A icles
h ps://doi.o g/10.1038/s41893-022-00940-6
A ull lis o a ilia ions appea s a he end o he pape .
Food p oduc ion is he main cause o en i onmen al impac s
such as biodi e si y loss1,2, eu ophica ion3 and o e exploi a-
ion o ma ine esou ces4. In pa icula , ag icul u e d i es 80%
o global de o es a ion and 70% o eshwa e wi hd awals5. Human
p essu es may ha e al eady pushed he Ea h sys em beyond
he sa e ope a ing space o humani y, as dema ca ed by he nine
plane a y bounda ies desc ibing c i ical Ea h sys em p ocesses6,7.
Beyond hese bounda ies, he isk o ab up o i e e sible global
en i onmen al change inc eases, wi h he po en ial o push he
Ea h sys em ou o i s s able Holocene condi ion, hus h ea ening
he capaci y o humani y o de elop and h i e7,8.
Recen li e a u e and modelling s udies9,10 p o ide he i s
quan i a i e es ima es o an a ay o in e ac ions among he Ea h
sys em p ocesses ep esen ed by he plane a y bounda ies. Thei
indings sugges ha he cascades and eedbacks o in e ac ions
ampli y human impac s on he Ea h sys em. Accoun ing o
hese in e ac ions may na ow he es ima ed global sa e ope a ing
space o human ac i i ies bu also illumina e syne gies: dec eas-
ing impac s on one Ea h sys em p ocess may dec ease impac s
on o he s. Gi en ha many in e ac ions ele an o ood p oduc-
ion a e no quan i ied by exis ing s udies9,10 and ha ma ine
p ocesses ha e ecei ed ela i ely li le a en ion wi hin he plan-
e a y bounda ies li e a u e11, achie ing sus ainable ood u u es
may equi e e en g ea e ood sys em e o ms han p e iously
sugges ed12.
In e ac ions among Ea h sys em p ocesses a e challenging o
accoun o in ood sys em analyses and spa ially disagg ega ed
models. The in e ac ions and in e ac ion s eng hs a e pa ially
unknown; hey can be con ex -speci ic and no all con ol a i-
ables a e sui able o spa ial models. Thus, esea ch on espec ing
he global sa e ope a ing space is complemen ed by explo ing how
o s ay wi hin c i ical limi s o ecosys ems a smalle scales13. The
e ec s o pe u ba ions on key Ea h sys em p ocesses a e o en
mani es ed locally1,13–15, aiding in de ec ing ecosys em-le el impac s
o in e ac ions. Fu he , ad ances in clima e, ocean and e es ial
modelling16–18 enable explo a ion o complex in e ac ions, imp o -
ing unde s anding o hei oles in clima e and sus ainabili y ac ions
and ou comes12.
He e, we concen a e on ou key Ea h sys em p ocesses el-
e an o ood p oduc ion12 ha p obably ha e al eady been ans-
g essed5: biogeochemical lows, biosphe e in eg i y (BI), eshwa e
use and land sys em change. We di ide BI in o land, eshwa e and
ocean componen s and eshwa e use in o blue and g een wa e
(Me hods; Figs. 1 and 2a). We hus e alua e in e ac ions among
se en e es ial and aqua ic con ol a iables— ha is, he unc-
ional indica o s o he unde lying Ea h sys em p ocesses (Fig. 2a).
All included Ea h sys em p ocesses a e bo om-up in na u e, as
opposed o, o example, clima e change, which is a op-down p o-
cess and o which in e ac ions wi h o he Ea h sys em p ocesses
ha e been explo ed p e iously9,10.
Quan i ying Ea h sys em in e ac ions o
sus ainable ood p oduc ion ia expe elici a ion
Anna Ch ysa i 1,2 ✉ , Vili Vi kki 1, Mika Jala a 1, Vilma Sands öm1, Johannes Piipponen 1,
Miina Po kka 1,3, S e en J. Lade 4,5, Kelsey La Me e6, Lan Wang-E landsson 4,7, Lau a Sche e 8,
Lau en S. Ande sen 9, Elena Benne 10,11, Ka e A. B auman 12, G ego y S. Coope 13,14, Ad iana De Palma15,
Pe a Döll 16,17, And ea S. Downing 3,4, Timo hy C. DuBois4, Ingo Fe ze 4,7, Elizabe h A. Ful on 18,19,
Die e Ge en 9,20, Hadi Jaa a 21, Jonas Jäge mey 9,22,23, Fe nando Ja amillo 24,25, Ma in Jung 26,
Helena Kahiluo o 27, Luis Lassale a 28, Anson W. Mackay29, Daniel Mason-D’C oz 30,31,32,
Mes in M. Mekonnen 33, Ki s y L. Nash 19,34, Amandine V. Pas o 35,36,37, Na in Ramanku y 38,39,
B ad Ridou 31,40, S e an Siebe 41,42, Benno I. Simmons 43, A ie S aal 44, Zhongxiao Sun 8,45,
A ne Tobian 4,9, A kai z Usubiaga-Liaño 46,47, Ruud J. an de En 48, A nou an Soesbe gen49,50,
Pe e H. Ve bu g 51, Yoshihide Wada 26, Sam Zippe 52 and Ma i Kummu 1 ✉
Se e al sa e bounda ies o c i ical Ea h sys em p ocesses ha e al eady been c ossed due o human pe u ba ions; no accoun -
ing o hei in e ac ions may u he na ow he sa e ope a ing space o humani y. Using expe knowledge elici a ion, we
explo ed in e ac ions among se en a iables ep esen ing Ea h sys em p ocesses ele an o ood p oduc ion, iden i ying
many in e ac ions li le explo ed in Ea h sys em li e a u e. We ound ha g een wa e and land sys em change a ec o he
Ea h sys em p ocesses s ongly, while land, eshwa e and ocean componen s o biosphe e in eg i y a e he mos impac ed
by o he Ea h sys em p ocesses, mos no ably blue wa e and biogeochemical lows. We also mapped a complex ne wo k o
mechanisms media ing hese in e ac ions and c ea ed a u u e esea ch p io i iza ion scheme based on in e ac ion s eng hs
and exis ing knowledge gaps. Ou s udy imp o es he unde s anding o Ea h sys em in e ac ions, wi h sus ainabili y implica-
ions including imp o ed Ea h sys em modelling and mo e explici biophysical limi s o u u e ood p oduc ion.
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain
830
A icles
Na U E SUS aiNabili y
Ou esea ch has wo majo con ibu ions: (1) unde s anding he
ichness o in e ac ions and in e ac ion mechanisms among Ea h
sys em p ocesses ele an o ood p oduc ion and (2) quan i ying
he s eng h o hese in e ac ions by expe elici a ion. Ou map o a
wide a ay o Ea h sys em p ocesses han p e iously a ailable ( o
example, e . 9) p o ides unp eceden ed in o ma ion on he in e -
ac ion mechanisms and bene i s Ea h sys em modelling among
o he ields. Ou quan i ica ion o in e ac ion s eng hs ad ances
ou unde s anding o he sa e ope a ing space o ood p oduc-
ion, as ew quan i ica ions o hese in e ac ions a e a ailable. While
e . 9 p o ide some li e a u e-based quan i ica ions, model-based
quan i ica ions a e a e because hey a e ime-consuming and
equi e ocusing on a e y limi ed se o in e ac ions a a ime. Fo
example, e . 10 quan i y in e ac ions among h ee Ea h sys em p o-
cesses a egional scale (clima e change, su ace uno and ege a-
ion co e ), while we include se en a local scale (Fig. 1). In sum,
ou wo k inc eases comp ehensi e quan i a i e knowledge o Ea h
sys em in e ac ions ele an o ood p oduc ion.
