Soil & Tillage Resea ch 241 (2024) 106125
A ailable online 26 Ap il 2024
0167-1987/© 2024 The Au ho s. Published by Else ie B.V. This is an open access a icle unde he CC BY license (h p://c ea i ecommons.o g/licenses/by/4.0/).
On he impac o soil ex u e on local scale o ganic ca bon quan i ica ion:
F om ai bo ne o spacebo ne sensing domains
Vahid Khos a i
a
,
*
, Asa Gholizadeh
a
, Daniel ˇ
Zíˇ
zala
a
,
b
, Radka Kodeˇ
so ´
a
a
,
Mohammadmehdi Sabe ioon
c
,
d
, P ince Chapman Agyeman
e
, Pe a Voku ko ´
a
a
,
Anna Juˇ
ico ´
a
b
, Ma ko Spasi´
c
a
, Luboˇ
s Bo ů ka
a
a
Depa men o Soil Science and Soil P o ec ion, Facul y o Ag obiology, Food and Na u al Resou ces, Czech Uni e si y o Li e Sciences P ague, Kamýck´
a 129, Suchdol,
P ague 16500, Czech Republic
b
Resea ch Ins i u e o Soil and Wa e Conse a ion, ˇ
Zabo ˇ
esk´
a 250, P ague 15600, Czech Republic
c
Helmhol z Cen e Po sdam GFZ Ge man Resea ch Cen e o Geosciences, Sec ion 1.4 Remo e Sensing and Geoin o ma ics, Teleg a enbe g, Po sdam 14473, Ge many
d
ILVO, Flande s Resea ch Ins i u e o Ag icul u e, Fishe ies and Food, Technology and Food Science-Ag icul u al Enginee ing, Me elbeke 9820, Belgium
e
Di ision o Plan Sciences and Technology, Uni e si y o Missou i-Columbia, Columbia, MO 65211, USA
ARTICLE INFO
Keywo ds:
Soil o ganic ca bon
Ai bo ne hype spec al da a
Sen inel-2
Soil ex u e
S a i ica ion
ABSTRACT
Soil o ganic ca bon (SOC) dis ibu ion and in e ac ion wi h ligh is in luenced by soil ex u e pa ame e s (clay,
sil and sand), which makes SOC p edic ion complica ed, especially in samples wi h conside able pedological
a iabili y. Hence, unde s anding he ela ionship be ween SOC and soil ex u e is impo an wi hin he con ex
o SOC p edic ion using emo e sensing da a. The main objec i e o his s udy was o ind he impac o soil
ex u e on he pe o mance o local SOC p edic ion models ha we e de eloped on Sen inel-2 (S2) mul ispec al
and CASI/SASI (CS) hype spec al ai bo ne da a as he main p edic o a iables. One app oach o ha objec i e
was o lowe ing he ex u e a iance by s a i ica ion o he samples. The e o e, soil samples collec ed om ou
ag icul u al si es in he Czech Republic we e seg ega ed based on he i) si e-based and ii) ex u e-based s a i-
ica ion s a egies. Random o es (RF) models we e hen de eloped on all s a i ied classes wi h and wi hou
conside ing he soil ex u e pa ame e s as p edic o a iables and esul s we e compa ed wi h hose ob ained by
he RF models de eloped on he non-s a i ied (NS) samples. Bo h s a i ica ion s a egies p o ided mo e ho-
mogeneous classes, which enhanced he accu acy o SOC p edic ion, compa ed o using he NS samples. In
addi ion, he ex u e-based RF models yielded highe accu acy p edic ions han he si e-based ones. Excep sand,
adding ex u e pa ame e s o he main p edic o s imp o ed accu acy o he models, so ha he highes p edic ion
pe o mance was ob ained by a ex u e-based model de eloped on clay added CS da a. O e all, ex u e-based
s a i ica ion could signi ican ly enhance he SOC p edic ion, when he ex u e pa ame e s we e added o he
S2 and CS da a as he main p edic o a iables.
1. In oduc ion
Soil o ganic ca bon (SOC) moni o ing is a c i ical s ep o sus ainable
land use p ac ices and alle ia ing he clima e change impac s (Gau am
e al., 2022). T adi ional SOC quan i ica ion me hods a e p ecise bu
equi e edious soil sampling campaigns, expensi e equipmen and
haza dous chemicals (Schillaci e al., 2017a). The e has hus been a
cons an demand o apid and inexpensi e al e na i e echniques o
compensa e o d awbacks o hose me hods.
Wi hin ecen decades, emo e sensing has been capable o quick and
non-des uc i e moni o ing o some speci ic soil p ope ies, including
SOC. Spacebo ne mul ispec al and hype spec al senso s ha e shown
huge capabili ies o b oad and con inuous co e age o he ea h su ace
and p oduc ion o soil su ace maps e en om he inaccessible and ha d
o each egions (Biney e al., 2021). As an example, Sen inel-2 (S2) wi h
e isi in e al o 5–7 days, p o ides mul ispec al op ical image y o
high spa ial esolu ion e es ial obse a ions. Encou aging SOC p e-
dic ion pe o mances ha e been ob ained using S2 da a o empe a e
soils in he Czech Republic (Gholizadeh e al., 2018; ˇ
Zíˇ
zala e al., 2019),
Belgium and Ge many (Cas aldi e al., 2019a) and No he n F ance
* Co esponding au ho .
E-mail add ess: [email p o ec ed] (V. Khos a i).
Con en s lis s a ailable a ScienceDi ec
Soil & Tillage Resea ch
jou nal homepage: www.else ie .com/loca e/s ill
h ps://doi.o g/10.1016/j.s ill.2024.106125
Recei ed 4 Oc obe 2023; Recei ed in e ised o m 30 Ma ch 2024; Accep ed 14 Ap il 2024
Soil & Tillage Resea ch 241 (2024) 106125
2
(Vaudou e al., 2019). The hype spec al da a acqui ed by he classic
spacebo ne senso s (e.g., Hype ion) su e om low spa ial and empo al
esolu ion and low signal o noise a io (SNR) (Gomez e al., 2008).
Endea o s ha e been made o add ess hese sho comings by he cu en
o o hcoming Ea h obse a ion missions such as En i onmen al
Mapping and Analysis P og am (EnMAP) (S o ch e al., 2023) and
Cope nicus Hype spec al Imaging Mission o he En i onmen
(CHIME) (Nieke and Ras , 2018). Success ul imaging o SOC has also
been epo ed in se e al s udies ha implemen ed mul i- o hype -
spec al senso s moun ed on ai c a s (Gholizadeh e al., 2018; Hong
e al., 2020; Guo e al., 2021; Majeed e al., 2023) o d ones (Weh han
and Somme , 2021; Biney e al., 2021; 2023). The lowe imaging al i-
ude o hese ae ial pla o ms has led o highe spa ial esolu ion da a
and hence mo e obus p edic ion models and highe p ecision and
quali y dis ibu ion maps.
