Relationships between the mineralogical and chemical ...geog.ucsb.edu/viper/viper_pubs/galvao_et_al08.pdfRelationships between the mineralogical and chemical composition of tropical

Available online at www.sciencedirect.com

emote Sensing 63 (2008) 259–271www.elsevier.com/locate/isprsjprs

ISPRS Journal of Photogrammetry & R

Relationships between the mineralogical and chemical compositionof tropical soils and topography from hyperspectral

remote sensing data

Lênio Soares Galvão a,⁎, Antônio Roberto Formaggio a,Eduardo Guimarães Couto b, Dar A. Roberts c

a Instituto Nacional de Pesquisas Espaciais (INPE), Divisão de Sensoriamento Remoto, Caixa Postal 515,12245-970, São José dos Campos, SP, Brazil

b Universidade Federal de Mato Grosso (UFMT), Faculdade de Agronomia e Medicina Veterinária,Av. Fernando Correa, s/n, Boa Esperança, 78060-900, Cuiabá, MT, Brazil

c University of California, Geography Department, Santa Barbara, CA, 93106-4060, USA

Received 23 January 2007; received in revised form 25 September 2007; accepted 27 September 2007Available online 7 November 2007

Abstract

Airborne Visible/InfraRed Imaging Spectrometer (AVIRIS) data acquired in 1995 in the vicinity of Campo Verde, central Brazil,were analyzed to investigate the relationships between the mineralogical and chemical composition of different soil types andtopography. Band depth analysis following the continuum removal separated areas of exposed soils from the other scene components.Principal component analysis (PCA) was applied to this subset of pixel spectra of exposed soils. The Spectral Feature Fitting (SFF)technique was used for mineral identification in the scene. Regression relationships between the silica/aluminum ratio and theabsorption band depth values at 2210 nm (kaolinite) and 2260 nm (gibbsite), calculated from laboratory spectra after the continuumremoval, were used to estimate the Ki index (1.7SiO2/Al2O3) in AVIRIS spectra, an indicator of the degree of soil weathering in soilsurveys. Results were plotted over an Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER)-derived digitalelevation model showing that several soil properties varied with surface elevation. Low reflectance soils (e.g., Rhodic Acrustox withnegative first component (PC1) scores, clay texture and higher organic carbon, SiO2, Al2O3, Fe2O3 and TiO2 content) predominated athigher elevations and high reflectance soils (e.g., Ustic Quartzipsamments with positive PC1 scores, sand texture and lower content ofthese constituents) at lower elevations. In some portions of the study area, soil composition changed gradually from Rhodic Acrustoxat higher elevations to Xanthic Acrustox at lower elevations, or from lower average soil reflectance (negative PC1 scores) and deeperkaolinite absorption band to higher average soil reflectance (positive PC1 scores) and deeper gibbsite absorption band. When appliedto AVIRIS data, the laboratory-derived relationship between the Ki index and the depth of the 2210 and 2260 nm absorption bandsshowed a good correspondence with spectral fits for kaolinite and gibbsite. Kaolinitic areas were associated with high Ki values due toa high SiO2 and lowAl2O3 content, whereas gibbsitic areas corresponded to low Ki values and highly weathered soil surfaces due to alow SiO2 and high Al2O3 content.© 2007 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved.

Keywords: Hyperspectral remote sensing; Tropical soils; AVIRIS; Topography; Mineral identification

⁎ Corresponding author.E-mail address: [email protected] (L.S. Galvão).

0924-2716/$ - see front matter © 2007 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V.All rights reserved.doi:10.1016/j.isprsjprs.2007.09.006

mailto:[email protected]

http://dx.doi.org/10.1016/j.isprsjprs.2007.09.006

260 L.S. Galvão et al. / ISPRS Journal of Photogrammetry & Remote Sensing 63 (2008) 259–271

1. Introduction

The genesis of tropical soils usually results in highlyweathered surfaces due to a desilification process or theremoval of silica and an accumulation of aluminum(Moniz et al., 1982; Curi and Franzmeier, 1984, 1987;Macedo and Bryant, 1987; Baptista et al., 1998). Tropicalweathered soils are characterized by the predominance ofa relatively simple mineralogy composed of quartz,goethite, hematite, kaolinite, gibbsite and opaque miner-als (e.g., ilmenite and magnetite) (Madeira Netto, 1993;Madeira et al., 1997). Except for quartz and opaqueminerals, the other minerals display well-defined absorp-tion bands in the Visible/Near Infrared (VNIR) region(goethite and hematite) and in the Shortwave Infrared(SWIR) interval (kaolinite and gibbsite) that can bedetected by laboratory/field instruments (reflectancespectroscopy) or airborne/orbital imaging systems (hyper-spectral remote sensing) with high spectral resolution.

Hyperspectral remote sensing provides new perspec-tives for the study of tropical soils due to the possibilityof measuring, on a per-pixel basis, iron oxide and claymineral absorption bands usually observed in laboratoryspectra. Since landscape position influences somechemical and physical properties of tropical soils andtheir mineralogy (Curi and Franzmeier, 1984; Uhlmannet al., 1997; da Motta et al., 2002; Cunha et al., 2005),these measurements can be related to topographic data toaid soil mapping. In spite of the potential of hyperspec-tral data to add information for soil mapping, these datahave not been widely used by the soil community.Examples of applications include Baptista et al. (1998),Palacios-Orueta and Ustin (1998), Ben-Dor et al. (2002),Galvão et al. (2001), Pizarro et al. (2001), Chabrillat et al.(2002) and Dehaan and Taylor (2004).

Most of the hyperspectral soil studies reported in theliterature did not consider topography in their data analysis.The combined use of different techniques of spectralanalysis (e.g., Principal Components Analysis (PCA), banddepth analysis and Spectral Feature Fitting (SFF)) and theirassociation with a Digital Elevation Model (DEM) mayprovide a better comprehension of the variation patterns ofsoil composition along toposequences. Such an approachmay also include the potential determination of the Ki ratio(1.7SiO2/Al2O3), a common index that indicates the degreeof alteration or weathering in tropical soils. For a Brazilianlatosol, this index is an indirect measure of the relativeproportions of kaolinite and gibbsite (Resende and Santana,1998) and can be estimated from reflectance valuesassociated with the 2210 nm (kaolinite) and 2260 nm(gibbsite) absorption bands (Madeira Netto, 1993; Madeiraet al., 1995; Baptista et al., 1998).

