Amazonian Dark Earths : pathways to sustainable development in tropical rainforests ?

Fertile dark anthrosols associated with pre-Columbian settlement across the Amazon Basin have sparked wide interest for their potential contribution to sustainable use and management of tropical soils and ecosystems. In the Upper Xingu region of the southern Amazon, research on archaeological settlements and among contemporary descendant populations provides critical new data on the formation and use of anthrosols. These findings provide a basis for describing the variability of soil modifications that result from diverse human activities and a general model for the formation of Amazonian anthrosols. They underscore the potential for indigenous systems of knowledge and resource management to inform efforts for conservation and sustainable development of Amazonian ecosystems.


InTRoDuCTIon
Amazonian dark earth (ADE) or terra preta (black earth) is a critical research topic for contemporary archaeology and historical ecology of Amazonia and potential strategies for sustainable development in tropical regions (Balée, 2010;Mann, 2002;Lehmann et al., 2003a;Glaser and Woods, 2004;Woods et al., 2009).Patches of human-modified, fertile ADE soil associated with pre-Columbian settlements provide vital clues as to the nature of prehistoric agriculture, landuse, and resource management that apparently sustained dense populations in some regions without significantly degrading environments or causing extinctions but rather enriching environments by concentrating fruit trees or creating patches of ADE (Balée, 1994;Graham, 2006).Amazonian dark earth also provides inspiration or a model for a developing controversial technology called biochar that is claimed will improve soil for agriculture and mitigate human-induced climate change by sequestering carbon in the soil (Ernsting and Smolker, 2009;Lehmann, 2007;Marris, 2006).The study of ADE can offer insights into sustainable environment and soil management in tropical regions mainly by revealing how ancient and modern indigenous Amazonians managed/manage their resources.
In most cases, the actual mechanisms of ADE formation are poorly understood due to limited archaeological research and a lack of contemporary ethnographic analogs for prehistoric agriculture and settlement patterns (Kern et al., 2009).Research in the Upper Xingu region of southeastern Amazonia (Mato Grosso state, Brazil), however, provides one of the clearest examples of large, densely settled pre-Columbian populations and intensive resource management in complex built environments with ethnographic analogs from descendant populations (Heckenberger, 1996(Heckenberger, , 1998(Heckenberger, , 2001(Heckenberger, , 2002(Heckenberger, , 2005;;Heckenberger et al., 1999Heckenberger et al., , 2003Heckenberger et al., , 2007Heckenberger et al., , 2008) ) (Figures 1-4).Results are presented here from analyses of ADE soil associated with these large archaeological settlements and on soils in contemporary indigenous villages (Schmidt, 2010a(Schmidt, , 2010b;;Schmidt andHeckenberger, 2009a, 2009b).This study shows that ADE is forming under current activities and management practices in the Upper Xingu and that results from current and historically abandoned villages correlate to findings from nearby large archaeological settlements1 .
There is continuing debate over the nature and extent of prehistoric human occupation in Amazonia (Barlow et al., 2012;Heckenberger et al., 1999Heckenberger et al., , 2001;;Lombardo et al., 2012;McMichael et al., 2012;Meggers, 2001Meggers, , 2003;;Roosevelt, 1991).Traditional views held that Amazonia was populated by small, mobile, and egalitarian horticultural tribes with relatively little impact on the environment and that, even on the várzea floodplains where populations were higher and social complexity greater, the environment (unpredictable floods in the várzea and infertile soils and sparse game in the terra firme) set a ceiling on village size and social complexity (Meggers, 1954(Meggers, , 1996;;Steward, 1946Steward, -1950)).Those views have undergone substantial revision as archaeology has documented large, settled populations in several parts of the Amazon, as suggested by 16 th to 17 th century accounts of dense populations in many areas (Heckenberger and Neves, 2009;McEwan et al., 2001;Moraes and Neves, 2012;Porro, 1994;Roosevelt, 1991;Viveiros de Castro, 1996).This indicates that much of the Amazonian tropical forest was substantially modified by pre-Columbian societies and post-1492 historical changes (Balée, 1994;Balée and Erickson, 2006;Heckenberger et al., 2003;Schaan et al., 2008).Although some researchers have cautioned that the anthropogenic impact must not be overestimated (Barlow et al., 2012;Bush and Silman, 2007;McMichael et al., 2012), large areas of Amazonia are virtually unknown archaeologically and ongoing research is uncovering substantial ancient settlements with deep deposits of ADE in areas more distant from major rivers (Guapindaia, 2008;Lima et al., 2011;Pereira et al., 2008).It remains to be seen whether there were large areas of Amazonia that were uninhabited (or lightly inhabited) by humans.
Integral to this debate are the patches of extremely fertile ADE soils noted by natural scientists since the 19 th century (Smith, 1879;Smith, 1980;Sombroek, 1966;Woods, 2003).Archaeological research has found an increase in the number and size of settlements with terra preta beginning around the second half of the 1 st millennium B.C. and believed to be associated with increased population and agricultural intensification (Moraes and Neves, 2012;Neves and Petersen, 2006;Neves et al., 2003Neves et al., , 2004;;Petersen et al., 2001).Patches of ADE contrast dramatically with the highly-weathered, acidic, and nutrient-poor oxisols (Latossolos) and ultisols (Argissolos) that dominate the region and have limited agricultural use without high inputs of fertilizer or long fallow periods (Lehmann et al., 2003b;Glaser and Woods, 2004;Woods et al., 2009;Smith, 1980;Sanchez and Buol, 1975).Amazonian farmers seek out ADE to grow crops because the extraordinarily high fertility allows them to grow a more diverse range of cultivars much more intensively than on the typical upland soils (Arroyo-Kalin, 2008;Fraser, 2010;Fraser et al., 2011aFraser et al., , 2012;;German, 2001;Hiraoka et al., 2003;Kawa et al., 2011;Lehmann et al., 2003b;Smith, 1980;Steiner et al., 2009;Woods and McCann, 1999).
The long-term resilience of these soils is explained by the large amount of charcoal and rich diversity of organisms in them (Glaser et al., 2000;Ruivo et al., 2009;Thies and Suzuki, 2003;Tsai et al., 2009;Woods and McCann, 1999).Two broad categories define distinctive ADE, including darker terra preta soils with abundant archaeological ceramics, primarily associated with refuse disposal, and lighter terra mulata with few artifacts, located on the periphery of ancient settlements and likely associated with agricultural practices that introduced ash, charcoal, and organic material into the soils (Sombroek, 1966;Woods and McCann, 1999).These two varieties, which differ not only in color but also in soil chemistry, intergrade in areas of archaeological sites and are a rough guide to general formation processes related to habitation and cultivation areas (Fraser et al., 2011b).It is still difficult to evaluate the discrete behaviors that produced ADE, including whether some are intentional products or byproducts of waste disposal or, likely, both (Sombroek, 1966).One way to address these questions is to carry out ethnoarchaeological research among contemporary indigenous groups to directly determine what impact their activities and cultivation practices have on the soil (Kern et al., 2009).
Before the present research (Schmidt, 2010a(Schmidt, , 2010b;;Schmidt andHeckenberger, 2009a, 2009b), a small number of studies had been published that were specifically focused on indigenous Amazonians' soil management practices.Hecht and Posey (1989) and Hecht (2003) provided observations on soil management practices by the Kayapó, a group that inhabits the middle Xingu River Basin.Relatively intensive agricultural methods such as mulching, in-field burning, soil amendments, and/or repeated cycles of shortfallow slash and burn have been proposed as the origin of brown anthrosols in peripheral areas of settlements (terra mulata).Hecht (2003) suggests that these soil management practices as used among the Kayapó may be similar to practices that, in the past, produced ADE (Hecht and Posey, 1989).Their soil analyses showed significant enrichment in samples collected from areas of Kayapó soil management.Silva and Rebellato (2004) and Silva (2009) discussed patterns of refuse disposal and the formation of anthrosols among the indigenous Asurini group of the middle Xingu River region, however, forthcoming soil analyses would be valuable for comparison with other data.
The Upper Xingu provides unique conditions to compare past with present-day activities.The present research uses ethnoarchaeological and pedoarchaeological approaches to address the question: how were ADE's formed?The Upper Xingu refers to the headwater basin of the Xingu River in the southern Amazon (Mato Grosso, Brazil).It is an ideal place to investigate the formation processes of ADE, since previous research documents cultural continuity in ceramic technology (reflecting manioc agriculture and fishing), settlement patterns (villages with circular central plazas), and overall use of the landscape (elaborate systems of roads, bridges, fields, and aquatic resource management) throughout a chronological sequence spanning over a millennium (Heckenberger, 2005).This presents a unique opportunity to address questions about the range of activities that produced the soil variation found in archaeological sites.
Pedoarchaeological research documents predictable changes in soil properties, including elemental concentrations, caused by human activities, even those occurring over a relatively short time (Barba and Ortiz, 1992;Beck, 2007;Cook and Heizer, 1965;Kern and Kämpf, 1989;Knudson et al., 2004;Middleton and Price, 1996).In particular, nutrients that are abundant in living organisms, i.e., Ca, Cu, K, Mg, Mn, N, P, Sr, and Zn, are enriched in soils and sediments from the remains of organic materials (Bowen, 1966).In the present case, analyses document that distinct patterns of elemental enrichment occur from specific domestic, public, and subsistence activities in the southeastern Amazon (Heckenberger, 1996;Heckenberger et al., 1999;Schmidt, 2010aSchmidt, , 2010b;;Schmidt andHeckenberger, 2009a, 2009b).