The accu acy o emo e sensing - based SOC es ima ion is usually
in luenced by ac o s such as soil ex u e and mois u e (Jiang e al.,
2016; Cas aldi e al., 2019a), we and d y ege a ion (D o ako a e al.,
2020; Cas aldi, 2021) and a mosphe ic condi ions (D o ako a e al.,
2020) which mus be aken in o accoun cau iously. The ela ionship
be ween SOC and soil ex u e pa ame e s (i.e., clay, sil and sand) is
complica ed and a iable (S enbe g e al., 2010), mainly depending on
some ac o s like soil ype, agg ega es, land use and managemen
p ac ices. Howe e , he e is gene ally a s ong co ela ion be ween SOC
and ine ex u e con en o soil, such as clay. Tha is because ine clay
pa icles p o ide a la ge su ace a ea wi h nega i e cha ge, allowing
hem o adso b SOC and p o ec i om mic obial de e io a ion and
decomposi ion (Ba ´
e e al., 2014; Han e al., 2016; Adiyah e al., 2022).
Sil pa icles also ha e la ge su aces which acili a e he agg ega ion
and e en ion o SOC, al hough he SOC and sil ha e a mo e complex
ela ionship since sil can be easily los by wind and wa e e osion. The
ela ionship be ween SOC and sand depends on a ious ac o s including
clima e as well as soil ype, zone, s uc u e and managemen p ac ices
(Yos and Ha emink, 2019). Gene ally, soils wi h coa se ex u es, such
as sand, s o e lowe con en o SOC (Yos and Ha emink, 2019). Tha is
mainly due o lowe su ace a ea (Almajmaie e al., 2017),
wa e -holding capaci y (ˇ
Simanský e al., 2019) and nu ien con en o
sandy soils (Almajmaie e al., 2017; ˇ
Simanský e al., 2019), which limi
hei plan g ow h capaci y and hence o ganic ma e inpu s. In addi-
ion, sandy soils ha e mo e coa se po es which cause be e ae a ion and
hence, as e decomposi ion o hei SOC con en s.
To o e come he abo e-men ioned complexi y be ween soil ex u e
pa ame e s and SOC, especially in la ge da ase s, soil samples can be
s a i ied (i.e., di ided) in o se e al classes based on a ce ain ea u e.
Models buil on s a i ied samples ha e shown imp o ed accu acies o
SOC es ima ion, which is because o lowe a iabili y and simila
cha ac e is ics o each sample g oup (We e lind and S enbe g, 2010;
Cas aldi e al., 2018). In a s udy in he Uni ed S a es, ex u e-based
s a i ica ion o nea ly 20000 na ionwide soil samples in o mo e ho-
mogeneous g oups led o a signi ican imp o emen in p edic ion o SOC
using isible–nea in a ed−sho wa e in a ed (Vis–NIR–SWIR;
350–2500 nm) e lec ance spec a (Wijewa dane e al., 2016). On a
na ion-wide scale s udy in Ge many, some soil p ope ies like ex u e
we e used o s a i ica ion o samples, which caused an inc ease in
accu acy o SOC model calib a ed on he NIR spec a. Sepa a ion o he
sandy soil samples om he o he samples caused a 14% educ ion in
oo mean squa e e o (RMSE) (Jaconi e al., 2017). In ano he s udy by
Mou a-Bueno e al. (2020), he s a i ica ion o a soil spec al lib a y
(SSL) esul ed in highe accu acy o SOC p edic ions han when using
gene al models. S a i ica ion can also be applied based on egional
(Igne e al., 2010) and geological (Udelho en e al., 2003) cha ac e is-
ics o he da ase . Region-based di ision o samples led o supe io local
models wi h enhanced SOC p edic ion pe o mances (Wijewa dane
e al., 2016).
Despi e nume ous li e a u es documen ing he ela ionship o
ex u e pa ame e s wi h SOC (Bosa a and Åg en, 1997; Dex e , 2004;
Wiesmeie e al., 2019; Wang e al., 2021; Johannes e al., 2023), only a
ew s udies ha e del ed in o hei impac on SOC p edic ion (Schillaci
e al., 2017b; Hamzehpou e al., 2019; Swe ha and Chak abo y, 2021;
Ga osi e al., 2022). Abo e-men ioned lacuna is mo e acu e, when mul i-
and hype spec al image y a e being used as he main SOC p edic o
a iables. To he bes o ou knowledge, no s udy has in es iga ed he
e ec o ex u e-based s a i ica ion on SOC p edic ion models, de el-
oped solely on he image spec a. The main pu pose o his s udy, hence,
was o explo e he in luence o soil ex u e on pe o mance o he SOC
p edic ion models de eloped using mul i- and hype spec al emo e
sensing da a. To do so, we s a i ied he soil samples in o di e en
classes based on i) geog aphical loca ion and ii) soil ex u e s a egies
and hen examined how well he S2 mul ispec al and CASI/SASI (CS)
ai bo ne hype spec al image y, can p edic SOC con en s wi hin each
de ined class, ei he wi h o wi hou in ol ing he ex u e pa ame e s as
p edic o a iables.
2. Ma e ials and me hods
2.1. S udy si es, soil sampling and analysis
This s udy was conduc ed in ou ag icul u al si es o No ´
a Ves nad
Popelkou (No ´
a Ves), Jiˇ
cín, Kluˇ
co and Pˇ
es a lky (Fig. 1), loca ed in
u al a eas o he Czech Republic. Oilseed ape, sp ing and win e ce-
eals, po a oes and maize a e mos ly being cul i a ed in hem. An
a emp was made o selec homogeneous soilscape ep esen a i e si es
wi h he same clima ic condi ions (p ecipi a ion and empe a u e), land
managemen and e ain cha ac e is ics (ˇ
Zíˇ
zala e al., 2017). Dissec ed
elie wi h ibu a y alleys, oe and back slopes and pla eaus a e some
o he main land o m ea u es o ou s udy si es.
Soils o he s udy si es, as desc ibed by he Wo ld e e ence base
(WRB) o soil esou ces (IUSS Wo king G oup WRB, 2014), a e mos ly
composed o Cambisols and S agnosols on c ys alline and sedimen a y
ocks and Che nozems and Lu isols on loess. De ails abou he sampling
si es and da a collec ion campaigns a e p esen ed in Table 1.
