosynthesis when measured at the same c i due to acclimation (Gunderson and Wullschleger 1994; Medlyn et al. 1999), which implies that assimilation rates may even have declined. Still, increases in assimilation rates have also been observed (Sage 1994), and we would need specific information about responses of the study species to increases in [CO2] to draw definite conclusions.

Potentially, one could use trends in diameter growth rates to evaluate whether assimilation rates changed over time. The constant diameter growth (cf. Fig. 2c) actually implies that biomass gains increased over time (as basal area increment increased), but it is difficult to separate ontogenetic effects from responses to increased [CO2] as most tree species increase in growth rates during ontogeny.

Finally, it is not clear how increases in W i manifest itself on the ecosystem. Assuming assimilation stayed constant we can estimate that stomatal conductance (g s) declined by ca. 30% (cf. Eq. 4). If this decrease in g s did result in reduced transpiration losses (E), this can have substantial consequences for the ecosystem and its hydrological cycle. Potentially there is a delayed water stress and the duration of assimilation may have increased either on a daily basis or by extension of the growing season length if soil water moistures increased (Henderson-Sellers et al. 1995). However, feedback mechanisms could exist. For example, lower stomatal conductance reduces leaf cooling effects and increases leaf temperature (Idso et al. 1993). This may lead to increases in leaf-to-air vapour pressure deficit and in turn increase in transpiration losses, but this is unlikely in a species with open canopies and small leaves where leaf and air temperatures are closely coupled. Reductions in transpiration losses may lead to a dryer atmosphere as well as increased soil moisture and runoff (Gedney et al. 2006) and changes in precipitation, which could significantly influence regional climate (Henderson-Sellers et al. 1995). For example, reduced stomatal conductance accounted for about one-fifth of the predicted rainfall decreases in the Amazon and is predicted to accelerate the rate of warming using global dynamic vegetation models (Betts et al. 2004). In contrast to large continental areas such as the Amazon, rainfall in Central America is largely of oceanic origin and not from regional transpiration (Taylor et al. 2002), thus reductions in stomatal conductance are unlikely to affect regional rainfall.

Clearly, more insight into physiological responses of trees to increasing [CO2] is required, especially for tropical forests as they will play a crucial role in the future evolution of climate change (Malhi and Grace 2000). To understand the conditions under which either c i or c i/c a tends to remain constant, and their implications for carbon cycling rates and true water use efficiencies, more studies on tree ring δ13C trends across major environmental gradients in the tropics, including dry and wet sites, would be helpful. Given the recent advances made on tropical tree ring studies (Brienen et al. 2009; Schöngart et al. 2004; Worbes 2002), tropical forests need no longer be excluded from tree ring studies including long-term annually resolved carbon isotope series. Results from such studies may be used to test DGVM’s which predict large-scale die-off of the Amazon rainforest (Betts et al. 2004), but remain highly simplistic due to lack of data and understanding of key processes.

Physiological responses to inter-annual variation in climate

Interpreting the long-term reactions to rising CO2 levels may be helped by understanding short-term reactions to factors other than [CO2]. Of all local climate variables, total annual rainfall is the dominant controlling factor of c i, and neither temperature, nor cloud cover or solar radiation influenced c i after controlling for the effect of rainfall on c i. The difference between simple and partial correlations of c i with climate variables shows that simple correlations can be misleading because of correlations between different climate variables. For example, El Niño years in the study region are not only drier, but also hotter and receive more solar radiation (Brienen et al. 2010). This emphasizes the importance of taking all local climate variables into account (cf. McCarroll and Loader 2004). A strong influence of rainfall on c i (and thereby tree ring δ13C) was also found in other studies of dry sites in the tropics (Fichtler et al. 2010; Gebrekirstos et al. 2009) or temperate regions (McCarroll and Loader 2004), and after correcting for changes in atmospheric [CO2], δ13C is mostly interpreted as a drought signal (McCarroll and Loader 2004). The relation between c i and rainfall is well understood for dry sites where moisture stress is limiting and can be explained by stomatal responses to soil moisture and relative humidity; during dry years, stomatal aperture decreases to prevent excessive water losses, leading to reduced influx of CO2 into the intercellular space, and thus a lower c i and lower 13C-discrimination (Farquhar and Sharkey 1982). Where drought stress is uncommon, the dominant factor controlling δ13C may be the photosynthetic rate as affected by irradiance and temperature (McCarroll and Loader 2004). However, temperature and irradiance may also indirectly influence stomatal behaviour, and thereby the carbon isotope signal, through their effects on the water vapour pressure deficit across stomata (Lloyd and Farquhar 2008; Seibt et al. 2008).

A main reason to analyse tree ring isotopes and indeed tree rings in general is to use correlations with past climate for climate reconstructions (McCarroll and Loader 2004). The reasonably high between-tree correlations and the correlations with rainfall are in the same order of magnitude as Gebrekirstos et al. (2009) reported for Ethiopia and show that the use of δ13C records is promising for tropical dry regions, although in our case limited by the short life-span of Mimosa acantholoba. Of particular interest is the high correlation with ENSO indices (SOI, SSTA-pacific), as palaeoclimatic proxies of ENSO from the tropics are scarce (Brienen et al. 2010; Schöngart et al. 2004). Our study shows that trees in Central America dry forests may be promising in this respect, although the signal of δ13C in tree rings was slightly lower than the signal in ring width in this species (Brienen et al. 2010). A multi-proxy approach combining ring width and isotope measurements (including water isotopes) may improve interpretation of climate signals and strengthen the overall palaeoclimatic potential compared to the use of one single proxy (McCarroll and Loader 2004).

Our results showed that δ13C in tree rings is a promising tool to evaluate long-term responses of tropical trees to increasing [CO2] and to variation in climate. Extending such studies to a larger number of tropical tree species with annual rings and to sites differing in rainfall, could improve our understanding of the responses of tropical forests to predicted changes in climate and atmospheric [CO2].