METHoDs
Soil samples were collected from a variety of contexts in current, historic, and prehistoric settlements and from off-site forest and agricultural contexts to determine how specific activities have impacted the soil in the study area.Occupational contexts that were sampled include the current village, two historic villages abandoned in the past 100 years, and three large prehistoric sites occupied by A.D. 850.Off-site contexts included forest, fields, and secondary forest or grass and scrub fallow.Areas that were sampled in the current village context included the plaza, house floors, backyards, fire hearths, manioc processing areas, and trash middens.Domestic areas, public areas, and trash middens were sampled in historic and prehistoric contexts.
Soil samples were collected using several different methods depending on the location.The majority of samples were collected using an 8 cm bucket auger to extract a core up to 2 m deep.These include all of the samples from the forest, fallow, and the current and historic villages.Samples were collected in some contexts using a 3 cm soil probe or shovel.A sampling strategy was adopted that employed transects with samples taken at one meter intervals in an attempt to capture small-scale variability.The data from transects can be used to characterize a soil in a given area as well as show transitions between areas with different soil properties and chemical concentrations.
Each sample location had a minimum of two samples at 0-5 cm and 5-10 cm depths.Deeper samples were taken at 10 cm depth intervals.Middens were always sampled to greater depths (minimum 60 cm) due to their height, mounded up to approximately 40 cm above the former soil surface, producing 30 to 40 cm of dark topsoil.Further samples were collected at 50 cm intervals from activity areas with restricted size such as manioc processing areas that are a few meters across.In prehistoric sites, samples were collected from archaeological excavations using a trowel from the ground surface to the base of cultural sediments and from greater depths and areas outside excavations using a bucket auger or soil probe.
Samples were transported to the Museu Paraense Emílio Goeldi (MPEG) in Belém, Brazil, for processing and analysis.A portion of each sample was stored in a permanent collection in the Department of Archaeology at MPEG.Soil analyses included measurements of pH, organic carbon (OC), and the elements Al, Ba, Ca, Co, Cr, Cu, Fe, K, Mg, Mn, Na, Ni, P, Pb, Sr, Ti, V, and Zn.Samples were air-dried at room temperature and screened through 2 mm mesh.Soil pH was determined in water 1:2.5 with each sample stirred for two minutes, allowed to set for 20 minutes, and stirred briefly again before measurement.Organic carbon (OC) was determined by Walkley-Black modified.
A closed-vessel microwave-assisted acid digestion was used to extract major, minor, and trace elements from the soil samples.Approximately 0.10 g of soil was weighed and placed in the microwave digestion vessel.A standard procedure of hydrofluoric acid digestion was used (USA, 1996).The acids added were: 1.2 ml of HF 40% (v/v), 3.5 ml of HNO 3 65% (v/v), 1 ml of HCl 37% (v/v) and 0.5 ml of H 2 O 2 at 30% (m/m).The samples were removed from the microwave and 2 ml of solution of 100 g/l H 3 BO 3 was added to fix the floridic acid in order to ameliorate its reaction with sensitive instrument components.A test was made for analytical recuperation using a standard reference material (SRM 2586) from the National Institute of Standards and Technology (NIST, USA).A Varian Vista Pro simultaneous Inductively Coupled Plasma Optical Emission Spectrometer2 (ICP OES) with axial viewing, a radio frequency of 40 MHz, and charge-coupled device (CCD) detection were used to measure the metals.
The twenty soil variables were used in descriptive statistics and discriminant analyses to examine differences between groups of soil samples from different activity areas, how well groups can be separated, the classification hit-rate of samples, and which variables contribute most to group separation.Discriminant analyses presented here were carried out using the program XLSTAT version 2009.1.01 (updated January 2009).The data from Levels 1 and 2 (0-5 cm and 5-10 cm depth, respectively) were combined for the multivariate statistical analyses in order to obtain a larger sample size for each zone or activity area.Except in one case where the linear method was employed using PASW software, the quadratic method was used in the discriminant analyses presented here3 .

