INTERNSHIP REPORT
Spa ial in e pola ion o Au oma ic Wea he
S a ion da a
GOYBET Thomas
In e nship supe iso : D . Ma hias Ba ay,
WSL Ins i u e o Snow and A alanche
Resea ch SLF (Da os, Swi ze land)
Uni e si y supe iso : P . Pa ick Rai oux,
Ins i u e o Ligh and Ma e (Lyon, F ance)
27/08/2025
”Bu since ou measu emen s and obse a ions a e no hing mo e han app oxima ions o he
u h, he same mus be ue o all calcula ions es ing upon hem” (Ka l F ied ich Gauss, 1809)
Abs ac
Spa ial in e pola ion algo i hms play a c ucial ole when wo king wi h Au oma ic Wea he S a-
ion (AWS) da a. Because AWS a e limi ed in numbe and expensi e o ins all in emo e moun-
ain egions, in e pola ion is necessa y o ex end hei measu emen s o loca ions wi hou di-
ec obse a ions. The pe o mance o he spa ial in e pola ion algo i hms di ec ly impac s he
quali y o he in e pola ed da a and he e o e he eliabili y o he esul s. In moun ain en i-
onmen s, howe e , hei pe o mance can be cons ained by complex opog aphy and mic o-
me eo ological condi ions. The quali y o he inpu da a, he numbe o AWS used, and hei
spa ial dis ibu ion in ele a ion and dis ance s ongly in luence he ou come o he in e pola-
ion. To p o ide guidance on how o choose and con igu e he algo i hms, a poin - o-poin
in e pola ion s udy was ca ied ou . The open-sou ce lib a y Me eoIO, de eloped by he WSL
Ins i u e o Snow and A alanche Resea ch SLF in Da os (Swi ze land), was used o pe o m
he in e pola ions. Based on he dense AWS ne wo k in he Da os egion and a cleaned da ase ,
a se o ecommenda ions has been es ablished. These ecommenda ions se e as guidelines o
suppo he se up o spa ial in e pola ions o AWS da a in new s udy a eas.
Key wo ds : Spa ial in e pola ion, Au oma ic Wea he S a ion (AWS), Me eoIO, INIshell, SNOW-
PACK, Me eo ological ield, Da os, SLF
Acknowledgemen s
Fi s o all, I wan o wa mly hank Ma hias o his supe ision all along he in e nship and o
he discussions we had om he i s in e iew o he las days o he in e nship. He ook he
ime o explain me he p ojec se e al imes, and o discuss he esul s I had and help me o
imp o e my wo k. I hope ou pa hs will c oss again in he u u e.
I also wan o hank he Snow P ocesses g oup o including me in he mee ings and gi ing me
some ad ice du ing he p esen a ions I did du ing some mee ings.
I wan o hank my o ice ma es, Anja and Se gi, wi h whom i was so com o able o wo k in
he same oom. I was always imp essed by hei p o essionalism on he subjec s hey handle,
and i makes me wan o be as good as he expe s hey a e one day .
A big hank you goes o all he mas e s uden s wi h whom I sha ed spo s momen s, lo s o
laughs and my hough s, so hank you Ma cel, Robin, Leah, Isa, bo h Yanniks, Al ed, Ida, Pia,
Alesch, A nod, Debo ah, No¨
el, Flo ian, Ch is, and all he o he s wi h whom we sha ed dinne s.
This Da os expe ience would no ha e been he same wi hou you.
A las , I wan o hank D . Alain Mi e o his ad ice be o e he in e nship, and he Mas e o
Ocean, A mosphe e and Clima e Sciences p og am ha allows me o di e in o one o my d eams
: wo king on Clima e Sciences in Alpine egions, whe e I lo e o spend ime.
1
CONTENTS CONTENTS
Con en s
1 In oduc ion 3
2 Me hods 4
2.1 Da aA ailabili y .................................... 4
2.2 Me eoIO......................................... 4
2.3 Wo k low ........................................ 5
2.4 How o se up he ecommenda ions . . . . . . . . . . . . . . . . . . . . . . . . . 6
3 Theo y 7
3.1 In e pola ionalgo i hm ................................ 7
3.1.1 Non physical in e pola ion algo i hm . . . . . . . . . . . . . . . . . . . . . 7
3.1.2 In e pola ion algo i hm o each pa ame e . . . . . . . . . . . . . . . . . 8
3.2 K-Combina ions..................................... 9
3.3 Co ela ion ac o R2.................................. 9
4 Resul s 10
5 Discussion 10
5.1 Spa ial in e pola ion algo i hm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.2 Discussion o each ecommenda ion . . . . . . . . . . . . . . . . . . . . . . . . . 12
5.2.1 Recommenda ion1............................... 12
5.2.2 Recommenda ion2............................... 12
5.2.3 Recommenda ion3............................... 14
5.2.4 Recommenda ion4............................... 15
5.2.5 Recommenda ion5............................... 15
6 Conclusion 16
A Appendix 17
A.1 IMIS Au oma ic Wea he S a ion . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
A.2 Wo k lowscheme.................................... 18
A.3 S udy case : Spa ial in e pola ion o a snowpack . . . . . . . . . . . . . . . . . . 18
B Re e ences 21
2
1 INTRODUCTION
1 In oduc ion
En i onmen al scien is s o en need some high quali y and high esolu ion me eo ological da a
o hei p ojec s o un physical models (Lis on & all, 2006)[1]. Bu i can be a challenge
ha is no easy o sol e. One way o ge ing some me eo ological da a may come om
poin measu emen s such as Au oma ic Wea he S a ions (AWS), and need o be spa ially
in e pola ed o he desi ed esolu ion. This way o ge ing me eo ological da a in Alpine
a eas was in es iga ed du ing his in e nship, wi h a ocus on he pe o mances o he spa ial
in e pola ion algo i hms in eg a ed in he SLF p oduc Me eoIO (Ba ay & Egge , 2014)[2].