O e all, 320 soil samples we e collec ed om he opsoil (0–20 cm)
o ou s udy si es in June 2021 (80 samples om each). S a i ied
andom s a egy condi ioned La in Hype cube Sampling (cLHS) was
used o he sampling design. A GeoXM global posi ioning sys em (GPS)
(T imble Inc., Sunny ale, Cali o nia, USA) wi h 1 m accu acy was
employed o eco d he posi ion o each sampling poin . Soil samples
we e aken as composi e samples, hen ai -d ied, g ound, sie ed (≤
2 mm) and ho oughly mixed be o e analyzing (ISO 11464:2006). SOC
was measu ed using he Walkley–Black me hod. Soil ex u e (i.e., clay,
sil and sand) con en o he samples was de e mined by he analysis o
pa icle size dis ibu ion (less han 0.002 mm, 0.002–0.01 mm,
0.01–0.05 mm, 0.05–0.25 mm and 0.25–2 mm ac ions) using he
pipe e me hod (ISO 11277:2009).
2.2. Ai bo ne hype spec al imaging, da a p e-p ocessing and co-
egis a ion
The ai bo ne hype spec al imaging campaign was ca ied ou on
June 3 d, 2021, a leas i e days a e he las ain. Da a acquisi ion ook
place using Vis–NIR (380–1050 nm) CASI (I es L d., Calga y, Canada)
and sho wa e in a ed (SWIR; 950–2450 nm) SASI (I es L d., Calga y,
Canada) senso s (Table 2), aboa d Cessna 208B G and Ca a an pho o-
g amme ic ai c a . A g ound su ey was also ca ied ou , on he day o
he campaign, o ob ain in o ma ion on soil co e o ege a ion and
plan esidues, mois u e and su ace oughness.
Da a p e-p ocessing ( adiome ic and geome ic co ec ions) we e
pe o med by Global Change Resea ch Ins i u e o he Czech Academy o
Sciences, ia lying labo a o y o imaging sys ems (FLIS). Radiome ic
co ec ions we e done in he RadCo Ve . 9.3.6.0 (I es L d., Calga y,
Canada) using labo a o y de e mined calib a ion pa ame e s. The basic
p ocedu e o adiome ic co ec ion consis ed o he da k sub ac ion
V. Khos a i e al.
Soil & Tillage Resea ch 241 (2024) 106125
3
and con e sion o aw alues aken by he senso (DN: digi al numbe s)
in o physically de ined adiance uni s. A mosphe ic co ec ions we e
implemen ed using he MODe a e esolu ion a mosphe ic TRANs-
mission (MODTRAN) adia i e ans e model (RTM) inco po a ed in
he ATCOR-4 Ve . 7.3 (ReSe Applica ions Schlap le Inc., Wil,
Swi ze land). The esul ing a mosphe ically co ec ed da a we e
exp essed as e lec ance a he su ace le el alues. Geome ic co ec-
ions, o ho ec i ica ion and geo e e encing (UTM zone 33 N, ETRS-89)
o da a we e ca ied ou using da a collec ed by he GNSS/IMU senso
and he digi al ele a ion model (DEM) in he GeoCo Ve . 3.7.2 (I es
L d., Calga y, Canada). The noisy bands a he wo edges o he senso s’
spec a we e elimina ed, esul ed in 64 ( om 383.05 o 981.98 nm) and
98 ( om 987.5 o 2442.5 nm) emaining bands o CASI and SASI sen-
so s, espec i ely. Be o e he main p ocessing s ep, he CASI da a we e
esampled o he spa ial esolu ion o he SASI senso (6 m). Da a usion
was hen pe o med o me ge he CASI and SASI image y in o one
hype spec al ull Vis–NIR–SWIR spec al ange da a cube. A e wa ds,
e lec ance (R) was ans o med o abso bance ia log (1/R) and he i s
de i a i e (FD) ans o ma ion was applied on he spec a o emo e he
baseline o se (Gholizadeh e al., 2015; Khos a i e al., 2020).
2.3. Mul ispec al sa elli e da a p e-p ocessing and indices e ie al
The wo cloud- ee bo om o a mosphe e (BOA) Sen inel-2 le el-2A
(S2) images we e downloaded om he Cope nicus open access hub as
close as possible o he sampling campaign da e (Table 1). Since he
downloaded S2 p oduc s we e geome ically and a mosphe ically co -
ec ed, he e was no need o p e-p ocessing o he downloaded images.
The six 20 m-pixel bands we e esampled o 10 m o ob ain a ine pixel
size S2 image and hence mo e p ecise selec ion o he ba e soil pixels and
d aw be e compa ison wi h he 6 m esolu ion CS da a. S2 ela ed
Fig. 1. Dis ibu ion o s udy si es in he a) Czech Republic and sampling ields in b) Kluˇ
co , c) Pˇ
es a lky, d) No ´
a Ves nad Popelkou and e) Jiˇ
cín.
Table 1
S udy si es, soil sampling and imaging campaign speci ica ions.
Si e A ea
(ha)
Dominan soil uni Pa en ma e ial Soil sampling
da e
Ai bo ne da a
acquisi ion da e
Sen inel-2 da a
acquisi ion da e
No ´
a Ves 129 Eu ic Cambisol Pe mian-Ca boni e ous ocks (sands one,
sil s one)
June, 2021 03.06.2021 04.06.2021
Jiˇ
cín 153 Lu isol, Albic Lu isol, Lu ic
Che nozem
Pleis ocene loess June, 2021 03.06.2021 04.06.2021
Kluˇ
co 175 Calcic Che nozem, Lu ic
Che nozem, Regosols
Pleis ocene loess, C e aceous ma l, e ace
g a els
June, 2021 03.06.2021 19.06.2021
Pˇ
es a lky 58 Haplic S agnosol, Eu ic and S agnic
Cambisol, Lep osol
Complex o P o e ozoic and Paleozoic
ocks (schis , g anodio i e)
June, 2021 03.06.2021 19.06.2021
Table 2
CASI and SASI senso s speci ica ions.
Senso Numbe o
bands
Spec al
ange (nm)
Spec al
esolu ion
(nm)
Field o
iew
(FOV)
FWHM
(nm)
CASI 72 Vis–NIR,
380–1050
3.2 40◦10
SASI 100 SWIR,
950–2450
15 40◦15
V. Khos a i e al.
Soil & Tillage Resea ch 241 (2024) 106125
4
a iables including en bands and 19 spec al indices we e ex ac ed and
calcula ed a each sampling poin and employed as co a ia es o de elop
he SOC p edic ion models. A Pea son co ela ion analysis was con-
duc ed o explo e he possible au oco ela ions be ween he inpu
indices and consequen ad e se e ec s such as inc easing he calib a ion
complexi y and educ ion in accu acy o he SOC es ima ions. Acco ding
o he esul s, he e we e signi ican co ela ions be ween some indices
such as NDVI and TVI, WDVI and V, BI2 and TSAVI, and EVI and SAVI.
