Sma Ag icul u al Technology 10 (2025) 100812
A ailable online 1 Feb ua y 2025
2772-3755/© 2025 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/).
Spec oscopic analysis (UV-VIS-NIR) o p edic i e modeling o mac o and
mic onu ien s in g ape ine lea es
J.I. Manzano
a,*
, M. Rod íguez-Febe ei o
b
, M. Fandi˜
no
b
, M. Vilano a
a,c,*
, J.J. Cancela
b,c
a
Ins i u o de Ciencias de la Vid y del Vino (ICVV), Consejo Supe io de In es igaciones Cien í icas-CSIC, Uni e sidad de La Rioja, Gobie no de La Rioja, Finca la G aje a,
Ca e e a de Bu gos, Log o˜
no 26080, Spain
b
GI-1716, P oyec os y Plani icaci´
on, Depa amen o Ingenie ía Ag o o es al, Escola Poli ´
ecnica Supe io de Enxe˜
na ía, Uni e sidade de San iago de Compos ela, Rúa
Benigno Ledo s/n, Lugo 27002, Spain
c
C opQuali y: C op S esses and Thei E ec s on Quali y (USC), Associa e Uni o Ins i u o de Ciencias de la Vid y del Vino (ICVV-CSIC), Spain
ARTICLE INFO
Keywo ds:
Nu i ional diagnosis
Vine
PLS-R
Chemome ics
Spec oscopy
NIR
Mac onu ien s
Mic onu ien s
ABSTRACT
Assessing nu ien concen a ions in g ape ines is c ucial no only o he o e all physiology o he plan bu also
o he quali y o he esul ing wine. Accu a e de e mina ions a e also ele an o enhancing nu ien use e -
iciency and o mula ing e ilize ecommenda ions. Hence, he e is a conside able demand o a swi echnique
o analyze ine o gans. Di use e lec ance spec oscopy coupled wi h chemome ic me hods eme ges as a po en ,
cos -e ec i e, and en i onmen ally iendly analy ical echnique o de e mining nu ien concen a ions in
plan s. The objec i e o his s udy is o asce ain he iabili y o wide ange spec um (190–2600 nm) spec-
oscopy in p o iding p ecise insigh s in o he nu i ional s a us o ines. Ou in es iga ion speci ically a ge s on
he de e mina ion o C, N, P, K, Ca, Mg, B, Cu, Fe, Mn, Zn, Na, and Al in ine lea es om di e en wine g owing
a eas, a ie ies and ha es yea s. Pa ial Leas Squa es Reg ession (PLS-R) was employed o cons uc models o
he concen a ions o hese nu ien s based on he e lec ance measu emen s o he lea es. The model was ained
using 70 % o he samples, while he emaining 30 % cons i u ed he independen alida ion. Resul s om he
alida ion se indica ed accu a e alida ion o mos nu ien s, wi h de e mina ion coe icien s (
2
) o 0.70 o C,
0.72 o N, 0.64 o P, 0.75 o K, 0.84 o Ca, 0.48 o Mg, 0.45 o B, 0.58 o Cu, 0.26 o Fe, 0.82 o Mn, 0.50
o Zn, 0.90 o Na, and 0.69 o Al. The indings e ealed ha e lec ances in he isible (VIS) egion o he
spec um played a key ole in p edic ing mic onu ien s like B, co esponding wi h pho osyn he ic pigmen s
(chlo ophylls and ca o enoids). In con as , e lec ances in he nea -in a ed egion (NIR) had a g ea e impac on
mac onu ien p edic ion, pa icula ly o P and Mg, due o hei s onge in e ac ion wi h o ganic compounds.
The ul a iole (UV) ange played a mino ole, highligh ing he p edominan impo ance o he VIS-NIR egions
in spec oscopic analyses.
Finally, he esul s suppo he po en ial o his echnique o swi ly and non-in asi ely p edic ing bo h mac o
and mic onu ien le els in g ape ine plan s, and acili a e he e iliza ion planning using a ie y-speci ic
e e ence le els, o p ecision i icul u e adap ed o si e-speci ic demands, including spa ial in a-plo a iabili y.
1. In oduc ion
The g ow h and de elopmen o g ape ines a e hea ily in luenced by
bo h mac o and mic onu ien s. De iciencies o imbalances in hese
nu ien s can lead o s un ed g ow h, educed ui quali y, and
inc eased suscep ibili y o diseases [11]. The e o e, g owe s u ilize
e iliza ion o modi y he nu i ional s a us o he ines, seeking o
achie e he igh balance be ween ege a i e g ow h, yield, and c op
quali y [6]. Hence, unde s anding he cu en nu i ional s a us o
g ape ines is c ucial o c ea ing an e ec i e e iliza ion plan. Reducing
excessi e e ilize applica ion no only cu s cos s bu also enhances
quali y and educes he isk o con amina ion [1]. Mo eo e , he exac
applica ion o a ce ain nu ien , h ough a e iliza ion schedule
adap ed o he equi emen s o each phenological s age, acco ding o
age, oo s ock-cul i a combina ion, and nu ien applica ion sys em
( e iga ion, sp aying machine, e c.) equi es knowledge o he
* Co esponding au ho s a : Ins i u o de Ciencias de la Vid y del Vino (ICVV), Consejo Supe io de In es igaciones Cien í icas-CSIC, Uni e sidad de La Rioja,
Gobie no de La Rioja, Finca la G aje a, Ca e e a de Bu gos, Log o˜
no 26080, Spain.
E-mail add esses: [email p o ec ed] (J.I. Manzano), [email p o ec ed] (M. Vilano a).
Con en s lis s a ailable a ScienceDi ec
Sma Ag icul u al Technology
jou nal homepage: www.jou nals.else ie .com/sma -ag icul u al- echnology
h ps://doi.o g/10.1016/j.a ech.2025.100812
Recei ed 30 Oc obe 2024; Recei ed in e ised o m 30 Janua y 2025; Accep ed 31 Janua y 2025
Sma Ag icul u al Technology 10 (2025) 100812
2
nu i ional s a us quickly and e icien ly, in o de o implemen p eci-
sion i icul u e concep s [37].