Some examples o spa ial in e pola ion can be ound in a alanche p edic ions (Mo in & all,
2020) [3], (Mon i & all, 2016) [4], a alanche dange le el (Rich e & all, 2021)[5], clima e
p edic ions o snow making and g owing in mode n ski eso s (Hanze & all, 2020)[6],
hyd ological models (Ca le i & all, 2022)[7] o i e empe a u e u u e unde clima e change
in Swi ze land (Michel & all, 2022)[8].
As pa o he i s yea o he Mas e o Ocean, A mosphe e and Clima e Sciences p og am
suppo ed by Lyon 1 Uni e si y and Ecole Cen ale de Lyon in F ance, I did a 4-mon h in e n-
ship unde he supe ision o D . Ma hias Ba ay a he WSL Ins i u e o Snow and A alanche
Resea ch SLF loca ed in Da os, Swi ze land, wi hin he Snow and A mosphe e uni in he Snow
P ocesses g oup.
The goal o his in e nship is o p o ide a lis o ecommenda ions o he use s o nume ical
models who need o pe o m spa ial in e pola ion. To do so, I will be using he e y ich and
dense AWS ne wo k o Da os ha coun s 72 AWS and ha s a ed eco ding in 1975. This will
allow me o explo e many di e en con igu a ions o s a ion dis ibu ions in moun ain a eas.
The eade will ind he me hods applied o es ablish he ecommenda ions, he needed heo y
o do so, he esul sec ion wi h he 5 ecommenda ions ha we e ound, a discussion o he
esul sec ion, some ou looks in he conclusion sec ion and a s udy case in he appendix.
3
2 METHODS
2 Me hods
The me hod sec ion highligh s whe e he da ase is coming om, gi es some insigh s abou
Me eoIO, highligh s he chosen wo k low, and highligh s he way he ecommenda ions we e
deduced.
2.1 Da a A ailabili y
Da os has one o he mos dense ne wo ks o AWS in he wo ld, wi h mo e han 70 AWS in
he a ea composed o IMIS AWS and non IMIS AWS. In 1996, he SLF and he Swiss Moun ain
can ons s a ed o build IMIS wea he s a ions, which a e s anda dized AWS wi h a sola panel,
and senso s o measu e snow dep h, ai and su ace empe a u e, as well as wind speed and
di ec ion, ela i e humidi y, e lec ed sho wa e adia ion, g ound empe a u e, snow empe -
a u e a 25, 50 and 100 cm abo e he g ound, and, a mos s a ions, ain all using a ain gauge.
See a pic u e o i in appendix A.1. IMIS AWS iles a e gene ally well s uc u ed, wi h clea
owne ship. Al hough some me ada a may be missing, such as imp ecise coo dina es o unce -
ain senso heigh s, bu hese issues a e ela i ely mino . In con as , non-IMIS AWS p esen ed
a g ea e challenges: he e was no s anda diza ion o pa ame e names, many coo dina es
we e missing, la ge amoun s o da a we e o go en o inaccessible (some imes le on old ha d
d i es), some da ase s lacked any con ex ega ding hei o igin, and o en no con ac pe son
was a ailable.
A huge wo k has been done by D . Ma hias Ba ay o clean all he iles and make hem s an-
da dized. He p o ided a clean da ase o 72 AWS da a unde he .sme o ma . A .sme ile
is a s anda d ile o me eo ological da a composed o some me ada a and he da a i sel . The
me ada a p o ides use ul in o ma ion abou he AWS, such as he name o he s a ion, he local-
iza ion o he AWS gi en 2 coo dina e sys ems, he al i ude, he uni s, he name o he measu ed
ields, he abb e ia ion used o each ield, and as well some in o ma ion abou he da a owne
o he .sme ile. The .sme iles we e p epa ed using he open sou ce lib a y Me eoIO, which
will be in oduced in sec ion 2.2. The da ase is no ye published. The e o e, i is no p esen
in he epo .
2.2 Me eoIO
Me eoIO was published by Ba ay & Egge (2014)[2] and is an open-sou ce lib a y designed o
p ocess me eo ological inpu and ou pu da a iles. I is widely used bo h in esea ch, as in his
in e nship, and in ope a ional con ex s, o example by he a alanche o ecas ing eam a SLF.
