Phytogeographic support tor the theory of Pleistocene forest refuges m the Amazon Basin , based on evidence from distributioh patterns in Caryocaraceae , Chrysobalanaceae , Dichapetalaceae and Lecythidaceae

In recent years it has generally been accepted that Amazonia was subject to long dry periods in the late Pleistocene and post-Pleistocene which induced forest cover to a few limited areas or refuges. It has been proposed that the subsequent genetic isolation into separate populations is a mnjor factor in the evolution of the species diversity within the lowland forest of Amazonia. Most of the previous evidence for this theory is based on studies of animals, for example: lizards, butterflies, and birds. Here data are presented to confirm the theory of forest refuges using evidence from phytogeography. Distribution patterns of the lowland species of the woody plant families Caryocaraceae, Chrysobalanaceae, Dichapetalaceae and Lecythidaceae are discussed and concur with the possibility of forest refuges. A map is given of the refuge areas that seem most likely, based on evidence from species distribution of the above plant families. The refuges proposed here correspond closely with the refuge areas proposed by Haffer and Brown rather than the extremely reduced areas proposed by Vanzolini.

In recent years it has generally been accepted t:bat Amazonia wns subject to long dry periods in the late Pleistocene and post-Pleistocene whtch reduced forcst cover to a fcw llmited areas or refnges.It has been propose<l that the subsequent gen'ltic isolation into separate populations is a mnjor factor in the evolution of the species diversity within the lowland forest of Amazonia .Most of the previous evidence for this theory is based on studies of animais, for example : lizards, butterflies, and birds.Here data are presented to confil m the theory of forest refuges using evidence !rom phytogeography.Distribution patterns of the lowland species of the woody plant families Caryocaraccae, Chl:-ysobalanaceae, Dichapetalaceae and Lecythidaceae are discussed and concur with the possibility of forest r efuges .A map is givcn of tbe refuge areas that seem most likely, based on evidence from species distribution of the ahove plant families.TCe refuges proposed here correspond closely with the refuge area.sproposed hy Haffer and Brown r ather than the extremely reduced areas proposed by Vanzolini.l.NTRODUCTION In recent years several researchers in various branches of zoology have discussed the interesting distribution patterns of organisms in Amazonia, in the light of the complex climatic history of the region.
The pioneers in this field of research are Haffer (1969), on speciation in Amazonian birds, and Vanzolini and associates in !lzards, Vanzolini and Williams (1970).Vanzolini (1970) and ( 1973).Recently their data h ave been backed up b' work in the Heliconian butterflies, Brown and Mielke (1972), Brown (1972).This GHU..LEAN T. PRANCE ( • ) w ork is also supported by the palynological evidence of Van der Hammen ( 1972).and most recently on linguistic and archeological evidence from lndian tribes by Meggers and Evan (1973).
This emphasis on Amazonian distribution patterns in other branches of biology prompted me to examine the various distribution patterns in the various plant families that I have been studying over the past few years.(For areas in which I have also collected plants see Fig .25).
The study of the taxonomy of any group must include ecologica l and phytogeographic considerations if it is to be a true interpretation of the biology of a group.White (1971), reminds us that phytogeography has two main aspects, descriptive and historie.During the course of monographic studies in the w oody plant families Chrysobalanaceae, Caryocaraceae, Dichapetalaceae and Lecythidaceae, I have been concerned with •he descriptive phase, considering the present day phenotype and distribution of these families, Prance (1972a), Prance (1972b), and Prance and Freitas (1973) , preparing distribution maps and pictorialising the geography of variation between and within closely related species.Since three of these families are predominantly families of trees centred in Amazonia, and the fourth, Dichapeta laceae, which contains vines as well as trees, is also we ll represented in the region, they provide material for the study of the vegetational history of the region which is largely covered by lowland rab-for est.The logical continuation of these monographic C*) -Instituto Nacional de Pesquisas da Amazônia, and B . A. Krukoff Curator of Amazonian Botany, the New York Botanical Garden.
studies is to use them as an aid to phytogeographic considerations.So far, botanists have made little comment about Haffer's theory on ~he contt ibution of forest refuges to the speciation duríng dry climatic periods in Amazonia.One of the few brief comments is that of Simpson (1972) who supports Haffer's generally accepted Peruvian refuges on evidence of Rubiaceae genera.
Briefly stated, Haffer has proposed that the generally accepted climatic changes of the Pleístocene and of the post-Pieistocene profoundly influenced the speciation of birds because of the severe depletion in forest during the dry periods.Haffer proposes that nine isolated are as, termed " refuges ", remained as forest while most of the rest of the area was covered with savanna.(See Figure 1).The bird species of the forest were forced into these refuges where the isolatíon of specíes occurred before the contínuous forest-cover returned.By the time that the re-•expanding forests came in contact agaín, some species were ísolated genetícally, and others hybrídísed again along the zone of contact .Thís theory ís generally endorsed by Brown and Vanzolíní, but each has hls own variations .
Since the tall raín-torest ís at the centre of Haffer's theory and of Vanzolini's theory, it should be possible to discuss the idea in terms of the distribution of species of trees in the present day forest.The four plant famílies wíth which I have been workíng are all predomínantly dístributed in lowlar;d forest , and are consequently discussed here withín the framework of Haffer's theory of refuges in an attempt to further our understanding of the complex plant species distríbution patterns found within the Amazon Basin, and further, to show the lack of uniformity of species throughout Amazonia.

