250
~j
u
Mo oh ,ver k ·R·300~l50 ~
'¡OOOm'
~
b)
1
I-Hills
10
IIl-Bottomlands
9-;'-"
I1-Uplands
5 2
1 100m
Figure 1 • Local palm study setup, iIIustrating (a) the localion 01 the areas sampled al Ihe hills (H). uplands (U) and bottomlands (B) 01 the Mokoti river basin and (b) the positioning olthe 10 plols 01 600 m' (60 x 10m) at each 01 the areas.
poines, and mose that tend to OCCUf in the same plots converge in the scatter plot, whereas those mat occur in differenr plots appear farmer aparto If the species are plotted onto the same diagram of floristic quadrates, they tend ro appear near the quadrates where they are most abundant. To reduce the importance of rare species, whose distributions are difficult to explain ecologically and could weaken che overall ordination results Oongman etal., 1995), only rhose palm species wirh more than 10 individuals were induded in rhe DCA.
LARGE·SCALE PAnERNS OF PALM DlSTRIBUTION
Regional patterns of palm richness were estimate<! from maps showing occurrence points of the 283 species, sub-species and varíeties of palms fuund within che Brazilian territory (Lorenzi et al.) 2004). The territory was divided into 47 units of approximately 150,000 km2 and species richness was inferred for the poínt corresponding to me center of each square. To
19 VOl. 37(1)2007 17· 26. SALMelal.
rACTA
~MAZONICA
CROSS-SCALE OETERMINANTS OF PALM SPECIES OISTRIBUTIONunderstand (he potential role of climate on palm richness and distribution we obtained (he envíronmental variables corresponding to each of me 47 points specified aboye. To (his end, we used worldwide climate maps generated by che interpolation of climaric information obrained [rom ground-based meteorological stations (New et al., 1999). The meao dimate surfaces were available for the period trom 1961 to 1990, with a monthly temporal resolution aod 0.5" (latitude) by 0.5"
(longitude) spatial resolution (New et al., 1999). The climatic variables used were: precipiration, maximum, minimum and mean temperarure saturation vapor pressure and wet-day frequency. The minimUffi and maxímum monthly temperature estimares were ca1culated froro che original dimate surfaces by suhtracting oc adding, respectivdy, half the diurna! temperature range trom the mean monthly temperature (N ew et al, 1999).
In arder to summarize (he information contained in chis time-series and capture (he seasonal features of (he data, each climatic variable was processed with temporal Fourier anaIysis (processed dataset provided by D.
J.
Rogers; see Rogers & Williams, 1994;Rogers eta/., 1996; Rogers, 2000), extracting in this process the
¡ntee-arrnual variability of (he monthly time-series ¡Uto
uncorrelated componenrs of cyeles rhat are repeared up (hree time§, peryeaT, We used the following parameters of each variable:
mean (aO) value; amplitude ofaonual (al), biaonual (al), aod triannual (a3) cyeles aod maximum (mx) and minimum (mn) values and
me
phase variables (reflectingme
tinUng of occurrence of cyelic everits). Tú explore che potential role of dimate on variation in paIm richness wirhin the country we men candueted a stepwise regression analysis ro determine which factors would better explain me observed patterns of richness. We scrutinized che dara to detennine whemer ir sacisfied the assumptia~ af me regressian,_and (log ar square-roor) transformed mase variables with distribution signiflcantly departing from normality, discaTding mase variables for which normalirywas nor achieved afrer transformation. Afrer derermining rhe eco-dima-tic constraints currenrly defining rhe distribution and richness of che paIm family wirhin the Brazilian territory, the resulting dimaric model was used ro generate a predicrive map of palm richness at a global scale (because rhe climatie variables are in continuous interpolated surfaces, ir was possible tú extrapolare the prediction trom the model ro aIl regions oftheworld).RESULTS
LOCAL PATTERNS OF PALM OISTRIBUTION
A total of 694 ·individual palms belonging ro 10 species were sampled at rhe Mokoti River basin: Artrocaryum aculeatum G. Mey., Astrocaryum gynacanthum Man., Attaka mdripa (Aubl.) Marr., BactTis acanthocarpa Mart., Bactris tomentosa Marr., Desmoncus polyacanthos Man., Euterpe precatona Man., Geonoma baculifera (Poit.) Kunth, Oenocarpusdistichus Mart. and Socmtea exorrhiza (Mart.) H. Wendl. Table 1 shows the mean values of
abundance Df each paim species observed at the hilis, uplands and bottomlands Df the Mokori river basin, togerher wirh me overaIl values of abundance, number of species and rree basal area at each of these sites. There was a significant ¡nerease in the abundance and riehness of paIms from
me
hilIs towards the uplaods, and in turo trom the uplands towards the bonomIands of the Mokoti basin civer. Thls inerease in richness and abundance follows a significant increase in tree basal area and in the number of sampled trees (Table 1).The palm species A. aculeatum, D. polyacanthos and
o.