One o i s key oles is o ans o m aw inpu da a in o il e ed, co ec ed, and s anda dized
da ase s. These e o s a e illus a ed o ins ance in Ba ay & all (2020)[9], and e lec s he
b idge om esea ch o ope a ional needs.
Many ea u es a e a ailable in his lib a y, bu only he spa ial in e pola ion ea u e was used in
his s udy, since he inpu da a a e al eady cleaned and ha he goal o he in e nship is o di e
in o he pe o mances o he spa ial in e pola ions. Each spa ial in e pola ion algo i hm has i s
de aul pa ame iza ion and can be changed i needed. In his s udy, he de aul pa ame iza ion
will be applied in o de o a oid in oducing any bias in o he algo i hms.
4
2.3 Wo k low 2 METHODS
Me eoIO eads .ini iles. We can ei he w i e he .ini iles and gi e hem o Me eoIO o use
INIshell (Ba ay & all, 2022)[10] which is a use - iendly in e ace o eading and w i ing .ini
iles. I is o pa icula help o he simplici y o na iga ion in Me eIO ea u es and hides he
complexi y o he model. In his s udy, INIshell was used o unde s and he s uc u e o he .ini
iles, o debug he wo k low and o launch Me eoIO simula ions.
To pe o m spa ial in e pola ions, Me eoIO equi es as inpu a Digi al Ele a ion Model (DEM)
o he s udy a ea, he e Da os. The DEM used in his wo k has a esolu ion o abou 10 me-
e s, which also de ines he esolu ion o he in e pola ions. Since he objec i e is o pe o m
poin - o-poin in e pola ions (see Sec ion 2.3), such a esolu ion is su icien . A ine DEM (1
o 2 me e s) would no signi ican ly imp o e he esul s bu would d as ically inc ease com-
pu a ional ime. Fo mos algo i hms, he in e pola ion is pe o med only o he DEM cell
con aining he e e ence AWS, wi h ele a ion, slope, and aspec aken om he DEM a he
han om he s a ion i sel . Fo algo i hms whe e he su ounding e ain plays a ole, such
as he LISTON WIND algo i hm o wind pa ame e s, he in e pola ion is ca ied ou o e he
whole DEM be o e ex ac ing he a ge cell.
In addi ion, his s udy elies on he i ual s a ion ( s a ion) ea u e o Me eoIO (Me eoIO
documen a ion)[19]. A s a ion is de ined as a poin o which Me eoIO econs uc s ime se ies
by il e ing and p ocessing he inpu da a, empo ally in e pola ing hem, and inally applying
he selec ed spa ial in e pola ion me hods. The use p o ides he coo dina es and ele a ion o
he i ual poin , as well as he pa ame e s o in e pola e, and Me eoIO ou pu s a .sme ile
con aining bo h me ada a and he in e pola ed da a. Beyond esea ch needs, s a ions a e also
applied in ope a ional con ex s, o example in a alanche wa ning sys ems (Mon i & all, 2016)
[4].
In his s udy, s a ions a e sys ema ically c ea ed a he exac loca ions o eal AWS, enabling
di ec compa ison be ween in e pola ed and obse ed alues. Only he spa ial in e pola ion ea-
u e will be used o econs uc he da a. This app oach is cen al o ou alida ion me hodology,
as i p o ides a obus assessmen o algo i hm pe o mance.
2.3 Wo k low
To achie e he objec i e o his in e nship, a wo k low was es ablished o ensu e ep oducibili y
o he me hod and o a oid andomness. The key idea is o apply a Lea e-One-Ou C oss-
Valida ion app oach: among he dis ibu ed AWS ne wo k, one s a ion is empo a ily excluded
om he in e pola ion and ea ed as a s a ion. I s coo dina es a e iden ical o hose o he
excluded e e ence AWS, which allows us o compa e in e pola ed alues agains ac ual mea-
su emen s a he same loca ion, see igu e 1. This se up p o ides a obus amewo k o e alua e
how close he in e pola ed alues a e o eal obse a ions, using he co ela ion ac o R2as a
pe o mance sco e o each spa ial in e pola ion. The compu a ion o R2will be de ailed in sec-
ion 3.3. Appendix A.2 shows a igu e o he wo k low ha was au oma ed. I s s eps would be
he ollowing :
1. Choose one me eo ological pa ame e : such as ai empe a u e, ela i e humidi y, adia-
ions, e c.
5
2.4 How o se up he ecommenda ions 2 METHODS
2. Choose a pool o s a ions o be used o he in e pola ion. The s a ion mus ha e he
chosen pa ame e , o he wise hey will no be used by he in e pola ion algo i hm. The size
o he pool o s a ions will inc ease o dec ease he compu a ional ime o he simula ion,
depending on he pool size.
3. Choose one s a ion : The s a ion mus be excluded om he s a ion pool used o pe o m
he in e pola ion algo i hm. O he wise, he esul s would be biased and appea a i icially
high, since many algo i hms end o p edic he local alue when a measu ed obse a ion
is al eady a ailable a he in e pola ion poin . The s a ion mus also con ain he chosen
pa ame e in i s .sme ile.
4. C ea e all he possible dis inc combina ions, ega dless o he o de o he s a ions, as i
will no change any hing in he in e pola ion. See sec ion 3.2.
