HAL Id: hal-01191676
https://hal.inria.fr/hal-01191676
Submitted on 2 Sep 2015
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entific research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
Determination of phytoplankton groups from ocean
color spectral measurements in the
Senegalo-Mauritanian upwelling
Ymane Taoufiq, Ousmane Farikou, Séverine Alvain, Julien Brajard, Michel
Crépon, Malick Ngom, Sylvie Thiria
To cite this version:
Ymane Taoufiq, Ousmane Farikou, Séverine Alvain, Julien Brajard, Michel Crépon, et al.. Determi-
nation of phytoplankton groups from ocean color spectral measurements in the Senegalo-Mauritanian
upwelling. Ocean Optics, The Oceanography Society (TOS), Oct 2014, Portland, Maine, United
States. �hal-01191676�
o ean olor spe tral measurements in the
Senegalo-Mauritanian upwelling
Ymane Taouq 1
, Ousmane Farikou 2
, Séverine Alvain 3
, Julien
Brajard 1,4
, Mi hel Crépon 1
, Mali k Ngom 1
, and Sylvie Thiria 1
1
Sorbonne Universités (UPMC, Univ Paris 06), CNRS, IRD, MNHN,
LOCEAN Laboratory, IPSL, 4 pla e Jussieu, F-75005, Paris, Fran e
2
Université Cheikh Anta Diop, Dakar, Sénégal
3
Laboratoire d'O eanologie et de Geos ien es, CNRS-ULCO-USTL,
Wimereux, Fran e
4
INRIA, Ro quen ourt, Fran e
20 o tobre 2014
1 Introdu tion
O ean olormeasurementshavebeenintensivelyusedtoestimate hlorophyll-a
on entration (Chl-a in abreviation) in the surfa e waters of the o ean, marginal
seas and lakes.
Phytoplankton is the rst element in the o ean food webs and onsequently
drivestheo eanprodu tivity.Italsoplaysafundamentalrolein limateregulation
by trappingatmospheri CO2through gas ex hanges atthe sea surfa e. Withthe
growing interest in limate hange, one may ask how the dierent phytoplankton
populationswillrespondto hangesino ean hara teristi s(temperature,salinity)
and nutrient supply.
Pigment analysis by High Performan e Liquid Chormatography (HPLC) has
been widelyused to ategorizedbroadPhytoplan ton size lasses (PSC)orphyto-
plan ton fun tional types (PFT) [Hirata et al., 2011℄. Ea h phytoplan ton group
(PSC/PFT) is asso iated with diagnosti pigmentsand a onvertion formula an
ments.
Thesein-situmeasurementswereusedtobuildrelationshipsbetweenPSC/PFT
and o ean properties that an be derived from satellite o ean olor sensors (e.g.
Chl-a on entration or water leaving-radian e), whi h is of fundamental interest
tounderstand the phytoplanktonbehavior andto modelitsevolution [Uitzetal.,
2006,Ciotti and Bri aud, 2006,Hirata et al., 2008,Sathyendranath et al., 2014,
Alvain etal.,2005,Alvainet al.,2012℄.
In the present work, we propose a regional algorithm based on PHYSAT [Al-
vain et al., 2012℄, that estimates diagnosti pigments asso iated with PFT/PSC
fromsatelliteo ean olormeasurements.Theregionofappli ationisthesenegalo-
mauritean upwelling and the results fo used on the relative on entration of Fu-
oxanthin (Fu o) whi h is the main diagnosti pigment for Mi roplankton (
>
20µm
)and Diatoms.2 Data
2.1 Satellite dataset
For this study, a satellite image ar hive of the senegalo-mauritean upwelling
(
8 o N
-24 o N
,14 o W − 20 o W
) obtained fromthe radiometerSeaWiFS wasused andthe year 2003 was hosen to be atest ase of the algorithm.
