Marine primary production
estimates from ocean color�
a comparative study of algorithms
Mary�Elena Carr�� Marjy Friedrichs�
�Jet Propulsion Laboratory
California Institute of Technology
�Old Dominion University
PRIMARY PRODUCTION ALGORITHM
ROUND�ROBIN � �PPARR��
Our goal is to provide a framework to systematically
compare algorithms which estimate primary produc�
tion from ocean color�OUTLINE
�� Introduction
�� Approach
�� Results
Global� basin� chlorophyll� SST
Southern Ocean� region� SST
�� Conclusions and future work
RESULTS FROM PPARR�
PPARR� was a blind intercomparison to in situ data�
��� stations worldwide PP� ����� � �� g C m�� d���
The best�performing algorithms were within a factor
of �� There are systematic biases� which may be ad�
dressable by �netuning the model parameterization�
Best performance in regions which have contributed
historically more data� The equatorial Paci�c and
Southern Ocean data presented higher biases�
The algorithms were highly correlated among them�
selves� irrespective of complexity�
Acknowledgements
We thank all participants �list of groups on next page�
for their hard work�
We thank the Goddard DAAC and the SeaWiFS Project
for providing SST� chlorophyll and PAR data�
We also thank John Marra and Chuck Trees for their
support�
This research was carried out at the Jet Propulsion
Laboratory� California Institute of Technology� under
a contract with the National Aeronautics and Space
Administration�
PPARR�
PART �� Annual cycle ������� Model output inter�
comparison�
PART �� Sensitivity analysis exploring biomass deter�
mination and parameterization of light utilization and
photo�adaptive physiology� Model intercomparison at
di�erent stages of PP estimation�
PART �� Comparison to in�situ ��C�uptake �ClimPP
��� tropical Paci�c stations�� Ground�truth compar�
ison�
��� David� IPCF� D� Antoine� B� Gentili and A� Morel�
�� Nick� BIO variant� N� Hoep�ner and F� Melin�
�� Kirk� K� Waters and B� Bidigare�
�� Tim� Modi�ed IPCF� T� Smyth and S� Groom�
� Mike� VGPM� M� Behrenfeld�
� Mike�� VGPM �Eppley PBopt�� M� Behrenfeld�
� ModisBF� VGPM� K� Turpie and W� Esaias�
�� Aurea� VGPM� A� Ciotti�
�� Joji� J� Ishizaka and Mr� Kameda�
��� Keith� K� Moore�
��� Heidi� Southern Ocean� H� Dierssen�
��� Heidi�� Southern Ocean chlorophyll� H� Dierssen�
��� Ichio� I� Asanuma�
��� Mark� Province�based� M� Dowell�
��� ModisHYR� HoYoRy� K� Turpie and W� Esaias�
��� RyYo� HoYoRy variant� J� Ryan�
� � HYRZe� HoYoRy variant �Ze�� M�E Carr�
��� Michele� Neural network� M� Scardi�
��� John� J� Marra�
��� SteveB� Hybrid WIM� VGPM PBopt� S� Lohrenz�
��� SteveB�� Hybrid WIM� VGPM PBopt� S� Lohrenz�
��� SteveA� Hybrid WIM� IPCF PBmax� S� Lohrenz�
��� OliCor� Ecosystem model� O� Aumont�
���� KeithE� Ecosystem model� K� Moore�
APPROACH
Given identical input �les �monthly mean�
Participants return integrated primary production�
m0 50 100 150 200
MLDJ
m0 50 100 150 200
MLDN
einstein m−2 d−10 15 30 45 60
PAR
oC
−1 9 19 29
SST
mg m−30.01 0.1 1
Chl
JAN
UA
RY
1998
g C m−2 day−10.1 0.32 1 3.2
MEAN MODELED PP
%0 25 50 75 100
STD AS % MEAN
Model spread quanti�ed as a function of �mean�
standard deviation of mean �x� as percentage
�xi � x��x� as percentage�
GLOBAL PRODUCTION
Mean global production for ���� is �� Gt C y�� �
Standard deviation of the mean is ���� Gt C y�� ���
Range of model estimates is �� Gt C y�� �
LOW� ��� Gt C y�� � models
HIGH� � � Gt C y�� �� models� ��E
INTERMEDIATE� ��� Gt C y�� ���� models� ��E
No VGPM or Spectral variant is in the high level� ���
models of each kind are found in the low level�
5 10 15 20
35
40
45
50
55
60
65
70
75
VGPMSpectral
DINTEcosystem
Gt C
y−1
MEAN GLOBAL PRODUCTION 1998
Spectral | VGPM Variants | DINT Variants | Ecosystem
BREAKOUT BY BASINS
BASIN AREA MEAN ��� MIN MAX
� Gt C y��
Paci�c �� �� ���� �� ��
Atlantic �� �� ��� �
Indian �� �� ��� ��
Southern �� � ���� � ��
Arctic ��� ���� ����� � ���
Med� �� ��� ��� ��� ���
Variability in model estimates has maximum impact in
small basins and in the Southern Ocean�
5 10 15 20
10−1
100
101
AN
NU
AL
PR
OD
UC
TIO
N /
Gt C
y−
1
Model number
Integrated Production
PacificAtlanticIndianSouthernArcticMed.