5. W i e a .ini ile o each combina ion.
6. Con e he .sme iles om he in e pola ed da a and he e e enced s a ion o . x iles
wi h only he heade wi h he educed name o he ield and he da a.
7. Fil e all he no da a ma ked as -999.
8. Check ha he wo . x iles om he in e pola ed da a and he e e enced s a ion ha e he
exac same da e and ime s ep, so we can compa e he da a o a simila da e. The goal o
his s ep is o a oid compa ing day ime da a o nigh ime da a, o example.
9. Once he ime s eps a e he same be ween he s a ion . x ile and he e e ence s a ion
. x ile, we can compu e he co ela ion ac o R2 o he p edic ed da a and he eal da a.
10. Then we ha e a esul ile ull o R2as he co esponding .ini iles and we can s a o lea n
some ules.
The wo k low was coded wi h Py hon and looped o each combina ions o s a ions. The com-
pu a ional ime was high, up o ens o hou s o e en o days. In o de o educe i , he Py hon
code used was pa allelized on he CPUs o he compu e and a clus e wi h much mo e compu-
a ional powe called Hype ion [20] was used. This helped o educe he compu a ional ime
om hou s o minu es.
2.4 How o se up he ecommenda ions
To se up a ecommenda ion, some analysis was done on he esul iles a he end o he wo k-
low. Th ee plo s we e made and had he same s uc u e : The co ela ion ac o R2as a unc ion
o he numbe o s a ions, he di e ence be ween he mean al i ude o he selec ed s a ions o
he in e pola ion and he al i ude o he e e ence s a ion, and he mean dis ance o he g oup
o s a ions o he e e ence s a ion.
These plo s highligh ed he simula ions wi h low and high R2and helped o unde s and wha
he common poin s o hese simula ions we e. Once he common poin s we e ound, hey we e
used o exp ess ecommenda ions, which a e a ailable in sec ion 4.
6
3 THEORY
Vi ual s a ion
Inpu s a ion
Re e ence s a ion
Figu e 1: Illus a ion o spa ial in e pola ion a he s a ion. Me eo ological ields a e in e pola ed
o he loca ion o he s a ion, which sha es he same coo dina es as he e e ence s a ion, enabling
c oss alida ion be ween obse ed and in e pola ed da a.
3 Theo y
The heo e ical pa highligh s he in e pola ion algo i hms used du ing he in e nship, as well
as he K-combina ions ha we e used and how he co ela ion ac o is desc ibed and in e p e ed
in his s udy.
3.1 In e pola ion algo i hm
Depending on he chosen pa ame e , se e al spa ial in e pola ion algo i hms can be used, bu
some a e mo e cons ained han o he s and may lead o dependencies. A i s app oach was s a-
is ical, which means ha ega dless o he physics o he pa ame e , many spa ial in e pola ion
algo i hms we e used, and he one ha ga e he bes esul s was chosen. A second app oach
was o ake in o accoun he physics behind each pa ame e and show ha he in e pola ion al-
go i hm ha is he closes o i s physics will gi e highe co ela ion ac o s and should be used.
The in e pola ion algo i hms can be ound in Me eoIO documen a ion [19]. In his subsec ion, a
i s pa will be dedica ed o he non-physical in e pola ion algo i hm used, hen a second pa
o each in e pola ion algo i hm ha ma ches he physics o he gi en pa ame e . In his s udy,
he esul s o he second app oach a e p esen ed.
3.1.1 Non physical in e pola ion algo i hm
I is possible o use no in e pola ion algo i hm o a pa ame e by selec ing he NONE op ion.
This can be use ul when a pa ame e only needs o be se up as inpu o ano he in e pola ion
algo i hm.
7
3.1 In e pola ion algo i hm 3 THEORY
The NEAREST algo i hm assigns he alue o he nea es inpu s a ion o he conside ed
pa ame e o he i ual s a ion.
The a e age (AVG) algo i hm calcula es, o each ime s ep, he a e age alue o all inpu
s a ions and uses i as he ou pu alue.
The In e se Dis ance Weigh ing (IDW) algo i hm wo ks as desc ibed in equa ion 1:
yp ed =Pi1
dαui
Pi1
dα
(1)
whe e yp ed is he p edic ed alue, uiis he alue om s a ion ia he gi en ime s ep, dis he
dis ance be ween he e e ence s a ion and he conside ed s a ion, and αcon ols he weigh o
he dis ance. A la ge αgi es mo e in luence o nea by s a ions. Inside Me eoIO, he de aul
alue o αis 1. I a physical in e pola ion algo i hm ails due o o e ly s ic cons ain s, IDW
was used as a allback.
I is also wo h no ing ha hese s a is ical algo i hms can be applied no only di ec ly o he aw
alues bu also in combina ion wi h al i udinal de ending. In his case, an ele a ion-dependen
end (e.g. lapse a e) is i s compu ed and emo ed om he da a, he de ended alues
a e in e pola ed wi h he chosen algo i hm, and he end is hen e-applied o econs uc he
in e pola ed se ies. This app oach helps o be e cap u e al i ude-dependen p ocesses and can
imp o e in e pola ion pe o mance in moun ainous e ain.