In this article, variations in soil types and in theirmineralogical and chemical composition, detected fromAirborne Visible/InfraRed Imaging Spectrometer(AVIRIS) data, were analyzed in relation to changes intopography obtained from an Advanced SpaceborneThermal Emission and Reflection Radiometer (ASTER)-derived DEM. Using reference spectra from a spectrallibrary, mineral identification of iron oxides and clayminerals was performed on a per-pixel basis with the SFFtechnique. Laboratory-derived regression relationshipsbetween the silica/aluminum ratio (Ki index) determinedfrom chemical analysis and absorption band depth valueswere applied to the AVIRIS scene to characterize thedegree of soil surface alteration and its relationship withmineralogy.

2. Study area

The study area (22×11 km) is located in the vicinity ofCampo Verde in the state of Mato Grosso in central Brazil(Fig. 1). According to the Brazilian system of soilclassification (Embrapa, 1999), the area is characterizedby the presence of the following major soil types:dystrophic Red Latosol (Rhodic Acrustox in the SoilTaxonomy), dystrophic Red–Yellow Latosol (TypicAcrustox), dystrophic Yellow Latosol (Xanthic Acrustox),orthic Quartzarenic Neosol (Ustic Quartzipsamments) andconcretionary Petric Plintosol (Petroferric Acrustox). Asmall occurrence of acriferric Red Latosol (AnionicAcrustox)was also observed in the study area. Henceforth,the Soil Taxonomy will be used to facilitate reading.

The natural vegetation is composed of savannas,which have been rapidly replaced by crops. The studyarea is dominated by agricultural activities that includethe growing of soybeans, cotton and corn. As a result,most of the areas of exposed soils are plowed fieldswithout the influence of physical or biogenic soil crusts.Dust also did not influence the spectral reflectance. Theclimate is tropical with a mean temperature of 21 °C.The annual rainfall may reach 1500 mm with a well-defined rainy season between December and May. Thetopography is gently undulating. The mean elevation is705 m with minimum and maximum values of 480 mand 850 m, respectively.

3. Methodology

3.1. Data collection

3.1.1. Soil analysesA total of 86 soil samples were collected in the study

area (22×11 km) from both the surface (0–20 cm; 43

Fig. 1. Location of the study area in the state of Mato Grosso (Campo Verde city). An AVIRIS image (band at 865 nm) is presented.

261L.S. Galvão et al. / ISPRS Journal of Photogrammetry & Remote Sensing 63 (2008) 259–271

samples) and subsurface (40–60 cm; 43 samples) of thesoil profiles. Bidirectional reflectance of 2-mm sieved,air-dried samples was measured in laboratory using anAnalytical Spectral Devices (ASD) FieldSpec instru-ment (Analytical Spectral Devices, Boulder, CO). TheASD collects spectra between 350 and 2500 nm at asampling interval of approximately 0.7 nm in the VNIRand 2 nm in the SWIR. Spectra were standardized with aspectralon panel (Labsphere, NH). A tungsten lamp wasused as a light source. The sensor acquired data at thenadir, with no foreoptic attached, at a height of 30 cmfrom the samples, which were placed in 40-cm diameterplates. The lamp was mounted at a zenith angle of 15° ata distance of 1 m from the samples. The signal-to-noiseratio of FieldSpec ranged from 10,000 in the visible(550 nm) and 6000 in the NIR (860 nm) to 400 in theSWIR (2200 nm).

Soil laboratory reflectance spectra were used toquantify color, one of the most important characteristics

used to differentiate tropical soils. For this purpose, the1976 CIE-UCS (Commission Internationale l´Eclairage-Uniform Chromaticity Scale) colorimetric system with aC-illuminant was chosen. To facilitate color interpretationfrom the chromaticity diagram, the two Helmholtzcoordinates, dominant wavelength and excitation purity,were calculated as described by Bedidi et al. (1992).

From the 86 soil samples, 50 samples (25 from thesurface and 25 from the subsurface) were submitted forphysico-chemical and mineralogical (X-ray diffraction)analysis (Agroanálise Laboratory, Cuiabá, MT), based onfieldwork information and on the inspection of color andreflectance spectra differences. Total sand, silt and claycontent were determined by the densimeter method(Camargo et al., 1986). Using the methodology ofEmbrapa (1997), the following constituentswere obtained:organic carbon (potassium dichromate oxidation andcolorimetric analysis), SiO2, Fe2O3, Al2O3, TiO2, MnOand P2O5 (sulfuric acid and colorimetric analysis),

Table 1Average and standard deviation (parenthesis) values of physico-chemical data of 50 soil samples from the surface (0–20 cm) and subsurface (40–60 cm) layers of the major soils types under study

Soiltype

Depth(cm)

Sand(%)

Silt(%)

Clay(%)

Organiccarbon(g/kg)

P2O5

(g/kg)SiO2

(g/kg)Fe2O3

(g/kg)MnO(g/kg)

Al2O3

(g/kg)TiO2

(g/kg)Mineralogy(X-ray diffraction)

N

RA 0–20 38.17 10.21 51.61 25.45 0.52 109.21 41.05 0.08 63.59 5.46 Q–K–H–G–Gt 8(29.22) (4.34) (24.93) (7.33) (0.22) (61.98) (23.88) (0.05) (34.28) (2.69)

40–60 34.44 9.39 56.17 12.42 0.33 122.87 67.47 0.05 120.41 8.51 Q–K–H–G–Gt 8(27.53) (2.49) (2.52) (3.47) (0.12) (68.19) (21.58) (0.05) (74.04) (4.03)

TA 0–20 41.70 10.45 47.85 26.22 0.55 63.67 45.27 0.07 92.37 6.65 Q–G–K–H–Gt 4(23.63) (3.97) (19.66) (5.63) (0.14) (25.25) (21.77) (0.04) (51.7) (3.03)