PRESENT DAY VEGETATION IN AMAZONIA
In order to discuss the vegetational history of Amazonia, it is also neceisary to be familiar with the present day vegetation types and habitats of the region.The tendency is to picture the region as a rather uniform rainforest, when there is, in fact, a great variety In vegetation types.Most of these have been discussed and summarised by previous workers, for example Ducke and Black (1954), Hueck (1966Hueck ( , 1972) ) , Pires (1973).Prance (in press).Since the main purpose of this paper is to refer to the history of this region, only a brief summary of the most important vegetation types of Amazonia is given below in  The brief summary given above of the most important vegetation types serves to show that there is considerable habitat diversity within the Amazon Region .The study of any large woody Amazonian genus sho-As how complex the present day distribution patterns are.Some of the habitats referred to above are highly specialised, (e. g. the Caatingas of the upper Rio Negro).and have a high rate of endemism of plants specially adapted to these areas.In this work, however, I am more concerned with the lowland rain-forest species which potentially could grow over a large part of the region.The study of the highland habitat and other specialised lowland habitats is of great lnterest, but is not of great relevance in considering the theory of forest refuges.The distribution of lowland species has tended to be neglected because of interest in areas of high endemism.such as mountaln-tops of the Guayana Highland area or the caatingas of the upper Rio Negro .However, the distribution of any Neo-tropical species is of interest, and gives some information about the phytogeography of the region.Although there are numerous combinations of distribution patterns which could occur, the same general patterns of distribution generally appear in different groups of plants.Consequently, we can gain information about the history of the vegetaticn and also about present ecological factors controlling the region.
While it is d.fficult to present the phytogeographic patterns in a readily understandable way, the following phytogeographic breakdown of the genus Hirtel/a (Chrysobalanaceae).Table 2, should serve to demonstrate the complexity and variation in distribution in a moderately large genus (ca.80 species).In order to find significant patterns.one must study more than one group.One would not expect exactly the same distribution in the same genus where some degree of genetic isolation must be involved to fi 11 many niches.When we find t he same pattern in many families then we may begin to draw some conclusions .
The reason for the vegetational diversity of Amazonia, a region that at first sight appears to be rather uniform are threefold :

. THE HISTORY OF THE REGION
As has already been pointed out, Haffer has recently attempted to explain the distribution of birds based or. the climate changes of the Pleistocene.There is now much evidence from ali over the world for considerable climate changes during the Pleistocene.So far little has been published for Amazonia, but there is an increasing amount of data such as Van der Hammen (1972).based on palynology.Simpson Vuilleumier (1971) gave a useful summary of data relating to climatic variatiotn in South America during the Pleistocene.This ties in with data from Africa, a continent whosc history is better known than the history of South Ame rica.Data on climatic changas in Africa were well summarised by Moreau ( 1966) in his study of the bird faunas of Africa.lt is interesting that in both Africa and South America ornithological data have tended to precede studies of the plants, which should yicld even more inform::ttion because of their ,• more static nature.However, we can now say with certainty that during the Pleistocene considerable climatic changes occurred in Amazonia.These changes included long dry periods which must have reduced the area of rain-forest and.consequently.increased the amount of savanna and other more xeric vegetation types.