3.1.2 In e pola ion algo i hm o each pa ame e
The in e pola ion algo i hms a e g ea ly inspi ed by he ones inside he Mic oMe model
de eloped by Lis on & all [1].
We know ha ai empe a u e TA changes in he al i ude wi h a lapse a e in he oposphe e.
The e o e, a LAPSE algo i hm will compu e a linea eg ession o he da a, de end he em-
pe a u e o sea-le el, hen compu e he in e pola ion algo i hm o his empe a u e whiche e
chosen spa ial in e pola ion algo i hm on he de end da a, hen e end he in e pola ed
empe a u e o he al i ude o he s a ion. This is a mo e physical app oach o in e pola ing
he empe a u e. Bu his can lead o some p oblems, see he discussion in sec ion 5.2.2.
Fo ela i e humidi y RH, he mos sui ed in e pola ion algo i hm is he one named LISTON RH.
This algo i hm i s de i es he dew poin empe a u e o he gi en RH, hen applies an in-
e pola ion algo i hm o he dew poin empe a u e, and he in e pola ed RH is again de i ed
om he in e pola ed dew poin empe a u e. The e o e, TA should be known and has o be
in e pola ed.
Fo p essu e P, he mos sui ed algo i hm is he s anda d a mosphe ic p essu e algo i hm
STD PRESS. This algo i hm compu es he p essu e a he desi ed ele a ion i no p essu e
is measu ed. I one s a ion measu es i , hen he o se be ween he local p essu e and he
s anda d a mosphe e p essu e is added o he compu ed p essu e o he desi ed ele a ion.
8
5.2 Discussion o each ecommenda ion 5 DISCUSSION
5.2.4 Recommenda ion 4
Recommenda ion 4 unde lines a o ing he la ges possible numbe o alid s a ions o imp o e
in e pola ion quali y. Bu he s a ions used mus ha e pass ecommenda ions 1, 2, and 3. Figu e
6(a) shows he e olu ion o he co ela ion ac o o di e en numbe s o s a ions o each
in e pola ion. As men ioned in sec ion 5.2.2, an in e pola ion wi h a lapse algo i hm should
ha e a leas 2 s a ions in inpu . Figu e 6(a) was il e ed wi h he pas ecommenda ions and
zoomed on 0.8 and 1.0 o he co ela ion ac o . We see as well ha 3 gi en s a ions o he
in e pola ion a e al eady enough o ha e a high pe o mance o he in e pola ion.
5.2.5 Recommenda ion 5
Recommenda ion 5 poin s ou limi ing he dis ance o s a ions used o a maximum o 25 km
om he in e pola ion poin . I comes om he limi a ion o he spa ial dis ibu ion o s a ions
in he da ase . All he s a ions a e in Da os a ea, he e o e no oo a om one ano he . Figu e
6(b) shows he maximum dis ance be ween he 2 u hes s a ions is 25km, and he sco e o
his in e pola ion is s ill high using a de ending algo i hm. So he dis ance h eshold is se a
25km. I migh be possible ha s a ions u he away om he in e pola ion poin wo k as long
as hey a e in moun ain a ea, bu i has o be alida ed.
Ne e heless, dis ance is a much mo e mino p oblem compa ed o al i ude. I is much mo e
impo an o check he al i ude dis ibu ion o he s a ions han he dis ance. Fo he same
se up as in igu e 6(a), we obse e in igu e 6(b) ha he dis ance has no impac on he sco es
dis ibu ion.
2 4 6 8 10
Numbe o s a ion
0.800
0.825
0.850
0.875
0.900
0.925
0.950
0.975
1.000
Co ele ion ac o R²
Co ela ion ac o as a unc ion o he numbe o s a ion Me hod : IDW_LAPSE
Indi iduel sco es
Mean
Mean + 3
Mean - 3
((a)) Co ela ion ac o as a unc ion o he numbe o
s a ions, zoomed on he ange 0.8–1.0 o R2.
10000 12000 14000 16000 18000 20000 22000 24000
Mean dis ance o he e e ence s a ion [m]
0.0
0.2
0.4
0.6
0.8
1.0
Co ela ion ac o R²
Co ela ion ac o as a unc ion o he mean dis ance o each simula ions o he e e ence s a ion Me hod : IDW_LAPSE
Indi idual sco es
Mean
Mean + 3
Mean - 3
((b)) Co ela ion ac o as a unc ion o dis ance o he
e e ence s a ion.
Figu e 6: Co ela ion ac o as a unc ion o he numbe o s a ions (a) and he dis ance o he e e -
ence s a ion (b). The in e pola ed pa ame e is he ai empe a u e and he in e pola ion algo i hm
used is IDW LAPSE. In bo h igu es, he blue do s ep esen he dis ibu ion o he sco es o each
numbe o s a ions, he ed poin s and line show he mean sco e, and he o ange a ea co esponds
o he mean alue ±3 imes he s anda d de ia ion.
