Professional Documents
Culture Documents
Demattê Et Al. (2017)
Demattê Et Al. (2017)
Demattê Et Al. (2017)
Research article
a r t i c l e i n f o a b s t r a c t
Article history: Wetlands are important ecosystems characterized by redoximorphic environments producing typical soil
Received 23 April 2016 forming processes and organic carbon accumulation. Assessments and management of these areas are
Received in revised form dependent on knowledge about soil characteristics and variability. By reflectance spectroscopy, infor-
3 March 2017
mation about soils can be obtained since their spectral behaviors are directly related to their chemical,
Accepted 4 March 2017
Available online 18 March 2017
physical, and mineralogical properties reflecting the pedogenetic processes and environment conditions.
Our aims were: (a) to characterize the main soil classes of wetlands regarding their spectral behaviors in
VIS-NIR-SWIR (350e2500 nm) and relate them to pedogenesis and environmental conditions, (b) to
Keywords:
Remote sensing
determine spectral ranges (bands) with greater expression of the main soil properties, (c) to identify
Reflectance spectroscopy spectral variations and similarities between hydromorphic soils from wetlands and other soils under
Soil spectral behavior different moisture conditions, and (d) to propose spectral models to quantify some chemical and physical
Soil characterization soil properties used as environmental quality indicators. Nine soil profiles from the Pantanal region
Spectral morphology (Mato Grosso State, Brazil) and one from the Serra do Espinhaço Meridional (Minas Gerais State, Brazil)
Environment monitoring were investigated. Spectral morphology interpretation allowed identifying horizon differences regarding
shape, absorption features and reflectance intensity. Some pedogenetic processes of wetland soils related
to organic carbon accumulation and oxide iron variation were identified by spectra. Principal Component
Analysis allowed discriminating soils from wetland and outside this area (oxidic environment). Quan-
tification of organic carbon was possible with R2 of 0.90 and low error. Quantification of clay content was
masked by soils with organic carbon content over 2% where it was not possible to quantify with high R2
and low error both properties when dataset has soil samples with high organic carbon content. By
reflectance spectroscopy, important characteristics of wetland soils can be identified and used to
distinguish from soils of different environments at low costs, reduced time, and with environmental
quality.
© 2017 Elsevier Ltd. All rights reserved.
1. Introduction
* Corresponding author.
^), ingrid.horak@ufvjm.edu.
E-mail addresses: jamdemat@usp.br (J.A.M. Dematte According to functional-factorial model proposed by Dokuchaev
br (I. Horak-Terra), rmbeirigo@yahoo.com.br (R.M. Beirigo), fabricio.terra@ufpel. (1883), Shaw (1930), and Jenny (1994), soils are formed from
edu.br (F.S. Terra), karina.marques@usp.br (K.P.P. Marques), caio.fongaro@gmail.
environmental factors such as parent material (rocks, mineral and
com (C.T. Fongaro), alexandre.christo@ufvjm.edu.br (A.C. Silva), pvidal@usp.br
(P. Vidal-Torrado).
organic sediments), climate, relief, organisms, and time. For a
http://dx.doi.org/10.1016/j.jenvman.2017.03.014
0301-4797/© 2017 Elsevier Ltd. All rights reserved.
J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62 51
particular relief (landscape position), the interaction between The link between pedogenesis and reflectance spectroscopy is
parent material and climate determines the presence of organisms, an important step to achieve consistency of spectral analysis and
which acts over time and occur soil formation (pedogenesis). Thus, understanding of soil behavior. In fact, soil characterization is
soils represent the result of a set of natural resources interaction, essential for environmental studies including its management plan,
and for that reason, it can be considered one of the best ecosystem risk analysis, monitoring and assessments. Besides that, important
components for environmental stratification. soil properties and environmental quality indicators, such as clay
Soils are considered the base of terrestrial ecosystems and and organic carbon (OC) content, have been properly modeled by
wetlands plays several environmental functions. Wetlands are reflectance spectroscopy, and their results have been largely
ecosystems with permanent or seasonal flooding where soils improved due to advances in statistical modeling (Guerrero et al.,
remain saturated to the surface long enough to allow physical, 2016; Shi et al., 2015).
chemical, and biological processes typical of anaerobic environ- Our aims were (i) to characterize the main soil classes of wet-
ments (National Research Council, 1995). Their ecological functions lands regarding their spectral behaviors in visible (VIS e
are essential for regulation of hydrological and biogeochemical 350e700 nm), near infrared (NIR e 700e1100 nm) and shortwave
cycles with fauna, flora and soil characteristics (Ramsar Convention infrared (SWIR e 1100e2500 nm) and relate them to pedogenesis
Secretariat, 2013). and environmental conditions, (ii) to determine spectral ranges
Wetland soils have been developed under strong influence of (bands) with greater expression of the main soil properties, (iii) to
redoximorphic processes, which cause Fe and Mn redistribution identify spectral variations and similarities between hydromorphic
and organic carbon accumulation. Properties of these soils are soils from wetland and other soils under different moisture con-
rather distinct from those developed under other hydrological dition, and (iv) to propose spectral models to quantify some
conditions, and the chemical and mineralogical features are asso- chemical and physical soil properties used as environmental quality
ciated to the typical pedogenetic processes, such as: paludization, indicators.
