Past Vegetation Changes in Amazon Savannas Determined Using Carbon Isotopes of Soil Organic Matter1

Abstract We investigated the variation of stable (δ13C) soil carbon isotopes in relation to depth in seven of the most important savanna areas to adjacent contiguous forests in the Amazon region. The δ13C of bulk organic matter in all profiles from forested sites increased with soil depth. In forest profiles from Amapá, Alter do Chão, and Roraima, the enrichment was less than 3.5‰ between deeper soil and surface layers, suggesting that C3 plants have remained the dominant vegetation cover. On the other hand, in forest soil profiles from Humaitá and Carolina sites, the δ13C enrichment was greater than 3.5‰, indicating the influence of past C4 vegetation or a mixture of C3/C4 vegetation (woody savanna). The surface δ13C values in the savanna profiles were 5–13‰ greater than the comparable forest profiles, indicating the influence of C4 vegetation. Two kinds of isotopic distribution were observed in deeper layers. The savanna profiles at Alter do Chão, Chapada dos Parecis, and Redenção had relatively constant δ13C values throughout the profile, suggesting minor past changes in the vegetation composition. In profiles at Amapá, Roraima, Humaitá, and Carolina, δ13C values decreased with depth from the surface and converged with comparable forest values, suggesting more woody savanna in the past than exists currently.

Several lines of evidence, including paleolimnological (van der Hammen 1972, Absy & van der Hammen 1976, van der Hammen 1983, Absy et al. 1991, paleofaunal (Rancy 1993), and the carbon isotopic composition of soil organic matter (Desjardins et al. 1996, Pessenda et al. 1996, Gouveia et al. 1997, suggest that the dynamics of expansion and contraction of savanna regions result from climatic fluctuations during the Quaternary period (Prance 1982, Sarmiento & Monasterio 1975. Several authors have pointed out that there were drier periods during the Pleistocene and the Holocene periods, when tropical forests were replaced by savanna-like vegetation having a predominance of grasses (van der Hammen 1974, Absy & van der Hammen 1976, Ab'Saber 1977, Absy 1980, Bigarella & Andrade-Lima 1982, Leyden 1985, Markgraf 1989, Bush and Colinvaux 1990, Bush et al. 1990, Absy et al. 1991, Markgraf 1991. The middle Holocene, from ca 6000 to 4000 years B.P., was identified as one of such drier periods in several places of South America, including the Amazon region (Absy 1980;Markgraf 1989;Servant et al. 1989;Absy et al. 1991;Ledru 1992Ledru , 1993Servant et al. 1993). The presence of charcoal dated from ca 3900 to 1800 years B.P. found in the northern Amazon (Desjardins et al. 1996) suggests the occurrence of more recent climatic changes in that region as well as in other regions of Amazonia (Servant et al. 1993).
There are seven major savanna areas in the Amazon region; in two of them (Roraima and Ronddnia), the past vegetation dynamics were investigated by using the carbon stable isotopic composition of soil organic matter (Desjardins et al. 1996, Gouveia et al. 1997, Pessenda, Gomes et al. 1998, Pessenda et al. 1998a. In both places, major vegetation changes occurred in the past; these studies inferred that savanna areas replaced forested areas of the early Holocene period during the middle Holocene. In this study, we investigated whether or not similar vegetation changes occurred in other major savanna areas of the Amazon. We compared soil carbon isotopic composition and its variation with depth in seven of the most important savanna areas with nearby contiguous forests in the Amazon region. Since the main control of the S13C in soil organic matter is plant litter inputs, and C3 plants  (7) Carolina.
dominant plants in the savannas) are isotopically distinct, it is possible to detect shifts in tropical forest zones to grassland (or vice versa) from the 6l3C signature of organic matter in soils (Desjardins et al. 1996, Neil1 et al. 1996, Bird & Pousai 1997.