15
6 CONCLUSION
6 Conclusion
Spa ial in e pola ions a e o e y g ea in o de o ge me eo ological da a whe e hey a e no
di ec ly measu ed by Au oma ic Wea he S a ions, in o de o make he mos ou o he al eady
a ailable AWS. This s udy showed ha spa ial in e pola ion pe o mance can be o high quali y
i some condi ions a e ul illed.
The key messages esul ing om his s udy a e ha o he ime pe iod o in e es , he quali y o
he inpu da a should be o high quali y, and ha he ele a ion dis ibu ion o he s a ions should
be wi hin some al i ude anges. The dis ance o he loca ion o in e pola ion has less e ec on
he pe o mances o he algo i hms han he ele a ion. High pe o mances o spa ial in e po-
la ion we e ound o ai empe a u e, ela i e humidi y, and sho and long wa es adia ions.
P ecipi a ion and wind ields a e s ill esea ch opics o ackle and may equi e o he echniques
o assess hem, such as machine lea ning and deep lea ning. I is also impo an o no ice ha
Da os is e y pa icula o he unusually la ge numbe o wea he s a ions ha ha e been se
up and a e s ill unning. In o he places, whe e he densi y o AWS is less, he ecommen-
da ions will help o choose an in e pola ion algo i hm and lead o high quali y in e pola ed da a.
The pu pose o highligh ing some ecommenda ions is o use hem in places whe e he e a e
ewe Au oma ic Wea he S a ions. The e o e, i would be e y use ul o apply he ecommen-
da ions somewhe e whe e he e a e ewe AWS and check i we can imp o e he quali y o he
in e pola ions o alida e wha was al eady done. These ecommenda ions can be also use ul
in mo e emo e a eas, in Alpine egions o in pola egions whe e i is ha d o se up la ge
AWS domain and whe e i is c ucial o ge high quali y da a o be e assess Clima e Change.
As well, since he SLF has a pu pose o wa ning and p e en ing a alanche na u al haza ds, i
would be in e es ing o y hese ecommenda ions o modelling snow co e p ope ies ela ed
o a alanche dange .
16
A APPENDIX
A Appendix
A.1 IMIS Au oma ic Wea he S a ion
Figu e 7: IMIS snow s a ion Belalp (can on Valais) a 2556 m: The ou poles isible in on o he
s a ion ca y a ence o p o ec he snow empe a u e senso s om wild animals and li es ock in he
summe pe iod. C edi s : h ps://www.sl .ch/en/a alanche-bulle in-and-snow-si ua ion/
measu ed- alues/desc ip ion-o -au oma ed-s a ions/
17
A.2 Wo k low scheme A APPENDIX
A.2 Wo k low scheme
Figu e 8: Scheme o he wo k low.
A.3 S udy case : Spa ial in e pola ion o a snowpack
In o de o e alua e he ecommenda ions es ablished in his wo k, spa ial in e pola ions o
selec ed me eo ological pa ame e s we e pe o med o he win e 2023–2024, ollowed by an
a emp o simula e he snowpack e olu ion.
The simula ions we e ca ied ou wi h he open-sou ce SNOWPACK model, de eloped by he
WSL Ins i u e o Snow and A alanche Resea ch SLF (Lehning & all, 1999) [16]. SNOWPACK
is designed o simula e he e olu ion o he snow co e and he g ound su ace based on
me eo ological da a (ai empe a u e, ela i e humidi y, wind speed, p ecipi a ion, snow heigh ,
adia ions, g ound empe a u e). I models snow mic os uc u e and he in e ac ions be ween
snow, he a mosphe e, he soil, and he snowpack i sel . I is widely used o bo h ope a ional
applica ions as o a alanche o ecas ing (Lehning & all, 1999) [16] and esea ch pu poses
(Oblei ne & Lehning, 2004)[17].
The pu pose o his s udy case was no o conduc a dep h analysis o SNOWPACK i sel ,
bu a he o es whe he a ealis ic snowpack could be econs uc ed o e one win e using
spa ially in e pola ed da a. Fo his, I used he example case p o ided in he SNOWPACK
documen a ion, which con ains an .ini con igu a ion ile and inpu da a om a e e ence
s a ion. The se up o he .ini ile would no be modi ied, excep o he inpu s a ions and
he spa ial in e pola ion algo i hms. The e e ence s a ion chosen is loca ed a Weiss luhjoch
(Da os, 2556m ele a ion), a he SLF measu emen si e. The simula ion was un o he
2023–2024 win e season. Based on he ecommenda ions es ablished du ing his in e nship,
12 su ounding AWS we e selec ed o p o ide in e pola ed inpu o ai empe a u e, ela i e
humidi y, and e lec ed sho wa e adia ion. Incoming sho wa e adia ion was de i ed om
e lec ed sho wa e alues and he albedo, while incoming longwa e adia ion could no
be in e pola ed due o insu icien co e age. Ins ead, he gene a o ea u e o Me eoIO was
18
A.3 S udy case : Spa ial in e pola ion o a snowpack A APPENDIX
applied, using he ALL SKY op ion and he Unswo h model (Me eoIO documen a ion) [19].
