This article was downloaded by: [Illinois State University Milner Library], [JonathanThayn]On: 01 August 2011, At: 12:46Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Journal of RemoteSensingPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/tres20

Locating Amazonian Dark Earths (ADE)using vegetation vigour as a surrogatefor soil typeJonathan B. Thayn a , Kevin P. Price b & William I. Woods ca Department of Geography and Geology, Illinois State University,Normal, IL, 61790-4400, USAb Department of Agronomy, Kansas State University, Manhattan,KS, 66506, USAc Department of Geography, University of Kansas, Lawrence, KS,66045-7613, USA

Available online: 01 Aug 2011

To cite this article: Jonathan B. Thayn, Kevin P. Price & William I. Woods (2011): LocatingAmazonian Dark Earths (ADE) using vegetation vigour as a surrogate for soil type, InternationalJournal of Remote Sensing, DOI:10.1080/01431161.2010.512941

To link to this article: http://dx.doi.org/10.1080/01431161.2010.512941

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching and private study purposes. Anysubstantial or systematic reproduction, re-distribution, re-selling, loan, sub-licensing,systematic supply or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representationthat the contents will be complete or accurate or up to date. The accuracy of anyinstructions, formulae and drug doses should be independently verified with primarysources. The publisher shall not be liable for any loss, actions, claims, proceedings,

demand or costs or damages whatsoever or howsoever caused arising directly orindirectly in connection with or arising out of the use of this material.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

International Journal of Remote SensingiFirst, 2011, 1–17

Locating Amazonian Dark Earths (ADE) using vegetation vigouras a surrogate for soil type

JONATHAN B. THAYN*†, KEVIN P. PRICE‡ and WILLIAM I. WOODS§†Department of Geography and Geology, Illinois State University, Normal,

IL 61790-4400, USA‡Department of Agronomy, Kansas State University, Manhattan, KS 66506, USA

§Department of Geography, University of Kansas, Lawrence, KS 66045-7613, USA

(Received 15 March 2010; in final form 12 July 2010)

Amazonian Dark Earths (ADE) are patches of archaeological soils scatteredthroughout the Amazon Basin. These soils are a mixture of charcoal, nutrientvegetable matter and the underlying Oxisol soil. ADE are extremely fertile incomparison to the surrounding soils and they are sought after by local residentsfor agricultural food production. Research is being conducted to learn how ADEwere created and to explore the possibility of replicating them to sequester carbonand to reclaim depleted soils in the Amazon Basin. A factor limiting the successof this research is our current inability to locate ADE sites hidden beneath thetropical forest canopy. We use annual time-series Moderate Resolution ImagingSpectroradiometer (MODIS) Enhanced Vegetation Index (EVI) satellite imageryfrom 2001 to 2005 and harmonic analysis (HA) to examine the spectral differ-ences between forest vegetation growing on ADE and forest vegetation growingon non-ADE. There is a significant difference between the reflectances of vegeta-tion growing on the two soil types, due primarily to lower EVI values over ADEduring the dry season (multiple analysis of variance (MANOVA) p-value = 0.040).A logistic model is used to create a predictive map of ADE location.

1. Introduction

Scattered throughout the typically nutrient-poor Oxisol soils of the Amazon Basinare relatively small patches of dark, very fertile soils called Amazonian Dark Earths(ADE; Glaser and Woods 2004). ADE contain highly elevated levels of organic mat-ter, mostly very slowly decomposing charcoal, which causes the soil’s dark coloration(Kern et al. 2003). The charcoal content of ADE soils is typically four times greaterthan that of neighbouring soils but can be as much as 70 times higher (Glaser et al.2001). The inert charcoal makes nutrients in the soil more recalcitrant (Glaser et al.2003, Lehmann et al. 2003b, Steiner et al. 2007) and, accordingly, ADE soils are someof the most fertile in the world (Tiessen et al. 1994, Kern et al. 2003, Lehmann et al.2003a). When productivity of plants grown on ADE soil was contrasted with that oftypical nutrient-poor Amazonian Oxisol soils, Major et al. (2005) found that maizeyields were as much as 63 times greater, weed cover was 45 times greater and plantspecies diversity was up to 11 times greater than for adjacent typical Amazonian soils.

*Corresponding author. Email: jthayn@ilstu.edu

International Journal of Remote SensingISSN 0143-1161 print/ISSN 1366-5901 online © 2011 Taylor & Francis

http://www.tandf.co.uk/journalsDOI: 10.1080/01431161.2010.512941

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

2 J. B. Thayn et al.

In a controlled experiment, Steiner et al. (2007) found that crop production on siteswhere fertilizer and charcoal had been applied was double that of sites where fertilizeralone had been used.

Although charcoal helps to retain nutrients that would otherwise be weathered fromthe soil, nutrient transfers from outside of ADE sites are necessary to explain currentnutrient levels in ADE (Woods and McCann 1999, Neves et al. 2003). This suggeststhat the formation of ADE soils ultimately became an intentional effort of prehis-toric Amerindian populations to improve the quality of their farmland (Woods andMcCann 1999, Neves et al. 2003). These nutrient sources may have been plant andanimal food wastes, fish bones and other unused fish matter, human excrement, andplant materials used for fuel and construction. The presence of algae in ADE fromc. 1150 BP and later suggests that silt from riverbanks was incorporated into the ADEsoils in at least one location (Mora et al. 1991).

Locating and studying ADE sites is important not only from an archaeological anda cultural heritage perspective but also because of their potential as a means for long-term carbon sequestration. To meet the challenges of possible global climate changecaused by greenhouse gases, atmospheric carbon concentrations must be reduced.Vegetation actively withdraws carbon from the atmosphere and stores it as organicmatter. Charcoal is created when organic matter is heated without oxygen and it con-tains twice the carbon content of ordinary biomass (Lehmann 2007). The additionof charcoal to the soil was part of the creation of ADE (Neves et al. 2003). Studiesof known ADE sites, which range in age from 500 to 2500 years (Neves et al. 2003),reveal that biochar is resistant to decay and can store carbon for centennial time-scales(Lehmann et al. 2006). This has led some to speculate on the viability of a charcoalcarbon sequestration industry that would reduce atmospheric carbon (Sombroek et al.2002, Lehmann et al. 2006, Marris 2006) and improve soil fertility (Lehmann et al.2003a, Glaser and Woods 2004).

Some maps of ADE exist for relatively small subregions (Heckenberger et al. 1999,Kern et al. 2003), but the geographic extent and location of ADE are unknown inthe major portion of the Amazon Basin (Woods 1995). Nonetheless, Sombroek et al.(2002) estimate that there is a patch of ADE for every 2 km2 along certain Brazilianriver corridors, and that they extend into Columbia, Venezuela, Peru, Bolivia and theGuianas. ADE patches range in size from 0.5 to 300 ha (Woods and McCann 1999,Sombroek et al. 2002), although 80% of known ADE sites are less than 2 ha (Kernet al. 2003).