distichus were represemed by less than 10 individuals and therefore not induded in the ordination analysis. In Axis 1 of the scatter-diagram resulting from the DCA rhe plots at the bottomlands of the Mokoti River are separared from those in che hills and uplands, which are in turn serapart in Axis 2 (Figure 2). G. baculifera and E precatoria"lere strongly assoeiated with
Table1 - Forest structure observed at the hills, uplands and bottomlands 01 tne Mokoti river basin, southeastem Pará, Brazi!. Different letters ("a" and
"b" or "a", "b" and "e"), within a line, represent significant differences between plot types. Conversely, there were no slgniflcant dlfferences among plots marked with the same later "a' or "b" (p<O.05) (e.g., the number 01 traes par hactara is not different between fue HiIIs and Uplands - same 'a' letter -, not different between the uplands and bottomlands - same 'b' letter -, but different between the HH1s and Bottomlands, as there ls no letter in common).
- - - . _ -NtJmber of trees per hectare Tres basal area (m2) per hectare
Figure 2 -Detrended Correspondence Analysis ordination of palm plots.
Sol id and apen symbols represent species and plots, respectively, with squares corresponding to plots on uplands, circles to plots on hills and triangles to plots on bottomlands (Lengths 01 gradient 2.483 and 2.61 9 SO-units; Eigenvalue Axis 1 ~0.4897 and Axis 2=0.1847).
20 VOL. 37(1) 2007: 17- 26 • SALM el al.
, ~ ACTA AMAZONICA
_ .. CROSS-SCAlE DETERMINANTS OF PAlM SPECIES DISTRIBUTION che bottomlands. The former is found in chis area al densities ofhundreds of indivíduals per hectare, while ir occurs at low densities in up1ands and is totally absent from the hills. The lattee is found in all sampled areas, bue ¡es density is significantly higher at (he bottornlallds of chis river. A. gynacanthum was found exclusively at che uplands. The two species of che genera Bactris appear to have opposite behavior: B. acanthocarpa was associated with (he upJands, where che species i5 found athigher densities (although chis difference was nor significam), whereas B. tomentosa was associated wieh che hilis, where ir i5 more abundam, followed by che bottomlands, being finally rarer at me uplands. It is possible to see in che scatter diagram matA.
manpa
lies ae an intermediare position beeween me uplands and bottomlands pIoes. The densiry of palms of this species is also significantly higher at the river bottomlands. S. exorrhiza, in contrase) was associared wirh rhe hills, where jr reached its highese density, followed by the borromlands and uplands, showing no significant difference in ahundance bev.veen ehe three sites (Table 2).LARGE-SCALE PATTERNS OF PALM DISTRIBUTlON
Within che Brazilian cerritory, paIm richness reaches its highesc values, with more than 6fty species found per 150)000 km2) in (wo areas of the Amazon, one at the center of the basin, around the city ofManaus, and che other at ics westernmost parts, in che St<lre of Acre. However, ir is important to note mat sorne of mese patterns may be due ro geographic differences in sampling effort (Nelson et al., 1990).