O he equi ed me eo ological pa ame e s such as wind speed, snow heigh , and g ound
empe a u e we e aken di ec ly om he e e ence s a ion. Fi s he in e pola ion would
ake place in Me eoIO, hen his ile would be me ged wi h he e e ence ile wi h he missing
pa ame e s (wind speed, snow heigh , p ecipi a ions). Ini ial esul s showed missing laye s
in he simula ed p o ile, which eappea ed once liquid p ecipi a ion da a om he e e ence
s a ion we e added. Figu es 9(a) and 9(b) p esen he compa ison be ween he e e ence
snowpack and he in e pola ed one.
((a)) Snowpack o e e ence
((b)) Snowpack in e pola ed
Figu e 9: Compa ison be ween a e e ence snowpack and an in e pola ed snowpack o win e
2023–2024 using he SNOWPACK model. The le y-axis shows he snow g ain ypes ( ed = mel ing
snow, ed-black line = mel - eeze c us , g een = new snow, pink = ounded g ains, and he blue
= ace ed c ys als, see Fie z & all, 2009 [18]), while he igh y-axis indica es snow dep h. The
x-axis ep esen s he ime pe iod om 01/09/2023 o 01/09/2024. Panel (a) displays he snowpack
simula ed wi h measu emen s om Weiss luhjoch, whe eas panel (b) shows he snowpack simula ed
wi h in e pola ed ai empe a u e, ela i e humidi y, and e lec ed sho wa e adia ion, combined
wi h measu ed snow heigh , wind speed, g ound empe a u e, and p ecipi a ion a Weiss luhjoch.
The esul s indica e ha he gene al snowpack s uc u e, om he i s snow p ecipi a ions o
he las mel , he heigh o he snow including i s main laye s, was ep oduced easonably well.
Howe e , some laye s we e missing, likely due o small disc epancies in in e pola ed empe -
a u e o humidi y a key momen s, p e en ing he o ma ion o ce ain s a ig aphic ea u es.
19
A.3 S udy case : Spa ial in e pola ion o a snowpack A APPENDIX
This highligh s ha , e en when in e pola ion pe o mance is high, ine-scale de ails may s ill
be los . Fu he mo e, he quali y o wind speed, snow heigh , and p ecipi a ion da a p o ed
essen ial, as hese a iables s ongly a ec p ocesses such as e eezing wi hin he snowpack.
Missing o inaccu a e alues led di ec ly o missing laye s. This unde lines he impo ance o
p o iding high-quali y inpu da a o wind, p ecipi a ion, and snow heigh when econs uc -
ing snowpack p o iles, which was a limi ing ac o in his s udy ocused on he pe o mance o
spa ial in e pola ions.
20
B Re e ences
[1] Lis on, G.E. & Elde , K., 2006. A me eo ological dis ibu ion sys em o high-
esolu ion e es ial modeling (Mic oMe ). Jou nal o Hyd ome eo ology, 7(2),
pp.217–234. h ps : / / jou nals . ame soc . o g / iew / jou nals / hyd / 7 / 2 /
jhm486_1.pd pages 3, 8
[2] Ba ay, M. and Egge , T.: Me eoIO 2.4.2: a p ep ocessing lib a y o me eo ological
da a, Geosci. Model De ., 7, 3135–3151, h ps://doi.o g/10.5194/gmd-7-3135-
2014, 2014. pages 3, 4, 10
[3] Samuel Mo in, Simon Ho on, F ank Techel, Ma hias Ba ay, C´
ecile Col´
eou, Cha les
Fie z, And eas Gobie , Pascal Hagenmulle , Ma hieu La aysse, Ma jaˇ
z Liˇ
za ,
Ch is oph Mi e e , Fabiano Mon i, Ka s en M¨
ulle , Ma c Ole s, John S. Snook, Alec
an He wijnen, Vincen Vionne (2020). Applica ion o physical snowpack models in
suppo o ope a ional a alanche haza d o ecas ing: A s a us epo on cu en im-
plemen a ions and p ospec s o he u u e, Cold Regions Science and Technology.
Cold Regions Science and Technology, 170, 102910. h ps://doi.o g/10.1016/j.
cold egions.2019.102910 pages 3
[4] Mon i, F., Mi e e , C., S einkogle , W., Ba ay, M. & Pozzi, A., 2016. Combining snow-
pack models and obse a ions o be e a alanche dange assessmen s. P oceed-
ings o he In e na ional Snow Science Wo kshop (ISSW), B ecken idge, Colo ado,
pp.343–347. h ps://a c.lib.mon ana.edu/snow-science/objec s/ISSW16_
O15.03.pd pages 3, 5
[5] Rich e , B., Schweize , J., Ro ach, M.W. & an He wijnen, A., 2021. Mod-
eling spa ially dis ibu ed snow ins abili y a a egional scale using Alpine3D.
Jou nal o Glaciology, 67(266), pp.1147–1162.h ps://www.camb idge.o g/
co e / jou nals / jou nal - o - glaciology / a icle / modeling - spa ially -
dis ibu ed - snow - ins abili y - a - a - egional - scale - using - alpine3d /
D8114EE9B45642A32BE7F06D70980802 pages 3
[6] Hanze , F., Ca magnola, C.M., Ebne , P.P., Koch, F., Mon i, F., Ba ay, M., Be nha d ,
M., La aysse, M., Lehning, M., S asse , U., F anc¸ois, H. & Mo in, S. (2020). Simu-
la ion o snow managemen in Alpine ski eso s using h ee di e en snow models.