Ea h data is a daily image of the water-leaving ree tan es (
ρ w
) at four wa-velengths (443nm, 490nm, 510nm and 555nm) and of the on entration of Chl-a
duringtheyear 2003.Theradian e at412nmwasnot retainedbe auseof thehigh
sensitivity of
ρ w
to olored dissolved organi matter (CDOM) at this wavelength.Due tothe presen e ofsaharandusts inthis region,very fewestimations of
ρ w
andChl-awere availableanditmaylead tostrongover-estimationsof hlorophyll-
a [Gregg and Casey, 2004℄. For that reason, an atmopsheri orre tion algorithm
dedi ated tosaharan dust [Dioufet al.,2013℄ wasused to obtaina urate
ρ w
andhl-adata.
As in [Ben Mustapha et al., 2014℄, the ree tan e ratio for ea h pixel was
omputed as follow:
Ra(λ) = ρ w (λ)/ρ wref (λ, chl − a)
(1)The on entration
ρ wref
depends on the Chl-a on entration only.ρ wref
wasal ulated for Chl-a values observed by SeaWiFS in the studied region using a
multilayer per eptron whi h is a lass of arti ial neural network able to model
any non linear fun tion. This is a dieren e ompared to [Ben Mustapha et al.,
2014℄ who used tabulated values. This permits to have a smoother
ρ wref
fun tioneven if the dependen y between
ρ wref
and Chl-a is not linear.The satellite dataset made of the
Ra(λ)
during the year 2003 is thereafterdenoted DSAT.
2.2 The pigment dataset
Phytoplanktonpigmentsare ommonlyusedtodis riminatePSCandPFT[Hi-
rata etal., 2011℄.The strong hypothesis madein this work isthat the orrelation
between the satellite measurement (
ρ w
, Chl-a) and pigment on entrations is not dependantonthelo ationorthedateofthemeasurement.Itmeansthatifasatel-litemeasurement anbeasso iatedwithapigment on entrationinoneparti ular
pla e, the asso iationmust stay relevantanywhere and atanytimein the o ean.
Forthatreason,itwasde ided touse alargeinsitu dataset ompiledatglobal
s ale during the whole SeaWiFS period. This dataset was ollo ated with the
ρ w
and Chl-a measured by SeaWiFSdata [BenMustaphaetal., 2014℄.Some missing
data was ompleted using aself-organizing map te hni s [Junninen etal., 2004℄.
The pigment dataset, denoted DPIG, is omposed of 1068 variables. Ea h va-
riable isa 10-dimensionalve tor dened as :
Component 1: hlorophyll-a on entration
Component 2: divinyl hlorophyll-a on entration ratio
Component 3: peridin on entration ratio
Component 4: fu oxanthin on entrationratio
Component 5: 19'hexanoyloxyfuxanthine on entration ratio
Component 6: zeaxanthin on entrationratio
Component7to10: SeaWiFSRaat4wavelengths : 443nm,490nm,510nm
and 555nm.
Thepigmentratioaredenedasin[Alvainetal.,2005℄bynormalizingthepigment
on entration by the Chl-a and divinyl hlorophyll-a on entration.
3 Method
The algorithmwasdivided intwophases.Therst phase onsistsin lustering
the DPIG dataset to retrieve the link between the ree tan e ration
Ra
and thepigment on entrationratio.The se ondphase onsistsinlabelingthe ree tan es
in DSAT interm of asso iated pigments.
The lusteringofDPIGwasdoneusingSelf-OrganizingMaps(herafterdenoted
SOM). The SOM [Kohonen, 1994℄ algorithmis a powerful non-linear lassier. It
aims at lusteringsamples ofamultidimensionaldataset(inour ase, DPIG) into
lasses represented by adedi ated network (the so- alledSOM map).
SOM is a neural lassier where ea h neuron of the map is asso iated with a
parti ularreferentve tor
V k
.ThedierentneuronsogtheSOMmapare onne tedtogether and determine a topologi al (neighbourhood) relationship between the
dierentneurons (subset ofsimilar data).