REsulTs AnD DIsCussIon
The composite results indicate that darker ADE is mainly formed in middens in backyard areas of dwellings, typically forming circular, rectilinear, or curvilinear mounds along backyard edges and trails (Figures 5 and 6).The degree of mounding follows a regular pattern that is dictated by the placement of houses, backyard activity areas, and trails, thinning out gradually with distance from these areas.The highest mounds form at locations between houses or at intersections of trails with backyards or with other trails.
Observations in the current village documented the frequently changing locations of activities within the domestic and refuse disposal areas such as when backyard activity areas shift on top of middens, where fruit trees that were planted earlier now provide shade (Figure 7).In the current village, houses are sometimes built in areas further from the plaza edge in positions behind other houses, sometimes forming family compounds.In several observations, new houses were built over middens, with the midden being either scraped aside or leveled to create a level pad for the house.In one case fresh soil was dug up from the area surrounding the house and put in place to make a new house pad (3-5 cm thick) on top of the former midden.When domestic areas (houses and backyards) shift on top of middens in the current village, the area of active refuse disposal shifts too, thereby expanding the area of middens.
Low fertility indicators that are typical of Amazonian soils -high exchangeable Al, high acidity, and nutrient deficiencies -are absent in midden soil, offering ideal conditions for plant growth and crop cultivation.Midden soils contain a high pH and high OC content, and much lower Al and Fe contents than surrounding soils.Of all the areas tested, middens were found to have greatly elevated levels of the widest range of nutrients typically elevated by human activities (Ba, Ca, Cu, K, Mg, Mn, Na, P, Sr, and Zn) (Table 1).
Kuikuro farmers refer to patches of terra preta as 'igepe', meaning cornfield in their Carib language (Carneiro, 1983;Heckenberger, 2005).They travel up to 15 km to open gardens in the terra preta of archaeological sites where they plant diverse crops that typically don't do well in normal soils (on normal soils they generally plant only manioc and sometimes piquí (Caryocar brasiliense) fruit trees).The crops planted in ADE include corn, bananas, sugarcane, sweet potatoes, cotton, and various fruit trees.Manioc is sometimes planted in these areas as well.Middens behind homes in the current village are used for cultivating the same diverse range of crops in home gardens and orchards on top of middens.
It is the fertile midden soil and the availability of abundant organic materials that act as fertilizer and mulch that makes the refuse disposal zone important cropland.Households typically plant nutrient demanding crops (annuals and perennials including fruit trees) in the trash middens behind their homes.Small patches of bitter and sweet manioc were also observed in some backyards on top of or adjacent to middens.Some old-world crops now have an important place in the agricultural system including banana, sugarcane, and mango.The home gardens of three households were surveyed for useful crops planted in the middens in 2005.The three gardens contained 21, 15, and ten useful species that had been planted.Species included sweet potatoes, squash, corn, beans, sweet and bitter manioc, chili peppers, bottle gourd, calabash gourd (cuia), sugar cane, mango, papaya, piquí, cashew, banana, guava, orange, Antilles cherry (acerola), cotton, tobacco, and several native fruits from the region including species from the gallery forest and fallow or floodplain savannah.
Sorting of refuse by type appears to be an important factor in the formation of darker, more nutrient rich soils nearer domestic areas (terra preta) and lighter, brown soils in more peripheral areas of settlements (terra mulata).Observations in the Kuikuro village suggest that materials such as hearth cleanings (ashes and charcoal), fish remains, and durable refuse such as broken ceramics are typically disposed of on the edges of the backyard while bulky organic waste, notably from manioc processing, is spread further away and over a wider area .Land on the periphery of the village is used for backyard homegardens or alternates between cultivated gardens, which are fertilized with organic refuse, mulched with weeds, and occasionally burned; and fallow areas used for defecation, refuse disposal, or processing plant materials (Figure 12).These activities result in darkened soil with elevated pH and OC and in nutrient enrichment similar to middens but at much lower magnitudes, thus analogous to terra mulata (see results for fallow area in Table 1).
Organic materials quickly break down in middens and midden material is mixed with the underlying soil.This is shown by increasing levels of Al, Fe, and V in the surface layers of middens as a result of bioturbation that moves underlying soil upwards, mixing it with midden material.Likewise, soil pH and some nutrients increase over time at greater depths underneath middens as midden material and nutrients are transported downward with bioturbation, leaching and other soil forming processes (Figure 13).In addition to middens and cultivation areas, groups of samples from domestic and public areas show significant nutrient enrichment and distinctive soil signatures (Table 1).Fire hearth soil displayed an elevated pH, very low OC, and exceptionally elevated K content.Manioc processing areas showed substantial nutrient enrichment, high pH, and high OC.Soils in current village domestic areas were visibly darkened, have an elevated pH, an elevated OC content, and are significantly enriched with nutrients.The re-building of houses in the same general location, but usually shifted forward or backward, causes domestic areas to shift positions and overlap each other as the house footprint changes locations.The burning of houses, whether in controlled or accidental fires, adds large amounts of ash and charcoal to the soil.After fires, much of the debris are hauled away to the midden area and dumped while some gets thrown in the holes left when burnt posts are removed.Repeated house fires likely have a proportionally large effect on soil darkening and enrichment in domestic areas.The plaza center showed soil enrichment on par with domestic areas but with differences due to the digging of graves, while the mid-range area of the plaza showed the least amount of enrichment in the village context.
Soil analyses from archaeological settlements, including recently abandoned Kuikuro villages (one occupied 1973-1983 and another ca. 1930-1940) and a large pre-Columbian settlement (Nokugu), showed that some properties of anthropically enriched soils significantly decrease over time in middens, including pH, OC, Ca, K, Mg, Na, Sr, P, and Zn (Table 1).Several soil variables were shown to be more stable in the soil over time -namely pH, Ba, Mn, and Sr -indicating that these are some of the most effective for detecting and documenting previous landuse and human impact on the soil4 .
In order to acquire further information about the differences between groups of samples from different activity areas and attempt to determine in more detail their unique soil signatures, multivariate discriminant analyses were employed.The objectives of the analyses were to: a) clarify differences between activity areas by determining which of the soil properties can best be used to separate samples from different areas; and b) classify the samples Figure 6.Midden in the current village at the intersection of the backyard with a trail where the highest mounds (~50 cm) are typically found.Source: Schmidt and Heckenberger (2009a).Table 1.Summary of group means for surface levels from the current and historic villages, offsite area (forest), and prehistoric sites.Notes: 1) midrange area of plaza (midplaza); 2) manioc processing area; 3) historic village; 4) prehistoric site Hialugihïtï, three transects in residential areas; 5) prehistoric site Akagahïtï,   Figure 10.The surface of a midden in the historic Kuikuro I village (occupied 1973-1983).In this mounded midden on the edge of the former backyard, large potsherds, a metal pot, and large fragments of charcoal are visible on the surface.Source: Schmidt (2010b).
into their respective groups based on the soil variables to determine how accurately groups can be classified and which configurations of variables perform best.Discriminant analyses generally achieved better than a 90% classification hit rate for groups of samples from activity areas in the current village and prehistoric sites, showing significant differences in soil chemistry based on a suite of variables (Table 2).Territorial maps plot the discriminant Z scores for each observation on two functions (Figures 14 and  15).They graphically show the distance between centroids, the clustering or dispersion of observations for each group, and the separation or overlap of each group.