The essen ial nu ien s a e six een and encompass ca bon (C), oxygen
(O), hyd ogen (H), ni ogen (N), phospho us (P), po assium (K), calcium
(Ca), magnesium (Mg), sul u (S), i on (Fe), manganese (Mn), zinc (Zn),
coppe (Cu), bo on (B), molybdenum (Mo), chlo ine (Cl), and nickel
(Ni). C, N, P, K, Ca, and Mg a e classi ied as mac onu ien s, essen ial in
la ge quan i ies, while Fe, Zn, Cu, B, Mn, Sodium (Na), Aluminium (Al)
and o he elemen s a e conside ed mic onu ien s, needed in smalle
amoun s, o he op imal g ow h and de elopmen o c op plan s [19].
C is no ypically ca ego ized as a adi ional nu ien o plan s in he
same sense as o he s like ni ogen, phospho us, and po assium, bu plays
a undamen al ole in hei g ow h and de elopmen . Plan s abso b
ca bon dioxide (CO
2
) om he a mosphe e du ing pho osyn hesis, u i-
lizing i o p oduce suga s and o he o ganic compounds necessa y o
hei me abolism and s uc u e. These o ganic compounds se e as en-
e gy sou ces o a ious physiological p ocesses and as building cellula
componen s such as cellulose, p o eins, and lipids [8]. N s ands ou as
he p ima y nu ien signi ican ly in luencing he ege a i e de elop-
men o plan s; i se es as a pi o al elemen o a ious physiological
p ocesses. De iciency in N mani es s as educed lea size and e iden
yellowing [4], whe e a sus ainable ni ogen e iliza ion plan is equi ed
depending on p oduc ion objec i es: yield o quali y [37]. P is indis-
pensable o he o ma ion o c ucial mac omolecules like nucleic acids,
phospholipids, and suga phospha es u ilized by plan s o he de el-
opmen o di e en o gans [28]. K is ele an in ag onomic p oduc ion,
being essen ial o enzyma ic eac ions, main enance o osmo ic po en-
ial, and acili a ion o wa e up ake du ing plan ma u a ion. A he
ineya d le el, po assium plays a c ucial ole in egula ing key physio-
logical mechanisms, pa icula ly he anspi a ion p ocess and he
opening and closing o s oma a [33]. In wine, howe e , excessi e po-
assium le els a e common and esul in dec eased acidi y and sho e
shel li e, leading o accele a ed oxida ion and uns able colo a ion [13].
Ca is mainly ela ed wi h oo g ow h and de elopmen , acili a ing
nu ien and wa e abso p ion om he soil. Ini ial signs o Ca de iciency
appea on young lea es, exhibi ing dis inc de o ma ions and chlo osis.
Mg p ima ily se es as a cons i uen o chlo ophyll, cell walls, and play a
i al ole in P ansloca ion and N assimila ion. Mg de iciency is ypi ied
by chlo osis along lea ma gins [1]. Rega ding mic onu ien s, Fe is
essen ial o plan playing a c ucial ole in a ious me abolic p ocesses,
pa icula ly in chlo ophyll syn hesis, pho osyn hesis, and ni ogen ix-
a ion. I on is a componen o se e al enzymes in ol ed in hese p o-
cesses, including ca alase, pe oxidase, and e edoxin. Mn con ibu e o
pe o m pho osyn hesis and in de ense agains oxida i e s ess. B is
in ol ed in cell wall o ma ion, ca bohyd a e me abolism, nucleic acid
syn hesis and ho mone egula ion. The de iciency can lead o symp oms
such as s un ed g ow h, dis o ed lea es, and poo lowe and ui
de elopmen , while excess o B can esul in oxici y symp oms like lea
bu n and nec osis. Zn is pa o chlo ophyll molecule and in consequence
is ela ed wi h he pho osyn he ic p ocess. Zn de iciency can lead o
symp oms such as s un ed g ow h, in e einal chlo osis, and educed
yields. Cu is in ol ed in ca bohyd a e me abolism and pho osyn hesis
[39]. Na is no conside ed an essen ial nu ien o mos plan s, bu
many plan s a e sensible o high concen a ion o his compound. The
excessi e le els o sodium dis up s he osmo ic balance wi hin plan
cells. This dis up ion can inhibi wa e up ake by plan oo s and lead o
dehyd a ion and wil ing. Addi ionally, high sodium le els can in e e e
wi h he up ake o essen ial nu ien s by plan s, u he exace ba ing
nu ien de iciencies [35]. Al is a complex elemen wi h bo h posi i e
and nega i e impac s on plan g ow h and de elopmen . I s e ec s
depend on ac o s such as soil pH, aluminium concen a ion, and he
speci ic plan species in ol ed. In ag icul u e, managing soil pH and
aluminium le els h ough p ac ices such as liming and soil amendmen
can help mi iga e aluminium oxici y and op imize plan g ow h [24].
U ilizing ul a iole (UV), isible (VIS), and nea -in a ed (NIR)
spec oscopy oge he wi h chemome ic me hods o e s a p omising
solu ion o apid and eliable de e mina ions o di e se physicochem-
ical pa ame e s. Spec oscopy equi es minimal sample p epa a ion,
ypically only necessi a ing d ying and g inding o mi iga e wa e ’s in-
luence on spec al abso bance.
NIR abso bance spec a o e insigh s in o he o ganic ma ix
h ough he de ec ion o C–H, O–H, and N–H ib a ional modes. While
mac o- and mic omine als lack hese speci ic ib a ional modes, hey
a e embedded wi hin he o ganic ma ix and in e ac wi h one o he
men ioned o ganic unc ional g oups. Consequen ly, hei concen a-
ions can be indi ec ly assessed h ough NIR spec a analysis [9]. The
calib a ion models exhibi ed g ea e accu acy o mac onu ien s,
a ibu ed o hei highe concen a ions. This ou come migh be sup-
po ed by he ac ha ele a ed nu ien le els in plan issue ypically
esul in a mo e p onounced signal- o-noise a io, he eby acili a ing
be e in e p e a ion o he esponse h ough chemome ic ools. Addi-
ionally, s onge co ela ions wi h o ganic compounds could con ibu e
o his phenomenon [16].