40–60 40.05 8.35 51.60 12.90 0.31 72.60 71.90 0.04 155.92 11.62 Q–G–K–H–Gt 4(22.44) (2.31) (20.14) (3.47) (0.09) (36.69) (33.46) (0.03) (111.69) (6.21)

XA 0–20 56.04 8.16 35.79 26.24 0.48 74.94 20.85 0.05 68.02 4.51 Q–G–K–H–Gt 11(20.40) (3.20) (17.46) (4.20) (0.19) (21.96) (10.45) (0.03) (33.74) (1.86)

40–60 50.32 7.83 41.85 13.75 0.26 82.36 47.04 0.03 118.44 7.15 Q–G–K–H–Gt 11(19.59) (2.24) (17.59) (3.62) (0.11) (25.26) (23.61) (0.02) (37.00) (1.93)

PA 0–20 50.90 8.30 40.80 24.50 0.40 63.30 34.90 0.05 95.40 6.30 Q–G–K 1– – – – – – – – – –

40–60 50.90 6.70 42.40 17.80 0.34 93.00 88.90 0.04 187.10 13.80 Q–H–K–G 1– – – – – – – – – –

UQ 0–20 87.30 3.40 9.30 14.70 0.28 28.20 5.10 0.04 10.40 1.00 Q–K 1– – – – – – – – – –

40–60 86.60 3.30 10.10 8.30 0.19 27.50 5.90 0.03 11.30 1.20 Q–K 1– – – – – – – – – –

Soil abbreviations are: RA (Rhodic Acrustox), TA (Typic Acrustox), XA (Xanthic Acrustox), PA (Petroferric Acrustox) and UQ (UsticQuartzipsamments). Mineral abbreviations are: Q (quartz), K (kaolinite), G (gibbsite), H (hematite) and Gt (goethite). N is the number of samples.

Fig. 2. Reflectance spectra of pairs of soil samples from the surface(black line) and subsurface (gray line) layers.


exchangeable Ca2+, Mg2+ (atomic absorption spectropho-tometry) and Al3+ (titration with NaOH), and K2O (flamephotometry). Water pH was determined with glasselectrodes using a soil/water mixture ratio of 1/2.5.

Soil classification was based on the field identificationof the soil attributes (e.g., color, texture, structure) of thesurface and subsurface horizons, and on the subsequentlaboratory analysis of chemical attributes. A Munsellcolor panel was also used in field to characterize the colorof samples in a wet condition with a clear sky.

3.1.2. Remote sensing dataAVIRIS data were collected on 7 September 1995,

during the dry season under clear sky conditions.AVIRIS is an imaging spectrometer that acquires spectrabetween 350 and 2500 nm in 224 bands with a nominalfull-width-half-maximum of 10 nm and an InstantaneousField of View of 1 mrad (Green et al., 1998). Whendeployed on an ER-2 from an altitude of 20 km, thisproduces a Ground Instantaneous Field of View ofapproximately 20 m at a mean ground altitude of zero.Signal-to-noise ratio for 1995 data ranged from 800 inthe visible (550 nm) to 360 in the SWIR (2200 nm)(Clark and Swayze, 1996). AVIRIS data were radiomet-rically calibrated by the Jet Propulsion Laboratory, then

converted to surface reflectance using the Fast Line-of-Sight Atmospheric Analysis of Spectral Hypercubes(FLAASH) algorithm, which is based on MODTRAN4radiative transfer code (Felde et al., 2003). Modelparameters included a Tropical Atmosphere and Ruralaerosol model with visibility set to 60 km. The algorithmincludes a correction for adjacency effects. AVIRIS

Table 2Average and standard deviation (parenthesis) values of color parameters and absorption band depth (ABD) of 50 soil samples from the surface (0–20 cm)and subsurface (40–60 cm) layers of the major soils types under study

Soil type Depth (cm) Purity (%) Dominant wavelength (nm) ABD 900 nm ABD 2210 nm ABD 2260 nm N

RA 0–20 49.76 (6.35) 587.87 (2.69) 0.079 (0.032) 0.119 (0.052) 0.154 (0.076) 840–60 57.99 (3.32) 591.50 (1.85) 0.154 (0.038) 0.146 (0.058) 0.204 (0.103) 8

TA 0–20 48.74 (10.54) 583.50 (3.41) 0.070 (0.028) 0.070 (0.022) 0.220 (0.029) 440–60 60.20 (8.82) 587.00 (2.58) 0.155 (0.029) 0.090 (0.020) 0.280 (0.029) 4

XA 0–20 40.07 (7.11) 580.18 (1.25) 0.053 (0.012) 0.097 (0.039) 0.208 (0.063) 1140–60 55.29 (6.64) 583.09 (1.64) 0.127 (0.025) 0.134 (0.069) 0.282 (0.069) 11

PA 0–20 45.93 (–) 581.00 (–) 0.075 (–) 0.082 (–) 0.214 (–) 140–60 50.06 (–) 586.00 (–) 0.173 (–) 0.113 (–) 0.250 (–) 1

UQ 0–20 25.13 (–) 580.00 (–) 0.001 (–) 0.100 (–) 0.031 (–) 140–60 30.65 (–) 582.00 (–) 0.001 (–) 0.160 (–) 0.023 (–) 1

Soil abbreviations are: RA (Rhodic Acrustox), TA (Typic Acrustox), XA (Xanthic Acrustox), PA (Petroferric Acrustox) and UQ (UsticQuartzipsamments). N is the number of samples.


bands covering the 1130-nmwater vapor band were usedto estimate precipitable water vapor on a per-pixel basis.

ASTER, a multispectral instrument onboard the Terrasatellite, has 14 bands in the VNIR, SWIR and thermalspectral intervals, variable spatial resolution (15 to 90 m),and along-track stereoscopic viewing capability (Yama-guchi et al., 1999). Assuming no topographic changesbetween the two data acquisition (AVIRIS and ASTER),an ASTER image acquired on 1 August 2001 was used toproduce a 15-m resolution DEM from the nadir- andbackward-looking images of the bands 3N and 3B usingthe ASTER_DTM algorithm (Sulsoft, 2004). Thisalgorithm converts these two bands into a pair of quasi-epipolar images, which have a pixel displacement in thesatellite flight direction proportional to the pixel elevation.A cross-correlation method is then used to determine thisdisplacement and to obtain elevation values with pre-cision of 14 m (Sulsoft, 2004).