PRESENT DAY RAINFALL AND CLIMA TE
Figure 2 shows a rainfall map for northern South America based on the work of Reinke , and also reproduced by Haffer, which shows the considerable variation in rainfall over the Amazon region.lt is, however.a rough estimation based on extremely variable and inconsistent data.This map was used by Haffer to help to determine the location of forest refuges.Obviously the difference between 1000mm and 3000mm of rain will have a profound effect on the vegetation.lt is noteworthy that the present day large savanna areas of Amazonia and Venezuela mostly fali within the drier areas under the 2000mm isohyet.Plant distribution studies of the region must take into consideration this rather large variation in rainfall, but at the same time we must remember that the rainfall data are scanty and we still have only an approx:mation of the isohyets.

OTHER FACTORS CAUSING HABITAT DIVERSITY
There are many other factors involved in forming habitat diversity in Amazonian species such as the geological features.e. g. the crystalline shield on the borders of the region, the distribution of sand, laterite.clay, etc.And perhaps one of the most important factors is water.My studies of woody species show that many plants grow only in temporarily flooded areas whereas others grow only in non-flooded areas.Species pairs with this type of distribution can be found in many woody families, for example the closely related species pair Caryocar microcarpum of flooded forest, and Caryocar glabrum of forest on terra firme.Looking at a distribution map based on collections made mainly near the major rivers one might assume that these two species are entirely sympatric.However this ;s not so.and in addition they are genetically isolated by having a different flowering season.I have found such pairs of species in ali four families of plants considered here : In the Chrysobalanaceae, Licania macrophylla on flooded ground and its closest relative in the family, Licania oblongifolia on terra firme .Some of the factors contributing to habitat diversity were summed up in Prance (in press).

DISTRIBUTION PATTERNS IN THE FOUR PLANT FAMILIES
My data for the Chrysobalanaceae are presented .in the form of a phytogeograph1c breakdown of the species of the two largest genera in Tables 2 and 3, anü in a selection of distribution maps with descr;~tive legends for ti'le Dichapetalace~.Caryocaraceae and Lecythidaceae.Many more maps could be given for the Chrysobalanaceae, but tiley would show much the same as the maps inciuded.
The refuge areas accepted here are based on the distribution of ali lowland forest species of South America in the four families.

. CHRYSOBALANACEAE
Data for the genera Hirtel/a and Ucania are given in Tables 2 and 3.As is typical of the large and predominantly lowland forest genera, there is much variety of distribution pattern, indicating a complex phytogeographic history for the region.There are severa I areas of particularly repeated endemism in each of the genera which correspond well with refuges proposed here in the final section.
One of the interesting features in Licania, a large genus of over 150 <>pecies, is the relationship of the savanna species within the genus.Table 3 lists, in addition to the phytogeographic areas, the distribution within the nine sections of the genus.The sections are listed in an arder which ranges from what I consider to be the more primitiva (section Moquilea) gradually developing to section Licania which is more advanced, (apetalous, many fewer stamens etc.).The few savanna speci es of Licania are ali members of the more advanced sections of the genus indicating that they are derived from what are basically forest species.The predominance of lowl and forest species shows that the divergence within the group has mostly taken pl ace within the lowland forest habitat.Something like the dry climatic periods with isolation of species into separate populations, and subsequently separate species, is needed to explain this lowland species diversity in an area where there are no real geographic barriers for genetic isolation.Some of the ochlospecies (species of wide distribution and a polymorphic phenotype) are easily explained by a certain degree of isolation into populations which did not develop sterility barriers during the time of isolation.Good examples of this are Lícania heteromorpha Benth.and L apetala (E.Mey.) Fritsch, see Prance ( 1972).Since there are no species between t hose on Figure 7, there was a probable isolation of a more continuous dist ribution at some stage of its history, possibly by t he drier climate of the Pleistocene.The present day distribution is only in some particularly wet areas of South America.There is much phenotypic plasticity in this species, typical of w idespread species that have probably been scattered into isolated popu!ations during the drier periods.3. CARYOCARACEAE Fig. 13-16.Ali species of the genus Caryocar.