The expe knowledge elici a ion was conduc ed ollowing
he IDEA (in es iga e, discuss, es ima e, agg ega e) s uc u ed
elici a ion p o ocol19 (Me hods). In an expe knowledge elici a-
ion, judgemen s o an unknown phenomenon can be exp essed
quan i a i ely in a s a is ical o ma 20 oge he wi h collec ing
quali a i e da a on he expe s’ easoning behind hei esponses.
Thus, he esul s can be used, o example, as model inpu o in
decision-making when limi ed o no o he in o ma ion is a ail-
able21,22. As we a e e y ea ly in he p ocess o explo ing in e ac-
ions among Ea h sys em p ocesses ele an o ood p oduc ion,
we a gue ha expe knowledge is an excellen i s s ep in ga h-
e ing in o ma ion accumula ed h ough expe s’ aining and
expe ience. The elici a ion was done o a hypo he ical s udy
a ea wi h a scena io-based echnique23 (Fig. 2b; Me hods), guid-
ing expe s in making in e ences ela i e o a ixed baseline. This
allowed us o accoun o s a e-dependen in e ac ions and explo e
a wide a ay o in e ac ions compa ed o s ic ly model-based
quan i ica ions.
Wi h his s udy, we p o ide a sys ems iew o he in e ac-
ions, hei s eng hs and he media ing mechanisms un il mo e
model-based esul s become a ailable. We en ision ha ou esul s
will bene i he iden i ica ion and p io i iza ion o syne gis ic
ac ions on de eloping ood sys ems and be use ul o modelling he
Ea h sys em as well as ood sys ems and ag icul u e, o na u al
esou ce manage s and o he de elopmen o he plane a y bound-
a ies amewo k.
(1) Ea h sys em p ocesses
o in e es
(2) Elici a ion p ocess
2nd elici a ion
ound
(3) Agg ega ion
Quan i a i e
da a
Quali a i e
da a
Media ing mechanisms
In e ac ion s eng hs
(4) Ea h sys em
p ocesses in e ac ions
1s elici a ion
ound
Discussion
ound
Aim: o assess he biophysical
in e ac ions be ween Ea h sys em
p ocesses ele an o ood p oduc ion P ojec
s eps
O
c
e
a
n
L
a
n
d
F
e
s
h
w
a
e
B
l
u
e
w
a
e
G
e
e
n
w
a
e
B
i
o
s
p
h
e
e
i
n
e
g
i
y
B
i
o
s
p
h
e
e
i
n
e
g
i
y
B
i
o
s
p
h
e
e
i
n
e
g
i
y
B
i
o
g
e
o
c
h
e
m
i
c
a
l
l
o
w
s
L
a
n
d
s
y
s
e
m
c
h
a
n
g
e
N
2
NH
4
NH
3
NO
3
NO
2
Fig. 1 | Schema ic ep esen a ion o he p ojec s eps. 1, ea h sys em p ocesses ele an o ood p oduc ion selec ed as he ocus o his wo k. 2, expe
elici a ion conduc ed ollowing a s uc u ed p o ocol. 3, Indi idual expe assessmen s agg ega ed o de i e he inal elici a ion esul s. 4, Agg ega ed da a
used o es ima e he in e ac ion s eng hs and build he ne wo k o media ing mechanisms.
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain 831
A icles Na U E SUS aiNabili y
Resul s
In he ollowing subsec ions, we p esen he iden i ied in e ac ions,
hei s eng hs and oles and we desc ibe he mechanisms media -
ing hem. Finally, we compa e ou indings wi h ele an li e a u e
and make sugges ions o u u e esea ch p io i iza ion.
Iden i ied in e ac ions and hei oles. Expe s iden i ied 37 di ec
biophysical in e ac ions be ween he selec ed con ol a iables (ou
o a possible o al o 54), which sugges s conside able local in e con-
nec ions o Ea h sys em p ocesses (Fig. 3). Some o hese in e ac-
ions, such as he impac s o land sys em change on BI land, a e
Biodi e si y in ac ness index (BII)*
A p oxy o unc ional di e si y. BII assesses change in popula ion abundance as a esul o
human impac s. The global bounda y is se a 90% BII ela i e o p e-indus ial le els.
Biomass o keys one ish species
Fo es ed land a ea*
Leached N concen a ion in uno o su ace wa e s
Seasonal i e discha ge
A p oxy o ecosys em unc ioning. Biomass ≥ 0.5 K is conside ed wi hin sa e biological
limi s. K is ca ying capaci y, he maximum size he popula ion can each in i s en i onmen .
Fo es co e ela i e o po en ial o es in biome scale. The bounda y is se a 85% o opical
and bo eal biomes and 50% o empe a e biomes. Global bounda y a 75% o o iginal o es .
A p oxy o N concen a ion in su ace wa e s. Concen a ion in uno should no exceed
1 mg N l–1 o p e en aqua ic ecosys em eu ophica ion.
Ri e discha ge should s ay abo e local en i onmen al low equi emen s, which ange
om 45 o 75% (low low season) and 15 o 45% (high low season) o p e-indus ial lows.
G owing season soil mois u e
G owing season oo -zone soil mois u e is used as a p oxy o a ious Ea h sys em unc ions
o g een wa e . I is assessed ela i e o a e age p e-indus ial condi ions.
Con ol a iables wi hin
local sa e space
Po en ially loca ed in
any egion
A ea: 100 km2
0.5 mg N l–1 in
uno
95% o p e-indus ial
soil mois u e
95% discha ge o low- and
65% o high- low season
Hypo he ical a ea
Con ol a iables
b
a
N2
NH4
NO2
NO3
NH3
G een wa e
Blue wa e
Biogeochemical lows
Land sys em change
BI eshwa e BI ocean
BI land
95% BII on land
0.8 K keys one
ish biomass
0.8 K keys one
ish biomass
90% o po en ial
o es co e
Fig. 2 | Con ol a iables and hypo he ical s udy a ea. a, he con ol a iables o each o he ea h sys em p ocesses used o he elici a ion pu poses.
con ol a iables indica ed wi h an as e isk a e he same as de ined in e . 7. Fo con ol a iables wi hou exis ing bounda y alues, sa e anges we e
de eloped and se (Supplemen a y Sec ion 2). b, A hypo he ical a ea o 100 km2 wi h con ol a iables wi hin sa e anges was used o assess he
in e ac ions among he ea h sys em p ocesses ele an o ood p oduc ion by expe s in his scena io-based elici a ion.