Mul iple esea ch endea ou s ha e alida ed spec oscopy’s capa-
bili y o de e mine nu ien concen a ions ac oss a ious c ops.
Ne e heless, he e a e s ill ela i ely ew s udies speci ically cen ed on
ine lea es o s ems. In his way, he le el o di e en nu ien s like P, K,
Ca, Mg, Mn, Fe, Cu, Zn, and B we e desc ibed in di e en ine o gans
(lea blades, pe ioles and be ies) using p edic i e models based in NIR
[11]. Thus, es ima ion o soil nu ien s (N), o ganic ma e , and clay in
ineya d pe o ming Pa ial Leas Squa es Reg ession (PLS-R) and
andom o es models wi h di e en spec um anges and di e en
p e ea men s was p e iously desc ibed [29]. Rega ding wine and mus ,
he e a e nume ous s udies on he use o NIR spec oscopy o he
analysis o a ious quali y- ela ed oenological pa ame e s, such as
alcohol con en , pH, ola ile acids, o ganic acids and educing suga s in
g apes and wine [9,18,27,36]. In o he ui ees, comple e nu i ional
diagnosis in ci us lea es was desc ibed in [1] s ablishing accu a e
models o he de e mina ion o P, K and B, al hough he lis o nu ien s
s udied included N, Ca, Mg, S, Fe, Cu, Mn, and Zn. Simila s udies we e
ealized in di e en species o ci us lea es de e mining N, K, Ca, Mg, B,
Fe, Cu, Mn, and Zn [2,14,23]. Fu he mo e, i has been desc ibed in [16]
ha in eg a ing NIR and mid-in a ed (MIR) spec a exhibi ed p omising
esul s o assessing bo h mac onu ien (N, P, K, Ca, Mg, and S) and
mic onu ien (Na, Fe, Mn, B, Cu, Mo, and Zn) concen a ions in ice
plan s. Finally, and con e sely o he heo y o he lack o speci ic
spec al signa u es o mic onu ien de e mina ions, some s udies ha e
demons a ed eliable p edic ion accu acy in es ima ing speci ic
mic onu ien s (Na, B, Al, Mn, Cu, and Zn) in ce ain plan species using
NIR spec oscopy [7,30].
The Eu opean Union (EU) is dedica ed o ad ancing sus ainable
ag icul u e and aims o educe e ilize usage by 2030, a key objec i e
ou lined in i s Eu opean G een Deal and Fa m o Fo k s a egy (F2F,
2020. Fa m o Fo k S a egy. Web add ess: h ps://ec.eu opa.eu/ ood/ a
m2 o k_en). To accomplish his objec i e, i is c ucial o de elop ools
and eliable p edic i e models o nu ien assessmen , which a e
essen ial o o mula ing in o med and sus ainable e iliza ion s a e-
gies. In his s udy, we success ully desc ibed he use o PLS-R gene a ed
models employing da a ob ained om spec oscopy o e a wide spec-
um ange, de e mining he nu i ional s a us o g ape ine plan s om
lea blades samples.
2. Ma e ials and me hods
2.1. Expe imen al design
The plo s used o his s udy we e loca ed in di e en wine g owing
a eas in Spain Galicia: Rías Baixas (Val do Ulla; Val do Saln´
es; O Rosal),
Ribei a Sac a (San Vic o io; O Ve xel; Alais; A E mida; San a Cubicia;
Qui oga), Ribei o (Lei o), b) Cas illa La Mancha: Albace e (ITAP), c)
Ex emadu a: Badajoz (CICYTEX), and d) Cas illa-Le´
on: Valladolid
(ITACYL), du ing he seasons o 2020, 2021 and 2022 (Fig. S1). The
J.I. Manzano e al.
Sma Ag icul u al Technology 10 (2025) 100812
3
s udy was ca ied ou on 339 lea samples om g ape ines o di e en
a ie ies: Alba i˜
no, T eixadu a, Lou ei a, Godello, B anco lexí imo,
Mencía, Cas a˜
nal, Sous´
on, Sy ah, Ga nacha and Temp anillo, du ing he
e aison s age o each a ie y and season (BBCH: 83–85). Speci ically,
he da ase we e spli in o a calib a ion se (70 %, co esponding o 234
samples) and a alida ion se (30 %, co esponding o 105 samples). The
dis ibu ion o whi e and ed a ie ies was ep esen a i e in bo h se s.
The calib a ion se included 91 whi e samples and 153 ed samples,
while he alida ion se included 40 whi e samples and 65 ed samples.
This p opo ional dis ibu ion e lec s he o e all composi ion o he
da ase , ensu ing ha he di ision is bo h ep esen a i e and unbiased.
20 lea es we e andomly aken om each sampling poin in each
plo unde s udy. Lea es we e selec ed om ui ing shoo s o medium
igo , loca ed on he opposi e side o he second clus e , on da es co -
esponding o he e aison s age o he ineya d. Each sample was
p ocess in i s ins ance in he nea es esea ch acili ies, and anspo
hem o he USC labo a o y o chemical de e mina ions.
2.2. Chemical de e mina ions on he lea es
Lea blades and pe ioles we e sepa a ed, washed wi h ap wa e and
insed wi h dis illed wa e , and inally o en-d ied (D y Big, J.P. Selec a,
Ba celona, Spain) a 70 ◦C o 48 h o ambien d ying, o cons an
weigh . Then hey we e g ound wi h a disc mill in o de o pass h ough
a 2-mm mesh and inally s o ed a oom empe a u e o be analyzed. N
concen a ion was de e mined by oxidizing he sample and hen quan-
i ying he gas p oduced du ing his combus ion using a he mal con-
duc i i y de ec o (T uSpec CHNS, Leco, S . Joseph, MI, USA) [17]. Fo
chemical analysis o he o he nu ien s, 1 g o sample was calcined a
500 ◦C o 8 h and subsequen ly we -diges ed wi h 1 mL o deionized
wa e and 5 mL 2 M HCl. N, C, Al, P, K, Ca, Na, Mg, Fe, Mn, Cu, Zn, and B
concen a ions we e de e mined wi h induc i ely coupled
plasma-op ical emission spec oscopy (ICP-OES) (Op ima 4300DV,
Pe kinElme , No walk, CT, USA), o e alua e mac oelemen s ICP
me hodology p esen a s anda d e o be ween 0.5 % and 5 % o he
mean alue, meanwhile, o ace le els o mic oelemen s, he s anda d
e o anges be ween 1 % and 10 % depending on he concen a ion and
ma ix in e e ences. Deionized wa e was used o all dilu ions. Con-
cen a ions we e exp essed in e ms o d y weigh (mg/kg) and/o
pe cen age (%).