DICHAPETALACEAE
• C. microcorpvm Ffg. 13 -Dlstrlbutlon of Caryocar microcarpum Oucke, a w idespread specles of annually flooded florest (varzea forest).Specles of flooded areas were not so atfected by drier periods because of rlverlne forest and rapid abillty to re-disperse by water.
The sub-specles lndicate that some diversificatlon has taken place through lsolation of individual populations.
-15    Kuntze, a coastal forest species which ls another example if a predomi nnntly Amazonian qenus with specles in the coastal forest also.estrellensis (Raddi) Kuntze with the interesting dlstribution from Acre to southern Brazil.This ls a species found In the Planalto and drier regions, and its distribution shows how it has malntained a more or less continuous dlstribution through gallery forest.Other species.e. g.Couratari macrosperma (Fig. 22) , have become dis-contlnuous, and in others there h as been further evolution resultlng In speciation.e. g.In Stephanopodium (Fig. 7 and 8) and Caryocar (Fig. 15 and 16) .-21

CONCLUSION REFUGES PROPOSED FROM THE STUDY OF THESE FOUR PLANT FAMILIES
A summary of the refuges proposed here is given in Fig. 24.The main difference from both Vanzolini and Haffer is that I think that they both have reduced the size of the poss ible refuges too much, (compare Figures 1 and 24), placing too much on the periphery of Amazonia.From plant distribution it certainly seems apparent that Vanzolini's f ive small areas of refuges could not explain the diversity of the present day distribution, nor could it have provided opportunity for the diversity of species to -evolve.Haffer's areas corresoond much more to the speciation cf lowland forest plant species.However, it seems dubi:)US whether the forest was ever as reduced at any

24-
one time, in the Pleistocene and post-Pieistocene climate changes, as Haffer proposes.lt should be rememberE::d that drier periods probably occurred at least tbree times, and therefore it is possib le that there was some variation in the refuge areas of forest, and that certain areas were smaller at different times.The sixteen refuges proposed here expand, rather than alter drastically, the refuges proposed by Haffer, and they certainly agree with his main hypothesis.The main differences which I find in plant distribution are : 1.The Napo area of Haffer is too small, and probably should be extended eastwards and to the north.Adjoining this a new refuge is proposed south of the Amazon River around São Paulo de Olivença, and another further east around Tefé.Both areas h ave an unusually large number of rather restricted lowland endemic species of pl;;~nts.
2. In ali four plant families studied there is very little endemism in or near the area of Haffer's proposed refuge Madeira-Tapajóz.Conversely, there are many local species in Rondônia and the neighbouring part of Amazonas as far as the Rio Roosevelt.Consequently, I have proposed a different refuge, Rondônia-Aripuanã, which fies to the west of Haffer's refuge of Madeira-Tapajóz.
3. There is an extremely high concentration of local species around Manaus which cannot be explained by migration of the species from the Guiana refuge.lt seems likely thé)t there was a large and i mportant refuge just to the north of the area flooded by sea-water.This area was directly north of Manaus.Tha receding sea-water brought southwards migration of the species to the present day position of Manaus.There is apparently little contact with the Guiana refuge, although today there are a number of plant species found in both the Guiana and the Manaus regions.A much commoner relationship for the Guiana species is with the Belém area, extending eastwards into the Amazonian part of Maranhão.The number of local species around Manaus is unlikely to have spread from a refuge south of the river, as the water has always acted as somewhat of <:! barrier in this respect.
4. Haffer's map of refuges left a few small areas along some of the main rivers as gallery forest refuges.Judging from plant distribu'•ions, I suspect that these gallery forest areas played a more important part than is indicated by Haffer in providing refuges and also some genetic contact between areas, as well as a route for the dispersai of the diaspores of various species.Today many of the species of the v::~rzea forests are the most widely distributed in Amazonia, partially because of the persistence of gallery forest in the dry times, and partly because of the ease of diaspore dispersai by water Today we have a similar situation in the area presently covered by cerrado vegetation in the Planalto of Centra l Brazil.A number of Amazonian species are distributed throughout the gallery forests of the Planalto and provide a link to the coasta l forests.Fig .19 shows the distribution of Cariniana estrellensis, a forest species in Acre and near to the coast but also occurring in gallery forest of the Planalto.Couratari macrospermum (Fig. 22), on the other hand, has not maintained a continuous contact through gallery forest, and now occurs in two disjunct and isolated populations in Amazonia and the coastal forest.
In conclusion it can be observed that plant distribution appears to concur well with zoological evidence for the reduction of forest cover during the last Pleistocene and post-Pieistocene period.lt is to be hoped tha1 further investigation can be ~arried out in branches of botany which will produce further evidence, e. g. palynology, and perhaps in studies of the soils, and most especially investigation which will pr;)duce further comments on the distribution of other plant families widely distributed throughout the region.