3.2. Data analysis

3.2.1. Spectral variations of soil samplesReflectance spectra of the surface and subsurface soil

samples were compared to understand how spectralchanges varied with soil composition. Correlation coeffi-cients between soil physico-chemical characteristics andspectral parameters (absorption band depth and color)were also calculated. Absorption band depth was cal-culated in the VNIR and SWIR following the continuumremoval described by Clark and Roush (1984).

An important laboratory relationship was obtained forthe Ki index (1.7SiO2/Al2O3), an indicator of the degreeof alteration of tropical soils, determined from chemicalanalysis. Low Ki values (b1.0) indicate a high degree ofsoil weathering and the predominance of aluminumoxides (gibbsite) over clay minerals such as kaolinite. In

this study, the Ki values determined from chemicalanalysis were regressed against the ratio of the 2210 nmband depth to the sum of the 2210 nm and 2260 nm banddepths calculated from laboratory spectra with thecontinuum removal method. Once this laboratory rela-tionshipwas established, it was applied toAVIRIS data, asdescribed below.

3.2.2. Variations in soil composition with topographyPCAwas applied to AVIRIS surface reflectance data to

characterize the variability in the reflectance of differentsoil types. The PCAwas preceded by band depth analysisfollowing the continuum removal to isolate pixels ofexposed soils from the other components in the AVIRISscene. Selected spectral intervals for the continuumremoval were: 760–1100 nm (goethite and hematite),2100–2265 nm (kaolinite) and 2100–2350 nm (gibbsite).However, the criterion used to separate exposed soil areaswas to select pixels in absorption band depth images withvalues higher than 0.04 (deep features) at spectral in-tervals of hydroxyl-bearing minerals (2210 nm or2260 nm). After identifying these pixels in the AVIRISsurface reflectance image and subsampling the data to 187bands to remove bands within strong water vapor absorp-tion intervals around 1400 nm and 1900 nm, the PCAwasperformed.

The SFF technique was applied to continuum removedAVIRIS spectra to identify goethite, hematite, kaoliniteand gibbsite using reference spectra from the UnitedStates Geological Survey (USGS) spectral library. SFF isan absorption feature-based method that matches pixelspectra to reference spectra using a least squares approachand continuum-removed data (Clark et al., 1991; Dehaanand Taylor, 2002). Scale/root mean square (RMS) valueswere used to locate areas that best matched the referencespectra. For more details on the continuum removal and

Table3

Matrixof

correlationbetweensoilconstituents,texture,colorparametersandAbsorptionBandDepth

(ABD)values

of50

soilsamples

Organic

carbon

SiO

2Fe 2O3

Al 2O3

TiO

2MnO

P2O5

Sand

Silt

Clay

Dom

inant

Wavelength

Purity

ABD

900nm

ABD

2210

nmABD

2260

nm

Organic

carbon

1−0.06

−0.16

−0.10

−0.14

+0.11

+0.71

⁎−0.32

+0.48

+0.29

−0.27

−0.19

−0.49

−0.58

⁎+0.07

SiO

21

+0.52

+0.47

+0.53

−0.09

+0.13

−0.61

⁎+0.55

⁎+0.62

⁎+0.49

+0.25

+0.14

+0.40

+0.20

Fe 2O3

1+0.74

⁎+0.79

⁎+0.02

+0.08

−0.69

⁎+0.48

+0.71

⁎+0.60

⁎+0.70

⁎+0.64

⁎−0.03

+0.47

Al 2O3

1+0.93

⁎−0.17

+0.05

−0.65

⁎+0.46

+0.67

⁎+0.28

+0.59

⁎+0.44

−0.01

+0.66

⁎

TiO

21

−0.06

+0.01

−0.60

⁎+0.42

+0.62

⁎+0.40

+0.59

⁎+0.48

+0.08

+0.56

⁎

MnO

1+0.13

+0.16

−0.15

−0.16

+0.10

−0.04

+0.09

+0.13

−0.41

P2O5

1−0.49

+0.63

⁎+0.46

+0.09

+0.01

−0.23

−0.50

+0.03

Sand

1−0.92

⁎−0.99

⁎−0.43

−0.59

⁎−0.28

+0.32

−0.60

⁎

Silt

1+0.90

⁎+0.33

+0.46

+0.11

−0.34

+0.48

Clay

1+0.44

+0.60

⁎+0.30

−0.31

+0.61

⁎

Dom

inant

wavelength

1+0.57

⁎+0.59

⁎+0.21

−0.06

Purity

1+0.81

⁎−0.12

+0.46

ABD

900nm

1+0.10

+0.31

ABD

2210

nm1

−0.33

ABD

2260

nm1

⁎Correlatio

nissign

ificantat

the0.01

level.


spectral fitting see Clark et al. (2003) and Debba et al.(2005).

The laboratory regression relationship between theKi index determined from chemical analysis andabsorption band depth values at 2210 nm and2260 nm calculated from laboratory soil spectra,described in the previous item, was applied to banddepths calculated from AVIRIS surface reflectanceimage following the continuum removal. The objectivewas to map the relative variations in SiO2 and Al2O3 inCampo Verde area and to compare AVIRIS estimated Kivalues to the identification of kaolinite and gibbsitederived from the SFF technique.

To investigate possible associations between hyper-spectral remote sensing information and topography,PCA, the spectral mineralogy and Ki index results wereplotted over a DEM. AVIRIS-detected variability in soilcomposition, mineralogy and in the intensity of soilsurface alteration was then analyzed as a function of theelevation data.

4. Results and discussion

4.1. Soil analyses

4.1.1. Physico-chemical analysesAverage and standard deviation values of physico-

chemical data of 50 soil samples from the surface (0–20 cm) and subsurface (40–60 cm) layers of the majorsoil types in the study area are reported (Table 1). In thesurface, Rhodic Acrustox has the largest weight of SiO2

and the highest clay fraction. Progressing from thesurface to the subsurface, all soil types showed adecrease in organic carbon, P2O5 and MnO content andan increase in clay fractions and in the other chemicalconstituents. Ustic Quartzipsamments is a sandy soiland sometimes is composed of more than 85% sand.