2.3. Chemome ic analysis
The lea samples we e scanned in e lec ance mode using a spec-
ome e (JASCO V-770, JASCO Co po a ion, Tokyo, Japan) wi h a
coupled sphe e uni (model ISN-923), and wo ligh sou ces (deu e ium
and halogen). The spec a we e collec ed in con inuous using Spec a
Manage so wa e o JASCO. The spec al cap u e co e ed he ange
om 190 o 2600 nm, encompassing he ul a iole (UV, 190–380 nm),
isible (VIS, 381–780 nm), and nea -in a ed (NIR, 781–2600 nm) e-
gions o he elec omagne ic spec um. The scanning speed was 1000
nm/min.
2.4. Spec al p e- ea men analysis
The h ee spec a ob ained om each olia sample we e a e aged o
ob ain a single alue and co ela ed wi h he nu ien concen a ions
de e mined by ICP-OES analysis. Di e en p e- ea men s we e applied
o he spec a: (i) S anda d No mal Va ia e (SNV), o educing he
dispe sion [3], (ii) Mean cen e (MC) in ol es cen e ing each a iable
by sub ac ing he mean o all i s elemen s om each elemen in he
a iable, (iii) Sa i zky-Golay i s de i a i e (FD) was u ilized o emo e
cons an baseline o se s and o se s ha exhibi linea a ia ion wi h
wa eleng h [1], (i ) Spec oscopic (SP) con e ing he e lec ance uni s
o loga i hm o 1/ e lec ance, and ( ) De ending in ol es emo ing
unwan ed ends, such as baseline shi s and linea a ia ions, o high-
ligh ele an spec al ea u es and enhance he accu acy o chemical o
quali a i e analysis [3]. A e an explo a o y s a is ical analysis, ou lie s
we e de ec ed and elimina ed o he s udy (4.98 % o o al samples; 12
samples in calib a ion se , and 5 samples in alida ion se ). Di e en
models, using ei he a single p e- ea men o a combina ion o se e al,
we e ained and es ed. T ans o ma ions we e calcula ed using he
Fig. 1. Flowcha o he me hodology used.
J.I. Manzano e al.
Sma Ag icul u al Technology 10 (2025) 100812
4
so wa e Unsc amble ® XI (CAMO So wa e Inc., Woodb idge, No way)
[5].
2.5. S a is ical analyses
Fo he eg ession analysis, we pe o med a PLS- eg ession (PLS-R)
o each gene a ed model. PLS-R is a common echnique o es ablish a
co ela ion be ween sample spec a and he a iables o in e es [29]. In
PLS-R, one o he objec i es is o de e mine he numbe o componen s
(called la en a iables in PLS) ha explain he maximum a ia ion in
he da a [40].
A e elimina ing ou lie s, 217 samples we e included in he cali-
b a ion se and 100 samples in he alida ion se o de eloping he PLS-
R models. Addi ionally, a Lea e-One-Ou c oss- alida ion (CV) was
pe o med wi h he calib a ion se in o de o de e mine he accu a e
numbe o componen s ha balances he p edic i e model, among o he
pa ame e s. The model wi h he lowes oo -mean-squa e e o (RMSE)
and he highes coe icien o de e mina ion (
2
) in he es o he model
was selec ed. Addi ionally, we e alua e bias and slope o he PLS-R
models. Bias e e s o he sys ema ic di e ence be ween p edic ed and
ac ual alues, whe e a bias nea ze o sugges s ha p edic ions a e, on
a e age, nei he o e es ima ed no unde es ima ed. Slope desc ibes he
ela ionship be ween p edic ed and ac ual alues, wi h a slope close o 1
indica ing s ong ag eemen be ween hem. Fu he mo e, he weigh ed
eg ession coe icien s (BW) o he PLS-R model we e employed o
iden i y he wa eleng hs c ucial o p edic ing nu ien le els. This
app oach e alua es he ela ionship be ween each wa eleng h and he
concen a ion o he elemen being analyzed, highligh ing wa eleng hs
wi h highe absolu e BW coe icien alues as he mos in luen ial in he
model [22]. Flowcha o he me hodology used is ep esen ed in Fig. 1.
3. Resul s and discussion
3.1. Desc ip i e s a is ics o he olia nu ien concen a ions
Table 1 p esen s he desc ip i e s a is ical da a ega ding he con-
cen a ions o olia nu ien s in he lea es collec ed om ineya ds, as
de e mined h ough classical p ocedu es. This able includes he alues
obse ed ac oss all samples, as well as hose o he calib a ion and
alida ion se s sepa a ely.
The minimum and maximum alues o he alida ion se samples lie
wi hin he ange o he calib a ion se ’s ex eme alues. This ensu es he
model can be e i ied unde app op ia e condi ions.
The samples collec ed in he 2021 season om he Lei o egion,
co esponding o he Sous´
on, Mencía, and T eixadu a a ie ies (da a no
shown), exhibi ed e lec ance spec a wi h a pe cen age o e lec ance
no ably di e en (highe o lowe ) compa ed o he mean o he
emaining samples. This di e ence could be a ibu ed o signi ican
a ia ions in he concen a ions o Cu (47.93 mg/kg in he o e all
da ase s. 178.50 mg/kg in hese samples), Mn (146.07 mg/kg s. 78.09
mg/kg), Zn (18.47 mg/kg s. 48.85 mg/kg), and Na (129.78 mg/kg s.
11.24 mg/kg). These elemen s, which a e p esen in no ably di e en
concen a ions in hese samples, a e likely esponsible o he obse ed
spec al di e ences.