redoximorphism, and bissialitization. These features are able to be
identified by spectral behavior and are directly related to their 2. Material and methods
chemical, physical, and mineralogical composition and environ-
ment conditions. 2.1. Study areas and their respective soil profiles
In Brazil, some tropical wetlands are located in the Pantanal and
in the Serra do Espinhaço Meridional, which are internationally The Reserva Particular do Patrimo ^nio Natural (RPPN) SESC
recognized for their ecological importance and considered Pantanal is located between the rivers Cuiaba and S~
ao Lourenço, in
biosphere reserves by United Nations Educational, Scientific and Bara~o de Melgaço e Mato Grosso State, Brazil and covers around
Cultural Organization (UNESCO) and Ramsar Convention on Wet- 106,644 ha (coordinates 16 32’ e 16 490 S and 56 03’ e 56 260 W).
lands (Ramsar Convention Secretariat, 2013). In particular, Pantanal € ppen, 1948) and characterized as
This climate is classified as Aw (Ko
has around 195.000 km2 of this important ecosystem, and infor- tropical humid with average annual precipitation of 1200 mm,
mation about wetland soils become essential for environmental eight months of water deficit, annual average temperature of 25 C
studies to develop management plans and carry out impact ana- (Hasenack et al., 2010), and altitudes between 100 and 150 m
lyses in these protected areas. (Nascimento et al., 2013). According to IUSS/WRB (2015), the nine
Reflectance spectroscopy is one part of the remote sensing soil profiles investigated in this region were classified as: Fluvic
techniques where information about properties of objects (soils, for Cambisol (CM1 and CM2), Plinthic Gleysol (GL1 and GL2), Cutanic
instance) have been obtained by reflected electromagnetic radia- Luvisol (LV), Endogleyic Arenosol (ARg), Stagnic Solonetz (SNj),
tion. There are many reasons to use this technique as an analytical Endogleyic Planosol (PLg) and Pisolithic Plinthosol (PT).
method in soil analysis, as minimal samples preparation, rapid The other area is placed in the Biribiri State Park, in Serra do
analysis, no waste, and simultaneous estimation of many soil Espinhaço Meridional, Diamantina e Minas Gerais State, Brazil, at
properties (Viscarra Rossel et al., 2009). Reflectance spectroscopy is 1.250 m above sea level, and the sampling place was at top of a
a non-destructive method that does not use toxic chemicals ele- planning surface in the head of the stream San Miguel (coordinates
ments to humans and environment and it allows estimating results 18º080 S and 43º350 W). The regional climate is classified as Cwb
equal to conventional wet chemical analyses (Cohen et al., 2007). (Ko€ppen, 1948) and characterized as tropical of altitude with cold
The graphic representation of reflectance data of objects along and dry winters, and mild and wet summers. The average annual
wavelength is known as curves behavior (or spectral signatures) temperature is 18.7 C and average annual precipitation is
and it is dependent on their inherent characteristics (Jensen, 2009). 1500 mm. The soil profile studied was classified as Sapric Histosol
Soil spectral behavior is directly related to its chemical, physical, (HS) (IUSS/WRB, 2015), which presents thick layers rich in organic
and mineralogical properties reflecting the pedogenetic processes material alternating with thin layers formed almost by sand,
and environment conditions. possibly by alternations of dry and wet periods due to paleoclimatic
Although many studies are related to quantitative spectroscopy changes. Because of mining activity, the lowering of the drainage
to model soil properties (Soriano-Disla et al., 2014), qualitative network (more than 3.5 m from the surface level) exposed the
assessments to investigate shapes and features of spectral behav- peatland in talvegue San Miguel Stream (Silva et al., 2009).
iors are also important for a rapid and efficient identification of soils
based on standards. Tropical wetland soils have not been appro- 2.2. Wet chemical, particle size distribution, and mineralogy
priately studied in terms of reflected energy yet. Indeed, there are analysis
not enough information about spectral behaviors of soils from
redoximorphic environments in Brazil. In this country, some The soil samples were air-dried, sieved at 2-mm mesh, and sent to
spectral libraries have been developed in order to establish stan- laboratory for wet chemical and particle size distribution analyses.
dards of spectral behaviors of typical tropical soil classes from oxic Chemical analyses consisted of pH in water, extraction of exchange-
environments (Terra et al., 2015; Dematte ^ and Terra, 2014; Belinaso able cations (Al3þ, Ca2þ, and Mg2þ) by KCl solution 1 mol L1 where
et al., 2010; Formaggio et al., 1996). Such information can help re- Ca2þ and Mg2þ were quantified by atomic absorption spectropho-
searches to understand soil characteristics and variability by tometry and Al3þ exchangeable by titration with NaOH solution
spectra making its monitoring better (Viscarra Rossel et al., 2016). 0.025 mol L1, extraction of Naþ and Kþ by Mehlich-1 which were
52 J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62
quantified by flame photometry and potential acidity (H þ Al) were carried out using the statistical software ParLes version 3.1
extracted by calcium acetate solution 0.5 mol L1 at pH 7 and deter- (Viscarra Rossel, 2008), which has implemented the Partial Least
mined by titration with NaOH solution 0.025 mol L1, according to Squares (PLS) as multiple regression technique. PLS was used to
Embrapa (2011). Organic carbon (OC) was determined by the Walkley determine the best correlation between the chemical and spectral
and Black method (1934). The iron by X-ray Fluorescence Analysis data since the models obtained by this method comprise a series of
method, Dithionite-Citrate-Bicarbonate and Acid Oxalate under “b” coefficients (loadings) presenting information that can be
darkness method inserted in this article was realized by Coringa et al. interpreted in terms of composition, structure, and concentration
(2014). The particle size distribution analysis was performed ac- of some soil properties. These loadings have been used to indicate
cording to Gee and Or (2002), which clay fraction (<0.002 mm) was the spectral ranges (or bands) that effectively contribute to pre-
measured by the hydrometer method, total sand fraction diction of soil properties (Min and Lee, 2005).