In an analysis using six groups from the current village and landscape, the territorial map (Figure 14) shows that: 1) forest samples are well separated from the other groups and fairly tightly clustered; 2) midden samples are more dispersed and partially overlap the other groups from the village; and 3) manioc processing and fire hearths in domestic areas, the mid range area of the plaza or mid-plaza, and the fallow area on the edge of the village all overlap one another.The close proximity of Fire hearths and Fallow Area 1 centroids and their distance from the midden centroid is noteworthy.Located in an area of cultivation and frequent burning with possible refuse disposal and defecation, the fallow area has a significantly modified soil signature (analogous to terra mulata) which is very different from the middens (analogous to terra preta).In a second analysis using five groups, the territorial map (Figure 15) shows the samples from historic middens clustered adjacent to and partially overlapping the samples from current village middens.The historic midden samples lie in an intermediate position between the current midden and forest, reflecting the slow return of the midden soil toward forest soil conditions after abandonment, due to bioturbation, leaching, and other soil processes.
In a discriminant analysis using 12 variables, the results for the average of 100 repetitions and a 33% random validation (hold out) sample show numerically which groups have greater separation or overlap (Table 2).This analysis, using the variables pH, OC, Al, Ba, Ca, Co, Fe, K, P, Sr, Ti, and V, performed best out of twenty tests using different numbers and combinations of variables.The ordering of the groups from highest to lowest by the classification hit rates (% correct) of the validation sample are as follows: 1) mid-plaza (96.8%); 2) forest (96.1%); 3) midden (95.4%); 4) fallow area (93.8%); 5) manioc processing (91.8%); and 6) fire hearth (87.6%).The mid-plaza group, representing the base line soil conditions within the village context, scored the highest -meaning it had the least number of samples misclassified into other groups.Fire hearth and manioc processing areas, on the other hand, scored the lowest and had the most misclassified samples.They were more frequently misclassified with each other, partially due to their close proximity, i.e., ash spills from the hearth onto the manioc processing areas.The fact that backyard areas sometimes lie on top of middens could also potentially lead to confusion between domestic samples and middens.
The confusion matrix with cross-validation results (Table 3) for the five group analysis using all variables (Figure 15) shows the groups into which samples were misclassified.Current and historic midden samples were misclassified with each other in four cases.The Fallow area samples were also misclassified with either current or historic middens in four cases.Forest samples were misclassified with historic middens in three cases and with current middens in one case.The confusion of forest samples with middens is partially due to the fact that forest soil has a high organic carbon (OC) content and relatively high levels of total phosphorous (P).The fact that more historic midden than current midden samples were misclassified as forest reflects the former's return, over time, toward forest soil conditions.Finally, midplaza samples were misclassified as historic midden in two cases.The relative frequencies of misclassifications reflect not only similarities between groups but also variability and anomalies within groups.For example, the mid-plaza has occasional pits where weeds were buried and/or burned during plaza cleaning events.Samples from these anomalous locations have higher pH, OC,  and nutrients, more similar to historic midden samples than typical mid-plaza samples.
Researchers employ a number of methods to judge the effectiveness or 'power' of variables in discriminating between groups, which can change depending on the analysis (Hair et al., 2006;Huberty and Olejnik, 2006).Perhaps one of the most reliable ways is the composite potency index whose results are presented here using two different discriminant analysis methods (Table 4).It is noteworthy that the variables pH and Ba consistently came out on top -two of the variables that also appear to be the most stable in the soil over time.A stretched vector plot shows graphically the effectiveness of the variables by another measure, the discriminant loadings (Figure 16).This graph shows the contributions of each variable to each of two functions (Function 1 contributes the most to group separation) as well as correlations between variables.In this case, it shows that the variables most increased by human impact (pH, OC, and soil nutrients) are highly correlated with one another, while those abundant in unmodified soil (Al, Fe, Ti, and V) are correlated with one another.
In summary, thirteen of the most significant conclusions pertinent to the formation of ADE that can be drawn from current and historic village results are: 1) groups of soil samples from all of the activity areas displayed significant soil modifications; 2) midden soils stood out from the rest with the most highly elevated levels of the greatest range of soil properties (pH, OC, and nutrients); 3) the underlying soil is rapidly mixed with midden material, as shown by relatively high Al, Fe, and Ti in the upper levels of middens; 4) middens affect soil properties to a considerable depth, shown by elevated pH and nutrients at depths of at least 1-2 m below the midden; 5) middens form distinctive patterns on the landscape consisting of linear mounds along backyard edges and trails; 6) middens are used for cultivating diverse home garden crops; 7) activity areas shift frequently, such as when fruit trees planted in middens later provide shade for domestic activities, or when middens are leveled and houses are built on top of them; 8) pH, OC, and nutrients (Ca, Cu, K, Mg, Na, P, Sr, and Zn) were shown to decrease in middens over time; 9) several variables, including pH, Ba, Mn, and Sr, appear to be more stable over time and are therefore among the best for determining anthropic impacts on the soil; 10) refuse disposal, soil management, and cultivation practices in peripheral areas of the settlement formed darkened soils with elevated pH, OC, and nutrients, but much less elevated than middens, analogous to terra mulata; 11) public and domestic areas were significantly impacted and displayed marked differences in soil properties from both middens and forest soil; 12) some of the activity areas displayed distinctive soil signatures, such as fire hearths with greatly elevated pH and K and a low OC content (however, the K in hearths appears to decrease rapidly over time); and 13) certain elements, including Cu, Cr, Na, Ni, and Pb, were less useful in discriminant analyses because of their high variability.
Studies of ADE in prehistoric archaeological landscapes show several features somewhat distinctive from contemporary patterns.These include broad contiguous macro-strata covering portions of larger settlements, along with ADE hamlets and possible wetland agricultural works5 (Erickson, 2008;Heckenberger, 2005).Distinctive mounds of ADE formed from construction and maintenance activities and from refuse disposal in middens in locations that were dictated by the use of domestic and public space, much as in the current Kuikuro village (Figure 17).In prehistoric sites linear mounds define plazas and roads, dividing public space from residential areas (Figure 18).Prehistoric landscape and soil modifications included the excavation of massive ditches at the larger sites.The ditches, up to 5 m deep and over 1 km long, Table 2. Results for a discriminant analysis using six groups from the current village and landscape.Results are of 100 repetitions using the quadratic method with 12 variables and a 33% random holdout sample (n = 356).Variables included in the analysis: pH, OC, Al, Ba, Ca, Co, Fe, K, P, Sr, Ti, V. were likely used as part of defensive works that probably included palisade walls.The largest mounds are found around the plaza(s) and on the roads leading out of them.Mounds generally decrease in size with distance from the settlement but they could still be identified and mapped for several kilometers in between sites.Mounds surrounding the circular plazas were excavated and found to contain dark soil and abundant material remains (mostly broken ceramics) giving them the aspect of middens.The mounds were presumably first formed with the opening and clearing of the plaza and roads.It is likely that middens in the area of the plaza and roads were scraped to the sides forming the linear mounds.It is hypothesized that the mounds continued to grow with periodic maintenance of the plaza and roads (scraping of weeds and debris to the edges and piling them on the mounds) and possibly deposition of refuse from plaza activities and/or from households surrounding the plaza.Table 3. Confusion matrix with cross-validation results for a five group analysis including historic midden samples (see Figure 15).Table 4. Ranking of variables by the composite potency index for the analysis with six groups from the current village and landscape.These were calculated from the structure [r]'s using ELSTAT software (quadratic method) and rotated structure [r]'s using PASW software (linear method).In the third column is a combined ranking from the two methods.16.Stretched vector plot of discriminant loadings for Functions 1 and 2 in the analysis using six groups from the current village and landscape.