5 10 15 20
100
102
% O
F T
OT
AL
|SPECTRAL VGPM DINT| ECO|
PACIFIC−Inf −Inf 55 54−Inf −Inf
−76 −Inf −Inf 52 55−63 −Inf −Inf 57 52−65 −Inf −Inf 54
−Inf −Inf−Inf −Inf
159
12ATLANTIC
−Inf −Inf−Inf −Inf
−72 −Inf −Inf 54−58 −Inf −Inf 59−60 −Inf −Inf
−Inf −Inf−Inf −Inf
159
12INDIAN
Mon
th o
f 199
8
−Inf −Inf 63 59−Inf −Inf 51 51 65
−77 −Inf −Inf 73−62 −Inf −Inf 77−66 −Inf −Inf 59
−Inf −Inf 62−Inf −Inf 64 52
159
12SOUTHERN OCEAN
−Inf −Inf −52 59−Inf −Inf −56 57
110 −61 −Inf −Inf −56 99 72−60 152 −Inf −Inf −56 −52 101 103
173 −55 −Inf −Inf 50 −53 −51 56 10082 −Inf −Inf −56 90
−Inf −Inf 59 −56 63
159
12ARCTIC
−Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf190 109 −Inf −Inf 66 −74 −76 −87 −Inf −Inf −Inf 90 342
313 −Inf −Inf −79 −75 −86−54 284 −Inf −Inf 158 −86 −94 −87 −50
59 201 −Inf −Inf −79 −89 −91−Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf−Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf
159
12
% O
F M
EA
N
−100
−80
−60
−40
−20
0
20
40
60
80
100
MEDITERRANEAN61 −Inf −Inf −63 −9764 −Inf −Inf −61 −96
−75 51 −Inf −Inf 86 −69 −73 66 53 −97−54 −62 −Inf −Inf 96 −69 −79 56 56 −98
−59 −Inf −Inf 68 −60 −66 −98−54 61 −Inf −Inf −66 −97−50 59 −Inf −Inf −65 −96
SPECTRAL | VGPM | DINT | ECO5 10 15 20
159
12
��� models diverge in the Paci�c� Indian� Atlantic�
Seven �anomalous� models � high� low� in the South�
ern Ocean� Ecosystem models very high�
��� models are anomalous in the Arctic and Mediter�
ranean �small basins��
CONCENTRATION LEVEL
LEVEL AREA MEAN ��� MIN MAX
� Gt C y��
Oligotrophic ��� ��� ��� �� ���
Mesotrophic ��� �� ��� ���� ���
Eutrophic ��� �� ���� ��� �
Models vary in relative importance of eutrophic and
oligotrophic waters �note spectral and ecosystem mod�
els and �����
5 10 15 2010
0
101
AN
NU
AL
PR
OD
UC
TIO
N /
Gt C
y−
1
Model number
Integrated Production
OligotrophicMesotrophicEutrophic
5 10 15 20
101
102
% O
F T
OT
AL
|SPECTRAL VGPM DINT| ECO|
OLIGOTROPHIC
−Inf −Inf 51 −54 66 71
−Inf −Inf −52 63 60
−83 −Inf −Inf 69 50
−72 −Inf −Inf 51 71
−74 −Inf −Inf 51 65 56
−Inf −Inf −51 61 67
−Inf −Inf 52 −55 62 73
1
5
9
12
MESOTROPHIC
Mon
th o
f 199
8 −Inf −Inf 52
−Inf −Inf
−74 −Inf −Inf 51 53
−61 −Inf −Inf 57
−63 −Inf −Inf 52
−Inf −Inf 53
−Inf −Inf 56
1
5
9
12
% O
F M
EA
N
−100
−80
−60
−40
−20
0
20
40
60
80
100
EUTROPHIC
55 −Inf −Inf −68 55 −57 −54
−Inf −Inf −66 62 −60 −55
−60 92 −Inf −Inf −54 −64 −53
69 −Inf −Inf −63 53 −65 −63
52 −Inf −Inf −67 61 −62 −56
−Inf −Inf −66 57 −62 −53
−Inf −Inf −65 −55
SPECTRAL | VGPM | DINT | ECO
5 10 15 20
1
5
9
12