(2e0.05 mm) by sieving and silt fraction (0.05e0.002 mm) was The Principal Component Analysis (PCA) was performed on the
obtained by difference. Calcium hexametaphosphate 0.1N and so- soil spectra in order to reduce the data dimensionality (Varmuza
dium hydroxide 0.1N were used as dispersing agents. These results and Filzmoser, 2009) and distinguish soils from different environ-
allowed calculating the cationic exchange capacity (CEC ¼ ments based on spectral characteristics of two datasets where one
Ca2þ þ Mg2þ þ Kþ þ Naþ þ Hþ þ Al3þ), clay fraction activity corresponds to wetland soils and the other to soils from distinct
(CFA ¼ [CEC*1000]/g kg1 of clay), and base saturation environmental conditions (oxic environmental). The PCA was per-
(V ¼ [Ca2þ þ Mg2þ þ Kþ þ Naþ/CEC]*100). formed on the soil absorbance spectra (log10[1/reflectance]). The
other spectral dataset (Terra et al., 2015) was compound by 15 soil
2.3. Soil spectral analysis profiles (total of 56 soil samples) of typical tropical soil classes (clay
content of diagnostic horizon between brackets), as follows: Red
The soil samples were air-dried, milled, and sieved through a 2- Yellow Ferralsol, dystric (LVAd - 37%); Red Ferralsol, dystric, ferric
mm mesh. Each sample corresponded to approximately 15 cm3 of (LVdf - 69%); Red Nitosol, dystric (NVdf - 78%); Ferralic Cambisol
soil that was placed in a petri dish. The reflectance data were ob- (CXbd - 79%); Red Ferrasol, dystric (LVd - 91%); Red Ferrasol, eutric
tained at laboratory using a spectroradiometer Field SpecPro (LVe - 34%); Red Yellow Acrisol, eutric (PVAe - 17%); Red Yellow
(Analytical Spectral Devices, Boulder, Colo.), which is an optical fi- Acrisol, dystric (PVAvd - 24%); Red Ferrasol, dystric (LVd - 26%); Red
ber device with spectral resolution of 1 nm from 350 to 1100 nm Ferrasol, eutric, ferric (LVef - 71%); Red Ferrasol, dystric, ferric (LVef
and 2 nm from 1100 to 2500 nm, and with a capacity of more than - 47%); Red Yellow Ferralsol, dystric (LVAd - 22%); Yellow Ferrasol,
30 scannings per second. The samples were analyzed in triplicate in dystric (LAd - 16%); Red Yellow Ferralsol, dystric (LVAd - 16%); and
order to increase the soil surface to be scanned by the sensor. The Arenosol (RQo - 14%). These profiles have been developing under
sensor was vertically placed at 8 cm from the sample, where the ferralitization and monossialitization processes and they have
reflected light was measured in an area of approximately 2 cm2 in consequently presented minerals as iron oxides and aluminum
the sample center. The light source was a 50-W halogen bulb with hydroxides: hematite, goethite, and gibbsite) and 1:1 phyllosili-
non-collimated beam to the target plane positioned at 35 cm of the cates (kaolinite), which have characteristic contributions on their
sample and at 30 zenithal angle. A Spectralon white plate that spectral behaviors.
reflects 100% was used as standard reference. The following soil properties were chosen to be quantified by
Spectral data of moist samples were also obtained for more spectral models: clay, sand, OC and CEC. From total soil data set (57
specific analyses related to the effect of water in clay fraction. These samples), 40 samples were randomly selected to calibrate the
data were graphically compared to assess the behavior of bands models, and the remaining samples (17 samples) to validate them.
regarding the clay fraction mineralogy, such as the presence of 1:1 Reflectance data were converted into absorbance values to reduce
and 2:1 minerals. Spectral behaviors were interpreted according to non-linearity in the spectrum. “Leave-one-out” cross-validation
methodology of Multiple Interpretation of Reflectance of Soils e was performed to assess the prediction models by calculating the
MIRS (Dematte ^ et al., 2014). MIRS has been used to determine following indexes: coefficient of determination (R2), Root Mean
differences between horizons based on the shape of spectral curves, Square Error (RMSE), Ratio Percentage of Deviation (RPD) and
reflectance intensity and absorption features associated with Standard Deviation of the Error (SDE).
analytical data. The interpretation of soil spectral curves was Since high OC content can mask some soil property features and
divided into seven steps, as follows: (a) the first step refers to the damage their modeling for a given soil population, we used the
intensity of the spectral curve, (b) the second step is to observe the database corresponding to the wetland soils to evaluate its influ-
general shape of the curves along the complete spectrum, (c) the ence in the quantification of clay. This procedure was carried out by
third step consists of evaluating absorption features, usually pro- removing soil samples with the highest OC content from the pop-
moted by mineralogy in specific wavelengths, (d) the fourth step is ulation, and, after that, clay content was quantified and evaluated
the comparison of spectral curves between horizons in the same by R2 and RMSE.
profile, (e) the fifth step is based on a quantitative view of the
spectral data, (f) the sixth step is related to a comparison between 3. Results and discussion
unknown profile spectra with a database already known, (g) the
seventh step is to collect all morphological and quantitative infor- 3.1. Spectral characterization of soil profile and their relation to
mation and observe the convergence of evidence that enables the pedogenesis
most probable classification for the soil to be reached.