MATERIAL AND METHODS
SOIL smPLrm.-Sampling locations for forest-savanna comparisons are identified in Figure 1. Four of these sites (Amapi, Alter do Ch5o, Humaiti, and Roraima) are "isolated" savanna pockets that are completely surrounded by forest, and three (Chapada dos Parecis, Redengo, and Carolina) are classified as non-isolated savannas. Soils in the sites generally showed a dystrophic character and with 1:l clay type (Table 1). Climatic conditions were similar at all sites, most of them classified as equatorial hot humid (Nimer 1989). Rainfdl varied from 1500 to 1750 mm in Roraima and Carolina up to 2750 mm in Amapi and Humaith (Table 1). The regional vegetation type was very uniform for forest sites, tropical dense forest, following the classification of RADNBRASIL (1 974), or terra Jrme forest according to Pires and Prance (1985). The regional vegetation types of savanna sites were more diverse, varying from savanna park to open woody savannas ( Table 2). The major difference between savanna park and open woody savannas is that in the latter type there is a higher number of trees (RADAMBRASIL 1974). The relative proportion of grasses (C4) and trees (C3) is important, since this proportion w i l l directly affect the soil sur-  Table 2 shows the number of trees per hectare with diameter at breast height (DBH) greater than 5 cm ( N ) and also the basal area (BNha) in the savanna sampling sites (Sanaiotti 1996). Nvaried from 62 (Amapi) to 312 individuals per hectare (Redengo). The lowest BA was observed in Redengo and the highest in Alter do Chso (Table 2). Soil cores were taken using a 4 m auger. In each study site, we sampled five depth intervals (0-0.05, 0.1-0.2, 0.5-0.6, 1.0-1.2, and 1.4-1.6 m) in two soil cores in the savanna and two in the surrounding or dosest forest. We sampled just one of the two cores in the savanna from the soil depth interval 1.4-1.6 m to the bottom of the soil pit. In contrast, in the forest we continued depth sampling at the following intervals: 1.8-2.0, 2.2-2.4, 2.6-2.8, 3.0-3.2, 3.4-3.6, and 3.8-4.0 m. The exception was Parecis, where we sampled only in the savanna. The samples were air-dried and sieved (<2 mm) to remove roots. The forest and savanna core sites were chosen on flat ground in undisturbed vegetation well away from the present ecotone (savanndforest boundary). The distance from the savanndforest boundary to the area of the forest sampled varied from 0.2 to 23 km. In the case of the isolated savanna sites, we sampled close to the center of the savanna vegetation. As isolated savannas varied in size from several hectares to tens of square km, the distance from the ecotone LABORATORY 613C AND '4C ISOTOPE MEASURE-MENX-A 10 g subsample of the sieved soil was combusted with CuO in sealed evacuated Pyrex tubes for 12 hours at 900°C. The resulting C 0 2 was purified cryogenically and stable isotope measurements were made with a Micros 602E mass spectrometer (Finnegan Mat, Bremen, Germany) fitted with dual inlet and double collector systems. The results are expressed in 6l3C relative to the PDB standard in the conventional 6 per mil notation as: where R denotes the 13C:12C ratio of the sample and of the standard (std). All results represent the mean of at least two replicate analyses that differed by less than 0.3%0.
Radiocarbon analysis was performed on bulk soil organic matter samples collected in AmapA, Alter do Ch5o, Carolina, and Roraima profiles. Variation in (a) sand content with depth activity was measured in an accelerator mass spectrometer at the Lawrence Livermore Laboratory, Livermore, California (Trumbore 1993). The conventional radiocarbon age, expressed as years before present (B.P.) was estimated assuming a 14C halflife of 5568 years.