4.1.2. Spectral variations of soil samplesExamples of reflectance spectra for pairs of soil

samples from the surface and subsurface layers areshown in Fig. 2. The reflectance increases from RhodicAcrustox to Ustic Quartzipsamments because theRhodic Acrustox has a larger clay fraction and higherorganic carbon, SiO2, Al2O3, Fe2O3 and TiO2 contentthan the Ustic Quartzipsamments (Table 1). For all soiltypes studied, VNIR reflectance increased from thesurface to the subsurface, due to a decrease in organiccarbon content with depth (Fig. 2; Table 1) andincreased expression of iron oxide features. For RhodicAcrustox, SWIR reflectance decreased with increasingdepth as the clay content increased (Table 1). These

Fig. 3. Laboratory-derived relationship between the Ki index (1.7SiO2/Al2O3) determined from chemical analysis and the depth (D) of theabsorption bands centered at 2210 nm (kaolinite) and 2260 nm (gibbsite)calculated after the continuum removal.

Fig. 4. First two eigenvectors from principal components analysisapplied to pixels of exposed soils in the AVIRIS scene.


changes in soil composition from surface to subsurfaceusually resulted in a decrease in the 2260/600 nm ratiowith depth, as also indicated by Galvão et al. (1997).

Several mineral absorption features are prominent inthese spectra, including kaolinite at 1400 nm and 2210 nm(Ustic Quartzipsamments), gibbsite at 1400 nm and2260 nm (Rhodic Acrustox and Xanthic Acrustox),hematite (steep reflectance increase at 580 nm and broadfeature at 870 nm in the Rhodic Acrustox spectrum) andgoethite (steep reflectance increase at 490 nm and broadfeature at 930 nm in the Xanthic Acrustox spectrum;Fig. 2). The Ustic Quartzipsamments spectrum showed astrong absorption feature at 2210 nm and no feature at2260 nm in concordance with laboratory mineralogicalanalyses (Table 1), which indicated the presence ofkaolinite and the absence of gibbsite for this soil.

Table 2 shows the average and standard deviationvalues of color parameters and the absorption band depth(ABD) values of the 0–20 cm and 40–60 cm depths of the50 soil samples from the major soils types under study. Asconcluded by the dominant wavelength results, RhodicAcrustox is essentially a red soil, whereas the other soiltypes can be characterized as red–yellow to yellow.Rhodic Acrustox showed the deepest band depths at900 nm (iron oxides) and 2210 nm (kaolinite) for thesurface samples. It also presented the longest wavelengthfor the dominant wavelength. Colors became moresaturated and more reddish with depth, as indicated by ahigher purity and a longer dominant wavelength in each

subsurface horizon, respectively. Except for Ustic Quart-zipsamments, soil spectra showed deeper absorptionbands for the 40–60 cm samples, especially the 900 nmspectral feature due to a decrease in organic carboncontent and an increase in Fe2O3 content (Table 2).

Correlation coefficients (r) between texture, soilconstituents, color parameters and band depth values ofthe 50 soil samples are presented in Table 3. Results forSiO2 show that this constituent is associated positivelywith the clay fraction (r=+0.62) and inversely correlatedwith the sand fraction (r=−0.61). Statistically signifi-cant positive correlations were observed for the relation-ships between Al2O3 and clay fraction (r=+0.67),organic carbon and P2O5 (r=+0.71), Fe2O3 and clayfraction (r=+0.71), Fe2O3 and Al2O3 (r=+0.74), Fe2O3

and TiO2 (r=+0.79), and Al2O3 and TiO2 (r=+0.93).Changes in dominant wavelength and purity values weremainly associated with variations in Fe2O3 (r of +0.60and +0.70, respectively). Band depth at 900 nm wascorrelated with Fe2O3 (r=+0.64), whereas the feature at2260 nm was associated with Al2O3 content (r=+0.66)and clay fraction (r=+0.61).

Fig. 3 presents the linear relationship between Kiindex values derived from chemical analyses of 33 soilsamples and laboratory-derived band depths at 2210 nmand 2260 nm, calculated after the continuum removal.To assure a more robust relationship, 17 samples weredeleted from the Ki correlation analysis since theypresented doubtful Ki results. The analysis was then runon the remaining 33 samples. The 17 samples left out of


the analysis present Ki values higher than 2.3, which isunusual in tropical soils and indicate that part of theSiO2 determined from sulfuric attack may come fromthe sand fraction. In Fig. 3, since Ki represents relativevariability in the SiO2 and Al2O3 content associated withkaolinite (Al2SiO5(OH)4) and gibbsite (Al2O3·3H2O), it ispositively correlated with an increase in the depth of theseabsorption bands. Ki values lower than 1.0 indicated thepredominance of gibbsite over kaolinite. A Pearsoncorrelation coefficient (r) of +0.60 and a residual meansquare value of 0.13 were obtained. Results of Fig. 3confirmed the existence of an association between Kivalues and the kaolinite and gibbsite absorption bands,which was previously reported by Madeira Netto (1993),Madeira Netto et al. (1995) and Baptista et al. (1998).However, the r value of +0.60 observed in Fig. 3 waslower than that obtained in previous studies. In spite of thespectral proximity of both absorption bands (2210 nm and

Fig. 5. (a) AVIRIS true color composite with the bands at 677 nm, 569 nm aASTER-derived digital elevation model; (b) Variations in PC1 scores for pixe2260 nm. Areas in white represent other scene components or spectral mixinexaggeration is 5 times.

2260 nm), correlation results of Fig. 3 may be affected insome extent by variations in opaque minerals and organicmatter content.