ELSTAT
Present day roads in the Upper Xingu are still maintained by scraping weeds (pulled out by the roots) to the edges, forming linear mounds of soil and dead vegetation.
In residential areas, mounds are found on the surface often overlying stratified deposits of dark earth and other cultural material (mainly ceramics).Curvilinear mounded middens were located and sampled in residential areas surrounding the plaza.In these areas, low mounds surrounding flat circular areas, some with a lowering in the mounds on opposite sides corresponding to the locations of doorways, apparently mark the locations of houses.Three groups of soil samples (domestic, hearth feature, and midden) collected from an extensive excavation in a domestic area were used in discriminant analyses and showed pronounced differences between the groups, similar to corresponding groups from the current village.In many cases dense concentrations of ceramic fragments were found lying on the surface often in linear arrangements, possibly marking the location of former paths of movement within the settlement.These fragments had apparently been deposited in middens that have since eroded away.In many cases, the surface middens appear to have been deposited on top of stratified archaeological  Soil analyses from samples collected in transects and select locations at three prehistoric sites (Nokugu (MT-FX-6), Hialugihïtï (MT-FX-13), and Akagahïtï (MT-FX-18) reflect the patterns of mound formation (Figure 19).Mounded middens and plaza mounds showed the highest OC and pH and greatest nutrient enrichment, most similar to current and historic middens in the contemporary context (Tables 5 and 6).Some locations were notably more enriched with nutrients, for example, the mound around the small plaza at Nokugu (Table 6) and the central area of Akagahïtï, a site that showed especially marked soil enrichment (Table 1).Samples from transects through residential areas displayed highly variable pH, OC, and nutrient enrichment.The variation along transects, with samples taken every 10 or 20 m, suggests alternating locations of habitations and middens.
Prehistoric settlement patterns include networks of closely spaced (3-8 km) large walled towns (30-50 ha), smaller plaza villages (≤15 ha), and non-plaza ADE hamlets, integrated by well-maintained roads, and indicative of a much more fixed and compartmentalized settlement and agricultural landscape (Heckenberger et al., 2008).Manioc agriculture and wetland management provided the economic foundation of these prehistoric societies, but landuse would have likely been more intensive, including more fixed agricultural areas and shorter-cropping, because of the dense settlement (Heckenberger, 2005;Schmidt, 2010b).Amazon dark earth was likely used to support intensive agriculture with crops grown within and surrounding villages.These findings indicate that prehistoric farmers were creating and using ADE in and around settlements likely as an important component of their agricultural system used to increase production and improve sustainability.However, not all terra preta was used for cultivation, as indicated by the deep mounding in some locations, the large size of potsherds, and the apparently densely packed habitation structures at many Amazonian sites with large deposits of terra preta (Schmidt, 2010b(Schmidt, , 2012a(Schmidt, , 2012b)).deposits, probably formed when middens were leveled to create pads for the construction of houses such as was observed in the current Kuikuro village.