RESULTS
Son cwcrERmmoN.-The sand content of all savanna soils averaged higher than 50 percent, with all but the RedenGo savanna profile having sand contents averaging more than 80 percent (Fig. 2). Forest soil profiles were generally lower in sand than in the savanna areas ( Fig. 2), but still had average sand contents higher than 60 percent. The exception was the forest profile of Alter do Chgo, which had an average sand content of only 10 percent. Therefore, with few exceptions, soils were predominantly sandy soils (Fig. 2). Textural changes with depth were observed mostly in forest soil profiles, which showed generally decreasing sand content with depth (Fig. 2B). The exception to this pattern was RedenGo, where the opposite trend was observed (Fig. 2B). At all sites, the soil chemical analyses indicated low pH and very low nutrient contents, as indicated by the low sum of exchangeable bases ( Table  2). The pH varied from 4.4 to 5.1 (not shown) and sum of exchangeable bases (SB) from 0.18 to 0.64 mmolc/lcg in savannas soils, and from 0.48 to 3.90 mmolJkg in forest soils (Table 2). These values were below the average SB value of 184 profiles for Ultisols with different textures sampled in the Amazon basin (9.7 mmol,/kg) by the RADAM-BRASIL project (Tognon 1997 N = 34), Humaiti (-28.9 ? 1.5%0; N = 3 9 , and Parecis (-28.5 ? 1.7%0; N = 34). The single largest difference between foliar 6'3C averages was 2.1%0, which was observed between Alter do Chiio and Carolina. Some variability was found among leaves of individuals of the same species collected at the same site (Appendix 1). Most of this variability was smaller than 2%0. Leaves from individuals of the same species differed by less than 3%0, except in five cases. The largest single difference (5.3%0) was found between two individuals of Sal-ve& convallariodora that were collected at Carolina savanna.
The 6l3C values for surface forest soils (04.2 m) were similar among different sites and within sites. The minimum and the maximum values were -30.0 (Redengiio) and -27.0%0 (Roraima), respectively. These values are within the range of -31.0 to -25.0Yw reported for similar forest surface soils (Fig. 3). Spatial variation in 6l3C values (both among different sites or within the same site) was greater for surface savanna soils than for s d c e forest soil samples. These larger variations observed in savannas appear to be common to other savanna areas of the world (Fig. 3) and probably are due to site-specific recent variations in the proportions and spatial distribution of C3 and C4 plants (Boutton et al. 1998). Among savannas, 6l3C values in organic matter ranged from a minimum of -26.5 (Redengo) to a maximum of -15.8% (Roraima ;  Table 3). Within savannas, the largest 6% differences between two soil profiles (5%0) were observed at Carolina, Humaiti, and Redengo (Table  3). The 6l3C difference between profiles within a savanna was smaller in the second depth interval (0.1-0.2 m) in relation to most surface samples (Table 3). There was a significant correlation between the density of C3 individuals in savanna sampling sites (number of individuals per hectare) and the 6I3C values for surface (0-0.05 m) and subsurface (0.10-0.20 m) soil samples (Fig. 4). The density of C3 individuals explained ca 50 percent of the variance in the 613C of surface soil samples from savanna sites. Excluding the Roraima sample, 85 percent of the variance in the 0.1-0.2 m layer can be explained by the density of C3 individuals (Fig. 4).

DEPTH V A R I A B I L I~ IN 6l3C VALUES OF SOIL ORGANIC
MA~TER.-AS several other studies have shown (Desjardins 1991, Balesdent et al. 1993, Desjardins et al. 1996, Gouveia et al. 1997, Pessenda, Gomes et al. 1998, the 6l3C of    soil organic matter increases with depth in forests profiles. In this study, the changes in 613C values with soil depth can be divided into two groups: Alter do Chiio, Amapi, and Roraima, where deeper portions of the soil organic matter (>1.5 m) showed little enrichment (<3.5%0) compared to surface layers in forests; and (2) a group represented by Humaiti, Carolina, and Reden60 sites, where soil enrichment greater than 3.5%0 was observed with soil depth (Fig. 5).