3.2. P incipal componen analysis: explo a o y analysis
Fig. 2 shows he p ojec ion o he spec al da a om he 341 lea
samples in o he i s wo p incipal componen s, which explain 57 % and
26 %, espec i ely, o he o al a iabili y be ween samples. The dis i-
bu ion o he samples chosen o ain he model (calib a ion se ) and
hose selec ed o alida ion ( alida ion se ) was simila , as can be
obse ed in Fig. 2. This is impo an when cons uc ing a model ha
add esses as many obse a ions as possible.
Table 1
Basic s a is ics o olia nu ien concen a ions ob ained h ough chemome ic analysis. Mac onu ien s (C, N, P, K, Ca, and Mg) a e exp essed as pe cen ages (%), while mic onu ien s (B, Cu, Fe, Mn, Zn, Na, and Al) a e
exp essed in millig ams pe kilog am (mg/kg).
Nu ien s To al sample se (n ¼339) Calib a ion se (n ¼234) Valida ion se (n ¼105)
Mean Max Min SD Median Mean Max Min SD Median Mean Max Min SD Median
Ca bon (C) 45.42 49.92 39.14 2.04 45.87 45.25 49.28 39.14 2.08 45.99 45.79 49.92 40.73 1.96 45.83
Ni ogen (N) 1.98 3.14 0.57 0.44 1.96 1.95 3.05 0.57 0.37 1.95 2.04 3.13 0.58 0.57 1.98
Phospho us (P) 0.17 0.45 0.01 0.06 0.16 0.17 0.45 0.01 0.06 0.16 0.17 0.34 0.02 0.05 0.17
Po assium (K) 0.97 4.24 0.22 0.57 0.85 0.89 4.24 0.22 0.55 0.80 1.17 4.11 0.53 0.57 1.02
Calcium (Ca) 1.88 4.65 0.45 0.70 1.88 1.97 4.65 0.45 0.72 1.92 1.71 3.07 0.57 0.64 1.73
Magnesium (Mg) 0.26 1.44 0.05 0.13 0.25 0.28 1.44 0.05 0.14 0.26 0.23 0.43 0.08 0.07 0.24
Bo on (B) 15.15 51.60 2.08 7.69 13.65 14.98 51.6 2.08 8.28 13.08 15.45 34.62 6.52 6.43 14.24
Coppe (Cu) 47.93 369.54 1.16 50.18 43.59 46.74 272.66 1.16 40.16 48.04 48.94 369.54 1.64 68.87 12.83
I on (Fe) 47.45 311.20 1.27 30.74 41.47 46.29 311.2 1.27 31.26 40.98 50.56 192.16 2.54 30.46 42.60
Manganese (Mn) 146.07 2863.59 4.36 323.51 79.64 119.15 2508.50 4.36 285.74 72.70 215.57 2863.59 12.88 398.81 91.40
Zinc (Zn) 18.47 92.21 0.96 17.29 9.64 18.34 92.21 0.96 15.99 10.92 16.90 88.07 2.18 19.05 6.32
Sodium (Na) 129.78 367.94 3.52 125.95 89.66 131.34 357.20 3.52 125.74 101.20 118.15 367.94 5.53 126.52 72.30
Aluminum (Al) 97.27 4848.61 0.08 374.68 48.08 100.01 4848.61 0.08 436.00 45.30 92.15 4617.90 4.80 190.26 57.98
SD: s anda d de ia ion.
J.I. Manzano e al.
Sma Ag icul u al Technology 10 (2025) 100812
5
3.3. PLS-R analysis
Fig. 3 shows he aw e lec ance spec a o he 341 lea samples. The
di e en p e- ea men s s udied we e applied o hese spec a and sub-
sequen ly used o gene a e p edic i e models using PLS-R.
Table 2 p esen s he p edic i e esul s o each elemen using PLS-R
wi h he op imal spec a p e- ea men , while he ela ionship be ween
he p edic ed and e e ence da a o alida ion se wi h an
2
>0.60 is
p esen ed in Fig. 4.
The
2
alues o he mac onu ien s (C, N, P, K, Ca and Mg) p e-
dic ion model anged om 0.48 o 0.84, wi h Ca ha ing he highes
2
=
0.84 in he alida ion se . Fo he mic onu ien s (B, Cu, Fe, Mn, Zn, Na,
and Al), he
2
alues anged om 0.45 o 0.90, wi h Na ha ing he bes
model wi h a
2
=0.90). Di e en p e- ea men s we e applied o he
spec um, as desc ibed abo e. The e was no an ideal p e- ea men o
he de e mina ion o all nu ien s s udied. Thus, he bes p e- ea men
u ned ou o be he use o he i s de i a i e, ei he alone o in com-
bina ion wi h MC o SNV. All PLS-R models gene a ed a e p esen ed in
Supplemen a y Ma e ial. In ineya d managemen , models wi h ² ≥0.7
p o ide p edic ions wi h su icien eliabili y o guide nu ien man-
agemen decisions, such as adjus ing e iliza ion s a egies o add ess-
ing de iciencies.
Fig. 2. PCA o aw spec a o he i s wo p incipal componen s om he lea blades samples. Blue poin s co espond o he samples om he calib a ion se , while
he ed poin s ep esen he samples om he alida ion se .
Fig. 3. Re lec ance spec a o all he ine lea es analyzed. The samples co esponding o ed g ape a ie ies a e shown in pu ple, while hose o whi e g ape
a ie ies a e shown in blue.
J.I. Manzano e al.
Sma Ag icul u al Technology 10 (2025) 100812
6
In his wo k, he
2
alue o he de e mina ion o C was 0.70 (0.83
o calib a ion se ). Fo C, p e ious s udies [12] achie ed an
2
=0.44 in
ine lea es. In a simila way, i has been desc ibed in soil nu ien s o he
ineya d an
2
=0.84 o he o ganic ma e de e mina ion [29]. N is
essen ial o ege a i e g ow h and chlo ophyll syn hesis, while P sup-
po s oo de elopmen and ene gy ans e [21]. The PLS-R model
de eloped o p edic N concen a ion assessed an
2
=0.83 and lowes
oo -mean-squa e e o p edic ion (RMSEP) =0.55. In p io esea ch on
g ape ine lea es, a model wi h a high alue o
2
(0.95) was de e mined,
when combining all samples in he same da abase (lea es blades and
pe ioles, and g ape be ies) [12]. In his way, in ci us lea es, a o able
esul s we e ob ained in se e al s udies wi h VIS-NIR spec oscopy
showing an
2
=0.78 in [2] and 0.91 (RMSE =1.06) a e e iliza ion
wi h a ious N doses [26]. In o he c ops, such as oli es, i has been
epo ed models wi h an
2
=0.91 using only he sho -wa e in a ed
pa o he spec um [31]. Finally, simila esul s o hose p esen ed in
his s udy conce ning N de e mina ion we e desc ibed in [25,38].