Derivate analyses (1st and 2nd derivatives) were also carried out Pantanal landscape presents several ecosystems distributed
in the Unscrambler program to find and enhance features related to over many geomorphic features producing different soil types,
mineralogy of Fe oxides (oxides and oxyhydroxides) in clay fraction which need to be protected due to them fragility. The geo-
in order to help spectral morphological descriptions. morphology of these ecosystems is compound by river plain, high
areas that do not receive floodwater, and low areas that receive
2.4. Descriptive and statistical assessment floodwater (Coringa et al., 2014). The main pedogenetic processes
related to these soils are: redoximorphic (gleization and plinth-
Multivariate statistical analyses of soil properties and spectra ization), melanization, aggregation, synthesis, eluviation,
J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62 53
illuviation, lessivage, sodification, solodization, ferrolysis and leu- along the spectra (Figs. 1 and 2, Tables 1 and 2). However, it was
cinization (Nascimento et al., 2015; Buol et al., 2011). The soils from more pronounced in wavelengths up to 1100 nm due to higher
Pantanal of Bara ~o de Melgaço have been from a complex Quater- absorption energy (Bartholomeus et al., 2008). In almost all sub-
nary sedimentation in alluvial megafan, whose parent materials surface horizons, there was presence of strong absorption features
derived from sedimentary and mafic rocks have resulted in very at 1400, 1900, and 2200 nm attributed to molecular vibration of
weathered soils. Kaolinite has been the dominant mineral, but OH groups (Stenberg et al., 2010; Whiting et al., 2004). Among
there also is the presence of illite, hydroxy-interlayered vermiculite, them, the feature at 1900 nm was even more evident, probably
and smectites (Nascimento et al., 2013, 2015; Coringa et al., 2014). because of the presence of H2O in interstratified minerals. Ac-
The spectral behavior of soil horizons varied mainly according to cording to Whiting et al. (2004), stronger absorption intensity at
OC content, particle size distribution, presence of Fe oxides (oxides 1900 nm has indicated predominance of structural H2O in 2:1
and oxyhydroxides), such as hematite and goethite, and features minerals, such as montmorillonite and vermiculite.
related to water and hydroxyls from phyllosilicates. Higher OC The absorption feature at 2200 nm in spectral curves was
content (2%) in surface horizons reduced reflectance intensity symmetric with greater openness toward the right and a “step” at
Fig. 1. Spectral behavior of soils horizons (a) Endogleic Planosol (PLg), (b) Stagnic Solonetz (SNj), (c) Endogleyic Arenosol (ARg), (d) Fluvic Cambisol (CM1), and (e) Sapric Histosol
(HS).
54 J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62
Fig. 2. Spectral behavior of soils horizons (a) Pisolithic Plinthosol (PT), (b) Plinthic Gleysol (GL1), (c) Fluvic Cambisol (CM2), (d) Plinthic Gleysol (GL2), and (e) Cutanic Luvisol (LV).
left, characterizing kaolinite predominance (Dematte ^ et al., 2014). (<0.2 mm), which are more likely to migrate in the soil due to high
The mineralogy of soil samples has been characterized by the content of exchangeable Na and Mg.
presence of 2:1 type clays (illite, smectite and vermiculite with The variation in OC and Fe oxide content between surface and
hydroxyinterlayers), however, most of them featured dominance of subsurface horizons conferred distinct behaviors to their spectral
1:1 type clays (Nascimento et al., 2013, 2015; Coringa et al., 2014). shapes (Figs. 1 and 2). This requires a different analysis between
The high clay fraction activity (CFA) in some horizons was not only horizons, as reported by Stoner and Baumgardner (1981) and
related to 2:1 clay, but also to 1:1 þ 2:1 clay (interstratified min- Formaggio et al. (1996). These authors have classified different soil
erals) and humified OM, as observed by Coringa et al. (2014). horizons in distinct spectral patterns (reflectance intensity and
2:1 minerals or interstratified ones showed, besides the feature spectral features) based on influence of soil properties. Spectral
at 1400 nm, one sharper feature at 1900 nm and another with more curves of surface horizons has been influenced by the OC content
elongated shape at 2200 nm. Both minerals can occur simulta- according to their soil complexity interaction, which is one of the
neously, which hinders identification. However, in these soils, 2:1 main factors that hinder the full understanding of interference from
clay minerals (smectites and vermiculite) are adsorbed at surface other minerals (Henderson et al., 1989).
by 1:1 minerals. The 2:1 minerals occur in the fine clay fraction Among the soil horizons with a textural gradient, greater
J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62 55
Table 1
Particle size distribution and chemical properties of Endogleic Planosol (PLg), Stagnic Solonetz (SNj), Endogleyic Arenosol (ARg), Fluvic Cambisol (CM1), and Sapric Histosol
(HS).
