Past vegetation changes in the Brazilian Pantanal arboreal-grassy savanna ecotone by using carbon isotopes in the soil organic matter

Measurements of the organic carbon inventory, its stable isotopic composition and radiocarbon content were used to deduce vegetation history from two soil profiles in arboreal and grassy savanna ecotones in the Brazilian Pantanal. The Pantanal is a large floodplain area with grass-dominated lowlands subject to seasonal flooding, and arboreal savanna uplands which are only rarely flooded. Organic carbon inventories were lower in the grassy savanna site than in the upland arboreal savanna site, with carbon decreasing exponentially with depth from the surface in both profiles. Changes in ^^C of soil organic matter (SOM) with depth differed markedly between the two sites. Differences in surface SOM "C values reflect the change from C3 to C4 plants between the sites, as confirmed by measurements of '^C of vegetation and the soil surface along a transect between the upland closed-canopy forest and lowland grassy savanna. Changes of ^^C in SOM with depth at both sites are larger than the 3-4 per mil increases expected from fractionation associated with organic matter decomposition. We interpret these as recording past changes in the relative abundance of C3 and C4 plants at these sites. Mass balances with ^'*C and "C suggest that past vegetational changes from C3 to C4 plants in the grassy savanna, and in the deeper part of the arboreal savanna, occurred between 4600 and 11 400 BP, when major climatic changes were also observed in several places of the South American Continent. The change from C4 to C3, observed only in the upper part of the arboreal savanna, was much more recent (1400 BP), and was probably caused by a local change in the flooding regime.


Introduction
The Pantanal, with an area of c. 140 000 km^, is one of the largest fioodplains in the world. It is a large depression formed at the end of the Cretaceous and filled by sediments deposited mainly by the Paraguay river and its major tributaries, especially during the Quaternary (Moreira 1977; Moreira Franco & Pinheiro 1982;Ab'Saber 1988). It is located at the centre of South America, bordering the main physiographic zones of the continent.
Correspondence: Reynaldo Luiz Victoria, fax +55-194-228339, e-mail reyna@pintado.ciagri.usp.br with the Amazon region to the north, the Cerrado to the east and the Chaco to the west. Due to the alternation of dry and wet climates that occurred at the end of the Pleistocene and during the Quaternary (Alvarenga et al. 1984), a mixed vegetation pattern was established, with species characteristic of the border areas (Amazon, Cerrado and Chaco) and with very few endemic species Ounk 1993). Drier periods also fostered the invasion of xeric species from the Caatinga of NE Brazil (Ab'Saber 1988). The distribution of the vegetation in the landscape is a function of the topography, which also governs the pattern of flooding. Consequently, small changes in the 1995 Blackwell Science Ltd. topography lead to a mosaic of habitats (Junk 1993), supporting numerous species of plants and animals, in one of the most biodiverse ecosystems of the world. The major vegetation type in the Pantanal is the savanna, in which different sub-units are established as a function of the local topography/hydrology (Loureiro et al. 1982). Plant communities may be classified based on the relative proportion of C4 (grasses) and C3 plants. The low-land areas are dominated by C4 grasses forming the so-called grassy savanna. Higher elevation areas have an increasing proportion of C3 plants with the progression from park savanna to open canopy arboreal savanna. Closed canopy arboreal savanna, formed almost exclusively by C3 plants, dominates the highest elevations.
Spatial distributions of these vegetation are expected to be dynamic in a system like the Pantanal, both because of periodic flooding, and past climate changes.
In this paper we deduce present and past changes in the vegetation composition of a typical savanna of the south-eastern part of the Pantanal, using '•'C isotopic composition and radiocarbon dating of soil organic matter (SOM). The bases for the use of the '•'C technique are: (i) the isotopic composition of the SOM is similar to the isotopic composition of the vegetation cover from which it is derived, and (ii) there is a large difference in the isotopic composition of C3 and C4 plants (Vitorello t' f al. 1989;Martin et al. 1990;Ambrose & Sikes 1991;Mariotti & Peterschmitt 1994). Generally, the SOM in deeper layers of a soil is enriched in ^-^C (Flexor & Volkoff 1977;Volkoff & Cerd 1987;GuiUet et al. 1988;Desjardins et al. 1991). Depending on the degree of this enrichment two basic processes can explain the observed trends: (i) increases smaller than 3-4%o are likely due to organic matter decomposition, which preferentially removes '•^, enriching deeper, older soil layers in '•'C; and (ii) larger values are likely the result of changes from C4 to C3 vegetation. Radiocarbon dating is used here to identify past vegetation change.
We collected samples from soil pits dug in grass and in closed canopy arboreal savannas. Virtually no grasses were present in the arboreal savanna. The absence of significant change in the stable carbon isotopic composition with depth, for any of the pits, would indicate that the system has been stable for the period over which the soil has formed. This period is determined from the radiocarbon age of SOM in the profile. A '^C enrichment with depth in the arboreal savanna, larger than the enrichment of 3-4 %o that typically results in decomposition, would indicate that the system evolved from a grassy savanna. Alternatively, a ''^C depletion with depth in the grassy savanna, would suggest that an arboreal savanna was present before being replaced by C4 grassy savanna.