Most known ADE sites were found by local caboclo residents who prefer ADEsoils for agricultural settlement (Sombroek et al. 2002). ADE are recognized based ontheir lower vegetation canopy, more closed understorey and unique species composi-tions, including Brazil nut (Bertholletia excelsa), cacao (Theobroma cacao), cupuaçu(Theobroma grandiflorum) and the giant Ceiba pentandra (Woods and McCann 1999).ADE soils also contain copious amounts of pottery shards (Sombroek 1966, Neveset al. 2003). Unfortunately, traditional field methods are unsuited for locating ADEfor two main reasons: (1) the extreme difficulties associated with fieldwork in the dense,inaccessible tropical forest; and (2) the time and expense that would be required tocover the enormous extent of the Amazon Basin. For these reasons, remote sensing-based models that predict the location of ADE sites are required. Such models wouldgreatly enhance researchers’ ability to find new sites, could contribute to preservingtropical forests, and would assist scientists’ efforts to study and replicate ADE forcarbon sequestration.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

Locating ADE 3

The main difficulty with developing such a model is that most known ADE siteshave been converted to agriculture with different crop types. Identifying ADE in thoseconditions is confounded by the varying spectral properties of different crops. Thosefew known sites that have not been converted to agriculture are located under densetropical forest canopies that completely occlude the underlying soil so that direct imag-ing of bare soil is impossible. The aim of the present study was to develop a remotesensing method for locating ADE sites currently located under forest canopy usingremotely sensed measures of vegetation as a surrogate for soil type.

1.1 Past research

Very little research has been undertaken exploring the possibility of using remotelysensed data to locate ADE. Most comparable research has focused on studying forestsuccession and mapping seral stages (Kimes et al. 1998, Steininger 2000, Lu et al. 2003,Roberts et al. 2003, Vieira et al. 2003, Salovaara et al. 2005). Thayn et al. (2008) andMeddens (2006) discuss possible methods for identifying ADE using remotely senseddata and present small, inconclusive pilot studies.

Russell (2005) predicted the location of archaeological sites on the upper XinguRiver in the southwestern portion of the Amazon Basin using the results of severalprincipal components analyses of various vegetation indices derived from a LandsatEnhanced Thematic Mapper Plus (ETM+) image. A maximum likelihood classifi-cation scheme was able to identify 12 different land-cover classes with an overallaccuracy of 95% and a Kappa of 0.903. Two of the classes were archaeological sitesnot under cultivation and archaeological sites under cultivation. The high accuracyassociated with classifying water bodies, Savannah, forest and cultural sites elevatedthe overall accuracy of the classification. For archaeological sites not under cultiva-tion and archaeological sites under cultivation, the respective producer accuracies were78% and 41% and the corresponding user accuracies were 41% and 62%.

Although Russell’s (2005) overall classification was fairly accurate, the method’sability to classify archaeological sites was low. In addition, Russell studied sites thatwere probably still being used on a rotational basis, interspersed with short fallow peri-ods rather than long periods of complete abandonment, so the archaeological sites notunder cultivation were still highly disturbed sites. We predicted that using a time-seriesapproach would increase classification accuracy and allow for classifying ADE sitesthat have been abandoned for much longer periods of time.

The effects of ADE on agricultural vegetation are well understood (Schlesinger1991, Glaser et al. 2001, Major et al. 2005); however, the effects on forest vegeta-tion are less studied, primarily because there are few known forest ADE sites. Severalstudies have reported colloquial evidence that ADE can be identified by indicator treespecies, which tend to be more exotic and more useful that the species growing on non-ADE (Moran 1981, Woods and McCann 1999, Sombroek et al. 2002), and at least onestudy has been able to support that evidence empirically (Junqueira 2008). Junqueira(2008) established 52 plots of size 10 m ! 25 m along the middle Madeira River inAmazonas State, Brazil. Twenty-six of these were on ADE soils and 26 were on non-ADE soils. In addition to interviewing local caboclo farmers to learn how ADE-basedand non-ADE-based forests were used, Junqueira conducted a census of all trees withdiameter at breast height (DBH) "5 cm and all palms taller than 1 m. An analysis ofthese data revealed several important findings (Junqueira 2008):

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

4 J. B. Thayn et al.

1. ADE sites showed higher density (169 individuals, p = 0.025) and greater rich-ness (25 species, p = 0.06) of tree species than non-ADE sites (113 individuals,14 species).

2. Six ADE indicator species (mostly palms) and three non-ADE indicator specieswere identified, which supports the evidence collected during earlier ethnob-otanic studies (Moran 1981, Woods and McCann 1999, Sombroek et al.2002).

3. Despite the different palm species present on ADE, there was no differencein the number of palms growing on ADE versus non-ADE, nor was there adifference in palm species richness between the two soil types.

4. The principal difference between vegetation growing on the two soil types isspecies composition, in both woody plants and palms.

5. A significant second difference relates to vegetation structure. In typical trop-ical forests there is a reduction in woody plant density as succession advancesand the canopy closes, reducing the amount of light that reaches the subcanopy.On ADE sites, woody plant density remained high during succession, possiblybecause the higher soil fertility encourages a greater number of pioneer specieswith a shorter life cycle. As these individuals die and fall, openings occur inthe canopy and more light reaches the subcanopy allowing for a denser under-storey (Junqueira 2008). This finding is consistent with that of Laurance et al.(1999), who attribute the decline in biomass on Amazonian secondary foreststo poor soil quality (see also de Castilho et al. (2006)).

The main driver of vegetation phenology in ever-moist tropical forests is incomingphotosynthetically active radiation (PAR; Van Schaik et al. 1993, Wright and VanSchaik 1994, Huete et al. 2002, Myneni et al. 2007, Zimmerman et al. 2007) ratherthan soil moisture or precipitation. Henderson et al. (2000) studied the floweringphenologies of palm trees at the Smithsonian Institution/Brazil’s National Instituteof Amazonian Research (INPA) Biological Dynamics of Forest Fragments Project(BDFFP) located about 41 km north of the present study site. They determinedthat palms show no preference for wet- or dry-season flowering as a community, butthat individual taxa and species tend to flower in either the wet or the dry seasons.Specifically, they report that taxa of the genus Bactris tend to flower during the rainyseason; Junqueira (2008) identified at least one Bactris species as an ADE indicator.Oenocarpus minor Mart. is an ADE indicator species and it also flowers during therainy season. Taxa of Astrocaryum flower during the dry season and Junqueira identi-fied three members of this genera as non-ADE indicator species. However, one speciesof Attalea and one of Geonoma were identified by Junqueira as ADE indicators andthese genera tend to flower during the dry season. It seems that, with a few exceptions,the ADE indicator palm species tend to flower in the rainy season and the non-ADEindicator palm species tend to flower in the dry season.