Soil horizons Depth pH OCa Kþ Ca2þ Mg2þ Al3þ HþAl Naþ Tb CFAc Vd Sand Silt Clay Fete FeDCBf FeOXg
reflectance intensity was observed for eluvial horizons of PLg and most develop of the study area where time factor had provided
LV soils (Figs. 1a and 2e), since these presented sandy and loamy lower levels for these mentioned characteristics.
textural class, respectively (Tables 1 and 2). C horizons showed SNj, GL2 and LV soils (Figs. 1b, 2d and 2e) showed low reflec-
higher reflectance intensity for CM1, PT, GL2 soils (Figs. 1d, 2a and tance intensity across the entire spectra with a concave starting up
2d), which could be attributed to the occurrence of Fe micronodules to 1300 nm. Although the OC content were not very high in surface
and/or the coating of grains that compose the sand fraction by Fe horizon (SNj: 19.1 g kg1, GL2: 52.8 g kg1 and LV: 29.6 g kg1)
oxides (loamy-sand textural class of C horizon from CM1, very (Tables 1 and 2), they presented spectral behaviors very close to the
clayey and loamy-sand of 6Clc horizon from PT and very clayey of soils with the high OC content and loamy-sand and loamy textural
4Cl horizon from GL2) (Tables 1 and 2). The only exception was HS soil.
(Fig. 1e), whose high OC and sand content in almost all profile did Luvisol (LV) and Cambisol (CM1) are eutrophic soils with base
not allow this differentiation. saturation (V) between 79 and 93%, clay activity between 25 and 73
Endogleyic Planosol (PLg), Fluvic Cambisol (CM1), and Cutanic mmolc kg1, and OC content between 20.3 and 29.6 g kg1 (Tables 1
Luvisol (LV) are located at paleolevees of alluvial fan of Sa~o Lour- and 2). High CFA is related to humic substances and the presence of
enço River, corresponding to the highest landscape position 2:1 minerals in surface horizons. The deep absorption at 1900 nm is
composed to Savanna Forest and Semidecidous Forest with palm due to H2O, although can also be inferred with CFA (Figs. 1 and 2,
(Attalea phalerata Mart.). This landscape position is not directly Tables 1 and 2), and the total iron content present in the CM1 and
affected by flood pulses (Beirigo et al., 2011). LV come from parent material and landscape position, which
Pedogenectic processes related to these soils are melanization, determine more stable redox conditions and iron forms (Tables 1
aggregation, synthesis, eluviation, illuviation, and lessivage in the and 2).
LV, and solodization, redoximorphic, ferrolysis, and leucinization in Plinthic Gleysol (GL1 and GL2) and Pisolithic Plinthosol (PT) are
the PLg. Enhanced by the range of base saturation between 14 and located at lower landscape position of Cuiaba River plain
34% (dystrophic soils), CFA between 13 and 31 mmolc kg1, and OC (Nascimento et al., 2015), which is composed by shurb vegetation
content between 4.1 and 9.3 g kg1 in PLg soil. This soil class is the with legumes and shurb with papyrus (Cyperus giganteus Vahl).
56 J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62
Table 2
Particle size distribution and chemical properties of Pisolithic Plinthosol (PT), Plinthic Gleysol (GL1), Fluvic Cambisol (CM2), Plinthic Gleysol (GL2), and Cutanic Luvisol (LV).
Soil horizons Depth pH OCa Kþ Ca2þ Mg2þ Al3þ HþAl Naþ Tb CFAc Vd Sand Silt Clay Fete FeDCBf FeOXg
This landscape position is directly affected by pulse flood (Junk, extracted by the solution of dithionite-citrate-bicarbonate of so-
2001) where redoximorphic process overlays others soils pro- dium of 8.6 g kg1 (Table 1). Similar effect was also observed for
cesses. In the PT, plinthization evolutes to petroplintization process, occurrence of 2:1 minerals in soils from the Pantanal, where Fe
however it has happened in a distinct hydrological scenario from oxyhydroxides that occurred by coating the quartz grains in C ho-
the current. These are clay soils with the highest clay content of this rizons of ARg. Galva ~o and Vitorello (1998) determined that the OC
study with a range between 14 and 66%. It has a different source of content enough to mask Fe spectral characteristics within the range
parent material than CM1, LV, and PLg (Nascimento et al., 2013), between 600 and 1100 nm was at least 1.7%. Therefore, the OC
transported by Cuiaba River. This material is predominantly content of the A horizon of ARg (21.5 g kg1 or 2.15%) was enough to
composed of fine sediments. The pedogenectic processes related to significantly reduce reflectance and mask the absorption features of
these soils are involved in iron, carbon, nitrogen and phosphorous Fe oxyhydroxides, even though they were present in small quan-
cycles (Reddy and Delaune, 2008). tities. Indeed, subsurface soil is yellow, related with goethite and
Most curves were classified in the second and third type, ac- indicated at the yellow reflectance band (Fig. 1c).