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain
832
A icles
Na U E SUS aiNabili y
well known and documen ed ( o example, e s. 1,14) bu many o
hem, such as he in e ac ions be ween he aqua ic (BI eshwa e
and BI ocean) and BI land componen s, ha e no been p e iously
quan i ied (Fig. 3 and Supplemen a y Table 5.1). Addi ionally, some
in e ac ions iden i ied in his wo k we e conside ed non-exis en
in e . 9.
Expe s es ima ed s ong impac s on aqua ic biodi e si y
(BI eshwa e and BI ocean) o igina ing om o he Ea h sys em
p ocesses. Fo example, BI eshwa e was seen o be especially
a ec ed by changes in blue wa e , biogeochemical lows and BI land
(Fig. 3). La ge nega i e impac s on BI eshwa e can be caused by,
o example, inc eased ni ogen concen a ion in su ace wa e s
In e ac ion om
ows o columns
No in e ac ion
No applicable
In e ac ion s eng h
Ampli ying di ec ion
A enua ing di ec ion
0
0
1
1
Low
Medium
High
Assessmen unce ain y
a
c
In e ac ion om
ows o columns
No in e ac ion
No applicable
b
d
0
.
8
6
3
.
2
4
B
l
u
e
w
a
e
H
F
B
l
u
e
w
a
e
L
F
G
e
e
n
w
a
e
B
I
l
a
n
d
B
I
e
s
h
w
a
e
B
I
o
c
e
a
n
L
a
n
d
s
y
s
e
m
c
h
a
n
g
e
1
.
3
3
0
.
9
2
1
.
7
8
1
.
4
0
1
.
8
6
1
.
5
0
2
.
6
8
0
.
7
5
2
.
9
0
1
.
3
1
2
.
9
4
1
.
7
5
0
.
3
0
1
.
2
0
B
l
u
e
w
a
e
H
F
B
l
u
e
w
a
e
L
F
G
e
e
n
w
a
e
B
I
l
a
n
d
B
I
e
s
h
w
a
e
B
I
o
c
e
a
n
L
a
n
d
s
y
s
e
m
c
h
a
n
g
e
2
.
6
8
0
.
7
5
1
.
3
1
2
.
9
4
1
.
7
5
0
.
8
6
1
.
3
3
0
.
9
2
1
.
7
8
1
.
4
0
1
.
8
6
1
.
5
0
0
.
3
0
1
.
2
0
3
.
2
4
2
.
9
0
BI land
BI eshwa e
Land sys em
change
BI ocean
Biogeochemical
lows
Blue wa e HF
Blue wa e LF
G een wa e
O igina ing om
BI land
BI eshwa e
Land sys em
change
BI ocean
Biogeochemical
lows
Blue wa e HF
Blue wa e LF
G een wa e
Impac ing
O igina ing om
Impac ing
BI land
BI eshwa e
Land sys em
change
BI ocean
Biogeochemical
lows
Blue wa e HF
Blue wa e LF
G een wa e
BI land
BI eshwa e
Land sys em
change
BI ocean
Biogeochemical
lows
Blue wa e HF
Blue wa e LF
G een wa e
B
i
o
g
e
o
c
h
e
m
i
c
a
l
l
o
w
s
B
i
o
g
e
o
c
h
e
m
i
c
a
l
l
o
w
s
Fig. 3 | Absolu e no malized biophysical in e ac ion s eng hs and associa ed unce ain y iden i ied wi h expe knowledge elici a ion. a, Iden i ied
in e ac ions and in e ac ion s eng hs be ween he selec ed con ol a iables, anging om he weakes (~0) o he s onges (1). b, Ne o igina ing and
ecei ing in e ac ion s eng hs o each con ol a iable. A ow wid h indica es he magni ude o he in e ac ion s eng h. Numbe s a ound he ci cle
indica e he sum o bo h ecei ing/o igina ing. c, Unce ain y ela ed o assessing he in e ac ions o indi idual in e ac ions. d, Unce ain y ela ed o
assessing he in e ac ions wi h ne o igina ing and ecei ing in e ac ion s eng hs. he unce ain y is e alua ed on he basis o expe ag eemen and he
numbe o esponses pe in e ac ion (Supplemen a y able 5.2 gi es unce ain y c i e ia and ca ego iza ion). bI, biosphe e in eg i y; HF, high- low season;
LF, low- low season. ‘No applicable’ e e s o he main diagonal in which he same con ol a iable is ep esen ed bo h in he espec i e column and ow.
‘No in e ac ion’ e e s o cases o which expe s did no iden i y an in e ac ion exis ing be ween hese wo con ol a iables.
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain 833
A icles Na U E SUS aiNabili y
and dec eased discha ge, while a dec ease in BI eshwa e can sub-
s an ially in luence BI land and BI ocean. This e lec s he impo -
ance o conside ing all h ee componen s in u u e BI assessmen s.
G een wa e and land sys em change a e in ol ed in many s ong
in e ac ions (Fig. 3 and Supplemen a y Table 5.1): o example, a
dec ease in soil mois u e can di ec ly impac blue wa e , BI land and
land sys em change. A he same ime, soil mois u e can be educed
when o es co e dec eases, ein o cing he eedback e ec s. Land
sys em change, as expec ed, is a majo cause o changes in o he
Ea h sys em p ocesses, no ably BI land and blue wa e (Fig. 3).
Ve y ew in e ac ions we e ound o be a enua ing, which would
mean ha when one a iable mo es owa ds he ou side o he sa e
space, he o he a iable mo es owa ds he sa e space. The in e ac-
ion om land sys em change o blue wa e was he only one ha
expe s iden i ied as s ongly a enua ing, whe eby a dec ease in
o es co e leads o inc eased i e discha ge (Fig. 3). Howe e ,
expe s ag eed ha his migh hold only a he local scale, as
la ge-scale dec eases in o es co e end o cause egionally d ie
condi ions due o dec eased p ecipi a ion24. Expe s also iden i ied a
weak a enua ing in luence o biogeochemical lows on land sys em
change, as inc eased ni ogen in uno can boos plan p oduc i -
i y; howe e , subs an ial inc eases in ni ogen may u n he di ec-
ion o his in e ac ion o ampli ying (P io i iza ion o in e ac ions).
As a enua ing in e ac ions a e a e, pa allel impac s o ampli y-
ing in e ac ions—in which u he pe u ba ion o a con ol a i-
able nega i ely pe u bs o he s—a e likely. Oppo unis ic ac ions
bene i ing om ade-o si ua ions, in which u he pe u ba ion
o a con ol a iable would imp o e he s a e o o he s, a e he e o e
no iable bu imp o ing he s a e o many con ol a iables h ough
ac ing on a single a iable is possible.
Fo se en o he in e ac ions, he in e ac ion s eng h was judged
ex emely weak (in e ac ion s eng h <0.005 on a scale 0–1). We
es ed whe he including he ex emely weak in e ac ions in he
o ce-di ec ed ne wo k p esen ed in Fig. 4 would change he esul s.