4.2. Variations in soil composition with topography

Eingenvector loadings for the first two principalcomponents (97.6% of the cumulative variance) ofAVIRIS surface reflectance data of exposed soils wereplotted as a function of wavelength (Fig. 4). Theloadings represent the contribution of each variable(reflectance values of the bands) to explain thecomponent. The first component (PC1) accounted for92.7% of the total variance and represents the meanreflectance (albedo or brightness) of the soil types, asindicated by positive loadings across the full spectralrange (Fig. 4). The second component (PC2) accountedfor 4.9% of the variance, and expressed the contrasting

nd 480 nm in red, green and blue colors, respectively, overlaid on thels of exposed soils or with deep mineral absorption bands at 2210 nm org. The Transect A–B indicated in (a) is discussed in the text. Vertical

Fig. 6. Relationships of ASTER-derived elevation data with AVIRISPC1 scores for 120 randomly selected pixels (40 per soil) of three soiltypes with increasing mean reflectance.


contributions of the SWIR (negative loadings between2000 and 2400 nm) and VNIR (positive loadingsbetween 600 and 800 nm) data or variations in theshape of the soil spectra. According to Galvão et al.(1997) and Palacios-Orueta and Ustin (1998), suchvariations associated with the second component areusually produced by changes in the content of ironoxides and organic matter.

Fig. 7. Variations in soil type, ground elevation data, PC1 scores, and in the(ABD) for the Transect A–B indicated in Fig. 5a.

Fig. 5a shows a true color composite of the study areawith the AVIRIS bands centered at 677 nm (red), 569 nm(green) and 480 nm (blue) overlaid on the ASTER-derived DEM. Examples of the occurrence of major soiltypes are indicated in this figure. Petroferric Acrustoxoccurs close to the streams and does not show a well-defined spatial pattern in the image. Fig. 5b presents thePC1 results. As mentioned in the methodology, PCAwasapplied considering only pixels of exposed soils selectedin absorption band depth images after the continuumremoval (pixels with deep absorption bands at 2210 nm(kaolinite) or 2260 nm (gibbsite); depth values higher than0.04). Thus, areas in white in Fig. 5b correspond to otherscene components or spectral mixing. Low reflectancesoils (red–yellow pixels in Fig. 5b) tend to occupy thehighest elevations, whereas high reflectance soils (bluepixels) predominate at lower elevations. This relationshipis confirmed in Fig. 6, which shows a negative correlation(r=−0.84) between AVIRIS-derived PC1 scores andASTER-derived elevation values for 120 randomlyselected pixels (40 per type) of three soil types withincreasing mean reflectance. According to this relation-ship, dark soils with high clay content (Rhodic Acrustox)tended to occur at higher slope positions and brighter soils(Ustic Quartzipsamments) with low clay content at lowerslope positions.

Along a toposequence (Transect A–B indicated inFig. 5a), strong spatial relationships were observedbetween soil types, surface elevation, PC1 scores, and

2210 nm (kaolinite) and 2260 nm (gibbsite) absorption band depths

Fig. 8. AVIRIS reflectance spectra representative of the soil typesindicated in Fig. 7.

Fig. 9. Relationships between absorption band depth values at2210 nm (kaolinite) and 2260 nm (gibbsite) for randomly selectedpixels (40 per soil) of dark soils at higher slope positions (RhodicAcrustox) and brighter soils at lower slope positions (UsticQuartzipsamments).


the 2210 nm (kaolinite) and 2260 nm (gibbsite)absorption band depths (Fig. 7). From the right to theleft of Fig. 7, soil composition changes gradually fromRhodic Acrustox at higher elevations (N750 m) toXanthic Acrustox at lower elevations (b725 m). Thischange corresponds to the transition from lower averagesoil reflectance (negative PC1 scores) and deeperkaolinite absorption band at 2210 nm to higher averagesoil reflectance (positive PC1 scores) and deepergibbsite absorption band at 2260 nm. Results for ironabsorption bands (goethite and hematite; not shown) didnot present a clear trend with topography or soil type.Elevational changes in soil composition are oftenobserved in other regions of Brazil. For example,Uhlmann et al. (1997) observed a transition from RhodicAcrustox at high elevations to Typic Acrustox at lowerelevations in a savanna area from southern Brazil. Curiand Franzmeier (1984) and Madeira Netto (1993)observed changes from reddish to yellowish soil colorstowards lower slope positions in toposequences ofoxisols from central Brazil.

AVIRIS reflectance spectra of soil types indicated inFig. 7 (Transect A–B) are shown in Fig. 8. The RhodicAcrustox AVIRIS spectrum showed both a prominentkaolinite spectral feature at 2210 nm and a shallowgibbsite absorption feature at 2260 nm. The TypicAcrustox and Xanthic Acrustox spectra presented aprominent gibbsite feature and almost no features at2210 nm.

The relationships between depth values of absorptionbands centered at 2210 nm and 2260 nm for the same120 pixels shown in Fig. 6 are presented in Fig. 9. Incomparison with Rhodic Acrustox and Ustic Quartzip-samments, Typic Acrustox shows deeper absorption

bands at 2260 nm (gibbsite) and shallower features at2210 nm (kaolinite). In spite of being a sandy soil, UsticQuartzipsamments (lower elevations in the study area)may show 2210 nm absorption bands. On the otherhand, it did not display features at 2260 nm (Fig. 9),which is in agreement with the results shown in Fig. 2.

As mentioned before, the Ki index is an indicator ofthe degree of decomposition of the clay fraction(desilification process). In a sequence of weathering,Ki values are higher for kaolinitic soils and lower forgibbsitic soils. In general, Ki values lower than 1.0indicate a high degree of soil weathering or the removalof kaolinite and a concentration of gibbsite. Fig. 10ashows a true color composite of a portion of the AVIRISscene. The position of the Transect A–B is indicated inthis figure only for reference. Fig. 10b presents the SFFresults for the identification of kaolinite and gibbsiteusing the reference spectra of these two minerals fromthe USGS spectral library. Fig. 10c illustrates variationsin Ki values derived from the laboratory regressionequation of Fig. 3 applied over AVIRIS absorption banddepth images (continuum removal method). Whencompared to each other, the results of the SFF mineralidentification (Fig. 10b) are consistent with variations inKi values (Fig. 10c). Kaolinitic areas (red color inFig. 10b) are associated with high Ki values (yellow andred colors in Fig. 10c) as expected, due to a high SiO2

and a low Al2O3 content. On the other hand, gibbsiticareas (yellow color in Fig. 10b) or weathered soil

Fig. 10. (a) AVIRIS true color composite with the 677 nm (red), 569 nm (green) and 480 nm (blue) bands overlaid on an ASTER-derived digitalelevation model; (b) Spectral Feature Fitting (SFF) identification of kaolinite and gibbsite; (c) Variations in Ki values for exposed soil pixels. Areas inwhite in (b) indicate unclassified pixels and in (c) spectral mixing or non-exposed soil surfaces. Vertical exaggeration is 5 times.