Study area
The Pantanal may be divided into 10 different geomorphological regions (Adamoli 1982). The Fazenda Nhumirim (18° 59'S and 56'' 39'W), our sampling station, is situated in the Nhecolandia region. The climate of the region is classified Aw, after the Koppen system. The annual average temperature is 25" C, and the annual average rainfall is 1280 mm. The dry season extends from May to September and the rainy season from October to April. The soil was classified as Areia Quartzoza according to the Brazilian classification, equivalent to an Inceptisol in the American classification system.
The samples were collected in a typical savanna complex. Grassy savanna, formed almost exclusively by C4 grasses, dominated the lower elevation terrain, which is periodically inundated. C3 plants as a proportion of the vegetation increased with elevation, to the exclusively C3 dosed arboreal savanna, situated in higher terrain c. 1.5 m above the grassy savanna and so avoiding periodical flooding. Two soil sampling procedures were used. The first was a 950 m transect from the grassy to the arboreal savanna, where surface soil samples (0-20 cm) were collected every 50 m. The second was performed in soil pits dug in both the grassy and the arboreal savanna. Samples were collected at 5 cm intervals from the surface to 20 cm deep, and then at 10 cm intervals to the bottom of the pits (110 cm in the grassy savanna and 170 cm in the arboreal savanna). Following a botanical survey of the area (Pott ct al 1986) leaves from the most common plants in the grassy savanna, in the forest and in the border of the ecotone were collected for ' -*C determination. In the arboreal savanna leaves were collected from the upper canopy (10-15 m) and also from the ground flora.

Analytical methods
Large roots and debris were separated from the soil using a 2 mm sieve. The sandy texture of the soil allowed manual separation of small roots by hand. Carbon concentrations were determined in a CHN analyser (Perkin Elmer, 2400 CHN), Sample treatment for isotope analysis was done through combustion with CuO in evacuated Pyrex tubes. Isotope measurements were performed with a Finnigan DELTA-E mass spectrometer fitted with double inlet and double collector systems. Results are expressed in 5^^C relative to the PDB standard, defined as: \ ^standard where Rgampie and Rstdnd.ird ^rt' the ratio '-'C:'^C of the sample and standard, respectively. Samples were analysed at least in duplicate with a maximum difference of 0.3 %o between replicates.
'*C activity was measured in an accelerator mass spectrometer at the Lawrence Livermore Laboratory, USA (Southon ct al. 1992;Trumbore 1993). The conventional radiocarbon age, expressed as years before present (BP) ± standard-deviation, called Apparent Mean Residence Time (AMRT), was estimated using the following equation, assuming a '"'C half-life of 5568 years (Stuiver & Pearson 1986): where / is the radiocarbon age expressed in years BI', and -^siimple 3nd '^stand.ud ^^-e the radiocarbon activity of the sample and standard, respectively.
For the ' •*C mass balance presented in the Discussion section, the '*C activity was expressed as A^'C per mil, estimated according to the following equation (Stuiver & Polach 1977): Carbon stocks in soils were calculated by multiplying the respective concentrations by soil density and by soil depth. Soil density was calculated by collecting a mass of soil in a small cylinder of known volume. After drying to constant mass, the density was calculated.

Soil organic carbon concentration and stock
Organic carbon concentrations were very tow in both areas, and showed the typical decrease with depth. For any particular depth the concentration was higher in the arboreal than in the grassy savanna (Fig. 1). The higher carbon stocks in the arboreal savanna, resulted from a higher carbon concentration, since the soil densities from both areas were not significantly different. The carbon stock to 1 m depth was equal to 3.4 kgC m~^ in the arboreal savanna and equal to 2.0 kgC m"^ in the grassy savanna.
in the forest floor (-31.1 ±1.5 %o, n = 6) vegetation. This enrichment was also present in the tropical environment (van der Merwe & Medina 1989;Stemberg et al. 1989;Kapos et ai 1993). The probable causes are, in order of importance, lower irradiance inside the forest, increasing the isotopic fractionation during photosynthesis, and to a minor extent, the use of isotopically depleted biogenic CO2 (Farquhar et al. 1982;Schleser & Jayasekera 1985;Sternberg et al. 1989). Surprisingly, the difference between the upper canopy and ground flora of the arboreal savanna was not statistically different, as usually found in tropical forests (van der Merwe & Medina 1989;Martinelli et al. 1991). The small number of forest floor vegetation samples, together with the high variance of the data, may explain this lack of significance. Also, the arboreal savanna at this site was more open with only two strata, in comparison to forest of higher rainfall areas, with less potential for recycling of C within the canopy.