Some combina ions o he indices we e used as inpu s o in es iga e
hei e ec s on he p edic ion. Conside ing he negligible di e ences
obse ed be ween he esul s, all indices we e inco po a ed in calib a-
ion o he SOC models o employ he maximum numbe o S2 ela ed
co a ia es. The S2 bands’ de ails and equa ions o he indices can be
ound in Table A1 and Table A2, espec i ely.
2.4. Selec ion o ba e soil pixels
The no malized di e ence ege a ion index (NDVI) and no malized
bu n a io 2 (NBR2) (Table A2) indices ha e been conside ably used o
delinea e g een ege a ion (Huang e al., 2021) and c op esidues
(D o ako a e al., 2023), espec i ely. Al hough he sensi i i y o NBR2
o de ec c op esidues on we soils is e y limi ed, hey o m a linea
ela ionship o d y soils (D o ako a e al., 2021), like he case in his
s udy. Acco dingly, bo h NDVI and NBR2 indices we e implemen ed o
bo h S2 and CS o emo e he pixels wi h g een and d y ege a ion and
c op esidue co e . The NDVI h eshold was de e mined by isual in-
spec ion o he ue colo composi e (TCC) o S2 (R: 664.6 nm, G:
559.8 nm and B: 492.4 nm) and ai bo ne (R: 658.8 nm, G: 554.2 nm and
B: 478.1 nm) images. Va ious NBR2 h eshold alues anging om 0.05
o 0.25, wi h inc emen o 0.01 we e selec ed and es ed o isola e he
ba e soil a eas. Resul s we e alida ed by compa ing he e ie ed pixels
wi h ield-based g ound u h obse a ions.
2.5. Sample s a i ica ion and co ela ion analysis
The samples co esponding o he emaining ba e soil a eas we e
s a i ied based on he s udy si es and he Uni ed S a es Depa men o
Ag icul u e (USDA) ex u e iangle. The ex u e iangle o he USDA is
one o he mos popula sys ems o soil classi ica ion, e ol ed om eigh
classes igh -angled iangle (Whi ney, 1911) and en classes equila e al
iangle (Da is, 1927) models o a wel e g oups inal e sion in o-
duced in 1951, which is cu en ly in e ec (USDA, 1951, 2017).
The Pea son co ela ion be ween SOC and each ex u e pa ame e
was calcula ed wi hin each s a i ied class o p o ide an insigh in o
hei ela ionship, while in es iga ing he possible impac s o in e ac ion
be ween he wo a iables on SOC p edic ion.
2.6. SOC modeling and pe o mance assessmen
Fo each s a i ied class, andom o es (RF) algo i hm (B eiman,
2001) was used o de elop he p edic ion models using S2 and CS im-
age y, wi h and wi hou sepa a e and combined inco po a ion o he
ex u e pa ame e s as addi ional p edic o a iables. Resul s we e hen
compa ed wi h hose ob ained by he RF models calib a ed on
non-s a i ied (NS) samples (i.e., on he whole da ase ). Th ee speci ic
pa ame e s used o he RF models in his s udy we e: i) he numbe o
ees in he o es (n
ee
), ii) quan i y o a iables used o each ee
(m
y
), and iii) he minimum da a pe node (nodesize). A g id sea ch was
conduc ed o de e mine he op imum quan i ies o he pa ame e s and
he alue o 500 was ob ained o n
ee
. One hi d o he o al numbe o
p edic o a iables was he op imal quan i y o he m
y
while he
nodesize was se o 5. The ‘ andomFo es ’ package in RS udio en i on-
men (R Co e Team, 2020) was used o un he RF model. Using a iable
impo ance in p ojec ion (VIP), he con ibu ion o all inpu p edic o
a iables, including he ex u e pa ame e s, we e assessed on ou come
o each de eloped class-model. This can help o de e mine he ex en o
which he ex u e can in luence he p edic ion accu acy.
To de elop he models, samples we e di ided in o aining and
es ing (75:25% a io) da ase s, using he Kenna d–S one (KS) algo-
i hm, which chooses ep esen a i e samples based on a dis ance mea-
su e (Kenna d and S one, 1969). A 5- old c oss- alida ion echnique was
used o calib a e he SOC p edic ion models on he aining da ase . The
wo mos common e alua ion c i e ia, RMSE and coe icien o de e -
mina ion (R
2
) (Eq. 1 and Eq. 2, espec i ely) we e selec ed o assess
pe o mance o he de eloped models.
RMSE =
1
n∑
n
i=1
(oi−pi)2
√(1)
R2=1−∑n
i=1(oi−pi)2
∑n
i=1(oi−
μ
o)2(2)
whe e, n is he numbe o samples, µ
o
is he mean o he obse ed alues
and o
i
and p
i
a e he obse ed and p edic ed alues, espec i ely. A
obus model, gene ally, has high R
2
and low RMSE. A lowcha o he
me hodology is p o ided in Fig. 2.
3. Resul s
3.1. Gene al s a is ics and co ela ions
A e igo ous pe usal, pixels wi h NDVI alues below 0.23 we e kep
o bo h sa elli e and ai bo ne images. Di e en models we e hen
de eloped on he samples emaining a e applying a ious NBR2
h esholds ( om 0.05 o 0.25 wi h inc emen o 0.01) and he op imal
alue o 0.09 was ob ained. By applying he op imal NDVI and NBR2
h esholds, 200 ba e soil pixels/samples which we e ully compa ible
be ween he wo da ase s we e inally e ie ed and used o he es o
he s udy.
Fig. 3 and Fig. 4 show he s a is ics, his og ams and he emaining
soil samples (based on he USDA ex u e iangles), o he si e-based
and ex u e-based s a i ied samples, espec i ely. The s a is ics o NS
samples ha e also been shown in Fig. 3. Acco ding o he esul s, SOC
con en s o he NS samples anged om he minimum alue o 0.31% o
he maximum alue o 2.59% wi h a e age and coe icien o a ia ion
(CV) o 1.19% and 0.27, espec i ely. Among he s udy si es, SOC in
Kluˇ
co had he highes CV o 0.29 ollowed by Pˇ
es a lky (CV =0.24)
and No ´
a Ves (CV =0.22), espec i ely. The highes CV o clay and sil
we e obse ed in Pˇ
es a lky, while Kluˇ
co showed he highes CV o
sand. Tha was while Jiˇ
cín indica ed he lowes CV alues o all h ee
ex u e pa ame e s. Acco ding o he his og am plo s and ela ed dis-
ibu ion cu es, SOC con en s o all classes almos ollowed he no mal
dis ibu ion. O he a iables did no ollow a ce ain dis ibu ion in
di e en classes (Fig. 3).