P was p edic ed wi h an
2
=0.64 and an RMSE =0.02. In ano he
s udy p esen ed by [11] in ine lea es, P was de e mined wi h a a io o
pe o mance o de ia ion (RPD) =1.02, bu wi h an
2
=0.77 using he
aw spec um o NIR in be ies. On ci us lea es, simila
2
we e
desc ibed anging 0.75–0.77 wi h low RMSE [2,26]. K pa icipa es in
enzyme ac i a ion and imp o es ui quali y and esis ance o se e al
diseases. The model gene a ed o K p edic ed his elemen wi h an
2
o
he alida ion se =0.75 and an RMSE =0.39. Using jus he NIR ange
o he spec um, an
2
=0.42 was epo ed in [11] o lea blades, and
0.76 and 0.79 in be ies and pe ioles, espec i ely. In hei s udy, Oli-
ei a and San ana [25] ob ained an ² =0.76 and an RMSE o 1.30 while
analyzing eucalyp us lea es.
The PLS-R model de eloped o p edic Ca showed an
2
=0.84 and a
RMSE o 0.35. Compa able esul s ha e been ound o lea blades wi h
and
2
=0.88, bu wi h highe alue o RMSE and lowe se size (n =
159) [11]. In his way, di e en models ob ained wi h sample lea es
om di e en c ops such as ci us ees, o eucalyp us, p edic ed Ca
concen a ion wi h
2
anging om 0.62 o 0.81 [2,25,26]. Rega ding
mic onu ien s (B, Cu, Fe, Mg, Mn, and Zn), Mg wasde e mined as well
as Fe wi h an
2
=0.48 and 0.26, espec i ely. O he esea che s we e
able o p edic he magnesium con en o lea blades wi h be e accu-
acy in ine ( ² =0.60) [11], and in suga cane, achie ing an ² =0.97
and an RMSEP =0.005, using mul iple linea eg ession and PLS-R in
he spec al ange o 780–2500 nm [41]. Conce ning Fe, he p edic i e
models a e poo in gene al in he li e a u e, being unable o p edic
success ully in he mos o he c ops analysed. Only in g ape ine lea
blades was obse ed a mode a e co ela ion wi h an
2
=0.58 in hese
o gans, which could be used o classi ica ion (excess, de iciency,…)
[11]. The p edic ion o B was de e mined wi h an
2
=0.45 and an
RMSEP =4.60, su p isingly he
2
in he calib a ion se was 0.62. This
model pe o med was be e in ine lea es samples han ano he
epo ed by [11] wi h a RPD =1.18. The bes model o B was p esen ed
in eucalyp us wi h a high le el o es ima ion wi h an
2
=0.83 and an
RMSEP =8.59 [25]. Mode a e p edic ion model was gene a ed o Cu
de e mina ion wi h an
2
=0.58 and an RMSEP =56.20. Simila esul s
we e ob aining by [11] using NIR spec oscopy in g ape ine lea es
eaching an
2
=0.61 in lea blades. The Mn concen a ion was p edic ed
wi h an
2
=0.58 and an RMSEP =56.20. Simila ou comes we e ach-
ie ed by [11]. Howe e , his elemen was p edic ed in ci us lea es (
2
=
0.69, RMSEP =40.75) [2]. The PLS-R model o Zn de e mina ion
p edic ed his elemen wi h an
2
=0.70 in he calib a ion se and an
2
=0.50 o he alida ion se wi h an RMSE =19.46 in he lea samples.
In a p e ious s udy, Zn was de e mined (
2
=0.82) in g ape be ies bu
wo se in lea blades (
2
=0.29) [11]. Conce ning Na, he e is no
nume ous s udies o p edic ion by spec oscopy o his elemen .
Fu he mo e, i has been desc ibed an
2
=0.83 in he p edic ion o Na in
legume plan s [10]. The bes PLS-R model p esen ed in his s udy is
p ecisely he one ela ed o he p edic ion o sodium (
2
=0.90; RMSE =
50.74). Finally, he model c ea ed o Al p edic ed his elemen qui e
well, achie ing a high ² alue =0.69. Ano he s udy in whi e pine and
ed oak using VIS-NIR e lec ance p edic ed his elemen wi h an
2
=
0.82 [15].
The esul s p esen ed demons a e good p edic i e pe o mance o
he elemen s analyzed, wi h ni ogen (N) and manganese (Mn) showing
ela i ely s ong esul s. Howe e , Fig. 3 e eals some a ia ion in he
p edic ions, especially a lowe concen a ions, whe e he poin s show
less con inui y along he line. This beha io is also obse ed o po as-
sium (K) and phospho us (P), which, despi e high coe icien s o de e -
mina ion, exhibi some p edic ion a iabili y, pa icula ly a lowe
alues. These indings sugges ha while ou models a e obus o
ce ain concen a ion anges, he e is oom o imp o emen , especially
a he ex emes. Fu u e wo k will ocus on e ining hese models,
explo ing al e na i e modelling echniques, and adjus ing he spec al
ange and p e- ea men s o enhance p edic ion accu acy ac oss all
concen a ion le els.
3.4. Analysis o signi ican wa eleng hs in he cons uc ion o p edic i e
models
Reg ession coe icien along he spec a, associa ed wi h he mos
accu a e p edic i e models o all he nu ien s s udied a e ep esen ed
in he Fig. 5. In his way, he mos ele an wa eleng hs a e indica ed in
Table 3.