ADE soils tend to exhibit greater density of woody species than do non-ADEsoils and this increased density remains as forest succession progresses, unlike thereduction in density that is typical of Amazonian forests (Junqueira 2008). EnhancedVegetation Index (EVI) values (Huete et al. 2002) and the near-infrared (NIR) bandsof the National Aeronautics and Space Administration (NASA) Moderate ResolutionImaging Spectroradiometer (MODIS) sensor have been shown to be sensitive to veg-etation structure (Gao et al. 2000). If the MODIS sensor is sensitive to the differentdensities of vegetation growing on ADE and on non-ADE, then a reliable model forpredicting the location of currently unknown ADE sites could be developed.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

Locating ADE 5

Compared to non-ADE sites, ADE sites tend to contain different tree species, par-ticularly palm species (Junqueira 2008). If these different species possess differentspectral reflectance properties, either consistently throughout the year or during spe-cific times of the year due to different flowering phenologies, then a remote sensingmodel for predicting the location of unknown ADE sites could be developed. Thedifficulty with testing this hypothesis is that most of these palm species are a part ofthe subcanopy and are therefore mostly occluded by taller woody species. However, astudy conducted 430 km east of the present study site found a slight increase in EVIvalues during the late dry season, which the authors linked to the phenology of theunderstorey (Huete et al. 2002). A similar increase was also identified by Myneni etal. (2007) and Xiao et al. (2006). If this increase is detected in our study site, it maycontribute to a classification scheme.

2. Methodology

2.1 Study site

The study site is a transect that begins in Manaus, Amazonas State, Brazil andruns west-southwest for nearly 400 km (figure 1). Petrobras, the Brazilian nationalpetroleum company, is constructing a gas line that will connect the city of Coari withthe city of Manaus. Once the new line is finished and connected to the existing 285-kmline that connects Urucu with Coari, it will transport 4.7 million m3 of natural gasto Manaus every day for electric power generation. An additional 125 km of gas linewill be constructed to connect the main line with the municipalities of Coari, Codajás,Anamã, Caapiranga, Manacapuru and Iranduba.

The surveyors for the new gas line were accompanied by archaeologists from theUniversity of São Paulo, who mapped and assessed the archaeological sites foundalong the route (Neves et al. 2007). Forty-one new ADE sites were discovered; ofthese, 28 where found along the new gas line that was cleared in 2006. Prior to 2006,these ADE sites were covered by forest. The majority of known ADE sites have longago been cleared of natural vegetation to take advantage of the soil’s high fertility for

Pre-existing Gas-Line

Figure 1. Map (Central Amazon River Basin) of the study site showing the gas line being con-structed by Petrobras and the numbered locations of the known ADE sites along the transect.The finished gas line will be 670 km long.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

6 J. B. Thayn et al.

agricultural purposes. These sites have typically been used for generations and underdifferent land management practices and the dates of clearing and fallow periods areoften difficult, if not impossible, to determine. The gas-line dataset uniquely providesthe location of ADE sites and the date of clearing so that pre-clearing imagery can beused to assess the effects of ADE on forest vegetation.

Non-ADE sites were selected at random intervals along the transect, at least 2 kmfrom known ADE sites. This ensured that there was no accidental overlap of an ADEsite.

2.2 Data

Complete annual time series of MODIS MOD13Q1 version 005 EVI (Huete et al.2002) imagery for 2001–2005 were downloaded from NASA’s WIST data gatewaywebsite (NASA 2010). MODIS surface reflectance data are processed as 16-day maxi-mum value composite (MVC) EVI data with 250-m spatial resolution. Pre-processinginvolves correcting for cloud and aerosol contamination as well as angular, Sun-target-sensor variations with an option to use bidirectional reflectance distribution function(BRDF) models. EVI exhibits less saturation in tropical regions than many vegetationindices (Didan 2002) and is related to forest stand biomass (Roberts et al. 2003), tropi-cal forest leaf litter-fall (Saleska et al. 2003, Xiao et al. 2005) and leaf canopy processes(Xiao et al. 2005), and is more sensitive to seasonal dynamics than other vegetationindices (Ferreira et al. 2003).

The traditional MVC process selects the highest pixel value as representative ofthe entire composite period, effectively reducing the effects of cloud interference andaerosol contamination in data that have not been atmospherically corrected (Holben1986). In data that have been atmospherically corrected prior to being composited(like MODIS), the MVC process tends to select pixels with large view and solar zenithangles, which may not be the most cloud-free pixels (Goward et al. 1991, Cihlar et al.1997). To correct for this problem, a constrained view angle maximum value compos-ite (CV-MVC) process is used for MODIS VI data, where the two highest VI valuesare compared and the observation with the view angle closest to nadir is selected torepresent the composite period (Huete et al. 2002). While this effectively limits cloudcontamination in most areas, Huete et al. (2002) found persistent cloud cover andcloud shadows near the Tapajos region of Brazil, which sits at approximately the samelongitude as the present study site.

To reduce the effects of any lingering cloud contamination, the time-series data weresmoothed prior to analysis using a modification of the mean value iteration (MVI)method introduced by Ma and Veroustraete (2006). MVI first identifies points alongthe time series that might be erroneous by comparing their values to the mean oftheir two neighbours such that if |(DNi–1 + DNi+1)/2 – DNi| > threshold, then thevalue at DNi is replaced with (DNi–1 + DNi+1)/2, where DNi is the digital numberin the ith position within the time series. This process is repeated until the absolutedifference between every point and the mean of its two neighbours is less than thethreshold. This method elevates downward troughs and compresses upward spikesto create a smoothed time series. As we were looking at tropical forest presumed tohave a slight unimodal annual phenological cycle (Van Schaik et al. 1993, Wright andVan Schaik 1994, Huete et al. 2002, Zimmerman et al. 2007), downward spikes wereassumed to be the result of cloud contamination. However, the values that sat onthe upper envelope were assumed to be accurate EVI values. It is desirable to replace

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

Locating ADE 7

any downward spikes but retain the upper envelope (Hird and McDermid 2009). Tomaintain the upper envelope we replaced only those values that were less than themean of their two neighbours. We used a threshold of 1000.

Of the 41 ADE sites discovered along the gas line, the first 13 were excluded fromthe analysis because they were located along the pre-existing gas line, which preventedpure forest pixels from being collected. Sites 26, 28 and 39 were also excluded becausethese sites are located along the banks of either the Manacapuru or Negro Rivers andat the 250-m resolution of MODIS these pixels contained water reflectance and werenot pure vegetation pixels (figure 1).

2.3 Overview of harmonic analysis

Many methods have been proposed for quantifying vegetation phenology using satel-lite imagery, such as curve fitting (Zhang et al. 2003) and moving windows (Ducheminet al. 1999). These methods, however, require a fairly strong seasonal pattern, whichis not present in the forest vegetation of the Amazon Basin. Harmonic analysis (HA),or discrete Fourier analysis, does not use windows, thresholds or curve fitting, so itreturns valid results from more subtle seasonal patterns. This makes HA much morerobust in the tropics, where differences in seasonal vegetation condition are very slight.