In the present ase, the SOM map is a two-dimensional (
13 × 12
) grid thatrepresents the partition of the DPIG dataset. Ea h lass is asso iated with a so-
alled referent ve tor
V k
(k = {1, 2, ...180}
).V k
are al ulated by a weigted meanofelementsinDPIG. Therefore,
V k
has thesame dimensionasea helementofthedataset(inour ase
V k ∈ R 10
)and ontains 5relativepigment on entrationratio, the remote sensing hlorophyll-a on entration andRa
at4 wavelengths.At the end of the lustering, ea h element of the dataset DPIG is asso iated
withthe referent
V k
(denotedthe BestMat hing Unit)whi histhe losestintermof the eu lidien distan e.
3.2 Labelisation
The labelisation phase onsists in asso iating ea h element of DSAT with a
pigment on entration ratio.
Ea h pixel
P j
of DSAT ontains the SeaWiFSRa
at the four wavelengths (443nm, 490nm, 510nm and 555nm). These elements an be dire tly omparedwith the omposants 7 to10 of the
V k
ve tor that representsRa
value inDPIG.The problem is thus to determine the Best Mat hing Unit
V k
using only fourvalues amongthe10usedintheSOM map.Atrun ateddistan e (TD)that onsi-
dereonlytheexistingvalueswasused.TheBestMat hingUnit
V k
wasdeterminedusing the TD.
Then,thepixel
P j
isdire tlyasso iatedwiththevaluesofthepigments on en-tration ratioof
V k
.With this method, ea h pixel
P j
is asso iated with 5 pigment on entration ratios. The underlying assumption is that the link between ree tan es and pig-ment ratios is the same in DPIG than for DSAT. In the following, we fo us on
the fu oxanthin on entration ratio whi h isa ara teristi ofdiatoms and mi ro-
planktons.
Chloro−a
n 0.0609
0.686
1.31
Rdiva
n 0.0103
0.162
0.313
Rperid
n 0.00915
0.0443
0.0794
Rfuco
n 0.0336
0.141
0.248
R19HF
n 0.0513
0.17
0.288
Rzeax
n 0.00738
0.179
0.351
Ra(443)
n 0.251
0.78
1.31
Ra(490)
n 0.434
0.868
1.3
Ra(510)
n 0.431
0.805
1.18
Ra(555)
n 0.383
0.955
1.53
Fig. 1 Representation of the value of the 7 omposants of
R k
on the Self-Organizing Map. Ea h image represent a omposant (6 pigments and 4
Ra
, ea hnode ofthe imagerepresent a lass. Here the dimension of the map is
13 × 12
4.1 Labelisation of the ree tan e spe tra
In this se tion, the asso iation between
Ra
spe tra and the pigment on en-tration ratioare presented.
First,inFig.1,weshowthevaluesofallthe
V k
omponentsofthe13×12
SOMmap. Ea h omponent was represented by the olor intensity of the grid point.It
an be noti e that values of the
V k
omponents were spatially well stru tured onSOM. Anotherimportantremark isthatthe values ofea h omponenthas alarge
range of variation of the same order as the range of variation of DPIG. It means
that the SOM map has aptured most of the variability of the dataset.
As ea h
V k
ontains a value for the pigment on entration ratio, it is possible toestimateseveral typi alindexof PSC orPFT.It isanimportantfeature ofthisalgorithmthat doesnot estimateone parti ularPSC/PFT but asso iatesdire tly
pigment on entration ratiosthat an be used asproxies.
As an illustrationand a validationof the approa h, we an ompute the per-
entage of mi roplankton(Mi ro) following the formula[Hirata etal., 2011℄ :
Micro = 1.41 × (F uco + P erid)
(2)where
F uco
(resp.P erid
) denotes the on entration ratio of Fu oxanthin (resp.Peridin).
In Fig. 2, we represent the per entage of mi roplankton ( al ulated using 2
with respe t with the hlorophyll-a on entration obtained for the referent ve -
tors
V k
. The relationshipbetween the Chl-a on entration and the Mi roplan ton per entage is onsistent with the relationship found in [Hirata etal.,2011℄ :Micro = [a 0 + exp(a 1 log 10 (Chl − a) + a 2 )] − 1
(3)where
a 0 = 0.9117
,a 1 = −2.7330
anda 2 = 0.4003
.We annoti ethat,in omparisonwiththisglobalrelationship,theregionalre-
lationshipfound usingSOM overestimatesthemi roplanktonper entageforsmall
value of Chl-a (
< 0.2mg/m3
). It would need further analyses to nd if it is anartifa t of the algorithmor aregional spe i ity.