ConClusIon
A new view of prehistoric Amazonia has emerged as a center for the development of agriculture and complex society with dense populations that significantly impacted their environment (Heckenberger and Neves, 2009;Roosevelt, 1999aRoosevelt, , 1999b;;Viveiros de Castro, 1996).This gives the region new importance for understanding the development of sociopolitical complexity, intensification of subsistence, and human-environment interaction.There is an intense interest in ADE for its promise to teach us about improving and managing the fertility of tropical soils, thereby allowing an increase in the productivity of agriculture in tropical areas of the world.The Upper Xingu presents a unique opportunity to study the genesis and distribution of anthrosols, as it is one of the few places in Amazonia where observations on contemporary indigenous agriculture, land use, and settlement patterns have been directly linked to the archaeological record.Pedoarchaeological research in the Upper Xingu informs long-running debates on the economic basis of large, settled populations in the Amazon, notably intensive agriculture, arboriculture, and management of aquatic resources (Heckenberger, 2005;Balée and Erickson, 2006;Heckenberger and Neves, 2009).This research is particularly relevant and timely in the Upper Xingu basin which lies in a unique transitional environment between the Amazon rainforest and the savanna (Cerrado) of central Brazil.It is in one of the regions of Amazonia hardest hit by deforestation and development -satellite images show Xingu Indigenous Park standing out as an island of forest and savanna surrounded by areas cleared for modern agriculture and pasture.The research adds new information about the prehistoric and current management of tropical environments and contributes to discussions on the extent of cultural versus 'pristine' forests in the tropics, particularly in the light of new data on the prehistoric human landscape (Heckenberger et al., 2003).The research yields information on fertility and management of tropical forest soils and may contribute to an understanding of the role of soils, particularly anthrosols, in acting as a sink for atmospheric carbon.Furthermore, the research promotes changing perspectives on tropical forest peoples and provides local Amerindians with information on their history and continuity in the region as well as training that gives them greater pride and power for protecting their archaeological sites, culture, environment, and territorial rights.
Concern over degraded agricultural land and how to feed a growing population have prompted efforts to develop new sustainable tropical production systems, sometimes modeled on traditional systems, such as agroforestry systems that incorporate soil management (National Research Council, 1993).Archaeology documents that large prehistoric populations in Amazonia enriched the soil on the land they lived and cultivated, resulting in substantial tracts of ADE.These fertile soils hold valuable lessons for resource management in the tropics (Lehmann et al., 2003a;Glaser and Woods, 2004;Glaser, 2007;Woods et al., 2009).The high carbon content of ADE has implications for the global carbon cycle and climate change (Sombroek et al., 2003).A controversial technology inspired by terra preta research called biochar is being promoted as a technique for simultaneously improving soil fertility and acting as a sink for atmospheric CO 2 (Leach et al., 2012;Mann, 2002;Marris, 2006).The way that indigenous populations managed their environments in the past has molded the patterns of biodiversity that are observed today in Amazonia and therefore have profound implications for the biological sciences (Balée, 1994(Balée, , 1998;;Balée and Erickson, 2006;Hecht and Posey, 1989;Heckenberger et al., 2007).Thus, ADE inform agriculture, conservation, and development efforts in tropical regions as well as plans for sequestering carbon in order to lessen the impact from human-induced global warming.Amazonian dark earths are a prime example for historical ecology, fundamental to understanding the ecology and biogeography in the South American Tropics.
The new data bolster hypotheses for the formation of ADE, namely that darker, artifact-rich terra preta was formed in habitation contexts mainly from refuse disposal in middens and that lighter brown soils (terra mulata) on the periphery of settlements formed in association with agricultural activities.The ethnoarchaeological context allows us to visualize the process of terra preta formation and make direct observations and measurements of soil parameters from areas where human activities are occurring.The cultural continuity in the Upper Xingu, shown by a continuation of settlement patterns, ceramic technology, and overall use of the landscape, adds a level of credibility when extrapolating processes back to the past (Heckenberger, 2005).Many of the processes that formed ancient terra preta are operating today in almost every Amerindian village and in most communities and households in the region (Hecht, 2003;Heckenberger, 1996;Silva, 2003;Silva and Rebellato, 2004;Smith, 1980;Woods and McCann, 1999;Wüst and Barreto, 1999).The research provides documentation that anthrosols are forming today in the Kuikuro village and suggests a workable model of prehistoric anthrosol development in Amazonia.
Beyond the anthropic soils of the Upper Xingu, this research is generally pertinent to pedoarchaeological studies everywhere because darkened soils are a common feature at many archaeological sites around the world.The ethnoarchaeological research in the Upper Xingu was found to be fundamental to understanding ADE formation in archaeological sites in other regions of the Amazon where recent research has been carried out including the Central Amazon and the lower Trombetas River region (Guapindaia, 2008;Neves et al., 2003;Schmidt, 2010aSchmidt, , 2012aSchmidt, , 2012b)).Anthrosols that share similarities with terra preta have been recorded throughout the tropics in Central America, the Caribbean, Africa, Southeast Asia, and the Pacific (Leach et al., 2012;Schmidt, 2010b;Sheil et al., 2012;Woods, 2003).The darkened soil can be formed in middens or by intensive agricultural activities.Shell middens from the U.S.A. to southern Brazil often show the characteristics of terra preta with the addition of shells.Darkened anthrosols or middens, some thousands of years old, are found throughout the American tropics in a great diversity of environmental and cultural contexts.
This research asks the question: 'How was terra preta formed?'.The objective was to develop soil analysis as a more useful tool in Amazonian archaeology by trying to understand the specific processes that formed terra preta.It is also hoped that the results will help those trying to understand the nature of terra preta and its high fertility in order to improve the fertility of soil in the tropics to help farmers increase their productivity and sustainability.The hope is that the ethnoarchaeological information about anthrosol formation will form a basis for further studies on terra preta formation processes and properties and may be applied to help interpret pedoarchaeological data from sites across the Amazon region and beyond.Further research on the collection of samples at the Museu Paraense Emílio Goeldi will focus on additional soil analyses to confirm the results and to understand in greater detail how soils are transformed by human activities in the Upper Xingu.