From chese areas, palm richness decreases towards dryer areas to the north, sourh and west. The major South American disjunction (Brieger, 1969), a wide transversal corridor roughly supeIposed ro che distribution of the Central Brazilian savannahs,
Table 2 ~ Abundance of each palm species, in individuals per hectare, observed at the hills, up/ands and oottomlands of too Mokoti river basin, southeastern Pará, Brazil. Different letters, within a line, represent significant differences between p/ot types (Mann-Whitney U, fJ<O.05). Converse/y, there were no significant differences among plots marked with the same later.
Species Hjlls Uplands BoHomlands
Desmoncus polyacaQtnos '* 1
Oenocarpus distichus *
Astrocaryum -¡/cuTeátúm * 6
Astrocaryum gynacanthum * 35
Att.a.lea marlpa * 6' .101 23"
* Individual with visible stem, **Individuals with OBH e" 5cm, *** Reproductive gene1s.
with lower annual rainfall, where the richness of palms falls to levels becween ten and twenty species, separares the Amazon and the Atlamic forests. The wetter Atlantic forest constitutes anorher center ofhigh species diversiey, wirh more rhan thirt)' species found in che Stare ofEspírito Santo and adjacem Southern Bahia. From rhis area, species richness deereases rowards the interior of che country and along che coast, ro che norrh and ro the sDuth. Less than ten species are found in che dry region in {he northeasc of the country. Within Brazil rhe effects of minirnlUU temperacure on palm species distriburion are more evident in the southernmost part of the country, wirh less than 10 paIm species per sampled region. These vaIues are tound up ro the Pantallal in Mato-Grosso do Su!, where low minimum temperatures are common during rhe austral winter due to invasíons of polar winds, which cross che Andes Dn a continental toute (Nimer, 1977).
The environmental model resulting from che stepwise regression analysís shows che limiting role of eco-dimatic facrors on paIm disrribution and richness, and provides us with additional insights (Table 3). The firsr predicror selected, which alone accountoo. for approximatdy 53 % of che variance in palm richness, was vapor pressure, namely che annual mean of che vaIues of vapor pressure registered monthly. We tested the assoóacion ben.veen richness and vapour pressure, and alone it explains 52.8% of the variability. Vapor pressure results from the nwnber of water vapor molecules in che air (the greater the moisture vapor content of air, rhe greater che vapor pressure), and is thus linear1y related to absolure humidity. Ir is a1so modulated by temperature, with higher ,ralues oE vapor pressure corresponding to warmet places. The selection of mis variable {herefore indicates that richness hics its highesc Ievels in simulranoously warmer and more humjd places, specifical1y in those places where che interacrion of temperamre and humidity (which is in rurn strongly influenced by precipitacion) does not allow vapor pressure to fall below certain levels.
TabIe 2 also shows that me second best predicror of richness (negatively related to ic) was che amplítude of che annual cycle of cemperarure (maximum temperature), meaning the degree of seasonal variar ion in the maximum observed values of temperature during any one year. This is an interesting resulc, as
Table 3 ~ Results of the stepwise regression analysis. Parameters are Interceptforlhe fun Model = 96.7. *unstandardiled coeflicients.
21 VOL. 37(1)2007: 17- 26 • SAlM el al.
CROSS-SCALE DETERMINANTS OF PAlM SPECIES DISTRIBUTlON
ir suggests rhar richness is higher in places with more srable maximum temperatures than in those regions characterized bya strong seasonaJity in rhis parameter. Finally, the last predictot selected was (he mean of maximum temperature (i.e.
me
averageof
me
maximurn temperatures registered per momh), which was negatÍvelyassociated ro richness, (hus suggesting [har in (hose places whefe maximum temperatures were highesr paIm richness was lower. Ir seems lO liS, however, mar (he latter association has resulted from che faer thar, wi[hin [he Brazilian territory, those regions with highesf maximum temperatures correspond to regions where there LS a very large variation in precipitation during me year (as indicated by [he significam carrelation berw-een "'maximum remperature" and "ampIitude ofthe annual cycle of precipitation": r = 0.53, p<O.OOl, where pis Bonferroni-eorreered). Areas with srronger seasonaliry in precipitation beis might have less palm species due to me palms' general vulnerabiliry to dimane hazards resuIting from me lack of reserve buds previously memioned (Tomlinson, 1990;Richards,1996).