Acco ding o he USDA ex u e iangle (Fig. 4), samples we e mos ly
plo ed in lowe posi ioned classes o sandy loam (SL), loam (L), sil loam
(SiL), sil y clay loam (SiCL) and clay loam (CL). Fo his eason, samples
we e s a i ied in o ou g oups o SL wi h 46 samples, L wi h 59 sam-
ples, SiL wi h 51 samples and SiCL_CL as combina ion o SiCL and CL
wi h 44 samples (Fig. 4). Acco dingly, he SiCL_CL and SiL classes had
highe amoun s o clay and sil , class L had ela i ely equal quan i ies o
sand and sil and he SL class included samples wi h high sand con en s.
Conside ing he SOC le el in each ex u e class, he SiCL_CL had he
highes a e age o SOC (1.23%) wi h highes CV o 0.31. SiL and SL
classes had he lowes SOC con en wi h a e ages o 1.15% and 1.17%,
espec i ely. Acco ding o he dis ibu ions o soil ex u es wi hin he
USDA ex u e iangle, mos o No ´
a Ves, Jiˇ
cín, Kluˇ
co and Pˇ
es a lky
samples we e in loam, sil loam, sil y clay loam and loam classes,
espec i ely (Fig. 4).
The Pea son co ela ion coe icien s be ween SOC and clay, sil and
sand in he NS and s a i ied classes a e p esen ed in Table 3. None o he
V. Khos a i e al.
Soil & Tillage Resea ch 241 (2024) 106125
5
ex u e pa ame e s showed signi ican co ela ion wi h SOC in he NS
class. Rega ding he si e-based s a i ied classes, SOC showed signi ican
co ela ion wi h all ex u e pa ame e s in No ´
a Ves, Kluˇ
co and
Pˇ
es a lky si es. In addi ion, he in e ela ionship was posi i e be ween
SOC and clay as well as SOC and sil and nega i e be ween SOC and sand
wi hin hose s udy si es. The highes posi i e co ela ion o SOC was
obse ed wi h clay in Kluˇ
co ( =0.84, P <0.001), while he highes
nega i e co ela ion was de ec ed be ween SOC and sand o Pˇ
es a lky
( = − 0.39, P <0.001). No signi ican co ela ions we e obse ed be-
ween SOC and ex u e pa ame e s in Jiˇ
cín.
Conside ing he ex u e-based s a i ied classes, SOC was posi i ely
co ela ed wi h he clay con en , while he co ela ion be ween SOC and
sand was nega i e in all classes excluding L. Though no qui e s ong, he
highes nega i e co ela ion be ween SOC and sand con en was in SL
class ( = − 0.32, p- alue <0.05), while SOC and clay showed he
highes posi i e co ela ion in SiCL_CL ( =0.75, p- alue <0.05).
3.2. Modeling esul s
3.2.1. Gene al compa ison
Tables A3 and 4 display he SOC p edic ion esul s o he models
c ea ed on he NS, si e- and ex u e-based s a i ied samples (calib a ion
and alida ion se s, espec i ely). Fig. 5 and Fig. A1 also show he
sca e plo s o he bes models ob ained o each s a egy and s a i ied
class on he S2 +clay and CS +clay da ase s, espec i ely. P edic ion
esul s o models de eloped on NS samples a e likewise shown. As can be
Fig. 2. Flowcha o he s udy.
V. Khos a i e al.
Soil & Tillage Resea ch 241 (2024) 106125
6
Fig. 3. His og ams and summa y s a is ics o SOC ( i s column), clay (second column), sil ( hi d column) and sand ( ou h column) wi hin all s a i ied and non-
s a i ied classes (NS: non-s a i ied, SL: Sandy Loam, L: Loam, SiL: Sil Loam, SiCL_CL: combina ion o SiCL (Sil y Clay Loam) and CL (Clay Loam)).
V. Khos a i e al.
Soil & Tillage Resea ch 241 (2024) 106125
7
obse ed, he models de eloped on he CS ou pe o med hose de el-
oped on he S2 da a. This supe io i y was mo e e iden (up o 60%) in
he models de eloped on NS samples (Table 4, Fig. 5 and Fig. A1).
E en hough ewe samples we e used o modeling, models de el-
oped on he s a i ied samples indica ed be e pe o mance han hose
de eloped on he NS samples. As an example, he si e-based s a i ica-
ion caused a minimum imp o emen o 7%, when using he CS da a as
he main p edic o a iable (+clay, sil and sand). This imp o emen
was mo e p onounced (a leas 29%), when he S2 image was conside ed
as he main p edic o a iable (+clay, sil and sand). The lowes model
pe o mances among all (RMSE >0.17% and R
2
<0.41), linked o hose
de eloped on he NS samples, ega dless o he employed da a ype and
p edic o a iables.
Compa ing he esul s ob ained unde he si e- and ex u e-based
s a i ica ion s a egies, he models de eloped on SiCL_CL and SiL
ou pe o med all he si e-based s a i ied ones, showing ha di iding
samples based on hei ex u e pa ame e s can lead o mo e success ul
SOC p edic ions. This conclusion holds ue, excep o some models
de eloped on he classes wi h highe sand con en , especially SL, which
pe o med e y poo ly (RMSE >0.16% and R2 <0.42). In o he wo ds,
he models de eloped o Kluˇ
co and Pˇ
es a lky, unde he si e-based
s a i ica ion, had highe pe o mances han he L and SL models,
de eloped unde he ex u e-based s a i ica ion. The pe o mance o
some models de eloped o Jiˇ
cín was also simila o hose de eloped on
he L class (like he model de eloped on he CS da a +clay wi h RMSE =
0.13% and R
2
=0.60).
Among all he ex u e pa ame e s, adding clay as he p edic o a -
iable caused he highes enhancemen in pe o mance o he models. The
bes imp o emen was ob ained o Jiˇ
cín, whe e inco po a ion o clay
enhanced he p edic ion pe o mance up o 30% compa ed o he
Fig. 3. (con inued).
V. Khos a i e al.
Soil & Tillage Resea ch 241 (2024) 106125
8
si ua ions ha no ex u e pa ame e added as he inpu a iable. Adding
all ex u e pa ame e s oge he also made subs an ial imp o emen s
compa ed o when he mul i- o hype spec al da a we e used as he only
p edic o s. Adding sil and sand indi idually, on he o he hand, did no
in oduce any imp o emen o he models’ p edic ion pe o mances. I is
no ewo hy o men ion ha adding he ex u e pa ame e s caused mo e
imp o emen o he models de eloped on he si e-based s a i ied
samples han hose de eloped on he ex u e-based s a i ied ones.
Fu he mo e, adding ex u e pa ame e s caused an a e age highe pe -
o mance o 4.73% o he models de eloped using he S2 da a, while he
a e age enhancemen was 3.78% conside ing he CS da a as he main
p edic o .