As i did no , we excluded hese in e ac ions om u he analysis
and he p io i iza ion scheme ou lined in P io i iza ion o in e ac-
ions. The weak ela ionships may e lec uly weak in e ac ions
bu i may also be he case ha hey a e p esen (and impo an )
only in speci ic en i onmen s. Fu he , hey may be mo e complex
han he o he s, no ollowing simpli ied linea i y assump ions
and he e o e ha e no been well-cha ac e ized in exis ing s udies
(Supplemen a y Sec ion 5). Fo u he in o ma ion, ele an li e a-
u e o he iden i ied in e ac ions and all indi idual answe s o he
second elici a ion ound a e p o ided in Supplemen a y Da a.
Iden i ied in e ac ion s eng hs in line wi h li e a u e. Se en o
he 37 iden i ied in e ac ions (Fig. 3) we e quan i ied a he global
scale by e . 9. Fo hese se en in e ac ions, ou s udy and e . 9 ag ee
on he di ec ion o in e ac ions, al hough wi h some di e ences in
s eng h. A s ic compa ison be ween he wo s udies is no possi-
ble due o di e ences in no maliza ion, he con ol a iables and he
spa ial scale bu ela i e compa isons a e shown in Supplemen a y
Ampli ying
A enua ing
NH4
N2
NH3
NO3
NO2
S eng h examples
1
0.5
0.1
1
0.5
0.1
L
a
n
d
O
c
e
a
n
F
e
s
h
w
a
e
B
l
u
e
w
a
e
B
l
u
e
w
a
e
G
e
e
n
w
a
e
B
i
o
s
p
h
e
e
i
n
e
g
i
y
L
a
n
d
s
y
s
e
m
c
h
a
n
g
e
B
i
o
g
e
o
c
h
e
m
i
c
a
l
l
o
w
s
B
i
o
s
p
h
e
e
i
n
e
g
i
y
H
i
g
h
l
o
w
L
o
w
l
o
w
B
i
o
s
p
h
e
e
i
n
e
g
i
y
Fig. 4 | A ne wo k diag am o he iden i ied in e ac ions wi h a o ce-di ec ed layou . Nodes a e a anged acco ding o he no malized in e ac ion
s eng hs, wi h s onge connec ions close oge he . In e ac ions wi h s eng h in he ange o −0.005–0.005 a e excluded om his igu e.
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain
834
A icles
Na U E SUS aiNabili y
Table 6.1. Fo i e in e ac ions, he in e ac ion s eng hs we e simi-
la ( om low o high in e ac ion s eng h ange). Fo wo in e ac-
ions (land sys em change on blue wa e and biogeochemical lows
on BI ocean), he di e ence in in e ac ion s eng h can be a ib-
u ed o he di e en con ol a iables and spa ial scale used. Fo
u he compa isons be ween ou esul s and quan i a i e es ima es
o selec ed in e ac ions, see Supplemen a y Sec ion 6. The adequa e
co espondence be ween ou esul s and li e a u e shows ha
expe opinions, elici ed ia a igo ous and o mal p o ocol, cap-
u e he a ia ion ha indi idual s udies ha e iden i ied. The e o e,
he ela i e in e ac ion s eng hs iden i ied he e a e obus bu he
in e ac ion s eng hs should be in e p e ed cau iously ou side his
wo k, p e e ably in conjunc ion wi h quan i a i e modelling.
Recei ing and o igina ing Ea h sys em p ocesses. The ole o di -
e en Ea h sys em p ocesses in he in e ac ions a ies, as shown
in he ne wo k o he iden i ied in e ac ions and hei s eng hs
(Fig. 4). The h ee Ea h sys em p ocesses wi h he g ea es num-
be o iden i ied connec ions in he ne wo k a e BI land, land sys-
em change and g een wa e (8–10 ou o a maximum 12; Fig. 4
and Supplemen a y Table 5.3). On he basis o he ne ecei ing
and o igina ing in e ac ion s eng hs (Supplemen a y Table 5.3),
h ee main ca ego ies eme ge: (1) p ocesses mainly on he ecei -
ing end, meaning ha hey a e a ec ed by o he s; (2) p ocesses
mainly on he o igina ing end, meaning ha hey a ec o he s;
and (3) p ocesses ha a e bo h ecei ing and o igina ing a simila
le els. This in o ma ion could help in iden i ying impac ul p ac-
ical ac ions; o example, alle ia ing g een wa e pe u ba ions
would be expec ed o alle ia e pe u ba ions on BI land, land sys em
change and blue wa e as hose ecei e s ong in e ac ions o igina -
ing om g een wa e .
The h ee BI componen s comp ise he i s g oup, as hey ecei e
he g ea es ne in e ac ion s eng h (Figs. 3 and 4 and Supplemen a y
Table 5.3). This is in line wi h wo s udies7,9 ha iden i y BI as one
o he wo co e Ea h sys em p ocesses conside ed in he plane a y
bounda ies amewo k. Land sys em change and g een wa e exem-
pli y he second ca ego y, as he g ea es ne in e ac ion s eng h
o igina es om hem (Figs. 3 and 4 and Supplemen a y Table 5.3).
These indings a e consis en wi h e . 9 iden i ying land sys em
change as a majo media o o in e ac ions among Ea h sys em
p ocesses, e s. 25,26 sugges ing ha ocusing only on blue wa e does
no cap u e all he c ucial Ea h sys em unc ions o eshwa e and
e . 27 p oposing o de ine a g een wa e plane a y bounda y. Finally,
in he hi d ca ego y, biogeochemical lows and blue wa e ha e
an impo an media ing ole in he iden i ied in e ac ions, as he
ecei ing and o igina ing in e ac ion s eng hs a e mo e balanced
(Figs. 3 and 4 and Supplemen a y Table 5.3).
Media ing mechanisms. The expe s iden i ied an a ay o p i-
ma y and case-speci ic mechanisms media ing he in e ac ions
(Supplemen a y Sec ion 7 and Supplemen a y Fig. 7.1). The main
mechanisms a e shown in Fig. 5, which illus a es ha he in e ac ions
G oundwa e
echa ge
Flooding
E apo anspi a ion
Te es ial
p oduc i i y
Seedlings
consump ion
Pollina ion, pes con ol,
decompos e s, seed dispe sal Aqua ic ecosys em in eg i y
( o example wa e quali y)
Food web/
biomass lux
Te es ial/ ipa ian/
coas al habi a in eg i y
Ni ogen ixa ion
In il a ion/wa e holding/
oo zone s o age capaci y
E osion
P ecipi a ion
Soil mois u e
de ici Su ace uno
Sedimen a ion and
sedimen low
Posi i e link
Nega i e link
Nu ien and ene gy ecycling/
ansloca ion ( ood webs)
Aqua ic
p oduc i i y
Eu ophica ion
N loss
N
2
NH
4
NH
3
NO
3
NO
2
Wild i es and
d ough s
F
e
s
h
w
a
e
S
o
i
l
m
o
i
s
u
e
R
i
e
d
i
s
c
h
a
g
e
K
e
y
s
o
n
e
i
s
h
b
i
o
m
a
s
s
K
e
y
s
o
n
e
i
s
h
b
i
o
m
a
s
s
O
c
e
a
n
N
i
o
g
e
n
i
n
u
n
o
L
a
n
d
B
i
o
d
i
e
s
i
y
i
n
a
c
n
e
s
s
F
o
e
s
c
o
e
Fig. 5 | Expe -iden i ied main mechanisms media ing he in e ac ions among Ea h sys em p ocesses ele an o ood p oduc ion. Posi i e links (blue)
indica e ha an inc ease (dec ease) in one a iable leads o an inc ease (dec ease) in ano he a iable. Nega i e links ( ed) indica e ha an inc ease
(dec ease) in one a iable leads o a dec ease (inc ease) in ano he a iable. he ne wo k o mechanisms is no exhaus i e because some links a e
unce ain (Fig. 6) and dependen on impac le el, spa ial scale, empo al dynamics and ea h sys em p ocesses beyond he scope o his s udy.