ACKnowlEDGEMEnTs
I would like to thank the Kuikuro Community, especially Afukaka, Tabata, Laqui, Kagito, Dunga, Masinua, Anselmo, Ticão and Jahila.I also thank Michael Heckenberger, my co-academic advisor and coordinator of the Southern Amazon Ethnoarchaeology project.All project maps and images (Figures 3,(5)(6)(7)(9)(10)(11)(12)(13)and 17) belong to the Kuikuro Community and their use was authorized by the Associação Indígena Kuikuro do Alto Xingu (AIKAX).Financial support was provided by the United States National Science Foundation, University of Florida (UF) Geography Department, and UF Tropical Conservation and Development Charles Wagley grants.Institutional support was given by the Museu Paraense Emílio Goeldi (MPEG), Coordenação de Ciências Humanas (CCH), Área de Arqueologia, and the National Museum of Rio de Janeiro.Soil analyses were carried out in the Coordenação de Ciências da Terra e Ecologia (CCTE) at MPEG.Laboratory analyses were coordinated by Paulo Sarmento and Leide Lemos and carried out by five dedicated interns: Patricia Oliveira da Silva, Marcelo Monteiro Farias, William Akira dos Santos, Daniel Alvino Mesquita, and Flavio Corrêa dos Santos.Nilson Gabas Jr., Ana Vilacy Galúcio, and Edithe Pereira provided institutional support for the project.I also wish to thank the following people for their help with the research and support at MPEG: in CCTE, Dirse Clara Kern, Maria de Lourdes Ruivo, Maria Thereza Prost, Maria Emília Sales, Cristina Senna and, finally, Francisco Juvenal Frazão who sadly recently passed away, helped me get the lab work started at the Goeldi Museum; in CCH, Vera Guapindaia, Daniel Lopes, Marcos Magalhães, Edna Moutinho, Paulo do Canto Lopes, João Ayres Fonseca, and Carlos Augusto Barbosa.I would also like to thank the following people for their help and guidance during PhD research: Carlos Fausto, Bruna Franchetto, Mara Santos, Quintino Araújo, Larry Swenson, Nigel Smith, Nick Comerford, Hugh Popenoe, Jim Petersen, Eduardo Neves, William Woods, Ken Sassaman, Denise Schaan, Joshua Toney, Christian Russell, Willie Harris, Emiliane Braga, and Robert and Sharon Schmidt.The Instituto Socioambiental (ISA) graciously provided the map of the Xingu Basin (Figure 1).I also thank two anonymous reviewers for their constructive comments.