As already discussed, organic matter 6l3C val-  5). consequently do not constitute an absolute age of mil organic matter. For example, radiocarbon dating of bulk soil organic matter was always younger &an humin extracted from the same samples collected in soil profiles at Humaitd (Gouveia et al. 1997). The same trend was observed between bulk soil organic matter and charcoal (Pessenda, Gomes et al. 1998). Therefore, the interpretation of radiocarbon ages in terms of the mean age of organic matter in soil profiles must be undertaken with caution (Trumbore et al. 1995). High values of radiocarbon (>Modern) reflect the incorporation of carbon fixed from the atmosphere since atomic weapons testing in the early 1960s, which nearly doubled the amount of 14C in the atmosphere. This "bomb" 14C signature is observed in the upper part of each soil profile (Fig.  5), indicating that the organic material in this layer is predominantly made up of constituents that cycle rapidly (decadal or shorter timescales; Trumbore et al. 1995). The I4C content of bulk organic matter declines rapidly with depth in the soil, indicating that organic matter on average resides long enough for significant radioactive decay of radiocarbon (half-life = 5730 yr). The oldest organic matter was found at the bottom of the h a p 5 forest soil profile (ca 10,300 yr B.P.). The radiocarbon ages in the other profiles were generally less than in the Amapd profiles, and varied from ca 2300 to ca 6400 years B.P. in the deepest depths (Fig. 5).
In h a p i and Alter do Chiio, the radiocarbon ages were higher in the forest than in the savanna profiles. The opposite trend, however, was observed in Roraima, and practically no difference between forest and savanna was observed in Carolina.

6l3C VALUES OF LEAVES IN SAVANNAS AND F0WS"S.-
The average 6l3C value of C4 grass found in the savannas (-13.2%) was higher (heavier) than the average value for C4 plants found by Desjardins et al. (1996) at Roraima savannas (-14.2%0) but lower (lighter) than the average value found by Pessenda, Gomes et al. (1998) at savannas in Ron-dBnia (-11.7%0). The 6I3C values of ca 100 C4 plants analyzed from the herbarium of the Instituto Nacional de Pesquisas da AmazBnia (INPA) ranged from -9.5 to -13.6%0 (Medina et al. 1999), with an average value of -11.7 ? 0.9%0 ( N = 102). On the other hand, the average S13C value for C3 plants of the savannas (-29.0%0) was similar to values found by Desjardins et al. (1996) and Pessenda, Gomes et al. (1998) in tree leaves collected in Roraima (-29.6%0) and RondGnia (-29.0 ? 1.8%0) savannas, respectively. As in the Brazilian Pantanal region (Victoria et al. 1995), the 6l3C values for tree leaves collected in savannas are enriched in 6l3C compared with leaves collected from trees in closed forests. For example, the average value of tree leaves collected in closed forest in Ron-dBnia was equal to -32.1%0, ca 2%0 more negative than tree leaves from savannas . The most probable cause for these differences are increased incorporation of isotopically depleted COZ in forest tree leaves compared to savanna tree leaves (Sternberg et a/. 1989, Kapos et al. 1993, Grace et al. 1995, Buchmann et al. 1996, Lloyd et al. 1996. Intraspecies variation in leaf 613C was greater than overall variation among different sites. Large intraspecies variation in 6I3C is common (Walcroft et al. 1996, Berry et al. 1997), but its causes are difficult to identi@. Characteristics such as leaf height, age, and position in the branch expose leaves to different light intensities, which can influence the photosynthetic rate, and consequently, leaf stable carbon isotopic composition (Farquhar et al. 1989).

SOURCES OF VARlABILITY IN 6l3C VALUES OF THE SOIL ORGANIC
ianER.--Generally, forest soil profiles showed an enrichment of 6l3C with depth throughout the profile (Fig. 5). Savanna soil profiles showed similar isotopic enrichment only to depths of 0.5 to 1 m, below which 6l3C values decreased in most of the profiles (Fig. 5). The magnitude of such isotopic changes varied, both in forest soil profiles and among savanna soil profiles. Several factors have been identified that cause isotopic changes with soil depth, including soil chemical composition and texture, organic matter decomposition processes, and past vegetation changes.
The sampled soils had similar chemical characteristics as expressed in terms of fertility, especially among savannas, where the s u m of bases was very low. Therefore, chemical composition does not seem to be the major cause of isotopic changes found in soil profiles.