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain 835
A icles Na U E SUS aiNabili y
a e complex and in e connec ed, e en when limi ing he scope o
Ea h sys em p ocesses ele an o ood p oduc ion. Mo eo e , di -
e en mechanisms can ha e coun e ac ing e ec s on he con ol
a iables, lea ing he esul ing in e ac ion o be de e mined by
he p esence, impo ance and cha ac e is ics o di e en mecha-
nisms. Fo example, o es co e is nega i ely impac ed by wild i es,
d ough s and seedling consump ion bu bene i ed by pollina o s,
seed dispe sal, decompose s, soil mois u e and e es ial p oduc-
i i y. These mechanisms a y subs an ially: o example, wild-
i es a ec o es co e ex emely quickly, whe eas o es g ow h is
a ec ed e y slowly. The ne in e ac ion s eng h is he e o e igh ly
dependen on which mechanisms p e ail a a ime. Due o his, we
aimed o explo e he ela i e impo ance o he di e en media ing
mechanisms du ing he second elici a ion ound. While we we e
unable o make obus in e ences on he ma e , he limi ed da a
ela ed o anking, oge he wi h de ails and ele an li e a u e on
he iden i ied mechanisms, a e a ailable in Supplemen a y Da a.
Unde s anding he media ing mechanisms is key o inding
ac ion poin s o es ablish bene icial syne gies. Fo example, dec eas-
ing o es co e (land sys em change) and soil mois u e (g een
wa e ) inc ease e osion, which u he inc eases N loss and sedi-
men low, bo h o which ul ima ely lead o nega i e changes in BI
eshwa e , BI ocean and biogeochemical lows. P e en ing e osion
o epai ing i s impac s, could hen be used as an e ec i e mecha-
nism o alle ia e he p essu e on many con ol a iables a once.
O he high-impac ac ions include, o example, imp o ing ipa -
ian and coas al habi a in eg i y h ough conse a ion measu es o
ag oecosys em in eg i y, which would di ec ly a ec BI eshwa e
and BI ocean and also p omo e mechanisms a ec ing soil mois u e,
i e discha ge and o es co e , p ima ily h ough inc eased e es-
ial p oduc i i y and i s e ec s on he hyd ological cycle. Al hough
impac s can be expec ed o diminish along long mechanism chains
and a y be ween con ex s, he ich ne wo k o mechanisms high-
ligh s ha ac ions o en a ec an ensemble o he con ol a iables
ins ead o only one a a ime.
P io i iza ion o in e ac ions. The expe s had high ag eemen
on in e ac ion magni udes; he coe icien o a ia ion in hei es i-
ma es was low in all bu one in e ac ion (Supplemen a y Table 5.2).
The numbe o esponses pe in e ac ion, howe e , a ied consid-
e ably, which inc eases he unce ain y o he esul s. Due o he
po en ial o inc eased bias and e o owing o he lowe numbe
o esponses, we c ea ed a p io i iza ion scheme o u u e esea ch
on he basis o he in e ac ion s eng h and unce ain y (Me hods;
Supplemen a y Table 5.2). We sugges ha ocus should i s be
gi en o s ong in e ac ions wi h limi ed cu en knowledge and/
o high unce ain y. Those in e ac ions could unexpec edly b each
he sa e ope a ing space, as he in e ac ion s eng h is high bu
ela i ely li le is known abou he media ing mechanisms. In
addi ion, we highligh ed in e ac ions wi h disc epancies in expe
opinions ega ding whe he hey we e ampli ying o a enua ing.
These disc epancies could be ela ed o empo al scales ( a iabil-
i y in he occu ence ime o p ocesses), di e en egional con ex s
( a iabili y in local en i onmen al ac o s) o di e en media ing
mechanisms ( a iabili y in which mechanisms domina e he ne
in e ac ion). Fo such in e ac ions, egional-scale s udies should
be p io i ized o iden i y he impo ance and empo al scales o he
in ol ed mechanisms. In cases o disc epancies, we de e mined he
in e ac ion di ec ion in Fig. 3 by he majo i y o esponses.
The high-unce ain y in e ac ions om biogeochemical lows o
BI eshwa e and BI ocean a e p ime examples o expe opinion
disc epancies (Fig. 6). Some expe s conside ed he posi i e impac s
on p ima y p oduc i i y and ecosys em unc ioning om added
nu ien s—be o e a c i ical limi is passed and impac becomes
nega i e—while o he s conside ed added nu ien s causing imme-
dia e nega i e impac s. This c i ical limi is e y con ex -speci ic,
including ac o s such as he deni i ica ion po en ial o ipa ian
we lands28. An app op ia e c i ical limi migh a y among en i-
onmen s— hus, his in e ac ion needs mo e case-by-case exami-
na ion. In addi ion, including elemen s such as phospho us29 could
subs an ially modi y he s eng h o biogeochemical lows in e ac-
ions on BI eshwa e and BI ocean30. O he no able unce ain ies
include, o example, he in e ac ion om blue wa e o BI ocean,
which was es ablished bu which is especially unce ain due o he
low numbe o esponses.
Medium-unce ain y in e ac ions ela ed o he BI componen s
dese e u he a en ion due o hem being o e all s ongly impac ed
(Fig. 6 and Recei ing and o igina ing Ea h sys em p ocesses)
Highe
In e ac ion s eng h
Lowe
Low HighMedium
Unce ain y
Disc epancies in in e ac ion
di ec ion
In e ac ion be ween wo
Ea h sys em p ocesses
— ead om le o igh
G een wa e
Blue wa e —low low
Blue wa e —high low
Biogeochemical lows
Land sys em change
BI ocean
BI eshwa e
BI land
Fig. 6 | In e ac ions iden i ied in he expe elici a ion g ouped by unce ain y and posi ioned ela i e o hei s eng h. he speci ic loca ion o an
in e ac ion wi hin an unce ain y zone is no signi ican , as unce ain y wi hin each ca ego y is conside ed equal. he unce ain y is e alua ed on he basis
o expe ag eemen and he numbe o esponses pe in e ac ion (Supplemen a y able 5.2 gi es unce ain y c i e ia and ca ego iza ion). Low unce ain y,
≥10 esponses pe in e ac ion; medium unce ain y, 6–9 esponses pe in e ac ion and one in e ac ion wi h high disag eemen (coe icien o a iance
>0.5) and 11 esponses; and high unce ain y, ≤5 esponses pe in e ac ion. In e ac ions wi h s eng h in he ange o −0.005–0.005 a e excluded om
his analysis.