3.2.2. Si e-based s a i ica ion
Conside ing Table 4, Pˇ
es a lky and Kluˇ
co yielded accep able SOC
p edic ion pe o mances on bo h S2 and ai bo ne da a. Howe e , he
bes esul s unde he si e-based s a i ica ion s a egy we e ob ained on
he ai bo ne hype spec al da a +clay wi h RMSE =0.11% and R
2
=
0.71, ollowed by he ai bo ne hype spec al da a +clay +sil +sand
wi h RMSE =0.12% and R
2
=0.70, bo h o he Kluˇ
co si e. Likewise,
o he Pˇ
es a lky si e, he models de eloped on he ai bo ne hype -
spec al da a +clay (RMSE =0.12% and R
2
=0.65) and he ai bo ne
hype spec al da a +clay +sil +sand (RMSE =0.12% and R
2
=0.63)
showed he highes pe o mances.
The same end was obse ed o he models de eloped using S2
(wi h and wi hou he ex u e pa ame e s) da ase . Acco dingly, Kluˇ
co
Fig. 4. USDA ex u e iangles o he non-s a i ied, si e- and ex u e-based s a i ied soil classes (NS: non-s a i ied, SL: Sandy Loam, L: Loam, SiL: Sil Loam,
SiCL_CL: combina ion o SiCL (Sil y Clay Loam) and CL (Clay Loam)).
V. Khos a i e al.
Soil & Tillage Resea ch 241 (2024) 106125
9
and Pˇ
es a lky had ela i ely accep able p edic ions, especially when
ex u e pa ame e s we e inco po a ed as he p edic o a iables (S2 +
clay wi h RMSE =0.13% and R
2
=0.64, S2 +clay +sil +sand wi h
RMSE =0.13% and R
2
=0.59, o Kluˇ
co , and S2 +clay wi h RMSE =
0.15% and R
2
=0.61 and S2 +clay +sil +sand wi h RMSE =0.14%
and R
2
=0.58, o Pˇ
es a lky). Fo o he si es (No ´
a Ves and Jiˇ
cín), he
pe o mance o si e-based s a i ica ion models d opped, ega dless o
using S2 o ai bo ne da ase s. Fo example, he bes model o Jiˇ
cín and
No ´
a Ves was de eloped using ai bo ne hype spec al da a +clay wi h
RMSE =0.13% and R
2
=0.60 and RMSE =0.16% and R
2
=0.46,
espec i ely.
3.2.3. Tex u e-based s a i ica ion
Among all models de eloped on he ex u e-based s a i ied samples,
hose de eloped on he SiCL_CL samples showed he highes p edic ion
pe o mances, ollowed by he SiL models (Table 4). This was ega dless
o he ype o da a and he ex u e pa ame e s added as inpu p edic o
a iables. The bes p edic ion accu acy was ob ained o he SiCL_CL
class model de eloped on he ai bo ne hype spec al da a +clay (RMSE
=0.08% and R
2
=0.81), ollowed by he same class model de eloped on
he ai bo ne hype spec al da a +clay +sil +sand (RMSE =0.10% and
R
2
=0.78). Acco ding o Table 4, i is highligh ed ha he SiCL_CL
models also yielded p omising esul s including RMSE =0.10% and R
2
=0.78 o he ai bo ne hype spec al da a +clay and RMSE =0.10%
and R
2
=0.73 o he ai bo ne hype spec al da a +clay +sil +sand.
Con a ily, pe o mance o models de eloped on he L (RMSE =0.12%
and R
2
=0.62) and SL (RMSE =0.16% and R
2
=0.42) classes was no
easonable.
3.2.4. Va iable impo ance
Va iable impo ance plo s o he RF models de eloped on di e en
classes o samples a e shown in Fig. 6. Acco dingly, he 30 highes
in luen ial p edic o a iables o each model a e p esen ed along wi h
hei pe cen age inc ease in mean squa ed e o (%IncMSE) as a mea-
su e o a iable impo ance ob ained du ing he calib a ion o he
models. As e idenced, clay was amongs he op impo an a iables in
almos all models (excluding No ´
a Ves CS). I also made he la ges
con ibu ion o he accu acy o he model de eloped on SiCL_CL class
wi h %IncMSE o 3.48 o he CS da ase . Compa ed o sand, sil be e
con ibu ed o de elopmen o he models, especially when combined
wi h he CS da a. The mos signi ican impac o sand was on he models
de eloped o SL and Pˇ
es a lky classes (Fig. 6).
4. Discussion
4.1. SOC p edic ion: e ec o co ela ion wi h he ex u e pa ame e s
I has gene ally been p o ed ha SOC con en is posi i ely and
nega i ely ela ed o clay and sand con en s o soil, espec i ely (Song
Table 3
Co ela ion ma ix be ween SOC con en and soil ex u e pa ame e s in he non-
s a i ied and s a i ied soil classes.
Co ela ion coe icien
Tex u e pa ame e NS No ´
a Ves Jiˇ
cín Kluˇ
co Pˇ
es a lky
Clay 0.05 0.46* 0.19 0.84** 0.48**
Sil -0.10 041** -0.08 0.73* 0.60***
Sand 0.05 -0.22 0.10 -0.23** -0.39***
Tex u e pa ame e SL L SiL SiCL_CL
Clay 0.60* 0.22* 0.30** 0.75*
Sil 0.50*** -0.10 0.36** -0.40**
Sand -0.32* 0.05 -0.04 -0.01
NS: non-s a i ied, SL: sandy loam, L: loam, SiL: sil loam, SiCL_CL: combina ion
o sil y clay loam (SiCL) and clay loam (CL).
*, ** and *** indica e signi icancy a P <0.05, P <0.01 and P <0.001,
espec i ely.
Table 4
SOC p edic ion esul s o he RF models de eloped on he non-s a i ied and s a i ied classes o samples ( alida ion da ase ).