Carbon isotopic composition of soil organic matter along the transect
The variation of 5^^C along the transect followed the anticipated pattern: as the proportion of C3 to C4 plants increased, the values became more negative, reflecting the increasing contribution of C3 derived organic matter in the SOM (Fig. 2). Two major changes were observed in the transect, the first after 350 m, where the 5''^C values dropped from -17 to -20 %o, and the second at 600 m, where a drop from -20 to -27 %o occurred. These changes are the results of changes in vegetation cover. The first drop indicates the fransition from grassy savanna to park savanna, and the second the transition from park to arboreal savanna.

Carbon composition of plant species
The h^^C values of the plant species collected along the transect are shown in Table 1. The dominant grass in the grassy savanna, Elyonurus miticus, showed a value typical of C4 plants (-12.4 %o). The trees from the border area showed an average 5''*C value of -28.8±0.9 %o (n = 10), which is significantly less negative than the average found in the upper canopy (-30.3±0.9 %<,, n = 11) and

Variation in carbon isotopic composition with soil depth and radiocarbon dating of the soil
The variations in d^Hl with depth, for both areas, were larger than the expected from fractionation during decomposition, suggesting that vegetation changes occurred in the past (Fig. 3). In the grassy savanna the S'' C became progressively more negative with depth, suggesting that a progressive replacement of C3 by C4 plants had occurred.   grassy savanna than in the arboreal savanna (Fig. 3). At 60-70 cm depth the SOM was 1940±40 BP for the grassy savanna, and approximately three times younger (640±40 BP) for the arboreal savanna. The same ratio was maintained at the 110-120 cm depth interval, with values of 4570+40 BP for the arboreal and 1480±50 BP for the grassy savanna. At the bottom of the arboreal savanna soil pit, the radiocarbon age was 2000±40 BP (Fig. 3).

Discussion
The name Patitatml (swamp) may give a false impression of an area permanently flooded, with poorly drained soils promoting the accumulation of organic matter. Actually, most of the Pantanal is only seasonally flooded, and some areas, like the arboreal savannas, are only rarely flooded. In addition, the sandy texture of the soil facilitates good drainage. These, combined with the alternation between aerobic and anaerobic conditions which fosters organic matter decomposition (Lugo et ai 1990), are the most likely reasons for the observed low organic carbon concentration in both environments. The arboreal savanna soil showed higher carbon concentrations and stocks, probably reflecting higher carbon inputs from vegetation, and less frequent flooding events. However, compared to other soils, the carbon stock was much lower. Eor instance, Areia Quartzoza soils collected in the Amazon basin (Moraes 1991) showed a carbon stock approximately three times higher than the arboreal savanna. The cause of vegetation changes in the past could be attributed either to changes in climate, or to a local change in the flooding regime. It is difficult to identify the exact cause of such changes. For Instance, the chronology of past climate change in the Pantanal is not well understood (Ab'Saber 1988), although there is evidence of past climatic fluctuations, from drier to wet regimes and vice-versa, a couple of times during the Quaternary (Alvarenga cl al. 1984). Obviously, a change caused by a climatic fluctuation would equally affect both areas, as they are adjacent to each other. Eor regions where the climate has become drier for a period, as in the Amazon, it is expected that the establishment of a C4-rich vegetation is facilitated (van der Hammen 1974;Absy & van der Hammen 1976;Absy 1980;Markgraf 1989;Bush ct al. 1990;Absy et al. 1991).
Soil organic matter is often modelled as a two-or threecomponent system, a mixture of more rapidly cycling fractions with turnover times ranging from less than one year to decades, and passive organic matter, with millennial or longer turnover times (Harrison et al. 1993;Jenkinson & Raynor 1977;O'Brien & Stout 1978;Parton el al. 1987;Schimel et al. 1994;Townsend cl al. 1995;Trumbore 1993). Trumbore et al. (in press) have shown that this picture holds true even at depths > 1 m in tropical soils, as deep roots feed an active carbon cycle (turnover times of < 40 years) while the bulk of the carbon is in old, refractory organic carbon. Thus, the AMRT associated with a ^^C age, even in deep soils is most likely a mixture of decadal and faster cycling organic matter (with '^C values > Modern) and old, refractory carbon (with '"'C values substantially less than Modern). In a soil profile where vegetation has changed with time from CJ to C4-dominated plants, we assume that the '\l! changes in the SOM may be used as a measure of the minimum amount of rapidly cycling carbon (as these have adjusted to a new steady state within a century of conversion), and that residual '^C-depleted carbon represents the passive carbon accumulated over millennia in the old vegetation regime. We used this approach to calculate the possible timing of vegetation change in this specific site of Pantanai. The change from arboreal savanna to grassy savanna (less negative S'^C towards the soil surface) can be analysed on the results from the grassy savanna soil profile. We may assume that the carbon present at the 110-120 cm depth is a mixture of carbon of different ages: an old carbon fraction, originated from C3 plants (AMRT > 4572 BP, which is the estimated age at this depth interval in the grassy savanna profile), and a younger fraction derived from C4 plants, added since the last vegetation change, and therefore with AMRT < 4572 BP. Therefore, the minimum AMRT of the old C3 carbon at the 110-120 cm would be c. 4572 BP, and no mixture would have taken place.
The maximum AMRT can be calculated through the following ^''C mass balance: where %C3-(Al^old,,rt,on) i| .carbon is the ^*C activity at the 110-120 cm soil depth interval in the grassy savarma profile (-435 %o), ' ' *^new-carbon is the '*C activity of the new C4 carbon, taken as the ''^C activity at the soil surface in the grassy savanna (+ 141 %«), and A'' Coij^.j-arbon is the ^""C activity of the old C3 carbon that we want to estimate. The %C4 and %C3, the relative percentage of C4 and C3 carbon in the soil profile, were estimated by '''C mass balance, through the following equation:

100,
where S^^Igampie is the carbon isotopic composition of the soil at the 110-120 cm depth in the grassy savanna (-22.3%i>), 8^^c4 is the carbon isotopic composition of the C4 material that is being added to the soil (-12.4 'Hm, or the *^C isotopic composition of the C4 vegetation cover), and 5'^Cc3 is the carbon isotopic composition of the C3 material in the soil (-27.9 %o, assumed to be 2 %o less negative than the average isotopic composition of C3 plants,-29.9±1.4 '%«>, n = 29). The proportion of C4 plants is 100 -%C3.
Using (4) and (5) we calculated a maximum AMRT equal to 11 420 BP. Therefore, the vegetation change from arboreal to grassy savanna probably occurred between 4572 and 11 420 BP. The two major dry periods reported to have occurred in tropical and sub-tropical regions of South America were from 4000 to 5000 BP, and from 10 400 to 10 500 BP (Absy & van der Hammen 1976;Absy 1980;Absy et ai 1991;Servant el ai 1993), which agree with the range found in the Pantanal soil.
The same approach was used to estimate the timing of the change from grassy savanna to arboreal savanna, indicated by the results of the 50-60 cm depth interval in the arboreal savanna soil profile (Fig. 3). In this case, however, as the C3 carbon increased 50-60 cm depth towards the soil surface, old carbon was assumed to be of C4 origin, and new carbon of C3 origin. In this case the data was analysed based on the results of the 30-40 cm depth interval in the arboreal savanna soil profile.
The following values were used in (4): + 51 %c. for S'*Cs,,jiH:aHx.n. + 141 %o for 5^ *Cnew-carbon-3"^ 5'*Co Id-carbon is the ''*C activity of the old C4 carbon that we want to estimate. In (5), the same values of S^^C for vegetation types were used, and the S^^C of the soil samples at 50-60 cm depth inten.'al was equal to -21.1 ' J(» The maximum AMRT was then estimated to be equal to 1390 BP, suggesting that the change from grassy savanna to arboreal savanna was a recent local event. In parallel with suggestions for other areas in the world (e.g. MacPerson et al. 1993), wtxKiland is expanding over the grassland at this specific site of the Pantanal. However, we also observed that the highly dynamic characteristic of the flooding regime may cause large variations in the stable carbon isotopic composition, and in the radiocarbon age as well. Therefore, the results presented here should not be extrapolated to other areas of the Pantanal region, without further studies. Overall, the findings reveal the dynamic nature of the spatial distribution of C3 and C4 savanna sub-communities and the coincidence of vegetation change with periods ot climate change.