cording Stoner and Baumgardner (1981), i.e., SNj, PT, GL1, CM2 and Endogleyic Arenosol (ARg), Fluvic Cambisol (CM2) and Stagnic
GL2 (Figs. 1b, 2a and 2b, 2c, 2d). For most A horizons, spectra shows ~o Lourenço River (ARg) and
Solonetz (SNj) are located at levees of Sa
low content of Fe oxyhydroxides and high sand content, besides Cuiab a River plan (CM2 and SNj), corresponding to a landscape
presenting convexity between 450 and 750 nm and concave position affected by pulse flood (Junk, 2001). These soils has similar
behavior from 750 to 1300 nm. soil processes present in GL1, GL2, and PT, however it happens in a
Endogleyic Arenosol (ARg - Fig. 1c) was homogenous in the less degree of magnitude because they are placed in a higher po-
entire profile and showed similar spectra for all horizons, classified sition and distinct hydrology. The variation of clay, silt, sand, and OC
as a second type of curve for both references used. The only dif- content at CM2 (Table 2) were observed because these soil are less
ference is in the appearance of the absorption feature of crystalline developed and still keeps most of the characteristics from sedi-
Fe at 850 nm in C horizons, not observed in the A horizon. The total mentation processes. Meanwhile, ARg was formed by coarser
Fe content expressed as Fe2O3 of 10.8 g kg1 in the C horizon of this sediment than others soils of this study, then it has sandy texture
soil are relatively low (Table 1). However, there is predominance of and good drainage, which implies in faster iron oxidation even
high-degree shapes of crystallinity, according to Fe content during the flooding.
J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62 57
For PLg, CM1 and LV soils (Figs. 1a, 1d and 2e), strong features of
Fe oxyhydroxides were identified as hydroxides at 850 nm, leading
to their classification as third type for Stoner and Baumgardner
(1981) and fourth type for Formaggio et al. (1996). These soils
had a higher expression of attributes related to Fe oxyhydroxides,
such as plinthic, chromic and other redoximorphic features. The
total Fe content in subsurface horizons ranged among 39.7 g kg1 in
the Btg horizon of PLg, 28.0 g kg1 in Bw2 horizon from CM1, and
51.5 g kg1 in Bt horizon from LV, and Fe content extracted by
dithionite-citrate-bicarbonate of sodium showed 18.5, 21.5, and
24.1 g kg1 (Tables 1 and 2), respectively. The OC content of these
soils were not the highest, however, they were enough to mask
features of Fe oxyhydroxides in these horizons.
The HS soil (Fig. 1e) displayed differentiated features because its
organic composition. The concave and ascending shape followed up
to 1100 nm, without centric concavity at 850 nm. The general
behavior of these spectra led to a reflectance reduction in the
beginning and final regions of the spectrum (up to 1100 nm and
after 1900 nm, respectively), besides there was almost absence of
absorption features, except at 1900 nm. Only in this soil class no
spectral feature of kaolinite was observed, because of the highest
OC content in the entire profile since it is not a mineral soil
(Table 1). The band at 1900 nm was attributed simultaneously to
the vibrational effects of water molecules by this wavelength and
the vibrational effect on the magnetic field of the OH of mineral
hydroxyls.
In the PLg and LV soils (Figs. 1a and 2e), there were inversions of Fig. 3. Spectral characterization of dry and wet samples of soil horizons (a) Oe-i2 -
spectra from A to B horizons. This inversion occurred around 1600 Sapric Histosol (HS), (b) Bwl2 - Plinthic Gleysol (GL1), and (c) Bt - Cutanic Luvisol (LV).
and 1700 nm, while curves of Bt horizons (which until then had
higher reflectance intensity), decreased continuously to a level
lower than that observed for the other horizons. This inversion was the coating of sand grains by OC, which tended to decrease
mainly attributed to the textural gradient between the surface and reflectance intensity, because both A horizons presented sand
subsurface horizon and where for PLg, with more expressive content of 82 and 84%, respectively (Tables 1 and 2). In the case of
gradient, the distance measured between the curves was greater ARg and HS soils (Fig. 1c and 1e), homogeneity of the spectral curve
than for LV, in agreement with Dematte ^ et al. (2014). Besides that, was attributed to particle size composition and OC content,
these soils contain Fe oxyhydroxides, which leads to a reduction in respectively, which hindered inversion of curves (Table 1).
reflectance. Soils with OC content greater than 2% mask features mainly in
In the CM1 and GL1 (Figs. 1d and 2b), inversion occurred in the VIS (Galva~o et al., 1997), which has been corroborated by Zheng
curves of A horizons, approximately between 1100 and 1200 nm, et al. (2016). In the horizons assessed, values ranged from 4.1 to
presenting had greater reflection intensity. From 1100 nm, OC 180 g kg1 (Tables 1 and 2), that is, some horizons presented OC
started to interfere less, raising intensity of A horizon. Thus, hori- content greater than 2%. All spectral curves of surface horizons
zons with higher clay content tended to have their reflectance showed lower reflectance intensity mainly at 400 and 1100 nm
reduced at the spectrum end. (VIS) (Figs. 1 and 2). On the other hand, OC also acted in the rest of
For the SNj and PT (Figs. 1b and 2a), spectra of A horizons the electromagnetic spectrum and absorbed energy. Changes in
remained below those belonging to subsurface horizons along the intensities between horizons were intrinsically related to attribute
wavelengths, and there was no inversion. This behavior is related to characteristics occurring in the horizons.