Da a ype Tex u e inpu a iable NS
(200 samples)
No ´
a Ves
(60 samples)
Jiˇ
cín
(18 samples)
Kluˇ
co
(59 samples)
Pˇ
es a lky
(63 samples)
SL
(46 samples)
L
(59 samples)
SiL
(51 samples)
SiCL_CL
(44 samples)
RMSE R
2
RMSE R
2
RMSE R
2
RMSE R
2
RMSE R
2
RMSE R
2
RMSE R
2
RMSE R
2
RMSE R
2
Sen inel-2 None 0.27 0.24 0.28 0.32 0.19 0.43 0.16 0.55 0.17 0.55 0.28 0.24 0.17 0.48 0.13 0.56 0.13 0.59
All 0.26 0.26 0.27 0.34 0.14 0.55 0.13 0.59 0.14 0.58 0.25 0.25 0.13 0.55 0.11 0.59 0.11 0.63
Clay 0.26 0.27 0.27 0.35 0.15 0.56 0.13 0.64 0.15 0.61 0.26 0.24 0.12 0.57 0.11 0.61 0.10 0.66
Sil 0.27 0.24 0.27 0.32 0.18 0.42 0.16 0.55 0.17 0.54 0.28 0.23 0.16 0.48 0.13 0.57 0.13 0.60
Sand 0.27 0.23 0.28 0.31 0.21 0.41 0.15 0.57 0.18 0.55 0.27 0.23 0.18 0.44 0.15 0.56 0.14 0.55
CS None 0.19 0.38 0.18 0.43 0.16 0.53 0.14 0.68 0.14 0.60 0.19 0.36 0.16 0.55 0.12 0.69 0.12 0.76
All 0.17 0.41 0.16 0.44 0.13 0.59 0.12 0.70 0.12 0.63 0.17 0.37 0.12 0.62 0.10 0.73 0.10 0.78
Clay 0.18 0.41 0.16 0.46 0.13 0.60 0.11 0.71 0.12 0.65 0.16 0.42 0.13 0.60 0.10 0.78 0.08 0.81
Sil 0.19 0.38 0.20 0.43 0.15 0.53 0.13 0.67 0.15 0.62 0.18 0.36 0.14 0.58 0.13 0.71 0.11 0.76
Sand 0.23 0.37 0.21 0.41 0.15 0.52 0.15 0.63 0.12 0.67 0.18 0.34 0.15 0.55 0.13 0.69 0.12 0.75
NS: non-s a i ied, SL: sandy loam, L: loam, SiL: sil loam, SiCL_CL: combina ion o sil y clay loam (SiCL) and clay loam (CL).
V. Khos a i e al.
Soil & Tillage Resea ch 241 (2024) 106125
16
da a o he es ima ion and mapping o opsoil o ganic ca bon: Fea u e selec ion
coupled wi h andom o es . Soil Tillage Res. 199, 104589.
Hong, H., Chen, S., Fang, Q., Algeo, T.J., Zhao, L., 2019. Adso p ion o o ganic ma e on
clay mine als in he Dajiuhu pea soil ch onosequence, Sou h China. Appl. Clay Sci.
178, 105125.
Hon o ia, C., Saa, A., Rod íguez-Mu illo, J.C., 1999. Rela ionships be ween soil o ganic
ca bon and si e cha ac e is ics in peninsula Spain. Soil Sci. Soc. Am. J. 63, 614–621.
Huang, S., Tang, L., Hupy, J.P., Wang, Y., Shao, G., 2021. A commen a y e iew on he
use o no malized di e ence ege a ion index (NDVI) in he e a o popula emo e
sensing. J. Res. 32, 1–6.
Igne, B., Ree es, J.B., McCa y, G., Hi ely, W.D., Lund, E., Hu bu gh, C.R., 2010.
E alua ion o spec al p e ea men s, pa ial leas squa es, leas squa es suppo
ec o machines and locally weigh ed eg ession o quan i a i e spec oscopic
analysis o soils. J. In a ed Spec osc. 18, 167–176.
IUSS Wo king G oup WRB, 2014. Wo ld e e ence base o soil esou ces. In:
In e na ional Soil Classi ica ion Sys em o Naming Soils and C ea ing Legends o
Soil Maps; Wo ld Soil Resou ces Repo s No. 106. FAO, Rome, I aly A ailable online.
h p:// www. ao.o g/3/a-i3794e.pd , Accessed da e: 1 Oc obe 2015.
Jaconi, A., Don, A., F eibaue , A., 2017. P edic ion o soil o ganic ca bon a he coun y
scale: s a i ica ion s a egies o nea -in a ed da a. Eu . J. Soil Sci. 68, 919–929.
Jiang, Q., Chen, Y., Guo, L., Fei, T., Qi, K., 2016. Es ima ing soil o ganic ca bon o
c opland soil a di e en le els o soil mois u e using VIS-NIR spec oscopy. Remo e
Sens. 8, 755.
Jindaluang, W., Kheo uen omne, I., Suddhip aka n, A., Singh, B.P., Singh, B., 2013.
In luence o soil ex u e and mine alogy on o ganic ma e con en and composi ion
in physically sepa a ed ac ions soils o Thailand. Geode ma 195, 207–219.
Johannes, A., Sauze , O., Ma e , A., Boi in, P., 2023. Soil o ganic ca bon con en and
soil s uc u e quali y o clayey c opland soils: A la ge-scale s udy in he Swiss Ju a
egion. Soil Use 39, 707–716.
Juˇ
ico ´
a, A., Chuman, T., ˇ
Zíˇ
zala, D., 2022. Soil o ganic ca bon con en and s ock change
a e hal a cen u y o in ensi e cul i a ion in a che nozem a ea. Ca ena 211,
105950.
Kenna d, R.W., S one, L.A., 1969. Compu e aided design o expe imen s. Technome ics
11, 137–148.
Kennedy, M.J., Pe ea , D.R., Hill, R.J., 2002. Mine al su ace con ol o o ganic ca bon in
black shale. Science 295, 657–660.
Khos a i, V., A dejani, F.D., A ya a , A., Youse i, S., Ka ami, S., 2020. P edic ion o
coppe con en in was e dump o Sa cheshmeh coppe mine using isible and nea -
in a ed e lec ance spec oscopy. En i on. Ea h Sci. 79, 165.
Konen, M.E., Bu as, C.L., Sando , J.A., 2003. O ganic Ca bon, Tex u e, and Quan i a i e
Colo Measu emen Rela ionships o Cul i a ed Soils in No h Cen al Iowa. Soil Sci.
Soc. Am. J. 67, 1823–1830.
Lei eld, J., Bassin, S., Fuh e , J., 2005. Ca bon s ocks in Swiss ag icul u al soils p edic ed
by land-use, soil cha ac e is ics, and al i ude. Ag ic. Ecosys . En i on. 105, 255–266.
Liu, F., Rossi e , D.G., Song, X.D., Zhang, G.L., Yang, R.M., Zhao, Y.G., Li, D.C., Ju, B.,
2014. A simila i y-based me hod o h ee-dimensional p edic ion o soil o ganic
ma e concen a ion. Geode ma 263, 254–263.
Majeed, I., Pu usho haman, N.K., Chak abo y, P., Panig ahi, N., Vasa a, H.B., Das, B.S.,
2023. Es ima ion o soil and c op esidue pa ame e s using AVIRIS-NG hype spec al
da a. In . J. Remo e Sens. 44 (6), 2005–2038.
Mau ya, S., Ab aham, J.S., Somasunda am, S., To eja, R., Gup a, R., Makhija, S., 2020.
Indica o s o assessmen o soil quali y: a mini- e iew. En i on. Moni . Assess. 192,
1–22.
McLauchlan, K.K., 2006. E ec s o soil ex u e on soil ca bon and ni ogen dynamics
a e cessa ion o ag icul u e. Geode ma 136, 289–299.