The dimaric model resulting from rhe regression analysis performed for the Brazilian territory was subsequendy used ro generare a predictive map of palm riehness at a global seale.
Given the discussed specificity of the Iase predicror selected (mean of maximum temperature) with the ecosystems of the BraziIian rerritory we decided ro excIude rhis factor from the model to generare rhe predicrive map for rhe globe, retaining rhe firsr twu best predictors of richoess. Figure 3 shows rhe resulting map.
The color-coded levels of richness effectively indicare the predicred number of palm species in each region as based on fiér environmental suitability (in rerffiS of vapour pressure and seasonaliry in temperature).
DISCUSSION
The finding rhar paIm richness in Brazil is besr predicred by vapor pressure and rhe amplirude oE the annua! cycle of temperature, with a marked presence of palms in stable areas of simulraneously high humidiry and low anoual oscilIation of
Figure 3 - Predicove map 01 palm fichness at a global scale. Color-coded levels 01 richness indicate 1he predicted number of species, sub-species and varieties of palms per 150,000 km2 units in each region as based on their environmental suitability in terms of vapor pressure and seasonality in temperature.
temperarures (Figure 4), reflecrs
me
basic physiological constraints characteristic of rhe palm family and ies vulnerabiliry tú dimatic hazards (Richard" 19%), particularly caused by me presence of a single, irreplaceable rerminal bud (Tomlinson, 1990). From an ecological perspecrive, these frndings are also compatible with the idea mat water availability is a key constraint of pimt richness, especia11y in those (tropical and subrropical) areas where energ)' is abundant (Hawkins et al., 2003). Alternativel)', insect and fungal pressure might increase with higher and more consrant levels of moisture, thus mediating palm tree diversiry through mechanisms of densÍ!y-dependent morrality as observed elsewhere (Givnish, 1999). In any of these cases, byaccounting for approximately 60% of the variarion in palm richness, (he climatic model strongly emphasizes the major influence of (he selected envíronmental predictors - and oE elimate in general-on diversiry gradients, and provides us wirh clues general-on palro tree communities' potemial response ro climare change.Within the tropies, the predictive map of global palm diversiry generated from the dimane model builr foc the Brazilian territory is eonsistentwíth the general knowledge of global palm disrribution - with the major cenrers of diversity aligned along the Equator in rheAmazon, the Congo basin and Sourheastern Asia, with the largest number of species found in che latter area (Comer, 1966). The island ofNewGuinea, forexample, which hosts 145 paIms species belonging to 32 genera wirhin its 808,510 km' of territory (Bachman et"L, 2003), is shown as a hotspot of palm diversity in ,he mode!.
a) """"'ef of ope~ie. so.t>-sp&cie. and va"et¡os '" p • ..", pe< 15D.[)(lO 11m2 lLore",. el al. 20041
Figure 4 • Variation in (a) the number of specie, sub-species and varieties 01 palms per 150.000 km' (lorenli el al. 1004) and the main climatic parameters. as fol/owing from our model, underlying the richness of palms across the Brazilian terrttory: (b) vapor pressure (hecta-Pascals) and (c) amplitude 01 the annual cycle 01 temperature ("C). Observe that while vapor pressure is positively related to palm nChness, the amplitude in temperature is negatively related to i1. The asterisk label the position 01 the Pinkaiti research station and the Mokoti river basin.
11 VOL. 37(1) 1007: 17- 26 • SALM el al
~ ACTA AMAZONICA
_. CROSS-SCAlE DETERMINANTS OF PAlM SPECIES DISTRIBUTION The model i, aimed at exploring the limits placed byeco-dimatic variables on che distribut10n and richness of che family.