HA expresses a complex curve, such as an annual EVI time series, as the sum ofa series of cosine waves (Bloomfield 1976, Broughton and Bryan 2009). Each of thecosine waves is defined by a unique wavelength, phase and amplitude. The wavelength,or harmonic term, designates the number of cycles completed by the time series over itschronological range. The term of each wave is supplied by the user and the amplitudeand phase values are calculated to return the cosine wave that best fits the originaltime series. The cosine curves are calculated in order of their respective terms, not indecreasing order of fit to the time series. The first term cosine function is the best-fitcurve constrained to a frequency of one, the second term cosine function is the best-fitcurve constrained to a frequency of two, and so on. Successive harmonic curves canbe added to produce a more complex curve. The sum of all possible harmonic curvesreproduces the original time series. The lower-order waves demonstrate trends in thedata, while the higher-order waves contain mostly noise (Jakubauskas et al. 2001).

HA summarizes vegetation dynamics in two values, the amplitude and thephase. The amplitude of the first harmonic indicates the variability of seasonalproductivity over the year, as expressed in a single pulse of net primary pro-duction. The phase of the first harmonic summarizes the timing of vegetationgreen-up and senescence. Subsequent harmonic values indicate the strength (ampli-tude) and timing (phase) of higher frequency patterns, such as secondary veg-etation types. The additive term of the harmonic series, or the mean valueof the time series, indicates overall productivity. The amplitude and phase ofthe lower-order harmonics have been used successfully in land cover/land useclassification and they work especially well in differentiating vegetation func-tional groups (Jakubauskas et al. 2001, Moody and Johnson 2001, Jakubauskaset al. 2002).

The equations for amplitude and phase (Jakubauskas et al. 2001, Broughton andBryan 2009) are:

amplitude =!

Cf (x) + Sf (x) (1)

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

8 J. B. Thayn et al.

phase = arctan"

Cf (x)Sf (x)

#(2)

If Cf(x) is less than zero then ! is added to the phase. The equations for Cf(x) andSf(x) are:

Cf (x) =n$

i=1

"i ! cos

"2!xf

n

##! 2

n(3)

Sf (x) =n$

i=1

"i ! sin

"2!xf

n

##! 2

n(4)

where n is the length of the time series, i is the data value from the time series, x isthe temporal unit of each i and f is the term (or frequency) of the harmonic beingcalculated.

The phase values returned by equation (2) range from zero to 2! and, since theyare circular values, a phase of zero is equivalent to a phase of 2!. The phase of thefirst harmonic indicates the position of the crest of the cosine wave. A phase angle of! indicates that the curve peaks at the centre of the time series, while a phase angleof 2! indicates that the peak occurs at the extremes of the period and the trough of

(a)

Amp = 521.1Phase = 4.74r2 = 0.55

(b)

2002

(c)

05

100

510

05

10

Amp = 513.1Phase = 5.01r2 = 0.62

Amp = 662.3Phase = 6.09r2 = 0.77

Enh

ance

d ve

geta

tion

inde

x (E

VI)

(th

ousa

nds)

J F M A M J J A S O N D

Figure 2. Three examples of annual EVI signatures from Manaus, Brazil superimposed bytheir first harmonic curves. The amplitude and phase of each curve are provided, as well as thecorrelation between the time series and the harmonic curves.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

Locating ADE 9

the curve is located at the middle of the time series. Figure 2 shows three examples offorest annual time series with their first harmonic term curves, amplitudes and phasevalues.

2.4 Use of HA in vegetation studies

HA is commonly used in smoothing noisy data (Bradley et al. 2007) and replac-ing clouded pixels in vegetation index (VI) time series (Jun et al. 2004). Olsson andElkundh (1994) used a least-squares fitting procedure to compare a VI time series toits first and second harmonic curves to determine whether vegetation in Africa exhib-ited a monomodal or bimodal pattern. Brown et al. (2007) used a similar method tomap agricultural intensification in Vilhena, Brazil; they calculated the amplitudes ofthe first three harmonic terms and used the data to determine whether pixels repre-sented single-, double- or triple-cropped agricultural sites. This simple classificationscheme mapped agricultural intensification with 80% accuracy.

Morton et al. (2006) calculated several descriptive statistics, including the amplitudeand phase values, of MODIS VI time series collected over Mato Grosso, Brazil. Thesevariables were entered into a decision tree classifier that successfully classified landcover as either cropland, pastureland or regrowth forest.

Lacruz and Sousa (2007) mapped the flood plain of the Taquari River in Brazilusing HA and a 2005 MODIS VI time series. They calculated the curve of the firstharmonic term and than compared it to the original time series using the coefficientof determination. Their study site was a grassland, which has a strong monomodalseasonal pattern, and therefore the coefficient of determination was typically high.In the floodplain, where rising waters disrupt the grasses’ monomodal pattern, thecoefficient of determination was low. They found that the flood plains were clearlyidentified when the coefficient of determination was #0.20. They were also able todifferentiate between farmland and pastureland using amplitude and phase values.

3. Analysis and results

To test whether EVI values collected over ADE are statistically different from thosecollected over non-ADE due to increased vegetation density and differences in speciesreflectance values, a multiple analysis of variance (MANOVA) was performed on theannual means of the data for each year. This resulted in 25 observations (site locations)and five variables (annual means) for each soil type. Although 25 sample observa-tions is relatively small for a remote sensing study, we were limited to the number ofADE sites discovered by the archaeological team from the University of São Paulowho surveyed the gas-line transect (Neves et al. 2007). The annual means of EVI val-ues from vegetation growing on ADE are significantly statistically different from EVIvalues from vegetation growing on non-ADE sites at an alpha of 0.05 (MANOVAPillai = 0.226, p = 0.040). A logistic regression was performed on these data to deter-mine whether each sample site was an ADE location. A logistic regression returnsthe percentage likelihood of the presence of ADE in a pixel. A jack-knife procedurewas used as a rough accuracy assessment. For the purposes of this assessment, wedeemed that any result "0.5 was an ADE site and any result <0.5 was a non-ADEsite. Accuracy scores are reported as the percentage of sites classified correctly as eitherADE or non-ADE. This model correctly predicted the presence or absence of ADE in58% of cases.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

10 J. B. Thayn et al.

Figure 3. The difference between a time series of EVI values collected over ADE and non-ADE: (a) the composite period mean of the 25 samples; (b) p-values from t-tests conducted foreach composite period. The precipitation line is the mean of data collected at the municipalitiesof Barro Alto, Anamã, San Antonio do Içã and Caapiranga.