58 J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62
GL2 with 52.8 g kg1 of OC content had higher reflectance in SWIR oxides was less intensely in VIS, agreeing with Clark and Roush
due to quartz (60% of sand content), while in the 2Bwlv horizon of (1984). Fig. 5c represents the loading variation for the three main
GL2 with OC content of 7.0 g kg1, there was less reflectance due to principal components (PC). The loading represented the covariance
the lower quartz content (2% of sand content) (Table 1). These between spectral samples. Therefore, the PC identified the spectral
factors indicated that even in soils with high OC content, it was bands that contributed to the spectral patterns related to soil at-
possible to detect spectral information related to other soil tributes. The soil samples analyzed, in most part, had high OC
properties. content (2%). The presence of this attribute changed the spectral
features primarily in the visible (VIS). Therefore, corroborating this
information, loading values showed important peaks in the region
3.4. Continuum removed and principal components analysis of OC features (520e620 nm).
A strong concavity was observed (between 350 and 1100 nm) in 3.5. Distinguishing soils from wetlands and dryland by PCA
soil samples with the high OC content (Fig. 5a and 5b). It also
highlighted the bands at 1400, 1900 and 200 nm due to OH groups The three first principal components were responsible by
and H2O (Fig. 5a). In the specific case of CM2 (Fig. 5b), the influence explaining 96.77% of the spectral data variability. According to the
of the OC content reached up to 950 nm, and the influence of iron eigenvectors (Fig. 6a), the first principal component (PC 1 ¼ 80.59%)
was related to the reflectance intensity variability (Galva ~o et al.,
1995) with almost the same contribution over wavelengths,
which has been assigned to the variations in soil particle distribu-
tions and soil OC content (Dematte ^ and Terra, 2014). The PC 2
explained 13.81% of spectral data variability and mainly repre-
sented changes in the absorption features regarding to electronic
transition of iron in goethite (related numbers with Fig. 6a []) (at [1]
425 nm, [2] 480 nm, and [4] 600 nm) and to non-fundamental vi-
brations of hydroxyl (OH1) in 2:1 and, mainly, 1:1 phyllosilicates
(at [7] 1414 nm and [9] 2205 nm) (Fig. 6a). The PC 3 was responsible
to explain (2.37%) variations in the absorptions related to electronic
transition of iron in hematite (at [5] 750 nm and [6] 1050 nm), to
non-fundamental vibrations of OH in 2:1 phyllosicate or structural
water (at [8] 1917 nm), to non-fundamental vibrations of aluminol
(Al-OH) in gibbsite (at [10] 2230 nm), and to some organic com-
pounds (at [11] 2316 nm and [14] 2448 nm) (Fig. 6a) (Terra et al.,
2015).
According to Fig. 6, by the PCA was possible to distinguish soils
from different environments, where samples from drylands
(although outside Pantanal) were visibly separated of samples from
wetlands. Some clusters could be observed as those composed by
high weathered Ferrasols with oxidic clay content (Fig. 6b and 6c)
and those composed of organic soils (Histosols) and some surface
horizons with high OC content (Fig. 6b and c). Even Quartzipsam-
ments from dry and wet environments were separated from each
other. There was some confusion for soils with increasing sand
content and with predominance of 2:1 phyllosilicates (smectite,
illite, and vermiculite) independent on the environment. This
confusion occurred only for some soil samples corresponding to
horizons of Acrisol, Cambisol, Luvisol and Alfisol.
Fig. 6. Principal component analysis of wetland and soils outside this area. PLg - Endogleic Planosol, SNj - Stagnic Solonetz, ARg - Endogleyic Arenosol, CM1 - Fluvic Cambisol, HS -
Sapric Histosol, PT - Pisolithic Plinthosol, GL1 - Plinthic Gleysol, CM2 - Fluvic Cambisol, GL2 - Plinthic Gleysol, LV - Cutanic Luvisol, LVAd - Red Yellow Ferralsol, dystric, LVdf - Red
Ferralsol, dystric, ferric, NVdf - Red Nitosol, dystric, CXbd - Ferralic Cambisol, LVd - Red Ferrasol, dystric, LVe - Red Ferrasol, eutric, PVAe - Red Yellow Acrisol, eutric, PVAvd - Red
Yellow Acrisol, dystric, LVd - Red Ferrasol, dystric, LVef - Red Ferrasol, eutric, ferric, LVef - Red Ferrasol, dystric, ferric, LVAd - Red Yellow Ferralsol, dystric, LAd - Yellow Ferrasol,
dystric, LVAd - Red Yellow Ferralsol, dystric and RQo e Arenosol.
Table 3
Statistical analysis of quantification of soil properties from wetland soils.
Soil property Coefficient of determination (R2) Root Mean Square Error Ratio of Percentage Deviation Standard Deviation of the Error
obtain an ideal model that related to the soil attribute. Still, despite An important explanation for the lower estimative of clay is
all the masking effect caused by OC, R2 of 0.50 for clay expressed the given by Zheng et al. (2016). The authors indicate that OC content
importance of particle size effects in SWIR, according to Madeira over 19 g kg1 (or 1.9%) masks many absorption features. As an
Netto and Baptista (2000) and White et al. (1997). example, we observe several values over this limit (Tables 1 and 2).