General tendency to overestimate oligotrophic PP�
��� overestimates mesotrophic�
���� �� overestimate while ��� and ECO underes�
timate eutrophic�
SST BINS
SST RANGE AREA MEAN ��� MIN MAX
� Gt C y��
SST� �oC ��� ���� ����� �� �
�o � �oC ��� �� �� � ���
�o � ��oC ��� ��� ��� �� �
� ��oC SST �� � ��� � ���
PP in SST���C is disproportionately lower than area�
Models di�er most ���C � and to a lesser degree ����C �
5 10 15 2010
−1
100
101
AN
NU
AL
PR
OD
UC
TIO
N /
Gt C
y−
1
Model number
Integrated Production
<0o
0−10o
10−20o
>20o
5 10 15 20
100
102
% O
F T
OT
AL
|SPECTRAL VGPM DINT| ECO|
SST<0oC
327 −Inf −Inf 81 −Inf −51 −78 120112 −61 −55 217 −Inf −Inf 58 −Inf −72 119 54
433 −Inf −Inf 135 −75 −54 −56 −51 59−55 −65 −65 −63 −63 −63 −61 −72 599 −Inf −Inf 323 −Inf −84 −93 −57 −71 77 −63 −65 −62 583 227
−53 182 218 −Inf −Inf −Inf −71 184 172132 −55 −53 218 −Inf −Inf 101 −Inf −64 310 6162 −54 292 −Inf −Inf 175 −Inf −71 357
1
5
9
12
0o>SST<10oC
Mon
th o
f 199
8
66 −Inf −Inf −59 −52 5486 69 −Inf −Inf −62
−54 94 −Inf −Inf −63−Inf −Inf −59
112 54 −Inf −Inf −62 55105 51 −Inf −Inf −60 −51 6463 58 −Inf −Inf −61
1
5
9
12
10o>SST<20oC
−Inf −Inf 51 56−Inf −Inf −51
−70 57 −Inf −Inf−52 64 −Inf −Inf−58 51 −Inf −Inf
−Inf −Inf−Inf −Inf 54 53
1
5
9
12
% O
F M
EA
N
−100
−80
−60
−40
−20
0
20
40
60
80
100
SST>20oC
−Inf −Inf 55 60−Inf −Inf 55 60
−78 −Inf −Inf 55 71−64 −Inf −Inf 60 59−66 −Inf −Inf 64 50 55
−Inf −Inf 53 54−Inf −Inf 55 51
SPECTRAL | VGPM | DINT | ECO5 10 15 20
1
5
9
12
SST ���C � nine models diverge� Many models un�
derestimate May or July� Ecosystem models overesti�
mate�
SST����C � �� and ��� ��� over��under�estimates
PP�
SST � ��C � ��� and � overestimates�
REGIONAL PRODUCTION IN THE
SOUTHERN OCEAN
The uncertainty of the role of the Southern Ocean in
the carbon cycle is aggravated by the di�culties of
�eld sampling�
There is a disagreement between in�situ and satellite�
based estimates of production�
Mean production �����S� for � is �� Gt C y�� �
Standard deviation of the mean is � Gt C y�� � ���
Range of model estimates is Gt C y�� � comparable
to the seasonal range for most models�
5 10 15 202
4
6
8
10
12
Model number
Gt C
y−1
MEAN SOUTHERN OCEAN PRODUCTION 1998
Spectral | VGPM Variants | DINT Variants | Ecosystem
BREAKOUT INTO LATITUDINAL BANDS
Latitudinal circles following average frontal location
�Orsi et al ����� Moore et al ������
REGIONS DEFINED BY FRONTAL POSITIONS
�� SubAntarctic Zone �SAZ�� ��S to SubAntarctic
Front �SAF�
a� North�Polar Front Zone �NPFZ�� SAF to Polar
Frontb� Polar Front �PF�� Within �� of the Polar Front�
c� South�Polar Front Zone �SPFZ�� Polar Front
to Southern Antarctic Circumpolar Current Front �SACCF��
�� South of the SACCF �SACCF�� South of the
SACCF subpolar waters and Weddell and Ross Seas�
SUMMARY OF MODEL PP
Mean model PP decreases with latitude� secondary
maximum in SACCF� Models diverge most moving
poleward�
0.1 0.2 0.3 0.4 0.5 0.6 0.7−70
−65
−60
−55
−50
−45
−40Model mean production for December 1998
g C m−2 d−1
Latit
ude
5 10 15 2010
−1
100
g C
m−
2 d−
1
Model number
December Production
Spectral | VGPM Variants | DINT Variants | Ecosystem
SAZNPFZPFSPFZSACCF
�� and �� SAZ is minimum ���� no di�erence�
Very low values south of PF ���� ��� � � ���
SOUTHERN OCEAN BREAKOUT IN SST
LEVELS
LEVEL AREA MEAN��� MIN MAX
� Gt C y��
��� ����� ��� ���� �� ������ �
�� ��� ����� � � ��� ���� ��� ��
�o���o ��� ���� ���� ��� ���� ���
� ��o � ���� � � �
PP in SST ���C is approximately half of area�
5 10 15 2010
−2
10−1
100
AN
NU
AL
PR
OD
UC
TIO
N /
Gt C
y−
1
Model number
Integrated Production Southern Ocean
<0oC0o−10oC10o−20oC
5 10 15 2010
0
101
102
% O
F T
OT
AL
Spectral | VGPM Variants | DINT Variants | Ecosystem
Relative apportioning same in all models� Greatest
divergence for SST���C �
SST<0oC
Mon
th o
f 199
8
310 97 71 −Inf −56 −79 107
115 −62 −56 192 −Inf −72 141
253 98 −65 −97 −55 −65 53 59 3.99e+032.02e+03
307 59 −70 −Inf −59 −67 4.69e+031.81e+03
−55 237 −52 −52 151 −Inf −60 −57 262 213
126 −57 −54 211 65 98 −Inf −64 298 51
57 −56 284 65 171 −Inf −51 −72 345
5 10 15 20
13579
1112
0o>SST<10oC
Mon
th o
f 199
8
78 −58 −56 61 51 61
75 76 −58 −62 67 54 72
170 −58 −60 −58 58 243 177
−58 191 −53 −65 95 −63 −61 395 288
214 54 −57 −62 64 117
135 77 −59 −56 85
69 73 52 −61 −53
5 10 15 20
13579
1112
% O
F M
EA
N
−100
−80
−60
−40
−20
0
20
40
60
80
100
10o>SST<20oC
Mon
th o
f 199
8
53 −59
−62
57 52 52 75 −51 −66
−71 121 55 −60 −57 81 −68 −52 102 91
−Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf −Inf
−63
51 −64 −51
MODEL NUMBER
Spectral | VGPM Variants | DINT Variants | Ecosystem
5 10 15 20
13579
1112
Nine anomalous models in SST ���C �
��� ��� �� uniformly high in SST����C � Model ���
was uniformly low everywhere and ��� low SST�
���C �di�erent from global breakout �
Some models �e�g� �� counter the seasonal cycle�
other �e�g� ��� reinforce it�
0 0.5−70
−68
−66
−64
−62
−60
−58
−56
−54Wavelength−resolved
Latit
ude
1=David2=Nick3=Kirk4=Tim
0 0.5
VGPM Variants
Primary Production /g C m−2 day−1
5=Mike6=Mike27=ModisBF8=Aurea9=Joji10=Keith11=Heidi12=Heidi2
0 0.5
Depth−integrated