Figure 1 .
Figure 1.Map of the Xingu River Basin showing the location of Xingu Indigenous Park.Source: Instituto Socioambiental.

Figure 3 .
Figure 3. Landsat TM image of the study area with overlying map of two prehistoric sites drawn from GPS data collected in 2002.Nokugu (MT-FX-6) in the upper left is connected to Hialugihïtï (Mt-FX-13) on the lower right by an ancient road.The brighter shades in the center are a forested peninsula of uplands surrounded by darker shades in the upper right and lower left representing floodplain areas dominated by rivers, lakes, and seasonally flooded savannah.The bright circle in the upper center is the current Kuikuro village.Source: adapted from Heckenberger et al. (2003).

Figure 5 .
Figure 5. Map of the Kuikuro village in 2002 drawn from GPS data.Black represents trash middens separated by trails leading from backyards out of the village.The domestic area lies in and around the houses and the square in the middle is the location of the men's house and ritual center of the village.The area between the plaza center and the houses is referred to as the mid-plaza.Source: Heckenberger (2005).

Figure 7 .
Figure 7. Backyard activity area and manioc processing location under a shady fruit tree on top of a midden.Active refuse disposal has shifted further back and extends along trails.Source: Schmidt (2010b).

Figure 8 .
Figure 8. Disposal of hearth ash and charcoal in a midden at the edge of a backyard.Source: Schmidt (2010b).

Figure 9 .
Figure 9. Bulky organic plant waste (leaves, manioc peelings, etc.) being carried away from the backyard to be dumped along the trail.Source: Schmidt (2010b).

Figure 11 .
Figure 11.Profile from a midden in the historic Kuikuro I village.Ceramic sherds are present in the 10-20 cm level.Dark soil extends to at least 30 cm and there is a transition zone between 30 and 40 cm.Source: Schmidt (2010b).

Figure 12 .
Figure 12.Agricultural work being done in a manioc garden at the edge of the village, adjacent to the fallow area where samples were collected.The soil here is visibly darkened.Source: Schmidt (2010b).

Figure 13 .
Figure 13.Soil pH by depth level in two-meter profiles in forest and three middens of different ages.A) Forest; B) Current midden; C) Historic midden; D) Prehistoric Nokugu midden.

Figure 15 .
Figure15.Territorial map of Functions 1 and 2 for a five group analysis including historic midden samples (using all twenty variables).

Figure 14 .
Figure14.Territorial map of Functions 1 and 2 in an analysis using all twenty variables and six groups from the current village and landscape.

Figure 17 .
Figure 17.Southern profile of a midden at Nokugu (40 cm depth).The darkest soil and ceramics extend to about 30 cm depth on the eastern half and there is a possible pit feature in the western half that extends to about 40 cm depth.Source: Schmidt (2010b).

Figure 19 .
Figure 19.Map of Nokugu showing the locations of transects and select samples with results for organic carbon (OC).Source: Schmidt (2010b).

Table 5 .
Nokugu midden profile results for pH, OC and seven nutrients.

Table 6 .
Nokugu small plaza mound profile results for pH, OC and seven nutrients.