Figure 3(a) shows that vegetation growing on ADE typically has a lower EVI valuethan vegetation growing on non-ADE. This relationship is consistent throughout theyear, although the dispersion of values around their means is so large that significantoverlap occurs between the two datasets. To test whether the phenology of vegetationgrowing on ADE is statistically different from that of vegetation growing on non-ADE sites, t-tests were used to compare the two sets of EVI values from each periodin the 5-year time series. More conservative two-tailed t-tests were applied, althoughthere seems to be a tendency for ADE-based vegetation to have lower EVI values. Thep-values from the t-tests were plotted to determine at which period of the year the twosoil types could be most easily distinguished.

The series of t-tests clearly demonstrates that the difference between vegetationgrowing on ADE and vegetation growing on non-ADE is most clear during the dryseason (figure 3). In this area the dry season occurs from October to June. The resultsdisplayed in figure 3(b) show that the difference between ADE- and non-ADE-basedvegetation is discernible (t-test p-values <0.05) at roughly the same time period. Theperiods of significant p-values seem to precede the dry season and encroach into thewet season slightly. This is probably because atmospheric correction pre-processing ismore successful at eliminating the effects of cloud interference when there are fewerclouds in the image. When the wet season is well under way, and cloud interference isat its maximum, the pre-processing algorithms are unable to correct for all of the con-tamination, which obscures the difference between vegetation growing on ADE andvegetation growing on non-ADE.

To test whether the potential difference in seasonal phenology was statistically sig-nificant enough to lead to a reliable classification of ADE sites, a second MANOVAwas undertaken using the amplitude values for each year of the data. Based on the

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

Locating ADE 11

results of the series of t-tests, we hypothesized that ADE-based vegetation would becharacterized by larger amplitude values. The results of the analysis were insignificant(MANOVA Pillai = 0.163, p = 0.151). A jack-knifed logistic model was able to predictthe presence or absence of ADE accurately in 54% of cases.

The harmonic phase values provide information regarding the timing of differencesin seasonal phenology. Phase angles are circular variables, that is 0$ equals 360$ (or32 December equals 1 January); therefore, the phase angles were summarized usingthe circular mean and circular variance. Circular variance ranges from zero to one; avalue of one indicates that the angles are dispersed uniformly around the circle anda value of zero indicates that the angles are clustered together (Jakubauskas et al.2002). The dry season, which occurs between October and June, typically has higherEVI values than the wet season (Van Schaik et al. 1993, Wright and Van Schaik 1994,Huete et al. 2002, Myneni et al. 2007, Zimmerman et al. 2007), so we expected thephase angles to cluster near 2!. Phase angles are circular (i.e. 2! equals zero), so thephase angles would either be a little less than 2! or a little larger than zero. The meanphase angles for 2001–2005 for ADE are clustered around their mean of 0.38 (circularvariance = 0.39). The mean phase angles for non-ADE are centred on 0.40 (circularvariance = 0.59).

ADE sites typically had more concentrated, less dispersed phase angles than didnon-ADE sites (values ranged from 0.19 to 0.48 for ADE and from 0.35 to 0.81 fornon-ADE sites). A circular ANOVA was performed for each year to determine if therewas a significant difference between the phase angles associated with ADE and thoseassociated with non-ADE (Mardia and Jupp 1999). There was no significant statisticaldifference for any year analysed. The circular ANOVA p-values were 0.239, 0.254,0.344, 0.706 and 0.571 for years 2001–2005, respectively.

As the annual mean and the annual amplitude values of the data seemed to cap-ture different properties of the vegetation, we performed a third MANOVA usingboth variables for each of the 5 study years. This analysis included 10 variables, theannual mean and the amplitude values of each of the 5 years, for each of the 25study locations. While this model was not the most statistically significant model wetested (MANOVA Pillai = 0.294, p = 0.151), the jack-knifed logistic model was themost successful as it correctly identified the presence or absence of ADE in 64% ofcases.

We submitted the EVI time series images to this logistic model to create a predictivemap of ADE locations throughout the study site. For this map we selected those pixelswith values "0.9. The resulting map contained a lot of ‘shot-gun’ noise or single, iso-lated pixels that were classed as ADE. To clean the map we used a 7 ! 7 focal majorityprocess that replaced each pixel with the majority value of its 48 neighbours. Thisremoved the small single-cell sites that were probably noise in the model and retainedthe larger contiguous clusters of pixels that have a much greater probability of beingsignificant ADE locations. The resulting map identifies 431 potential ADE locations(figure 4).

4. Discussion

This research has shown that there is a difference in reflectance properties of vegetationgrowing on ADE and vegetation growing on non-ADE. There are two main reasonsfor this. The first, and probably most significant, is the difference in species compo-sition between the two soil types. ADE-based vegetation has higher species richness

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

12 J. B. Thayn et al.

Figure 4. Predictive map of ADE locations based on a logistic model using the mean andamplitude of time-series EVI data for each of the 5 years (2001–2005) as independent variables.Only those pixels whose likelihood of containing ADE was " 0.90 were included. The final mapwas cleaned using a 7 ! 7 focal majority process.

and different assemblages of palm species, although the number of palm trees growingper unit area seems to be consistent across the two soil types (Junqueira 2008). Theresults of this project suggest that, in general, ADE-based vegetation has lower ADEvalues than the nearby non-ADE-based vegetation; that is ADE-based vegetation haslower NIR reflectance and/or higher red reflectance. Future fieldwork using a portablespectroradiometer will further develop this idea.

Second, while ADE-based vegetation has consistently lower EVI values, the differ-ence is more pronounced during and immediately after the dry season. The work ofHenderson et al. (2000) indicates that the palm assemblages common to ADE soilstend to flower during the rainy season whereas the palm species common to non-ADEsoils tend to flower during the dry season. This variation in seasonal phenology prob-ably contributes to the more pronounced difference between the reflectance patternsduring the dry season. Variation in seasonal phenology may also be related to thepulse of tropical forest understorey discussed by Huete et al. (2002), Xiao et al. (2006)and Myneni et al. (2007).

As mentioned earlier, an indirect cause of the more pronounced difference betweenthe reflectance patterns may be the increased cloud contamination present during thewet season. Because of the increased presence of clouds, the CV-MVC process and theMVI smoothing process may be less able to remove the effects of cloud contaminationin wet season images than in dry season images. The lingering cloud contaminationpresent in the wet season images may diminish the difference between the spectralpatterns.

Although the differences in annual mean EVI values for ADE-based and non-ADE-based vegetation were statistically significant, the simple jack-knifed accuracyassessment of the model contains large amounts of potential error. It would take yearsof effort to perform a traditional, field-based accuracy assessment of the map that wasproduced. These locations are remote and inaccessible. It is this inaccessibility that hasprotected these ADE sites from clearing for agriculture. While the sites located alongthe Amazon/Solimões River may be accurate, the dense cluster of sites to the west ofthe Negro River seems unlikely (figure 4).