This is a rst validation of the orrelations found in the SOM map, and it
demonstrates the potentiallity of the approa h. The asso iation between remote
sensingree tan e spe tra andpigment on entrationsisane ient way toiden-
tify phytoplankton groups.
4.2 Labelisation of images
Usingthe trun ateddistan e (TD) des ribed inthe previous se tion,itis pos-
sible to asso iate a pixelof animage to areferent
V k
and thus to allthe pigmenton entration ratios. In this work, we present results of this asso iation with the
Fu oxanthin. AllpixelsfromDSATwere asso iatedwiththe Fu oxanthin on en-
tration ratio. In g 3, the mean ratio for ea h sequen e of 3 months is presented
(January toMar h JFM, April toJune AMJ, July to Septembre JAS and nally
O tober toDe ember OND).
We an observethe seasonalvariability of the asso iated Fu oxanthin on en-
tration ratio, with a maximum on entration and a southernmost extent at the
beginning of the year. Knowing that high Fu oxanthin on entration ratio are
hara teristi of the presen e of diatoms, this seasonal variability is onsistent
with previous studies of the senegalo-mauritanean upwelling variability [Farikou
10 −2 10 −1 10 0 10 1
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Mikroplankton proportion
Chl−a concentration (mg/m3)
Fig. 2 Per entage of Mi roplankton following Eq. 2 with respe t with the
hlorophyll-a on entration. Ea h ir le represents a referent ve tor of the SOM
map and the solid bla k-line represents the relationship in Eq.3
et al., 2013,Lathuilière et al., 2008℄, and the observations done during the EU-
MELI ruises [Claustre and Marty, 1995℄ and the Atlanti Meridional Transe
(AMT) [Aiken etal.,2009℄.
5 Con lusion
A regional lassi ation te hnique derived from [Ben Mustapha et al., 2014℄
that asso iates ree tan e spe tra with pigment on entration ratio was develop-
ped andtestedduringthe year2003inthe senegalo-mauritanianandtheFu oxan-
thin on entration ratio. It was shown that results was oherent with relations
found in literature. It also allows to retrieve the seasonal variability of the as-
so iated Fu oxanthin on entration ratio. It is thus possible to have signi ant
several period : January to Mar h JFM, April to June AMJ, July to Septembre
JAS and nallyO tober toDe ember OND
quantitativeindi es ( on entration ratio) ofphytoplankton groups.
This approa h gives a lot of opportunity to study variability of pytoplankton
groups (PSC or PFT) in the region of the senegal-mauritanean upwelling. This
study ouldbe ontinued forother lassi ationgroupsandvalidatedusingin-situ
data in this region.
A knowledgement
Theauthorsa knowledgeNASA/GSFC/DAACforprovidinga esstodailyL2
SeaWiFSprodu ts.Thisworkwasfundedbythefren hspa eagen yCNES/Tos a
program.
[Aiken et al.,2009℄ Aiken,J.,Pradhan, Y.,Barlow, R.,Lavender, S.,Poulton,A.,
Holligan,P.,andHardman-Mountford,N. (2009). Phytoplanktonpigmentsand
fun tionaltypes in the atlanti o ean :a de adal assessment, 19952005. Deep
Sea Resear h Part II : Topi alStudies in O eanography, 56(15) :899917.
[Alvain etal.,2005℄ Alvain, S., Moulin, C., Dandonneau, Y., and Bréon, F.-M.
(2005). Remote sensing of phytoplankton groups in ase 1 waters from global
seawifs imagery. Deep Sea Resear h Part I : O eanographi Resear h Papers,
52(11):19892004.
[Alvain etal.,2012℄ Alvain, S., Vantrepotte,V., Uitz,J.,and Duforêt-Gaurier, L.
(2012). Use of global satellite observations to olle t information in marine
e ology. Sensors for e ology, page 227.