In lea es, he pho osyn he ic pigmen s (chlo ophylls and ca o en-
oids) abso b abou 90 % o mo e o he ligh si ua ed in he isible egion
(351–700 nm). Con e sely, he NIR lacks molecules wi h s ong ab-
so p ion, leading o plan s e ac ing o ansmi ing almos all adia ion
in his ange, e aining only abou 10 % [32]. These essen ial pho o-
syn he ic pigmen s in lea es play a c i ical ole in plan pho osyn hesis,
Table 2
Resul s o calib a ion, c oss- alida ion, and alida ion se using PLS-R o he bes spec oscopic models.
Nu ien Symbol P e-T ea men Componen s Calib a ion Lea e-One-Ou C oss-Valida ion Valida ion Se
2
RMSE
2
RMSE
2
RMSE Bias Slope
C MC+FD 7 0.83 0.85 0.66 1.21 0.70 1.54 −0.62 0.68
N SNV 7 0.72 0.30 0.55 0.38 0.83 0.55 −0.16 0.75
P Raw 7 0.47 0.04 0.38 0.04 0.64 0.02 0.02 0.81
K FD 7 0.68 0.31 0.56 0.47 0.75 0.39 0.03 0.60
Ca Raw 7 0.47 0.52 0.41 0.55 0.84 0.35 0.07 0.78
Mg SNV 7 0.35 0.08 0.26 0.08 0.54 0.07 0.04 0.74
B MC+FD 6 0.62 4.99 0.43 6.19 0.45 4.60 0.71 0.45
Cu Raw 6 0.52 27.66 0.43 30.18 0.58 56.20 0.86 0.28
Fe FD 2 0.26 10.08 0.14 11.59 0.41 12.89 −3.36 0.25
Mn FD+SNV 7 0.80 126.86 0.54 229.00 0.82 229.79 0.38 0.69
Zn FD 5 0.70 9.02 0.54 11.16 0.50 19.46 0.90 0.39
Na FD+SNV 7 0.82 53.74 0.51 88.46 0.90 50.74 0.30 0.77
Al SNV 7 0.54 14.72 0.47 15.85 0.69 18.29 0.20 0.50
RMSE: Roo mean squa e e o ; MC: Mean cen e ; FD: i s de i a i e; SNV: s anda d no mal a ia e.
J.I. Manzano e al.
Sma Ag icul u al Technology 10 (2025) 100812
7
Fig. 4. Rela ionship be ween p edic ed and e e ence alues o he bes PLS-R models (
2
>0.6) in he alida ion se .
J.I. Manzano e al.
Sma Ag icul u al Technology 10 (2025) 100812
8
con e ing sunligh and CO
2
in o suga s. Chlo ophylls a e di ided in o
chlo ophyll a (Chl-a) and chlo ophyll b (Chl-b), esponsible o he
cha ac e is ic g een colo o lea es, wi h abso p ion peaks a ound 450
and 680 nm. Ca o enoids a e di ided in o ca o ene a, ca o ene b and
xan hophylls, and exhibi s ong ligh abso p ion in he blue egion o
he spec um (450–500 nm). Almos all he elemen s de e mined in his
s udy a e ei he di ec ly o indi ec ly ela ed o chlo ophyll and he
pho osyn he ic p ocess. In his way, he isible pa o he spec um is
ele an ega ding he con ibu ion o he gene a ion o he p oposed
models (190 – 2600 nm) wi h 61 e lec ance peaks p o iding in o ma-
ion o he di e en models, especially o mic onu ien s models wi h
(32 main peaks in VIS egion) (Fig. 4). In he case o B, he isible
spec um is he only pa aken in o accoun o he c ea ion o he PLS-R
model, while he es o elemen s use a leas wo- ange a ea (UV, VIS
and NIR). On he con a y, wi hin he NIR egion o he cha ac e is ic
plan e lec ance spec um, compounds such as p o eins, a y acids, and
s a ch exhibi p ominen abso p ion ea u es be ween 800 and 1000 nm
[34]. Meanwhile, cellulose and lignin display dis inc i e abso p ion
peaks a ound 1250–1800 nm, while ni a e abso p ion becomes no able
a 1600 nm, and suga con en is disce nible a 2300 nm [20]. The
con ibu ion o NIR ange o he spec um was ele an con ibu ing
wi h 52 selec ed peaks in he de e mina ion o he elemen s s udied, and
being mos impo an o he de e mina ion o mac onu ien s wi h 29
peaks, p edic ably due o he p esence in highe concen a ions in
plan s. In his way, he ela ionship wi h he o ganic compounds is
highe han mic onu ien s, enhancing signal- o-noise a io, and acili-
a ing he in e p e a ion by PLS-R models. The elemen s mo e in lu-
enced by he NIR egion o he spec um we e P (7 peaks), Mg (6 peaks),
Cu (6 peaks), and Ca (5 peaks). The mos complex PLS-R model was
c ea ed o C elemen wi h 14 ele an wa eleng hs (Table 3). This is
consis en because i is he elemen in he highes concen a ion and he
one ha gene a es he g ea es in e ac ion wi h o ganic ma e , and i is
Fig. 5. Weigh eg ession coe icien s om he PLS-R models o all he nu ien .
J.I. Manzano e al.
Sma Ag icul u al Technology 10 (2025) 100812
9
also an elemen di ec ly in ol ed in he ib a ion o C
–
H bonds. Finally,
he con ibu ion o he UV ange in he de e mina ion o nu ien s ends
up being ma ginal, wi h peaks in he p edic ion o P (2), K (1), Mg (1),
and Mn (1). Thus, Mn is he unique mic onu ien wi h a ele an
wa eleng h in he UV egion (371 nm).
4. Conclusions
In he ealm o nu ien analysis, he e’s a g owing demand o an
e icien and economically iable echnique ha minimizes sample
handling, educes eagen s, and allows eal- ime assessmen on p o-
duc ion lines.
This s udy highligh s he p omising po en ial o UV–VIS-NIR spec-
oscopy combined wi h PLS-R models in p edic ing mac o- and
mic onu ien concen a ions in g ape ine lea es wi h conside able
accu acy. The models achie ed high coe icien s o de e mina ion o
key elemen s such as calcium (Ca) and sodium (Na), showcasing he
obus ness o hese echniques o suppo e iliza ion and nu ien
managemen decisions in ineya ds. Howe e , he a iabili y in model
pe o mance ac oss di e en nu ien s also e lec s he complexi y o
nu ien in e ac ions in plan s. In his way, ionomics, he comp ehensi e
s udy o he elemen al composi ion o li ing o ganisms, o e s p omising
ad ancemen s in he ield o plan science and has he po en ial o
e olu ionize ag icul u al p ac ices.