ACKNOWLEDGEMEN'T'S
The field studies and research on the four plant families were supported by by the National Science Foundation Grants GB4641, GB7356, GB18655, and GB32575 which are gratefully acknowledged.I am gratefu l to the Director of the Instituto Nacional de Pesquisas da Amazônia for much assistance from the Instituto while in Brazil, and to the President and various staff members of the New York Botanical Garden.I am also grateful to Dr. Keith Brown for discussions concerning the forest refuges.

Fig. 3 -Tapura guianensis
Aubl.A typical widespread forest species of Amazonia.As is common in many widespread species there is considerable variation in leaf-size and shape, and other variable characters, perhaps reflecting isolation of various populations during drier periods

Fig. 4 -
Fig.4-Distribution of Tapura amazonica.A subsp.amazonica A Subsp. manausensis Prance.This distribution shows a widespread species on ~•he fringet of Amazonia.on hl.;h nonflooded fores"(.Subspecies manausensis ouly occurs in central Amazonia, and has become isolated frorn the rest of the species.

Fig. 7 - 11 Fig . 8 -
Fig. 7 -Distribution of the northern species of Stephanopodium.O S. aptotum Wheeler corrE:sponds with an area of high plant endemism, the Santa Marta refuge.A S. venezuelanum Prance is from the Paria refuge, and.6 S. angulatum (Little) Prance from the southernmost part of the Chocó refuge.e S. peruvianum is rather widespread In the Napo, Peruvian and Chocó refuge areas.

Fig. 15 -
Fig. 15 -Distribution of species of Ca••yocar.+ C. amygdaliferum Mutis was probably isolated in the Nechi refuge.A C. amygdaliforme G. Don is in the eastern Peruvian refuge A C. gracile Wittm, is confined to t he white sand forests of western Amazonia.()) C. pallidum A. C .Smith is a widespreod species in Central Amazonia.
e C. 1,uciferum L. is in the Guiana refuge area.O C. <Jdule Casar, of the coastal forest In Brazil, shows the lsolation of species in coastal forests.

Fig. 16 -Fig. 17 -
Fig. 16 -Dístribution of specíes of Caryocar.O C. ~1(HWin Prance is in the lmataca refuge area.;. C. villosum (Aubl.. Pers), ls a widespread species on non-flooded forest.The other species of the genus are outside the Amazon region In dtler areas in Central and southern Brasil.
of Cariniana.Á C. pyriformis M iers a species of the Nechi and Cat umbo refuge area (probably lntroducad to Trinidad).e C. domestica (Mart.)M iers, a southern Amazon specles.0 C. sp.nov. a species o f the Rondônia refuge area.O C. rubra Gardn.ex M lers a Planalto species, and.Q C. janierensis Knuth, a coastal species .

Fig. 21 -
Fig. 21 -Pictorialised distribution map o f Couratari ri paria Sandw, and C. tenuicarpa A. C .Smith showing a pai r of closely riverine species which have dlHerentiated sufficlently to be recognized as species, and were probably lsolated during drier periods .

Fig. 22 -Fig. 24 -
Fig. 22 -Oistribution ot species of Couratari.Q C. stellata A .C .Smith, a species confined to northern Amazonia.e C. macrosperma A. C. Smith , the species most closely related to C. stellata which ls confined tosouthern Amazonía and the coastal area.This shows probable isolation of the •two species during drier periods, and also the interesting isolation between Amazonia and the Brazil ian coastal forests of two allopatric populations within the same species.A C. pyramidata, another closely related species, is endemic to t he Rio de Janeiro area.
Fig .3-Tapuraguianensis Aubl .A typical widespread forest species of Amazonia .As is common in many widespread species there is considerable variation in Jeaf-size and shape, and other variable characters, perhaps reflecting iso• lation of various populations during drier periods.

Table J •
Main types of prescnt day habitats in thc Amazon Basin.