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain
836
A icles
Na U E SUS aiNabili y
and hei cen al ole in he Ea h sys em7,9. This applies in pa icu-
la o in e ac ions in ol ing he aqua ic BI componen s, as hey ha e
only ecen ly been discussed and explo ed in ele an li e a u e ( o
example, e . 11). Mos in e ac ions ela ed o land sys em change,
blue wa e and g een wa e a e low-unce ain y in e ac ions (Fig. 6),
as expec ed gi en ha hese Ea h sys em p ocesses and hei ela-
ionships ha e been widely explo ed (see Supplemen a y Da a o
ele an li e a u e). Howe e , some in e ac ions wi h a g ea e num-
be o esponses ( o example, land sys em change on blue wa e )
show disc epancies ela ed o he di ec ion o he in e ac ions, e en
hough ag eemen on he magni ude o he in e ac ion s eng h is
high (Fig. 6). Again, his could be a ibu ed o di e ences in con-
ex s and imescales, which highligh s ha case-by-case app oaches
a e equi ed o ope a ionalizing ou indings.
Discussion
We ha e iden i ied in e ac ions and media ing mechanisms be ween
Ea h sys em p ocesses ele an o ood p oduc ion a he local scale
and quan i ied hei s eng hs using expe knowledge elici a ion.
Ou o 54 possible in e ac ions, 37 we e iden i ied by he expe s.
BI eshwa e , BI ocean and g een wa e we e ound o ha e c ucial
oles in he in e ac ions. In addi ion, ou esul s highligh he majo
ole o land sys em change impac ing o he Ea h sys em p ocesses
and he high impac o biogeochemical lows on BI eshwa e and
BI ocean (Figs. 3 and 4). Ou s udy maps he mechanisms in ol ed
in in e ac ions among hese many Ea h sys em p ocesses in de ail,
e ealing a complex and in e connec ed mechanisms ne wo k
(Fig. 5). Taking ad an age o syne gis ic mechanisms, h ough which
many con ol a iables can be a ec ed by change in one, is key o
limi ing u he an h opogenic pe u ba ions on he Ea h sys em.
Ou ca ego iza ion o he in e ac ions on he basis o hei s eng hs
and associa ed unce ain y (Fig. 6) can guide u u e esea ch.
B idging he local and global scales. By illumina ing he complex
in e ac ions be ween he Ea h sys em p ocesses ele an o ood
p oduc ion, ou esul s highligh he need o an holis ic app oach
o en i onmen ally sus ainable ood p oduc ion and sugges po en-
ial u u e de elopmen s o he plane a y bounda ies amewo k.
The plane a y bounda ies we e de eloped o unde s and he Ea h
sys em limi s wi hin which humani y can h i e6,7 and he ame-
wo k has o en been seen as a connec o be ween Ea h sys em and
sus ainabili y sciences31. A he same ime, he amewo k has been
c i icized o being a s ic ly op-down concep , while many o he
ele an Ea h sys em p ocesses and s esses occu locally—al hough
wi h global impo ance32–34. Mo eo e , in e ac ions be ween many
o he Ea h sys em p ocesses ake place on a local- o- egional
scale. Wi hou unde s anding hese in e ac ions and he mecha-
nisms media ing hem, go e nance ha aims o keep us wi hin
sa e global bounda ies could be c i ically misdi ec ed and de ea
i s pu pose.
Ou indings e eal he di ec ions and s eng hs o many in e -
ac ions be ween Ea h sys em p ocesses ele an o ood p oduc-
ion. Since mos in e ac ions iden i ied he e we e ampli ying ones
(Fig. 3), he sa e ope a ing space may be na owed by one Ea h
sys em p ocess deg ading o he Ea h sys em p ocesses. Howe e ,
inding posi i e syne gies is also possible: alle ia ing p essu es on
one Ea h sys em p ocess, such as land sys em change, can simul a-
neously alle ia e p essu es on o he s, such as BI and biogeochemi-
cal lows, h ough he complex web o in e ac ions among hem
(Figs. 3–5; e . 9). By be e unde s anding bo h he s eng h and
he media ing mechanisms o hese in e ac ions, ou esul s ha e
clea implica ions o sus ainabili y managemen in (1) a oiding
unin ended consequences o ac ions; (2) emphasizing syne gis ic
solu ions o sus ainabili y challenges; and (3) iden i ying and p i-
o i izing managemen o he co e p ocesses ha mos impac and
a e impac ed by o he Ea h sys em p ocesses. Ou s udy is hus
an impo an s ep owa ds enabling mo e comp ehensi e conside -
a ion o Ea h sys em p ocesses ac oss sec o s and disciplines un il
mo e model-based in o ma ion on in e ac ions becomes a ailable
o en ich and imp o e ou indings.
Ou wo k could help o adap he plane a y bounda ies ame-
wo k o he scale a which managemen o Ea h sys em p o-
cesses—such as conse a ion o na u al a eas, limi a ion o ni ogen
applica ion and egula ion o wa e wi hd awals— ypically occu s.
S aying wi hin he plane a y bounda ies and he eby keeping
humani y wi hin he global sa e ope a ing space, equi es adjus -
ing local sa e ope a ing spaces wi h espec o he ela ed Ea h sys-
em p ocesses and in e ac ions shown he e. Ou indings may also
augmen local and egional ood sys em models and assessmen s.
De ailed, model-based quan i ica ion o in e ac ions be ween Ea h
sys em p ocesses is ime-consuming and ew a e cu en ly a ail-
able10. The e o e, inco po a ing ou indings in o models, un il o he
sou ces o in o ma ion become a ailable, could enable quan i ying
aspec s beyond he use o esou ces only—such as he subs an ial
impac s on BI. Taking ad an age o all da a—while acknowledging
he ela ed unce ain ies—should be p e e ed o wai ing o he
‘pe ec da a’ and po en ially delaying ac ion, he eby causing i e-
e sible damage.
A he same ime, i should be no ed ha he mechanisms medi-
a ing he in e ac ions, in e ac ion s eng hs and e en he in e ac ion
di ec ions a y in di e en con ex s, while he esul s p esen ed he e
show only hei agg ega e ou comes. Fu he , a be e unde s anding
o local-scale mechanisms po en ially cascading o plane a y-scale
eedbacks is needed o p io i izing managemen ac ions. Globally
consis en da ase s o many o he con ol a iables a e al eady
a ailable ( o example, e s. 1,35–40). These da ase s can be used
in local-scale models o alida e ou in e ac ion s eng hs while
e aining global consis ency, which would educe he unce ain y
in compa ing esul s be ween di e en local con ex s. Howe e , his
endea ou would equi e case-speci ic knowledge and sophis ica ed
models. The e o e, while we ha e alida ed ou esul s wi h com-
pa isons o e . 9 and a ious case s udies (Supplemen a y Sec ion
6), we lea e ex ensi e local-scale alida ion o u u e wo k. He e,
we p o ide bo h he in e ac ion s eng hs and mechanisms oge he
wi h a oadmap o p io i iza ion ha can guide u u e e o s.