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

Locating ADE 13

This is the first known attempt to map ADE locations hidden under dense forestcanopy in the Amazon Basin, and is therefore the first step of a longer research initia-tive. Subsequent attempts will build on this work until a more refined model, with ahigher level of classification accuracy, is developed. Possible future improvements onthis analysis include the following:

1. Although differences in the understorey species composition are indicatorsof ADE, the effect of these species was muted by the forest canopy. Futureattempts to use remotely sensed data to predict ADE location should focuson the spectral properties of the upper canopy rather than the understorey.Unfortunately, the effects of ADE on upper canopy species is less understoodthan the effects on the understorey. Junqueira (2008) found that there is adifference in the woody species assemblages between the two soil types, so itis possible that there are significantly difference spectral reflectance patterns.Additional botanical studies and fieldwork using a portable spectroradiometerare needed to determine whether this can be used to strengthen the predictivemodel.

2. Eighty per cent of known ADE patches are smaller than 2 ha. At the spa-tial resolution of MODIS (roughly 6 ha per pixel), these smaller sites are mixedpixels. The use of imagery with finer spatial resolution would reduce mixed pix-els and could result in better classification accuracy. Russell’s (2005) moderatesuccess with Landsat imagery for identifying archaeological sites suggests thatthis may be the case. The results presented here indicate that a loss in temporalresolution, in exchange for an increase in spatial resolution, may not be as detri-mental as previously thought. The Advanced Spaceborne Thermal Emissionand Reflection Radiometer (ASTER) sensor aboard the Terra satellite has15-m spatial resolution but much reduced, and irregularly spaced, temporalresolution. Future research is needed to determine if such drastically reducedtemporal resolution captures the seasonal variation used in this study to locatepotential ADE sites.

3. The geographic coordinates of the ADE sites used in this study were collectedby archaeologists from the University of São Paulo, who collected the data forbasic mapping purposes. We suspect that an effort to ensure points were takenin the centre of the ADE patches, which would reduce mixed pixels, wouldresult in increased classification accuracy. Unfortunately, forest clearing andconstruction is currently under way, limiting access for research purposes.

4. Another source of classification confusion may be the soil moisture capacityof clayey soils in the region. Earlier studies have indicated that ADE soils arebetter able to retain moisture than non-ADE soils (Lehmann et al. 2003b).However, Teixeira (personal communication, 2008) found that, even thoughADE has slightly higher moisture retention than non-ADE soils with simi-lar microstructure, the increased soil moisture capacity of ADE is primarily afunction of its clay content, so that clayey Oxisols and ADE share similar soilmoisture capacities. It is likely, then, that at least some clayey non-ADE sitesmay be misclassified as ADE soils.

A satellite-based model with the ability to accurately predict the location ofcurrently unknown ADE sites would greatly benefit the archaeological and biogeo-graphical research communities as well as the cultural heritage of the region. The

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

14 J. B. Thayn et al.

research presented here provides a step in the right direction, but additional workis required to create a suitable model.

ReferencesBLOOMFIELD, P., 1976, Fourier Analysis of Time Series: An Introduction (New York, NY: John

Wiley & Sons).BRADLEY, B.A., JACOB, R.W., HERMANCE, J.F. and MUSTARD, J.F., 2007, A curve fitting pro-

cedure to derive inter-annual phenologies from time series of noisy satellite NDVI data.Remote Sensing of Environment, 106, pp. 137–145.

BROUGHTON, S.A. and BRYAN, K., 2009, Discrete Fourier Analysis and Wavelets: Applicationsto Signal and Image Processing (Hoboken, NJ: John Wiley & Sons).

BROWN, J.C., JEPSON, W.E., KASTENS, J.H., LOMAS, J.M. and PRICE, K.P., 2007, Multitemporal,moderate-spatial-resolution remote sensing of modern agricultural production and landmodification in the Brazilian Amazon. GIScience and Remote Sensing, 44, pp. 117–148.

CIHLAR, J., LY, H., LI, Z., CHEN, J., POKRANT, H. and HUANG, F., 1997, Multi-temporal, multi-channel, AVHRR data sets for land biosphere studies: artifacts and corrections. RemoteSensing of Environment, 60, pp. 35–57.

DE CASTILHO, C.V., MAGNUSSON, W.E., DE ARAUJO, R.N.O., LUIZAO, R.C.C., LUIZAO, F.J.,LIMA, A.P. and HIGUCHI, N., 2006, Variation in aboveground tree live biomass in a cen-tral Amazonian forest: effects of soil and topography. Forest Ecology and Management,234, pp. 85–96.

DIDAN, K., 2002, MODIS vegetation index production algorithms. In MODIS Outreachon MODIS Vegetation Variables (VI/LAI/FPAR/NPP), 15–18 July, University ofMontana, Missoula, MT.

DUCHEMIN, B., GOUBIER, J. and COURRIER, G., 1999, Monitoring phenological key stages andcycle duration of temperate deciduous forest ecosystems with NOAA/AVHRR data.Remote Sensing of Environment, 67, pp. 68–82.

FERREIRA, L., YOSHIOKA, H., HUETE, A. and SANO, E., 2003, Seasonal landscape and spectralvegetation index dynamics in the Brazilian Cerrado: an analysis within the large-scalebiosphere-atmosphere experiment in Amazônia (LBA). Remote Sensing of Environment,87, pp. 534–550.

GAO, X., HUETE, A., NI, W. and MIURA, T., 2000, Optical-biophysical relationships of vege-tation spectra without background contamination. Remote Sensing of Environment, 74,pp. 609–620.

GLASER, B., GUGGENBERGER, G. and ZECH, W., 2001, The terra preta phenomenon: a modelfor sustainable agriculture in the humid tropics. Naturwissenschaften, 88, pp. 37–41.

GLASER, B., GUGGENBERGER, G., ZECH, W. and RUIVO, M.D.L., 2003, Soil organic matterstability in Amazonian dark earths. In Amazonian Dark Earths: Origin, Properties,Management, J. Lehmann, D. Kern, B. Glaser and W.I. Woods (Eds.), pp. 141–158 (TheNetherlands: Kluwer Academic Publishers).

GLASER, B. and WOODS, W.I., 2004, Amazonian Dark Earths: Explorations in Space and Time(Berlin: Springer-Verlag).

GOWARD, S.N., MARKHAM, B., DYE, D.G., DULANEY, W. and YANG, J., 1991, Normalizeddifference vegetation index measurements for the advanced very high resolutionradiometer. Remote Sensing of Environment, 35, pp. 257–277.

HECKENBERGER, M.J., PETERSON, J.B. and NEVES, E.G., 1999, Village size and permanencein Amazonia: two archaeological examples from Brazil. Latin American Antiquity, 10,pp. 353–376.