Fig. 11. Relationship between the estimated Ki values of Fig. 10b andthe corresponding Spectral Feature Fitting (SFF) values for gibbsite.The circle indicates pixels whose spectra are dominated by the2260 nm gibbsite absorption band.


surfaces correspond to low Ki values (blue and greencolors in Fig. 10c) due to a low SiO2 and a high Al2O3

content. In the gibbsitic domain, highly soil surfaces(low SiO2/Al2O3 ratios) are indicated by the blue colors(the lowest Ki values) in Fig. 10c. When the estimatedKi values of Fig. 10b were plotted against thecorresponding SFF values for gibbsite, a statisticallysignificant non-linear relationship was obtained with anr value of −0.78 and a residual mean square value of0.13 (Fig. 11). Despite the consistency between results,variations in opaque minerals and organic mattercontent and shadow effects are potential factors thatmay affect the accuracy of Ki estimates in the scene,even for closely spaced kaolinite and gibbsite absorptionbands.

5. Conclusions

This VNIR-SWIR hyperspectral study of the relation-ships between soil types, mineralogical and chemicalcomposition, mean reflectance, and surface elevationshowed the existence of an association between ASTER-derived elevation data and AVIRIS PC1 scores (meanreflectance from the 400–2500 nm interval). Results fromPCA applied over anAVIRIS surface reflectance image ofexposed soils showed the predominance of low reflec-tance soils (e.g., Rhodic Acrustox with negative PC1scores, clay texture and higher content of organic carbon,SiO2, Al2O3, Fe2O3 and TiO2) at higher elevationsaltitudes and of high reflectance soils (e.g., Ustic

Quartzipsamments with positive PC1 scores, sand textureand lower content of these constituents) at lowerelevations. In some portions of the study area, RhodicAcrustox gradually changes into Typic Acrustox andXanthic Acrustox from high to low elevations in atoposequence, which is equivalent to the transition from alowmean reflectance (negativeAVIRIS PC1 scores) and adeeper kaolinite absorption band at 2210 nm in pixelspectra to a high mean reflectance (positive AVIRIS PC1scores) and a deeper gibbsite absorption band at 2260 nm.

Results of the identification of kaolinite and gibbsite intheAVIRIS scenewith the SFF techniquewere concordantwith variations in Ki values calculated on a per pixel basisfrom a laboratory-derived regression relationship betweenchemical data and absorption band depth values at2210 nm and 2260 nm. Kaolinitic areas were associatedwith high Ki values due to a high SiO2 content and a lowAl2O3 content, whereas gibbsitic areas corresponded tolow Ki values and highly weathered soil surfaces due to alow SiO2 content and a high Al2O3 content.

In spite of this concordance, further field validation isnecessary to evaluate the impact of the variations inopaque minerals and organic matter content and of theshadow effects on Ki estimates from hyperspectral data.

Acknowledgments

The authors are grateful to FAPESP (Fundação deAmparo à Pesquisa do Estado de São Paulo) for financialsupport of this project (05/01737-0). Thanks are also dueto CNPq (Conselho Nacional de DesenvolvimentoCientífico eTecnológico), and to the anonymous reviewersfor useful suggestions that improved the manuscript.

References

Baptista, G.M.M., Madeira Netto, J.S., Meneses, P.R., 1998. Determi-nação da relação sílica–alumina a partir dos dados do sensor AVIRIS(JPL/NASA) para discretização espacial do grau de intemperismo desolos tropicais. Proc. Simpósio Brasileiro de Sensoriamento Remoto,Santos, São Paulo, 11–18 September 1998, pp. 1345–1355.

Bedidi, A., Cervelle, B.,Madeira, J., Pouget,M., 1992.Moisture effects onvisible spectral characteristics of lateritic soils. Soil Science 153 (2),129–141.

Ben-Dor, E., Patkin, K., Banin, A., Karnieli, A., 2002. Mapping ofseveral soil properties using DAIS-7915 hyperspectral scanner data:a case study over clayey soils in Israel. International Journal ofRemote Sensing 23 (6), 1043–1062.

Camargo, O.A., Moniz, A.C., Jorge, J.A., Valadares, J.M., 1986.Métodos de Análise Química, Mineralógica e Física de Solos doIAC. Instituto Agronômico de Campinas, São Paulo.

Chabrillat, S., Goetz, A.F.H., Krosley, L., Olsen, H.W., 2002. Use ofhyperspectral images in the identification and mapping of expansiveclay soils and the role of spatial resolution. Remote Sensing ofEnvironment 82 (2–3), 431–445.


Clark, R.N., Roush, T.L., 1984. Reflectance spectroscopy: quantitativeanalysis techniques for remote sensing applications. Journal ofGeophysical Research 89 (7), 6329–6340.

Clark, R.N., Swayze, G.A., 1996. Evolution in imaging spectroscopyanalysis and sensor signal-to-noise: an examination on how far wehave come. Proc. Sixth Annual JPL Airborne Earth ScienceWorkshop, Pasadena, California, 4–8 March 1996, pp. 49–53.

Clark, R.N., Swayze, G.A., Gorelick, N., Kruse, F.A., 1991. Mappingwith imaging spectrometer data using the complete band shapeleast-squares algorithm simultaneously fit to multiple spectralfeatures from multiple materials. Proc. Third Airborne Visible/InfraRed Imaging Spectrometer (AVIRIS) Workshop, Pasadena,California, 20–21 May 1991, pp. 2–3.

Clark, R.N., Swayze, G.A., Livo,K.E., Kokaly, R.F., Sutley, S.J., Dalton,J.B., McDougal, R.R., Gent, C.A., 2003. Imaging spectroscopy:Earth and planetary remote sensing with the USGS Tetracorder andexpert systems. Journal of Geophysical Research 108 (E12), 5131.