Despice che potentiallimitations (har heterogeneous effofts in the collecrion of da,a on palm ticbnes, along the Brazilian terri,ory couId impose on che final climatic model generated, (he overall consistency between irs predictions and general reports of paIm diversiry arourrd me globe highligh,s me majar role played by climate - and parucularly
me
variables selected in our analyses-on palm diversity. Future studies reporting observed palm [ichness levels in differenr localities of me globe couId enable a furiliee test ofme
model with independent data sers. Moreover,me
madel can be used to identif)r (hose poorly studied areas po,en,iaIly marked by a high diversiry of palm species, providing additional crirecia ro ditect sampling effofts. For example, Figure 4 suggests thar rhe northwesrern region of the Amazon is eurrendy under-sampled, as there would not be any climatie fae(Or - as given by our model- íustifYing the drop in richness in this region following (he peak in diversity observed in the surroundings ofManaus, the capital of the Amazon state.The model is additionally airoed at serving as a heuristie tool for the investigation of those cases in which faetors other than climate might underlie the discrepancy between the model's predicrion and reality. For example, aIrhough palms can survive in very dry regions (such as sorne regions of the Middle Easc and Central Australia), rhey do so only in those areas where the ground water is near (O che soil surface. Sinee chis lattervariable was nor made available (O our analyses, ir would be thereby possible ro find a higher diversity of paIms than that predieted by the model in dry regions vvirh chese conditions. The dynamics ofisland biogeography (MacArthur & Wilson, 1967) also seero ro be me cause of sorne of
me
differences between che predicted and observed diversity of paIms in sorne tiny islands of Micronesia: although che model predicts areas vvim richness values as high as 40 species per 150,000 km', well-known effects of insularity such as ¡sland size and distance from continentallands seern te underlie che less pronounced diversity obsenred in mese islands, Within Brazil, che effects ofinsulariey and the presence of subterranean water may respectively explain rhe strueture and diversiey of palm communities in eeosystems subjected ro human induced or natural fragmemation (Scariot, 1999) and rhe formation of palm-dominared foresrs in afeas thac are dimarically unsuitable for rheir occurrence (Lorenzi et al., 2004).Ac a local sca1e, che disrriburion maps ofLorenzi et al (2004) indica'e tha, me 150.000 km' area including the Pinkaití research starton, located ar che umits of me Amazon forestwith che Cerrado of Central Brazil, has a number of speeies typical of rhe dryer regions ofCenrral Brazil (with riehness rangingfrom 10 ro 20 species in areas ofsimilar size). Five palms species were however found at rhe Mokori basin beyond the distribution limies proposed by rhese authors: Attalea marípa, AstrocaY)'um aculeatum, Bactns tomentosa, Euterpe precatoria and Geonoma
baculifera. With these species, che richness of the region dimbs ro rhose ranges typical of che seasonally dry forest ar rhe limits of theAmazon (20 ro 40 speeies in areas of similar size).
Within the field study site, the palm family beeame more abundant and rieher from the open and drier forest of che Mokori hilIs ro the denser forests at the uplands and, subsequently, ro che moister bottornlands of chis river basin. AJthough on the one hand ir is dear that factors influencing richness are likely (O be scale-dependent, on the other this result is consistent with che associarion between moisture and richness as captured in che climatie model at a larger sparial scale, Ir is important ro notice, however, that it is nor currendy possible ro determine che role played by orher faerors differing among me plors, such as their edaphic and topographicfeatures and panerns ofvariation (e.g.
Vormisro et al., 2004a, 2004b), as welI as their potential interaccion with moisture differences. Further research involving the use of a larger number of pIots and study regions where chese [actors are measured and combined should help separaring rhe individual effecr of each of them upon palm riehness and abundanee.
Our field data also indicare that moisture is associated wieh forest structure at che Mokoti basin, as suggesred by che ¡nerease in total tree basal area observed from rhe hilis towards the bottomlands, AIternatively, the steepness of slopes in che hiUs could have inereased the likelihood oflarge trees falling with a
Our field data also indicare that moisture is associated wieh forest structure at che Mokoti basin, as suggesred by che ¡nerease in total tree basal area observed from rhe hilis towards the bottomlands, AIternatively, the steepness of slopes in che hiUs could have inereased the likelihood oflarge trees falling with a