HENDERSON, A., FISCHER, B., SCARIOR, A., PACHECO, M.A.W. and PARDINI, R., 2000,Flowering phenology of a palm community in a Central Amazon forest. Brittonia, 52,pp. 149–159.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

Locating ADE 15

HIRD, J.N. and MCDERMID, G.J., 2009, Noise reduction of NDVI time series: an empiricalcomparison of selected techniques. Remote Sensing of Environment, 113, pp. 248–258.

HOLBEN, B.N., 1986, Characteristics of maximum value composite images from temporalAVHRR data. International Journal of Remote Sensing, 7, pp. 1417–1434.

HUETE, A., DIDAN, K., MIURA, T., RODRIGUES, E.P., GAO, X. and FERREIRA, L.G., 2002,Overview of the radiometric and biophysical performance of the MODIS vegetationindices. Remote Sensing of Environment, 83, pp. 195–213.

JAKUBAUSKAS, M.E., LEGATES, D.R. and KASTENS, J.H., 2001, Harmonic analysis of time-series AVHRR NDVI data. Photogrammetric Engineering and Remote Sensing, 67,pp. 461–470.

JAKUBAUSKAS, M.E., PETERSON, D. and KASTENS, J.H., 2002, Time series remote sensingof landscape-vegetation interactions in the southern Great Plains. PhotogrammetricEngineering and Remote Sensing, 68, pp. 1021–1030.

JUN, W., ZHONGBO, S. and YAOMING, M., 2004, Reconstruction of a cloud-free vegetationindex time series for the Tibetan plateau. Mountain Research and Development, 24,pp. 348–353.

JUNQUEIRA, A.B., 2008, The Use and Management of Secondary Forest over Terra Preta byTraditional Communities in the Central Region of the Madeira River, Amazonas, Brazil(Manaus, Brazil: INPA/UFAM).

KERN, D., D’AQUINO, G., RODRIGUES, T., FRAZÃO, F.J.L., SOMBROEK, W., MYERS, T.P. andNEVES, E.G., 2003, Distribution of Amazonian dark earths in the Brazilian Amazon.In Amazonian Dark Earths: Origin, Properties, Management, C.L. Lehman, D. Kern,B. Glaser and W.I. Woods (Eds.), pp. 51–75 (The Netherlands: Kluwer AcademicPublishers).

KIMES, D., NELSON, R., SKOLE, D. and SALAS, W., 1998, Accuracies in mapping secondarytropical forest age from sequential satellite imagery. Remote Sensing of Environment,65, pp. 112–120.

LACRUZ, M.S.P. and SOUSA, M.D.A.J., 2007, The use of MODIS EVI time series and har-monic analysis to study the Taquari River basin. Revista Brasileira de Cartografia, 59,pp. 9–15.

LAURANCE, W.F., LAURANCE, S.G., DELAMONICA, P., RANKIN-DE MERONA, J.M., GASCON,C., FEARNSIDE, P.M., LOVEJOY, T.E. and CHAMBERS, J.Q., 1999, Relationship betweensoils and Amazon forest biomass: a landscape-scale study. Forest Ecology andManagement, 118, pp. 127–138.

LEHMANN, J., 2007, A handful of carbon. Nature, 447, pp. 143–144.LEHMANN, J., DA SILVA, J.P., Jr., STEINER, C., NEHLS, T., ZECH, W. and GLASER, B., 2003a,

Nutrient availability and leaching in an archaeological Anthrosol and a Ferrasol in theCentral Amazon basin: fertilizer, manure and charcoal amendments. Plant and Soil,249, pp. 343–357.

LEHMANN, J., GAUNT, J. and RONDON, M., 2006, Bio-char sequestration in terrestrialecosystems – a review. Mitigation and Adaptation Strategies for Global Change, 11,pp. 395–419.

LEHMANN, J., KERN, D., GERMAN, L., MCCANN, J.M., MARTINES, G.C. and MORIERA, A.,2003b, Soil fertility and production potential. In Amazonian Dark Earths: Origin,Properties, Management, J. Lehmann, D. Kern, B. Glaser and W.I. Woods (Eds.),pp. 105–124 (The Netherlands: Kluwer Academic Publishers).

LU, D., MORAN, E. and BATISTELLA, M., 2003, Linear mixture model applied to Amazonianvegetation classification. Remote Sensing of Environment, 87, pp. 456–469.

MA, M. and VEROUSTRAETE, F., 2006, Reconstructing pathfinder AVHRR land NDVI time-series data for the northwest of China. Advances in Space Research, 37, pp. 835–840.

MAJOR, J., DITOMMASO, A., LEHMANN, J. and FALÇÃO, N.P.D.S., 2005, Weed dynamicson Amazonian dark earth and adjacent soils of Brazil. Agriculture, Ecosystems andEnvironment, 111, pp. 1–12.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

16 J. B. Thayn et al.

MARDIA, K. and JUPP, P., 1999, Directional Statistics, Section 7.4 (Chichester: John Wiley &Sons).

MARRIS, E., 2006, Black is the new green. Nature, 442, pp. 624–626.MEDDENS, A.J.H., 2006, Possibilities of mapping Amazonian Dark Earths using remote sensing

techniques. Wageningen University, Wageningen, The Netherlands.MOODY, A. and JOHNSON, D., 2001, Land-surface phenologies from AVHRR using the discrete

Fourier transform. Remote Sensing of Environment, 57, pp. 305–323.MORA, S., HERRERA, L., CAVELIER, I. and RODRIGUES, C., 1991, Cultivars, Anthropic Soils and

Stability: A Preliminary Report of Archaeological Research in Araracuara, ColombianAmazonia. Latin American Archaeological Report No. 2 (Pittsburgh, PA: University ofPittsburgh).

MORAN, E.F., 1981, Developing the Amazon (Bloomington,IN: Indiana University Press).MORTON, D.C., DEFRIES, R.S., SHIMABUKURO, Y.E., ANDERSON, L.O., ARAI, E., DEL BON

ESPIRITO-SANTO, F., FREITAS, R. and MORISETTE, J., 2006, Cropland expansionchanges deforestation dynamics in the southern Brazilian Amazon. Proceedings of theNational Academy of Sciences of the United States of America, 103, pp. 14637–14641.

MYNENI, R., YANG, W., NEMANI, R.R., HUETE, A., DICKINSON, R.E., KNYAZIKHIN, Y.,DIDAN, K., FU, R., JUÁREZ, R.I.N., SAATCHI, S.S., HASHIMOTO, H., ICHII, K.,SHABANOV, N.V., TAN, B., RATANA, P., PRIVETTE, J.L., MORISETTE, J., VERMOTE, E.F.,ROY, D.P., WOLFE, R.E., FRIEDL, M.A., RUNNING, S.W., VOTAVA, P., EL-SALEOUS, N.,DEVADIGA, S., SU, Y. and SALOMONSON, V.V., 2007, Large seasonal swings in leaf areaof Amazon rainforests. Proceedings of the National Academy of Sciences of the UnitedStates of America, 104, pp. 4820–4823.