Cunha, P., Marques, J., Curi, N., Pereira, G.T., Lepsch, I.F., 2005.Geomorphic surfaces and latosol (oxisol) characteristics on asandstone/basalt sequence from the Jaboticabal region, Sao Paulostate, Brazil. Revista Brasileira de Ciência do Solo 29 (1), 81–90(In Portuguese).

Curi, N., Franzmeier, D.P., 1984. Toposequence of oxisols from theCentral Plateau of Brazil. Soil Science Society of America Journal48 (2), 341–346.

Curi, N., Franzmeier, D.P., 1987. Effect of parent rocks on chemicaland mineralogical properties of some oxisols in Brazil. SoilScience Society of America Journal 51 (1), 153–158.

da Motta, P.E.F., de Carvalho Filho, A., Ker, J.C., Pereira, N.R., deCarvalho Júnior,W., Blancaneaux, P., 2002. Soil-geomorphic surfacerelationships and landscape evolution in an area of the BrazilianCentral Plateau. Pesquisa Agropecuária Brasileira 37 (6), 869–878(In Portuguese).

Debba, P., van Ruitenbeek, F.J.A., van der Meer, F.D., Carranza, E.J.M.,Stein, A., 2005. Optimal field sampling for targeting minerals usinghyperspectral data. Remote Sensing of Environment 99 (4), 373–386.

Dehaan, R.L., Taylor, G.R., 2002. Field-derived spectra of salinizedsoils and vegetation as indicators of irrigation-induced soilsalinization. Remote Sensing of Environment 80 (3), 406–417.

Dehaan, R.L., Taylor, G.R., 2004. A remote-sensing method ofmapping soils and surficial lags from a deeply weathered aridregion, near Cobar, NSW, Australia. Geochemistry: Exploration,Environment, Analysis 4 (2), 99–112.

Embrapa, 1997. Manual de Métodos de Análise de Solos. EmpresaBrasileira de Pesquisa Agropecuária, Rio de Janeiro.

Embrapa, 1999. Sistema Brasileiro de Classificação de Solos. EmpresaBrasileira de Pesquisa Agropecuária, Brasília.

Felde, G.W., Anderson, G.P., Adler-Golden, S.M., Matthew, N.W.,Berk, A., 2003. Analysis of Hyperion data with the FLAASHatmospheric correction algorithm. Proc. International Geoscienceand Remote Sensing Symposium (IGARSS), Toulouse, 21–25July 2003, pp. 90–92.

Galvão, L.S., Vitorello, I., Formaggio, A.R., 1997. Relationships ofspectral reflectance and color among surface and subsurfacehorizons of tropical soil profiles. Remote Sensing of Environment61 (1), 24–33.

Galvão, L.S., Pizarro, M.A., Epiphanio, J.C.N., 2001. Variations inreflectance of tropical soils: spectral-chemical composition relation-ships from AVIRIS data. Remote Sensing of Environment 75 (2),245–255.

Green, R.O., Eastwood, M.L., Sarture, C.M., Chrien, T.G., Aronsson, M.,Chippendale, B.J., Faust, J.A., Pavri, B.E., Chovit, C.J., Solis, M.S.,Olah, M.R., Williams, O., 1998. Imaging spectroscopy and theAirborne Visible Infrared Imaging Spectrometer. Remote Sensing ofEnvironment 65 (3), 227–248.

Macedo, J., Bryant, R.B., 1987. Morphology, mineralogy, and genesisof a hydrosequence of oxisols in Brazil. Soil Science Society ofAmerica Journal 51 (3), 690–698.

Madeira, J., Bédidi, A., Pouget, M., Cervelle, B., Flay, N., 1995. Spectral(MIR) determination of kaolinite and gibbsite contents in lateriticsoils. Comptes Rendus de L'Academie des Sciences 321 (2),119–127.

Madeira, J., Bédidi, A., Cervelle, B., Pouget, M., Flay, N., 1997.Visible spectrometric indices of hematite (Hm) and goethite (Gt)content in lateritic soils: the application of a Thematic Mapper(TM) image for soil-mapping in Brasília, Brazil. InternationalJournal of Remote Sensing 18 (13), 2835–2852.

Madeira Netto, J.S., 1993. Étude quantitative des relations constituantsminéralogiques-réflectance diffuse des latosols Brésiliens. Thèse deDoctorat, ORSTOM,Université Pierre etMarie Curie, Paris, pp. 236.

Moniz, A.C., Buol, S.W., Weed, S.B., 1982. Formation of an Oxisol–Ultisol transition in São Paulo, Brazil: II. Lateral dynamics ofchemical weathering. Soil Science Society ofAmerica Journal 46 (6),1234–1239.

Palacios-Orueta, A., Ustin, S.L., 1998. Remote sensing of soilproperties in the Santa Monica mountains: I. spectral analysis.Remote Sensing of Environment 65 (2), 170–183.

Pizarro, M.A., Epiphanio, J.C.N., Galvão, L.S., 2001. Mineralogicalcharacterization of tropical soils by hyperspectral remote sensing.Pesquisa Agropecuária Brasileira 36 (10), 1277–1286 (In Portuguese).

Resende,M., Santana, D.P., 1998. Uso das relações Ki e Kr na estimativada mineralogia para classificação dos latossolos. In: Embrapa (Ed.),Reunião de Classificação, Correlação dos Solos e Interpretação daAptidãoAgrícola. Empresa Brasileira de Pesquisa Agropecuária, Riode Janeiro, pp. 225–232.

Sulsoft, 2004. ASTERDTM2.2: Installation and User's Guide. Sulsoft,Porto Alegre.

Uhlmann, A., Curcio, G.R., Galvão, F., Silva, F.M., 1997. Relationshipsbetween distribution of vegetation physiognomic types and geomor-phic and pedological patterns in a savanna area (cerrado) in southernBrazil. Arquivos de Biologia e Tecnologia 40 (2), 473–483.

Yamaguchi, Y., Fujisada, H., Kudoch, M., Kawakami, T., Tsu, H., Kahle,A.B., Pnief, M., 1999. ASTER instrument characterization andoperation scenario. Advances in Space Research 23 (8), 1415–1424.

Documents

Relationships between the mineralogical and chemical ...geog.ucsb.edu/viper/viper_pubs/galvao_et_al08.pdfRelationships between the mineralogical and chemical composition of tropical