In forest stands, it has been observed that clay fractions are slightly enriched in W C , compared to coarser fractions of surface soil layers (Balesdent et al. 1993). For example, a difference of ca 4%0 was found between clay and sand fractions of a surface soil under tropical forest in southeast Brazil (Vitorello et al. 1989). Smaller enrichments were observed in surface soils of other regions in Brazil (Desjardins et al. 1991), the Congo (Martin et al. 1990), and France (Balesdent et al. 1993). Even pure stands of native C4 savannas showed a small enrichments (1-2%0) in fine compared to coarser fractions at the soil surface (Martin et al. 1990).
On the other hand, in situations where C3 stands were artificially replaced by C4 vegetation (Vitore-110 et al. 1989, Desjardins et al. 1991 or vice versa (Balesdent et al. 1988, Martin et al. 1990), larger differences between fine and coarse soil factions have been observed (up to 10%0; Boutton et al. 1998). Among soil profiles that had changes in texture with depth, no significant change in the 6'3C of the bulk soil organic matter was observed in studies reported by Desjardins and collaborators (Desjardins et al. 1991(Desjardins et al. , 1996. In our study, the same was true. In the few forest profiles that had depth changes of textural fractions, such as Roraima (Fig. 3), we could not detect any major change in 6l3C with depth (Fig. 5). Therefore, it seems that soil texture was not the main factor controlling the isotopic composition of soil organic matter.
Another possible cause of a13C variation with soil depth is the organic matter decomposition process, which favors the loss of 12C (Boutton 1996). Generally, it is accepted that up to a 3.5-4.0%0 isotopic enrichment in organic matter 6l3C with depth in soils is due to decomposition processes (Mariotti & Peterschmitt 1994, Desjardins et al. 1996. Increases in the average radiocarbon age of organic matter with depth demonstrate that organic matter in deeper soils has been exposed longer to decomposition processes.
Vegetation changes in the past may be another important factor in explaining isotopic changes with soil depth (Schwartz et al. 1986 . In forest soils, isotopic enrichment larger than 3.54.0%0 has been explained by the presence of remaining old C4 vegetation. On the other hand, in savanna soils, an impoverishment of 6'3C with depth may suggest the presence of relict organic matter derived from C3 vegetation. It is important to note that different rooting depths for trees and C4 grasses may partly account for changes with soil depth in woody savanna, especially in Reden@o, Roraima, and Alter do Chgo, where the number of C3 trees per hectare was high ( Table 2). Most of the studies dealing with changes in 6l3C with soil depth in savanna areas do not take into account the possible rooting effect on observed changes. One exception is the study conducted by Boutton et al. (1998) in a savanna area of south Texas. In that study, they found that 613C of roots were not in equilibrium with 6% values of coexisting soil organic matter. The 6% of roots in woodlands were ca -25%0, while 6I3C values of the soil organic matter varied from -24 to -17%0. The 6l3C of roots in grasslands varied from -24 to -21%0 and S13C values of soil organic matter varied from -16 to -20%0. Obviously, the results found by Boutton et al. (1998) cannot be extrapolated directly to our study areas. If the contribution of roots in our study is important, past climate changes may be overestimated or underestimated depending on the proportion of C3 and C4 roots.
The 6l3C values observed in the forest soil profiles of Alter do Chgo, Amapi, and Roraima are in the range of the expected values for soils where C3 plants have remained the dominant vegetation cover. In these profiles, the maximum 613C enrichment was ca 3%0 (Fig. 5). This trend was also shown for nine soil profiles collected at six forested sites not bordered by savannas in the central and southeast Amazon region (Desjardins et al. 1991, Valencia 1993, Trumbore et al. 1995. Enrichment in 6% values for forested sites greater than those that may reasonably be expected from decomposition processes, such as those found in Humaiti and Carolina profiles, have been interpreted as indicating the influence of past C4 vegetation, or a mixture of C3 and C4 vegetation. Desjardins et al. (1996) found soil enrichment higher than 3.5% in forested sites at depths varying from 0.5 to 2.0 m in areas near the forestsavanna boundary in Roraima. Below 2 m depth, the enrichment predominantly observed was less than 3.5%0 different than surface values. In another forest profile in the same area but far from the forest-savanna boundary, the isotopic enrichment was greater than 3.5%0 only between 0.5 and 1.0 m depth. The latter site was similar to the study at Roraima, where an enrichment higher than 3.5%0 was observed only for the 0.1 to 0.2 m depth interval. Pessenda et al. (1998a) observed similar 6I3C distribution with depth in the Humaid area. A transect of 6l3C values from savanna to forest area indicated decreased contributions of C4 plant inputs toward the forest.