Mou a-Bueno, J.M., Dalmolin, R.S.D., Ho s -Heinen, T.Z., en Ca en, A., Vasques, G.M.,
Do o, A.C., G unwald, S., 2020. When does s a i ica ion o a sub opical soil
spec al lib a y imp o e p edic ions o soil o ganic ca bon con en ? Sci. To al
En i on. 737, 139895.
Nieke, J. and Ras , M., 2018, July. Towa ds he cope nicus hype spec al imaging
mission o he en i onmen (CHIME). In Iga ss 2018-2018 ieee in e na ional
geoscience and emo e sensing symposium (pp. 157-159). IEEE.
Pan, G., Li, L., Zhang, X., Dai Jingyu, Z.Y., Zhang, P., 2003. Soil o ganic ca bon s o age o
China and he seques a ion dynamics in ag icul u al lands. Ad . Ea h Sci. 18, 609.
Schillaci, C., Acu is, M., Lomba do, L., Lipani, A., Fan appi`
e, M., M¨
a ke , M., Saia, S.,
2017b. Spa io- empo al opsoil o ganic ca bon mapping o a semi-a id
Medi e anean egion: The ole o land use, soil ex u e, opog aphic indices and he
in luence o emo e sensing da a o modelling. Sci. To al En i on. 601–602.
Schillaci, C., Lomba do, L., Saia, S., Fan appi`
e, M., M¨
a ke , M., Acu is, M., 2017a.
Modelling he opsoil ca bon s ock o ag icul u al lands wi h he S ochas ic G adien
T eeboos in a semi-a id Medi e anean egion. Geode ma 286, 35–45.
ˇ
Simanský, V., Ju iga, M., Jonczak, J., Uza owicz, Ł., S ępie´
n, W., 2019. How
ela ionships be ween soil o ganic ma e pa ame e s and soil s uc u e
cha ac e is ics a e a ec ed by he long- e m e iliza ion o a sandy soil. Geode ma
342, 75–84.
Sollins, P., Homann, P., Caldwell, B.A., 1996. S abiliza ion and des abiliza ion o soil
o ganic ma e : Mechanisms and con ols. Geode ma 74, 65–105.
Song, J., Gao, J., Zhang, Y., Li, F., Man, W., Liu, M., Wang, J., Li, M., Zheng, H., Yang, X.,
Li, C., 2022. Es ima ion o Soil O ganic Ca bon Con en in Coas al We lands wi h
Measu ed VIS-NIR Spec oscopy Using Op imized Suppo Vec o Machines and
Random Fo es s. Remo e Sens (Basel) 14, 4372.
S enbe g, B., Visca a Rossel, R.A., Mouazen, A.M., We e lind, J., 2010. Visible and Nea
In a ed Spec oscopy in Soil Science. pp. 163–215.
S o ch, T., Honold, H.P., Chab illa , S., Habe meye , M., Tucke , P., B ell, M.,
Ohndo , A., Wi h, K., Be z, M., Kuchle , M., Mühle, H., 2023. The EnMAP imaging
spec oscopy mission owa ds ope a ions. Remo e Sens En i on. 294, 113632.
Swe ha, R.K., Chak abo y, S., 2021. Combina ion o soil ex u e wi h Nix colo senso
can imp o e soil o ganic ca bon p edic ion. Geode ma 382, 114775.
Thu i`
es, L., Pansu, M., Felle , C., He mann, P., R´
emy, J.C., 2001. Kine ics o added
o ganic ma e decomposi ion in a Medi e anean sandy soil. Soil Biol. Biochem 33,
997–1010.
Udelho en, T., Emme ling, C., Ja me , T., 2003. Quan i a i e analysis o soil chemical
p ope ies wi h di use e lec ance spec ome y and pa ial leas -squa e eg ession:
A easibili y s udy. Plan Soil 251, 319–329.
USDA, 1951. Soil su ey manual. Soil Conse a ion Se ice Volume Handbook, U.S.
Depa men o Ag icul u e 18.
USDA, 2017. Soil su ey manual. Soil Conse a ion Se ice Volume Handbook, U.S.
Depa men o Ag icul u e 18.
Vaudou , E., Gomez, C., Loiseau, T., Baghdadi, N., Loube , B., A ouays, D., Ali, L.,
Lagache ie, P., 2019. The impac o acquisi ion da e on he p edic ion pe o mance
o opsoil o ganic ca bon om Sen inel-2 o c oplands. Remo e Sens 11 (18), 2143.
Wang, Q., Wen, Y., Zhao, B., Hong, H., Liao, R., Li, J., Liu, J., Lu, H., Yan, C., 2021.
Coas al soil ex u e con ols soil o ganic ca bon dis ibu ion and s o age o
mang o es in China. Ca ena 207, 105709.
Wang, S.Q., Yu, G.R., Zhao, Q.J., Niu, D., Chen, Q.M., Wu, Z.F., 2005. Spa ial
cha ac e is ics o soil o ganic ca bon s o age in China’s c oplands. Pedosphe e 15,
417–423.
Weh han, M., Somme , M., 2021. A pa simonious app oach o es ima e soil o ganic
ca bon applying unmanned ae ial sys em (UAS) mul ispec al image y and he
opog aphic posi ion index in a he e ogeneous soil landscape. Remo e Sens. 13 (18),
3557.
We e lind, J., S enbe g, B., 2010. Nea -in a ed spec oscopy o wi hin- ield soil
cha ac e iza ion: small local calib a ions compa ed wi h na ional lib a ies spiked
wi h local samples. Eu . J. Soil Sci. 61, 823–843.
Whi ney, M., 1911. Use o soils eas o he G ea Plains egion.
Wiesmeie , M., U banski, L., Hobley, E., Lang, B., on Lü zow, M., Ma in-Spio a, E., an
Wesemael, B., Rabo , E., Ließ, M., Ga cia-F anco, N., Wollschl¨
age , U., Vogel, H.J.,
K¨
ogel-Knabne , I., 2019. Soil o ganic ca bon s o age as a key unc ion o soils - A
e iew o d i e s and indica o s a a ious scales. Geode ma 333, 149–162.
Wijewa dane, N.K., Ge, Y., Wills, S., Loecke, T., 2016. P edic ion o soil ca bon in he
con e minous Uni ed S a es: isible and nea in a ed e lec ance spec oscopy
analysis o he apid ca bon assessmen p ojec . Soil Sci. Soc. Am. J. 80, 973–982.
ˇ
Zíˇ
zala, D., Minaˇ
ík, R., Z´
ado o ´
a, T., 2019. Soil o ganic ca bon mapping using
mul ispec al emo e sensing da a: P edic ion abili y o da a wi h di e en spa ial
and spec al esolu ions. Remo e Sens. 11 (24), 2947.
Yos , J.L., Ha emink, A.E., 2019. Soil o ganic ca bon in sandy soils: A e iew. pp.
217–310.
V. Khos a i e al.