[Ben Mustaphaet al.,2014℄ Ben Mustapha, Z., Alvain, S., Jamet, C., Loisel, H.,
andDessailly,D. (2014).Automati lassi ationofwater-leavingradian eano-
maliesfromglobalseawifsimagery:Appli ationtothedete tionofphytoplank-
tongroupsinopen o eanwaters. Remote Sensingof Environment,146:97112.
[Ciotti and Bri aud, 2006℄ Ciotti, A. M. and Bri aud, A. (2006). Retrievals of
a size parameter for phytoplankton and spe tral light absorption by olored
detritalmatterfromwater-leavingradian esatseawifs hannelsina ontinental
shelfregion o brazil. Limnologyand O eanography : Methods, 4 :237253.
[Claustre and Marty, 1995℄ Claustre, H. and Marty, J.-C. (1995). Spe i phy-
toplankton biomasses and their relation to primary produ tion in the tropi al
north atlanti . Deep Sea Resear h Part I : O eanographi Resear h Papers,
42(8):14751493.
[Diouf etal., 2013℄ Diouf, D., Niang, A., Brajard, J., Crepon, M., and Thiria, S.
(2013).Retrievingaerosol hara teristi sandsea-surfa e hlorophyllfromsatel-
liteo ean olormulti-spe tralsensorsusinganeural-variationalmethod.Remote
Sensing of Environment, 130 :7486.
[Farikou etal.,2013℄ Farikou, O., Sawadogo, S., Niang, A., Brajard, J., Mejia,
C., Crépon, M., and Thiria, S. (2013). Multivariate analysis of the senegalo-
mauritanianareaby mergingsatelliteremotesensingo ean olor and sstobser-
vations. Resear h Journal of EnvironmentalandEarh S ien es,5(12):756768.
[Gregg and Casey, 2004℄ Gregg, W. W. and Casey, N. W. (2004). Global and
regionalevaluation of the seawifs hlorophyll data set. Remote Sensing of En-
vironment,93(4) :463479.
[Hirata etal., 2008℄ Hirata, T., Aiken, J., Hardman-Mountford, N., Smyth, T.,
and Barlow, R. (2008). An absorption model to determine phytoplankton
112(6):31533159.
[Hirata etal., 2011℄ Hirata, T., Hardman-Mountford, N., Brewin, R., Aiken, J.,
Barlow, R., Suzuki, K., Isada, T., Howell, E., Hashioka, T., Nogu hi-Aita, M.,
etal.(2011).Synopti relationshipsbetweensurfa e hlorophyll-aanddiagnosti
pigmentsspe i tophytoplanktonfun tionaltypes. Biogeos ien es, 8(2):311
327.
[Junninen et al.,2004℄ Junninen, H., Niska, H., Tuppurainen, K., Ruuskanen, J.,
and Kolehmainen, M. (2004). Methods for imputation of missing values in air
quality data sets. Atmospheri Environment, 38(18) :28952907.
[Kohonen, 1994℄ Kohonen, T. (1994). Self-organizing map. Springer-Verlag.
[Lathuilièreet al.,2008℄ Lathuilière, C., E hevin, V., and Lévy, M. (2008). Sea-
sonal and intraseasonal surfa e hlorophyll-a variability along the northwest
afri an oast. Journal of Geophysi al Resear h : O eans (19782012), 113(C5).
[Sathyendranath et al.,2014℄ Sathyendranath, S., Aiken, J., Alvain, S., Barlow,
R., Bouman, H., Bra her, A., Brewin, R., Bri aud, A., Brown, C., Ciotti, A.,
et al. (2014). Phytoplankton fun tional types from spa e. (Reports of the In-
ternationalO ean-Colour Coordinating Group (IOCCG); 15), pages 1156.
[Uitzet al.,2006℄ Uitz,J.,Claustre, H.,Morel,A.,andHooker, S.B.(2006). Ver-
ti aldistribution of phytoplankton ommunitiesinopen o ean: An assessment
based onsurfa e hlorophyll. Journal of Geophysi al Resear h : O eans (1978
2012),111(C8).