Therefore, for OC, the band 400e800 nm in the spectral curve For example, spectral curve of horizon A of PT (Fig. 2a) has low
had greater weight in the estimation model of this attribute, reflectance because high OC content (19.1 g kg1) (Table 2).
agreeing with Krishnan et al. (1980), who observed that the On the other hand, the quantification of OC and clay content
wavelengths of 632.6 and 564.4 nm were related to OC. For Pirie using a population of 57 samples was possible with a 0.8 and 0.61
et al. (2005), bands from 615 to 1000 nm showed the best corre- R2 and 20.08 and 161.94 RMSE, respectively (Table 4). The
lation for this attribute, whereas for Islam et al. (2003) was from morphological evaluation showed that high OC content display
587 to 585 nm. On the other hand, the spectral range between 1850 unique shapes and intensities on spectra. In fact, it decreases
and 2350 nm showed higher contribution to the estimation model reflectance intensity and promote a concave behavior. While OC
of CEC. content decreases in soil, shapes alter. Indeed, we observed that,
J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62 61
Table 4
Predictions of clay and OC taking into account a reduction in the soil database by removing maximum values of organic carbon.
Soil sample Maximum organic Coefficient of Root Mean Square Coefficient of determination Root Mean Square Error of
carbon (g kg1) determination of clay Error of clay of organic carbon organic carbon
while we take out samples with higher OC content of the popula- environmental monitoring at low costs, reduced time, and with
tion, R2 increases from 0.61 to 0.71 on the quantification of clay environmental quality.
content (Table 4). This increase started to get stable with an OC 5 The quantification of soil properties showed R2 of 0.90, 0.68,
content maximum of 124.13 and 19.72 g kg1. Indeed, 20 g kg1 is 0.55, and 0.50 for organic carbon, cation exchange capacity, sand
the value of OC content that authors such as Stoner and and clay, respectively. The population of soil samples with
Baumgardner (1981) indicated as a maximum, from with more organic carbon content over 2% would mask the quantification
than these spectra would be altered. On the other hand, OC had 0.8 of soil properties, especially clay. Below this value of carbon
R2 of quantification with all samples in population. When taking organic, clay raises from to 0.97 R2 with low error.
out the same samples, we found the best shape with on about 0.97
R2 with a population with OC maximum of 124.13 g kg1. After this, Acknowledgments
results start to get stable and lower. Thus results indicate that, in
this population, it would be the best model using the maximum of The authors wish to thank SESC Pantanal, S~ ao Paulo Research
124.13 g kg1 of OC to estimate this attribute and clay in simulta- Foundation (FAPESP) e Pantanal Project (2009/50276-0). We also
neous. A higher OC content seems to alter significantly spectra and thank the Research Group of Geotechnologies in Soil Science
lose the ability on quantification. (GeoCis - http://esalqgeocis.wixsite.com/english), Coordination for
the Improvement of Higher Education Personnel (CAPES) and
Brazilian National Council for Scientific and Technological Devel-
4. Conclusions
opment (CNPq).
1 A relationship between spectral behaviors of soils from wetland
Appendix A. Supplementary data
and their pedogenesis and landscape positions is observed by
typical spectral morphology where oxi-reduction environments
Supplementary data related to this article can be found at http://
can be identified by reflectance spectroscopy making it an useful
dx.doi.org/10.1016/j.jenvman.2017.03.014.
tool for environmental monitoring.
2 The variations observed in the soil spectral behaviors of wet-
References
lands occurred due to the soil properties according to the
following order of importance organic carbon content, particle Araújo, S.R., Wetterlind, J., Dematte ^, J.A.M., Stenberg, B., 2014. Improving the pre-
size distribution, and mineralogical composition, in special Fe diction performance of a large tropical VIS-NIR spectroscopic soil library from
oxides. The distribution of smectites (2:1 clay) and Fe oxides, Brazil by clustering into smaller subsets or use of data mining calibration
techniques. Eur. J. Soil Sci. 65, 718e729. http://dx.doi.org/10.1111/ejss.12165.
forming coatings on other minerals, may have influenced some Baldock, J.A., Hawke, B., Sanderman, J., Macdonald, L.M., 2013. Predicting contents
behaviors and led to overestimation of these minerals in soils of carbon and its component fractions in Australian soils from diffuse reflec-
from Pantanal. tance mid-infrared spectra. Soil Res. 51, 577e595. http://dx.doi.org/10.1071/
SR13077.
3 Spectral data of wet soil samples presented the following Bartholomeus, H.M., Schaepman, M.E., Kooistra, L., Stevens, A., Hoogmoed, W.B.,
spectral characteristics: low-intensity reflectance across the Spaargaren, O.S.P., 2008. Spectral reflectance based indices for soil organic
spectrum, concave shape between 400 and 1100 nm, increased carbon quantification. Geoderma 145, 28e36. http://dx.doi.org/10.1016/
j.geoderma.2008.01.010.
intensity and shapes at 1400 and 1900 nm bands, masked
Beirigo, R.M., Vidal-Torrado, P., Stape, J.L., Couto, E.G., Andrade, G.R.P., 2011. Solos da
feature at 2200 nm, and high energy absorption, especially in reserva particular do patrimo ^ nio natural SESC Pantanal. SESC, Rio de Janeiro (in
SWIR. portuguese).
Belinaso, H., Dematte ^, J.A.M., Romeiro, S.A., 2010. Soil spectral library and its use in
4 It is possible to discriminate soils developed in wetlands of soil
soil classification. Rev. Bras. Cie^nc. Solo 34, 861e870. http://dx.doi.org/10.1590/
from outside areas by reflectance spectroscopy, and identify S0100-06832010000300027.
their specific signatures. That indicates their importance to Buol, S.W., Southard, R.J., Graham, R.C., McDaniel, P.A., 2011. Soil Genesis and
62 J.A.M. Dematt^e et al. / Journal of Environmental Management 197 (2017) 50e62