The abili y o p edic a wide ange o nu ien s wi h ² alues ≥0.7
o se e al key elemen s ensu es hese models can se e as eliable ools
o ineya d nu ien moni o ing and decision-making. Beyond p ac ical
u ili y, he app oach exempli ies he po en ial o spec oscopy o
con ibu e o he b oade goals o p ecision ag icul u e, including
op imized esou ce use and enhanced c op quali y. On he o he hand, a
key insigh om his s udy is he di e en ial con ibu ion o spec al
egions o p edic i e models. The isible (VIS) egion was key o p e-
dic ing mic onu ien s like bo on (B), linked o pho osyn he ic pig-
men s. In con as , he nea -in a ed (NIR) egion was mo e in luen ial
o mac onu ien s like phospho us (P) and magnesium (Mg), due o
hei associa ion wi h o ganic compounds and plan s uc u es. The ul-
a iole (UV) ange had a mino ole, emphasizing he impo ance o
VIS-NIR egions in spec oscopic analyses.
The indings achie ed ein o ces he use o UV–VIS-NIR spec oscopy
o nu i ional diagnosis in ineya ds. This me hod signi ican ly educes
he cos s associa ed wi h adi ional chemical analyses. Addi ionally,
spec oscopy echnologies acili a es eal- ime decisions, allowing
ineya d manage s o add ess nu i ional de iciencies p omp ly, op i-
mize e ilize use, and ul ima ely imp o e c op yield and quali y.
Equally impo an is he con ibu ion o in e na ional da abases o
he gene a ed models, en iching he scien i ic communi y’s knowledge
base and encou aging in e disciplina y esea ch.
Fu u e wo ks should be ocus in he ansla ing his echnology o he
ineya d in o de o imp o e he nu ien use e iciency and quali y o
c op plan s. Addi ionally, explo ing new non-linea models, such as
a i icial neu al ne wo ks, suppo ec o machines, o a i icial in elli-
gence, should be conside ed as al e na i es o enhance he ob ained
esul s, especially o hose elemen s wi h wo se esul s employing
linea models (Mg, Fe, and B). In esponse o sugges ions ega ding
p ac ical applica ion, de eloping a po able de ice along wi h a s an-
da dized p o ocol o ineya d g owe s will also be a key aspec o u u e
wo k. Finally, u u e s udies will ocus on explo ing na owe spec al
anges o imp o e he pe o mance and accu acy o he models o
speci ic nu ien s.
Funding
IRRIVITIS-PID2019–105039RR-C44 (Funding by MCIN / AEI
/10.13039/501,100,011,033, Spain)
E hics s a emen
No applicable: This manusc ip does no include human o animal
esea ch.
I his manusc ip in ol es esea ch on animals o humans, i is
impe a i e o disclose all app o al de ails.
I Yes, please p o ide you ex he e:
Suppo ing in o ma ion
Compa a i e s a is ics and unce ain y indices (
2
and RMSEP) on
e lec ance o he alida ion se s o he model gene a ed by PLS-R o all
he nu ien s and igu e showing he loca ion o s udy plo s in he Ibe-
ian Peninsula.
CRediT au ho ship con ibu ion s a emen
J.I. Manzano: W i ing – e iew & edi ing, W i ing – o iginal d a ,
Visualiza ion, Me hodology, In es iga ion, Fo mal analysis, In es iga-
ion, Concep ualiza ion. M. Rod íguez-Febe ei o: W i ing – e iew &
edi ing, Visualiza ion, Me hodology, Fo mal analysis, Concep ualiza-
ion. M. Fandi˜
no: W i ing – e iew & edi ing, Visualiza ion, Me hod-
ology, Fo mal analysis, Concep ualiza ion. M. Vilano a: W i ing –
e iew & edi ing, Visualiza ion, Supe ision, In es iga ion, Concep u-
aliza ion. J.J. Cancela: W i ing – e iew & edi ing, Visualiza ion, Su-
pe ision, In es iga ion, Funding acquisi ion, Concep ualiza ion.
Decla a ion o compe ing in e es
The au ho s decla e ha hey ha e no known compe ing inancial
in e es s o pe sonal ela ionships ha could ha e appea ed o in luence
he wo k epo ed in his pape .
Acknowledgmen s
To esea che s which collabo a e wi h he collec ion o lea samples
in he p ojec s: IRRIVITIS-PID2019–105039RR-C4 (Funding by MCIN /
AEI /10.13039/501100011033, Spain), AROMAVID (Funding by MCIU
/ CDTI, Spain), ALBASOUL (Funding by MCIU / CDTI, Spain) and
Table 3
Analysis o ele an wa eleng hs o PLS-R models.
Nu ien
code
PLS-R selec ed wa elengh s (nm)
UV VIS NIR
C–450, 478, 512, 544, 601,
629, 651, 702, 733, 769
1389, 1440, 1891, 2032
N–412, 461, 581, 595, 639,
669, 703, 741
1153, 1222, 1290, 2006,
2079
P 267,
303
420, 454, 478, 603 794, 1173, 1401, 1582,
1932, 2108, 2487
K 329 413, 450, 601, 708 1399, 1901
Ca –554, 737 1095, 1273, 1582, 1926,
2104
Mg 302 692 899, 1267, 1423, 1948,
2118, 2402
B–514, 545, 601, 653, 707 –
Cu –491, 715 1113, 1292, 1581, 1826,
1932, 2110
Fe –446, 583, 600, 632, 651,
705, 775
1422
Mn 371 411, 534, 599, 649, 683,
745, 770
851, 1145, 1388, 1430,
1875, 2024
Zn –437, 583, 686 793, 1144, 1388, 1900,
2028
Na –514, 545, 600, 699, 776 1418, 1900
Al –545, 603, 694 781, 1388, 1900
The wa eleng hs o g ea es ele ance o he gene a ion o he p edic i e model
a e highligh ed in bold.
J.I. Manzano e al.