If we assume that past vegetation changes remain recorded in soil organic matter, it may be concluded that in the Alter do Chgo and Chapada dos Parecis profiles appear not to have undergone major past changes in their vegetation cover, while major past vegetation changes are recorded in the Amapi, Roraima, Humaiti, RedenGb, and Carolina profiles. Interestingly, among the sites where carbon isotopes indicate past vegetation changes, the degree and manner of change was different for each area (Fig. 5). The most common pattern is that the S13C signatures of organic matter in the deepest soil layers of savanna profiles appear to record more woody (C3) vegetation than is present today. The same trend was found by earlier studies conducted in Roraima and Humaiti, especially in those profiles that were collected from the forestsavanna boundaries (Desjardins et al. 1996, Gouveia et al. 1997. Radiocarbon dating of bulk soil organic matter does not allow us to infer with precision the chronology of past vegetation changes because it represents a mixture of younger and older material (Trumbore et al. 1995). For example, the bulk radiocarbon age may be influenced by factors such as soil texture, since clays tend to stabilize organic carbon in soils for a longer time than sands. Despite these complications, radiocarbon data can be used to roughly relate the timing of vegetation change with climatic events that occurred in the past. Paleoecological studies have suggested that a maximum in the proportion of grass pollen was found in the middle Holocene (6000-4000 yr B.P.; Absy etal. 1991, Ledru 1993, Servant etal. 1993). Desjardins et al. (1996 found charcoal with radiocarbon ages ca 6000-7000 years B.P. in Roraima savanna profiles, and according to them, those would be indirect evidence that present savannas were formed during that period. The radiocarbon ages in the deepest savanna profiles of our study are not old enough to support or reject this hypothesis (Fig. 5); however, at 3.4 to 3.6 m depth intervals, the 8l3C savanna profile in Roraima clearly indicates a strong presence of C3 vegetation, and the radiocarbon age of the bulk soil organic matter was ca 6400 years B.P. Gouveia et al. (1 997) found at 90 cm depth in Humaiti, humin dating 6000 years B.P. with a corresponding 8l3C of soil organic matter of ca -21.00/0, which indicates a mixture of C3 and C4 vegetation. Therefore, it could be that even a widespread event like the suggested climatic fluctuations during the Holocene period did not produce changes in vegetation dis-tribution that were uniformly recorded in soil profiles. Probably other factors, such as microclimate, soil characteristics, root distribution, fire regime, and human actions, may have contributed to the composition of the vegetation cover in the forestsavanna ecotones of the Amazon. For example, based on charcoal dated from 3230 to 1790 years B.P., Desjardins et al. (1996) suggested that fires which occurred in the late Holocene period were a key process to define the forest-savanna dynamics in Roraima. In fact, in the majority of the savannas areas studied in the Amazon, maximum 8l3C values (higher proportion of C4 plants) were found near the surface, and where radiocarbon dating was available, they have always indicated that this maximum occurred in the late Holocene (Desjardins et al. 1996, Gouveia et al. 1997, Pessenda, Gomes et al. 1998). In addition, Roscoe et al. (2000), working in a savanna area of central Brazil, showed that in a period of only ca 20 years, the C4 grass population increased in areas with higher fire incidence, and the 8l3C of the soil became higher, suggesting the increasing influence of C4 vegetation in the soil organic matter.