Impo ance o expe knowledge. In modelling, p ocess dynamics
should be ep esen ed in ma hema ical e ms bu an exac ep e-
sen a ion o he Ea h sys em is beyond ou cu en capabili ies and
models will necessa ily be based on simpli ied p ocess desc ip ions.
Deciding which subp ocesses o model, wha assump ions o make
and how o ep esen in e ac ions be ween he p ocesses when all
media ing mechanisms canno be ully modelled a e all expe deci-
sions41. Fu he mo e, as we ha e shown he e, he media ing mech-
anisms a y wi h local con ex . This is also e lec ed in an expe
elici a ion; he expe s ha e hei own backg ounds ha a ec hei
iews and decisions42—such as di e en ields o s udy o amilia i y
wi h di e en na u al en i onmen s—which may lead o appa en
disc epancies.
This expe knowledge elici a ion e ealed ha no all expe s
ag eed on he di ec ion o some in e ac ions (Fig. 6). Such occu -
ences a e aluable, as hey documen di e ing iews and belie s
abou he same p ocess as well as s a is ical a iabili y. Combined,
expe esponses p o ide holis ic iews—beyond s a is ical a iabil-
i y—o complex phenomena ha a e di icul o s udy in quan i a-
i e e ms. E en hough modelling has adi ionally been based on
collec ed quan i a i e da a, he u he we inc ease he complexi y
o wha we aim o model, he highe he da a demand becomes.
Thus, using expe knowledge has become common and is applied
in di e en ields and o a ious pu poses— o example, in ecosys-
em modelling43, isk assessmen 44 and e en augmen ing machine
lea ning models45. Ou expe elici a ion on complex Ea h sys em
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain 837
A icles Na U E SUS aiNabili y
p ocess in e ac ions is a no able s ep o wa d in be e unde s and-
ing o p ocesses ha would o he wise emain unknown.
Me hods
The ollowing sec ion desc ibes he main s eps in ou me hodology as illus a ed
in Fig. 1. We i s explain he con ol a iables we used o each o he Ea h
sys em p ocesses o in e es (Fig. 2a) and he hypo he ical s udy a ea (Fig. 2b) and
in oduce he s uc u ed elici a ion p o ocol. We u he desc ibe he me hods
used o agg ega e and no malize he elici a ion esul s o quan i y he in e ac ion
s eng hs. Fo u he de ails, especially ela ed o he elici a ion p ocess and
p o ocol, see Supplemen a y Sec ions 1–4.
De ini ions o Ea h sys em p ocess con ol a iables. Fo he BI land
componen , we e ained he con ol a iable used by e . 7, he biodi e si y
in ac ness index, a p oxy a iable o unc ional di e si y (Fig. 2a). Ag icul u al
impac s on BI ha e been demons a ed a a local scale1 and e aining he exis ing
a iable he e o e i s ou pu poses well. The BI eshwa e and BI ocean
componen s we e included on he basis o ecen sugges ions o hei impo ance
in BI assessmen s9,11. Fo bo h BI eshwa e and BI ocean, we used he s a us
o keys one ish species biomass as a con ol a iable. O he s9 assign ecosys em
unc ioning as he con ol a iable and use global ishe ies s a us as a p oxy o
some o he in e ac ions hey iden i y; hus, ou con ol a iable is simila as hey
bo h assess biomass le els. In aqua ic en i onmen s, keys one ish species ac as a
obus indica o o ecosys em unc ioning and play a c i ical ole in de e mining
communi y s uc u e46–49. In addi ion, eshwa e habi a s in pa icula ha e
expe ienced a subs an ial decline in biodi e si y due o human ac i i ies and
en i onmen al change50,51. The e o e, keys one ish biomass can ac as a con ol
a iable o assess he aqua ic componen s o BI.
The con ol a iable o land sys em change is o es ed land a ea ela i e o
po en ial o es co e ( ha is, assuming no human land co e change; Fig. 2a)—a
a iable e ained om e . 7. Fo biogeochemical lows, we assessed ni ogen,
using leached ino ganic N concen a ion in uno o su ace wa e s as he con ol
a iable52. Fo blue wa e , he con ol a iable used is i e discha ge, ela i e o he
p e-indus ial a e age. In addi ion, o accoun o seasonal a ia ion in i e lows,
we sepa a ed he blue wa e in e ac ions in o e ec s du ing high- and low- low
pe iods. While o he s7 p opose maximum allowable wa e wi hd awals, we
ocused on he low emaining in i e s a e any discha ge al e a ion. Ex ending
he con ol a iable beyond wi hd awals cap u es discha ge al e a ion due o bo h
di ec human impac s, such as wa e ex ac ion53 and indi ec human impac s, such
as clima e change54 and changes in a mosphe ic mois u e ecycling55. Fo g een
wa e , he con ol a iable we used is oo -zone soil mois u e du ing he g owing
season ela i e o he p e-indus ial g owing season a e age (Fig. 2a), simila o
he con ol a iable sugges ed in e . 27. Though g een wa e is no iden i ied as
a sepa a e con ol a iable wi hin he eshwa e use bounda y o he o iginal
plane a y bounda ies amewo k, ecen esea ch25,26 p oposes ha ocusing only
on blue wa e does no cap u e all c ucial Ea h sys em unc ions o eshwa e and
hus we conside ed g een wa e o be indispensable. Fo mo e de ails on he con ol
a iables used, see Supplemen a y Sec ion 2.
Elici a ion p ocess. Expe knowledge elici a ion has been applied wi hin a ious
ields o en i onmen al sus ainabili y- ocused esea ch ( o example, e s. 56–60) and
i s sui abili y o na u al esou ces managemen has been demons a ed (see o
example, applica ions in e s. 61–63). Wi h an elici a ion, we can o mula e expe
knowledge and belie s abou po en ial unce ain ies in o a p obabilis ic o m64 ha
can subsequen ly be ea ed as empi ical da a63 and used o modelling pu poses
when o he da a a e una ailable.
He e, we ollowed he s uc u ed IDEA elici a ion p o ocol19 o a emo e
expe knowledge elici a ion. S uc u ed elici a ion p o ocols help educe bias and
e o associa ed wi h heu is ics ha expe s use when making judgemen s65,66. The
IDEA p o ocol is a s uc u ed modi ied Delphi app oach ha leads o imp o ed
judgemen s when a di e se g oup o engaged expe s pa icipa e21. I combines he
bene i s o Delphi61,67 and ou -s ep elici a ion p ocesses67–69, which in combina ion
has been shown o imp o e judgemen s65,70,71. Ou elici a ion p ocess consis ed
o wo anonymous elici a ion ounds, be ween which was an online discussion
ound using pseudonyms (Fig. 1). De ails o he elici a ion p o ocol a e a ailable
in Supplemen a y Sec ion 1.1 and Supplemen a y Table 1.2. The discussion ound
is a c i ical pa o he p ocess, as i dec eases linguis ic ambigui y, p omo es
c i ical hinking and sha es e idence. The IDEA p o ocol in eg a es elici a ion and
discussion because he e is e idence ha , when a discussion s age is included in
a s anda d Delphi p ocess, he esponse accu acy o he second elici a ion ound
inc eases72. The discussion phase also p o ed o be bene icial as i dec eased
ambigui y ega ding he ques ions and inc eased ag eemen among expe s
(Supplemen a y Sec ion 1.4 and Supplemen a y Fig. 1.1 wi h de ails ela ed o he
discussion ound and Supplemen a y Table 1.5 wi h he ag eemen me ic in he
wo elici a ion ounds).