NASA, 2010, NASA Warehouse Inventory Search Tool. Available online at: https://wist.echo.nasa.gov/api/ (accessed 28 June 2010).

NEVES, E.G., DA SILVA, C.A., LIMA, H.P., CARNEIRO, C.G. and SAMPAIO, A., 2007,Archaeological survey of the Coari-Manaus gas line: second partial report, survey ofclearings, the transect, and preliminary results of the excavations. Report prepared bythe archaeological team, University of São Paulo.

NEVES, E.G., PETERSON, J.B., BARTONE, R.N. and DA SILVA, C.A., 2003, Historical andsocio-cultural origins of Amazonian Dark Earths. In Amazonian Dark Earths: Origin,Properties, Management, J. Lehmann, D. Kern, B. Glaser and W.I. Woods (Eds.),pp. 297–50 (The Netherlands: Kluwer Academic Publishers).

OLSSON, L. and EKLUNDH, L., 1994, Fourier series for analysis of temporal sequences of satellitesensor imagery. International Journal of Remote Sensing, 15, pp. 3735–3741.

ROBERTS, D.A., KELLER, M. and SOARES, J.V., 2003, Studies of land-cover, land-use, and bio-physical properties of vegetation in the large scale biosphere atmosphere experiment inAmazônia. Remote Sensing of Environment, 87, pp. 377–388.

RUSSELL, J.C., 2005, Integrated approach to predictive modeling: a case study from theUpper Xingu (Matto Grosso, Brazil). PhD dissertation, Department of Anthropology,University of Florida.

SALESKA, S., MILLER, S., MATROSS, D., GOULDEN, M., WOFSY, S., DA ROCHA, H., DE

CARMARGO, M., CRILL, P., DAUBE, B., DE FREITAS, H., HUTYA, L., KELLER, M.E.,KIRCHHOFF, V., MENTON, M., MUNGER, J., PYLE, E., RICE, A. and SILVA, H., 2003,Carbon in amazon forests: unexpected seasonal fluxes and disturbance-induced losses.Science, 60, pp. 315–355.

SALOVAARA, K.J., THESSLER, S., MALIK, R.N. and TUOMISTO, H., 2005, Classification ofAmazonian primary rain forest vegetation using Landsat ETM+ satellite imagery.Remote Sensing of Environment, 97, pp. 39–51.

SCHLESINGER, W.H., 1991, Biogeochemistry: An Analysis of Global Change (San Diego, CA:Academic Press).

SOMBROEK, W., 1966, Amazon Soil: A Reconnaissance of the Soils of the Brazilian AmazonRegion (Wageningen: Centre for Agricultural Publications and Documentation).

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1

Locating ADE 17

SOMBROEK, W., KERN, D., RODRIGUES, T., DA SILVO CRAVO, M., JARBAS, T.C., WOODS, W.I.and GLASER, B., 2002, Terra Preta and Terra Mulata: pre-Columbian Amazon kitchenmiddens and agricultural fields, their sustainability and their replication. In 17th WorldCongress of Soil Science, 14–21 August, Bangkok, Thailand.

STEINER, C., TEIXEIRA, W.G., LEHMANN, J., NEHLS, T., VASCONCELOS DE MACÊDO, J.L.,BLUM, W.E.H. and ZECH, W., 2007, Long term effects of manure, charcoal and mineralfertilization on crop production and fertility on a highly weathered Central Amazonianupland soil. Plant and Soil, 291, pp. 275–290.

STEININGER, M., 2000, Satellite estimation of tropical secondary forest above-groundbiomass: data from Brazil and Bolivia. International Journal of Remote Sensing, 21,pp. 1139–1157.

THAYN, J.B., PRICE, K.P. and WOODS, W.I., 2008, Locating Amazonian Dark Earths (ADE)using satellite remote sensing – a possible approach. In Amazonian Dark Earths: WimSombroek’s Vision, W.I. Woods, W.G. Teixeira, J. Lehmann, C. Steiner, A. WinklerPrinsand L. Rebellato (Eds.), pp. 279–298 (Berlin: Springer-Verlag).

TIESSEN, H., CUEVAS, E. and CHACON, P., 1994, The role of soil organic matter in sustainingsoil fertility. Nature, 371, pp. 783–785.

VAN SCHAIK, C., TERBOUGH, J. and WRIGHT, S., 1993, The phenology of tropical forests-adaptive significance and consequences for primary consumers. Annual Review ofEcology and Systematics, 24, pp. 353–377.

VIEIRA, I.C., SILVA DE ALMEIDA, A., DAVISON, E.A., STONE, T.A., RIES DE CARVALHO, C.J. andGUERRERO, J.B., 2003, Classifying successional forests using Landsat spectral proper-ties and ecological characteristics in eastern Amazônia. Remote Sensing of Environment,87, pp. 470–481.

WOODS, W.I., 1995, Comments on the Black Earths of Amazonia. Applied GeographyConferences, 17, pp. 159–165.

WOODS, W.I. and MCCANN, J.M., 1999, The anthropogenic origin and persistenceof Amazonian Dark Earths. Yearbook 1999 – Conference of Latin AmericanistGeographers, 25, pp. 7–14.

WRIGHT, S. and VAN SCHAIK, C., 1994, Light and phenology of tropical trees. The AmericanNaturalist, 143, pp. 192–199.

XIAO, X., HAGEN, S., ZHANG, Q., KELLER, M.E. and MOORE, B.I., 2006, Detecting leaf phenol-ogy of seasonally moist tropical forests in South America with multi-temporal MODISimages. Remote Sensing of Environment, 103, pp. 465–473.

XIAO, X., ZHANG, Q., SALESKA, S., HUTYRA, L., DE CAMARGO, P., WOFSY, S.C., FROLKING,S., BOLES, S., KELLER, M.E. and MOORE III, B., 2005, Satellite-based modeling of grossprimary production in a seasonally moist tropical evergreen forest. Remote Sensing ofEnvironment, 94, pp. 105–122.

ZHANG, X., FRIEDL, M.A., SCHAAF, C.B., STRAHLER, A.H., HODGES, J.C., GAO, F., REED,B.C. and HUETE, A., 2003, Monitoring vegetation phenology using MODIS. RemoteSensing of Environment, 84, pp. 471–475.

ZIMMERMAN, J.K., WRIGHT, S.J., CALDERÓN, O., PAGAN, M.A. and PATON, S., 2007, Floweringand fruiting phenologies of seasonal and aseasonal neotropical forests: the role ofannual changes in irradiance. Journal of Tropical Ecology, 23, pp. 231–251.

Dow

nloa

ded

by [I

llino

is St

ate

Uni

vers

ity M

ilner

Lib

rary

], [J

onat

han

Thay

n] a

t 12:

46 0

1 A

ugus

t 201

1