Cold Regions Science and Technology, 172, 102995. h ps://doi.o g/10.1016/j.
cold egions.2020.102995 pages 3
[7] Ca le i, F., Michel, A., Casale, F., Bu i, A., Bocchiola, D., Ba ay, M., and Lehning,
M.: A compa ison o hyd ological models wi h di e en le el o complexi y in Alpine
egions in he con ex o clima e change, Hyd ol. Ea h Sys . Sci., 26, 3447–3475,
h ps://doi.o g/10.5194/hess-26-3447-2022, 2022. pages 3
21
[8] Michel, A., Schae li, B., We e , N., Zekolla i, H., Lehning, M., and Huwald, H.:
Fu u e wa e empe a u e o i e s in Swi ze land unde clima e change in es iga ed
wi h physics-based models, Hyd ol. Ea h Sys . Sci., 26, 1063–1087, h ps://doi.
o g/10.5194/hess-26-1063-2022, 2022. pages 3
[9] Ba ay, M, e al. 2020. Au oma ic Da a S anda diza ion o he Global C yosphe e
Wa ch Da a Po al. Da a Science Jou nal, 19: 6, pp. 1–8. h ps://doi.o g/10.
5334/dsj-2020-006 pages 4
[10] Ba ay, M., Reisecke , M., Egge , T., and Ko hamme , D.: Inishell 2.0: seman i-
cally d i en au oma ic GUI gene a ion o scien i ic models, Geosci. Model De ., 15,
365–378, h ps://doi.o g/10.5194/gmd-15-365-2022, 2022. pages 5
[11] Duja din, J. e Lehning, M. (2022) ’Wind-Topo : Downscaling nea -su ace wind
ields o high- esolu ion opog aphy in highly complex e ain wi h deep lea ning’,
Qua e ly Jou nal O The Royal Me eo ological Socie y, 148(744), p. 1368-1388.
h ps://doi.o g/10.1002/qj.4265 pages 11
[12] K alse h, T. O. (1985). Cau iona y No e abou R 2. The Ame ican S a is ician, 39(4),
279. h ps://doi.o g/10.2307/2683704 pages 9
[13] A anzi, F., E colani, G., Gabellani, S., C emonese, E., Poglio i, P., Filippa, G., Mo a
di Cella, U., Ra o, S., S e enin, H., Caudu o, M., and Juglai , S.: Lea ning abou
p ecipi a ion lapse a es om snow cou se da a imp o es wa e balance modeling,
Hyd ol. Ea h Sys . Sci., 25, 2109–2131, h ps://doi.o g/10.5194/hess-25-
2109-2021, 2021. pages 11
[14] La son, L. W. and Peck, E. L.: Accu acy o p ecipi a ion measu emen s o hyd ologic
modeling, Wa e Resou ces Resea ch, 10, 857–863, h ps://doi.o g/10.1029/
WR010i004p00857, 1974 pages 11
[15] Ba ay, M., We e , N., Fie z, C., Lehning, M. (2024, Sep embe 30). Looking back a
he las 15 yea s o ope a ional a alanche wa ning wi h he SNOWPACK model in
Swi ze land. In e na ional Snow Science Wo kshop 2024 (ISSW), T omsø. h ps:
//doi.o g/10.5281/zenodo.13861453 pages 11
[16] Lehning, M., Ba el , P., B own, R.L., Russi, T., S ¨
ockli, U., Zimme li, M., Snowpack
Model Calcula ions o A alanche Wa ning based upon a new Ne wo k o Wea he
and Snow S a ions, 1999, Cold Reg. Sci. Technol., 30, 145-157 h ps://doi.o g/
10.1016/S0165-232X(99)00022-1 pages 11, 18
[17] Oblei ne , F., and M. Lehning (2004), Measu emen and simula ion o snow and
supe imposed ice a he Kongs egen glacie , S alba d (Spi zbe gen), J. Geophys.
Res., 109, D04106, h ps://doi.o g/10.1029/2003JD003945 pages 18
[18] Fie z, C., A ms ong, R.L., Du and, Y., E che e s, P., G eene, E., McClung, D.M.,
Nishimu a, K., Sa yawali, P.K. and Sok a o , S.A. 2009. The In e na ional Classi ica-
ion o Seasonal Snow on he G ound. IHP-VII Technical Documen s in Hyd ology
N°83, IACS Con ibu ion N°1, UNESCO-IHP, Pa is. h ps://c yosphe icsciences.
22
o g/wp-con en /uploads/2019/02/snowclass_2009-11-23- agged-high es.
pd pages 19
[19] Me eoIO documen a ion |h ps : / / me eoio . sl . ch / doc - elease / h ml /
in e pol2d.h ml pages 5, 7, 19
[20] Hype ion documen a ion |h ps://hype ion.wsl.ch/documen a ion-mas e /
build/h ml/ pages 6
23