Pa icipan s we e ec ui ed on he basis o hei expe ise in any o he Ea h
sys em p ocesses conside ed in his s udy and hei knowledge o he plane a y
bounda ies amewo k. Fo he ec ui men p ocess, ele an li e a u e on Web o
Science was sea ched and, once a lis o 200 po en ial pa icipan s was eached, he
li e a u e-based ec ui men was concluded (Supplemen a y Sec ion 1.3). All 200
po en ial pa icipan s we e in i ed o pa icipa e in he elici a ion ia e-mail. In
addi ion, he ‘snowballing me hod’ was used: when po en ial pa icipan s we e i s
con ac ed, hey we e asked o sugges u he sui able pa icipan s. This esul ed
in in i ing 31 addi ional po en ial pa icipan s. Ea lie plane a y bounda ies wo k
has ocused on ce ain Ea h sys em p ocesses (especially land sys em change,
blue wa e and BI land) and, hus, ep esen a ion o all desi ed disciplines was no
equal and was e lec ed in he numbe o answe s ecei ed o each in e ac ion.
In o al, 37 expe s comple ed he elici a ion p ocess, esul ing in 5–19 answe s
o each o he iden i ied in e ac ions. Li e a u e sugges s ha a minimum o ou
o six expe s should be included in an elici a ion73,74, wi h empi ical e idence
sugges ing ha only mino imp o emen s a e gained when ha ing mo e han 6–12
pa icipan s74–76. Fo de ails on he ec ui men p ocess and expe s’ backg ound
see Supplemen a y Table 1.3 and o o al numbe o answe s pe in e ac ion see
Supplemen a y Table 1.4.
The emo e expe elici a ion was pe o med wi h a web-based applica ion
specially de eloped o his pu pose using he ‘Shiny’ R package77. Al hough i
comes wi h i s own challenges ela ed especially o usabili y and use expe ience,
he bene i s o a cus om-made applica ion a e ha i minimizes he amoun o
ma e ial sha ed wi h pa icipan s and can be ully ailo ed o a speci ic ask. The
web applica ion (a ailable a h ps://ch ysa i1.shinyapps.io/shiny_exp_elic/)
displayed e e y hing a pa icipan needed o comple e he ull elici a ion p ocess,
consis ing o a consen o m, backg ound in o ma ion on he elici a ion p ocess,
he Ea h sys em p ocesses, he con ol a iables o be assessed, a ques ion example
and a dashboa d o selec ing speci ic in e ac ions and collec ing he inpu s.
Expe s we e asked o e alua e he in e ac ions wi hin he hypo he ical
a ea and o elabo a e on hei hinking p ocess behind he p o ided answe s.
A scena io-based app oach is pa icula ly cons uc i e when knowledge is
unde eloped a a heo e ical le el23, such as in ou case, in which explo a ion o he
complex in e ac ions be ween Ea h sys em p ocesses is s ill a an ea ly s age. Due
o he la ge numbe o po en ial in e ac ions explo ed and o minimize complexi y
and ime equi ed om he pa icipan s o comple e he elici a ion, we ocused
only on one scena io in which all Ea h sys em p ocesses a e wi hin sa e limi s
(Supplemen a y Sec ion 2). Addi ionally, expe s had he possibili y o se his
hypo he ical scena io in a egion o hei choice o hei esponses.
The ques ions asked o he expe s ollowed a ou -s ep o ma . An example
ques ion is p o ided in Supplemen a y Sec ion 1.2 and all elici a ion ques ions can
be explo ed by accessing he web applica ion. Fo each in e ac ion, we asked how
a change ΔX in he con ol a iable (X) would al e he cu en le el o ano he
con ol a iable (Y). The expe s i s ga e hei es ima es o he lowe and uppe
plausible alues and hen hei bes es ima e. In addi ion, a con idence in e al
(CI) o he p o ided es ima es was asked. The uppe and lowe plausible alues
desc ibe he limi s o an expe ’s CI; o example, assigning a 70% CI means ha
he expe belie es ha he e is a 70% p obabili y ha an in e ac ion s eng h alue
would all wi hin he in e al o he uppe and lowe alue, wi h he bes es ima e
as he mos likely alue. This o ma helps expe s o cons uc and con e hei
knowledge in o a quan i a i e o m19. Pa icipan s we e encou aged o p o ide
inpu only o he in e ac ions hey el bes i hei expe ise.
Agg ega ion o expe opinions. Commonly, applica ions o he IDEA p o ocol
use quan ile agg ega ion in de e mining he agg ega e alue o many indi idual
expe esponses19,71. Al hough e . 71 showed ha o he agg ega ion me hods could
lead o less o e con iden es ima es, hey also no e ha quan ile agg ega ion is
simple and in o ma i e in p o iding he bes es ima e wi hou en ailing addi ional
dis ibu ion assump ions. As ou main esul s ocus on he bes es ima es, we
deemed quan ile agg ega ion an app op ia e me hod o agg ega e he expe
esponses. Be o e he quan ile agg ega ion, we i expe esponses wi h a PERT
dis ibu ion78 o explo e egional di e gence among expe opinions. The PERT
dis ibu ion, which c ea es a smoo h dis ibu ion based on h ee pa ame e s— he
lowe , uppe and bes (mos likely) alue—is a sui able dis ibu ion o i when
ollowing he ou -s ep ques ion o ma ha speci ically asks o hese es ima es.
All expe answe s ha did no show subs an ial egional di e gence based on he
PERT dis ibu ion we e hen used o pe o m a quan ile agg ega ion o he lowe ,
bes and uppe alues wi h a CI o 80% (Supplemen a y Table 4.2), as desc ibed in
he IDEA p o ocol p ac ical guide19.
Calib a ing he expe s would ha e been a e y challenging endea ou , as da a
o a simila si ua ion om a known sys em a e no a ailable o ou knowledge.
Addi ionally, inding a single o ew es ques ions ha could i all expe
backg ounds would also ha e been di icul . One op ion ha we did no wan
o op in o was weigh ing expe s by hei academic posi ion, as i is a highly
ambiguous c i e ion o pe o mance. Thus, expe opinions we e agg ega ed
wi h an unweigh ed median o he PERT dis ibu ion i ed on he basis o he
indi idual esponses. The median was selec ed ins ead o he mean o minimize
he e ec o ou lie s when equally weigh ing small g oups19. Expe s could
also p o ide an example egion and ee- o m de ails in hei assessmen when
quan i ying an in e ac ion (Supplemen a y Sec ion 1.2). This possibili y was made
a ailable wi h he ini ial goal ha i su icien egional inpu became a ailable,
egion-speci ic in e ac ion es ima es could be in es iga ed. Howe e , egional
NATURE SUSTAINABILITY | VOL 5 | Oc Obe 2022 | 830–842 | www.na u e.com/na sus ain
838
