Ecology Letters, (2007) 10: 461–469 doi: 10.1111/j.1461-0248.2007.01033.x LETTER Decelerating growth in tropical forest trees Abstract Kenneth J. Feeley,1* S. Joseph The impacts of global change on tropical forests remain poorly understood. We Wright,2 M. N. Nur Supardi,3 examined changes in tree growth rates over the past two decades for all species occurring Abd Rahman Kassim3 and Stuart in large (50-ha) forest dynamics plots in Panama and Malaysia. Stem growth rates 2,4 J. Davies declined significantly at both forests regardless of initial size or organizational level 1 Center for Tropical Forest (species, community or stand). Decreasing growth rates were widespread, occurring in Science, Arnold Arboretum Asia 24–71% of species at Barro Colorado Island, Panama (BCI) and in 58–95% of species at Program, Harvard University, 22 Pasoh, Malaysia (depending on the sizes of stems included). Changes in growth were not Divinity Avenue, Cambridge, consistently associated with initial growth rate, adult stature, or wood density. Changes in MA 02138, USA 2Smithsonian Tropical Research growth were significantly associated with regional climate changes: at both sites growth Institute, Panama City, Panama was negatively correlated with annual mean daily minimum temperatures, and at BCI 3Forest Environment Division, growth was positively correlated with annual precipitation and number of rainfree days Forest Research Institute (a measure of relative insolation). While the underlying cause(s) of decelerating growth is Malaysia, Kuala Lumpur, still unresolved, these patterns strongly contradict the hypothesized pantropical increase Malaysia 52109 in tree growth rates caused by carbon fertilization. Decelerating tree growth will have 4Center for Tropical Forest important economic and environmental implications. Science, Panama City, Panama *Correspondence: E-mail: Keywords [email protected] Carbon cycling, carbon fertilization, climate change, forest dynamics, tree growth rates. Ecology Letters (2007) 10: 461–469 Manaus, Brazil, and found that the growth rates of most INTRODUCTION genera (87%) had increased over time. This increase was It is well documented that human activities are affecting most pronounced in genera of large fast-growing canopy environmental conditions world-wide. Given the high trees and was therefore considered a response to carbon diversity as well as ecological and economical importance fertilization. In contrast, in the only species-level study to of lowland tropical rainforests it is critical that we date, Clark et al. (2003) looked at changes in the diameter understand their responses to these changes (Malhi & growth of six canopy tree species at La Selva, Costa Rica, Phillips 2004; Wright 2005; Lewis 2006). Recent studies and found that annual growth rates decreased steadily have reported that growth rates of tropical rainforests have over time in all species. Growth rates were negatively accelerated over the past several decades (Laurance et al. correlated with daily minimum temperatures and therefore 2004; Lewis et al. 2004b) consistent with the hypothesis of the declines were attributed to increasing nighttime increased productivity caused by rising concentrations of temperatures and the associated increases in respiratory atmospheric CO2 and carbon fertilization (Melillo et al. costs relative to photosynthetic gains. However, this study 1993; Laurance et al. 2004; Lewis et al. 2004a; Lewis 2006). tracked growth rates only within a single cohort of trees However, these studies have focused primarily on changes (i.e. new recruits were not incorporated; Clark et al. 2003) in stand-level growth rates (i.e. change in total basal area or and thus decreased growth rates may alternatively be biomass per unit area). One potential limitation of this attributable to a decrease in the relative growth of larger approach is that stand-level growth rates provide little trees. Clearly additional work is needed if we hope to information about the dynamics of most species as changes understand how the dynamics of tropical trees respond to may be driven by compositional shifts and/or the responses global anthropogenic disturbances such as rising concen- of just a few dominant species. trations of CO2 and climate change (Clark 2004; Nelson Laurance et al. (2004) investigated changes in tree 2005; Wright 2005). Understanding these responses will dynamics at the genera-level in forest plots located near have important implications for the world’s carbon Ó 2007 Blackwell Publishing Ltd/CNRS 462 K. J. Feeley et al. Letter budget; for example, models show that accelerated stem collaboratively with the Center for Tropical Forest Science growth may lead to persistent increases in carbon which ensures strict standardization of methods. sequestration, potentially helping to offset emissions (Chambers et al. 2001). Stand-level growth rates Here we use long-term tree census data from New- and Old-World 50-ha tree plots [Barro Colorado Island, In order to estimate stand-level growth rates at BCI and Panama (BCI), and Pasoh, Malaysia] to examine changes Pasoh, each plot was subdivided into 50 non-overlapping in relative basal area growth rates over the past several 1-ha (100 · 100 m) subplots. Relative stand-level growth decades for various size classes of stems (saplings, poles, rates (RGRstand) were then calculated for each hectare and canopy trees) at the species-, community- and stand- subplot during each census period as: level. Species-level growth describes the mean growth lnðBAt ÞlnðBA0Þ across individuals within each species; community-level RGRstand ¼ growth describes the mean relative growth across all t 2 species in the plot (i.e. species weighted equally); and where BA is the total basal area of all trees in mm , stand-level growth rates are the relative change in total regardless of species, that were alive and measurable at both basal area (all stems combined regardless of species) caused the start (BA0) and end (BAt) of the census period annual- by growth per hectare. Based on the species-specific ized by the number of years (t, measured in days) between growth rates, we then test if there are consistent censuses. Changes in basal area due to mortality or differences between decelerating vs. non-decelerating spe- recruitment are excluded. Estimates of RGRstand were log- cies in several species traits related to life-history strategy normally distributed and thus we used log transformed including initial growth rate, adult stature, and wood values for all analyses. density. Finally, using long-term meteorological data from In the first two censuses at BCI and the first census at BCI we test the relationship between the observed changes Pasoh the diameter of stems < 55 mm dbh were recorded in community-wide growth rates and changes in temper- at 5 mm increments (always rounded down). To account for ature and precipitation. this, the growth rates of these stems for the first two intervals at BCI and the first interval at Pasoh were based on diameter measurements rounded down to the nearest 5 mm METHODS (Condit et al. 2006). To correct for measurement errors, the ) maximum diameter growth rate was set to 75 mm year 1.In Study sites addition, we assumed zero growth (i.e. BA0 ¼ BAt) for any This study was conducted using data from 50-ha Forest individuals in which the point of measurement (POM) Dynamics Plots located on BCI and at Pasoh, Malaysia changed between censuses (e.g. due to growth of buttresses) (Table 1). These plots differ greatly in their local weather (Condit et al. 2006). POM corrections affected only a small regimes, soil characteristics, and history (Leigh et al. 2004; percentage of individuals at BCI (0.3–6.7% depending on Manokaran et al. 2004). In both plots, all woody stems the census interval and size of stems included (Appen- ‡ 10 mm dbh (excluding lianas) have been identified to dix S1), and no individuals at Pasoh. Still, we recalculated species, mapped, tagged and measured to the nearest mm growth rates using several methods to estimate growth rates (but see below) in repeated censuses. The plot on BCI of trees for which POM changed (Appendix S1). Our was initiated in 1981/1982 and has been recensused in results are robust to these corrections (Appendix S1). Here 1985, 1990, 1995, 2000, and 2005. The plot at Pasoh was we only present results based on the correction method initiated in 1986 and has been recensused in 1990, 1995, described above. and 2000 (Pasoh was recensused in 2005 but these data The rates of change in RGRstand through time were are currently unavailable). Both plots are administered determined by calculating the linear least-squares regression Table 1 Characteristics of study plots Mean No. of months Number of species Location(latitude, Elevation annual with <100 mm * Site longitude) (m) precipitation(mm) precipitation Total Saplings Poles Trees BCI 9° 09¢N, 79° 51¢E 120–160 2551 3 – 4 (January – April) 320 242 219 217 Pasoh 2° 58¢N, 102°18¢E 80–104 1788 0 – 1 (January) 823 775 692 651 *Includes all species, even those present in only a single census. Includes only those species occurring in all censuses. Ó 2007 Blackwell Publishing Ltd/CNRS Letter Decelerating tropical tree growth 463 coefficient (b) of RGRstand (log transformed) vs. date (b) of RGRcomm and RGRspecies (log transformed) vs. date at separately for each of the 50 subplots. The 95% confidence Pasoh and BCI (RGRspecies was calculated for each species intervals (CIs) for the median b at BCI and Pasoh were that occurred in all censuses). We calculated the significance determined through bootstrapping (5000 resamples) and of b for each individual species through a stratified compared with the null expectation of no consistent change bootstrapping procedure: an estimate of the species-specific (i.e. b ¼ 0). growth rate was sampled with replacement from the As growth rates are strongly related to tree size, we also corresponding posterior distribution of RGRspecies for each calculated stand-level growth rates and the associated rates of of the census intervals and b was recalculated.