13=Ichio14=Mark15=ModisHYR16=RyYo17=HYRZe18=Mich19=John20=SteveB21=SteveB222=SteveA
0 0.5
Ecosystem models
Primary Production /g C m−2 day−1
23=OliCor24=Keith
All model�types bracket the observations� a few �high�
values� tendency to underestimate�
Comparable� �� �SPEC� ���� � and � �VGPM�
and ���� � � and ���� �� �DINT� Ecosystem mod�
els not great�
0 0.1 0.2 0.3 0.4 0.5 0.6−72
−70
−68
−66
−64
−62
−60
−58
−56
−54
VGPM
Spectral
Other DIM
Ecosystem
TRANSECT AT 170oW, MARCH 1998
Primary Production/ g C/(m2 day)
Latit
ude
C14Mean
The mean of the modeled estimates is less than �but
� ���� g C m��d��� the measured primary production
�Hiscock et al ����� in all � regions�
VGPM variants tend to underestimate PP�
40 60 80 100 120 140 160−72
−70
−68
−66
−64
−62
−60
−58
−56
−54
−52
−50SEASONAL PRODUCTION ALONG TRANSECT AT 170oW
Primary Production/ g C m−2
Latit
ude
C14, Nelson et alPPModel, Buesseler et alMean 98, this studySPECVGPMDINTECO
Oct�May comparison for non�coincident years�
Model PP � C�� estimate � Optimized PP model
�except south of ���S��
Outlier models
Anomalies �H� L� are a function of the mean model�
A� P� I� S � basins�
T��T�� SST bins�
O� M� E� �chl�
S � Southern Ocean anomaly� s� SO and global
� anomaly only May�Sep� �boreal winter
A P I S T� T� T� T� O M E
� H Hs Hs
� L L L L L L L L
� Hs Hs H
� H H H H H H H
� Ls Hs L
� LS Hs
� L Ls Ls LS L
� H H
� L LS LS
�� H HS H
�� H L H H
�� H HS Hs H L
�� H HS Hs L
CONCLUSIONS
The spread between models is considerable �almost a
factor of ���
Peak disagreement for Southern Ocean� small basins�
SST ����C � and oligotrophic and eutrophic waters�
Generally divergence between models is greater mov�
ing south� and for waters ���C �
Comparison with concurrent March ���� measure�
ments indicate tendency of models to underestimate�
Seasonal PP estimate is less than derived from C��
or from ocean color using a tuned parameter�
We need to use more Southern Ocean data to param�
eterize the models� Issues such as chlorophyll deter�
mination and parameterization of photosynthesis at
low SST are likely crucial�
Future work� Uncover the reasons behind these di�er
ences �Part �� and comparison to in�situ data �Part
��Stay tuned���
0 50 100 150 200
MLD JPL
SAZ
NAPFZ
PF
SAPFZ
SACCF
CONDITIONS IN 1998
1 3 5 7 9 11 12
0 15 30 45 60
PAR
SAZ
NAPFZ
PF
SAPFZ
SACCF
1 3 5 7 9 11 12
−2 2 6 10 14
SST
SAZ
NAPFZ
PF
SAPFZ
SACCF
1 3 5 7 9 11 12
0.1 0.18 0.32 0.56 1
Chlorophyll
SAZ
NAPFZ
PF
SAPFZ
SACCF
1 3 5 7 9 11 12
0.03 0.06 0.1 0.18
MEAN PP
SAZ
NAPFZ
PF
SAPFZ
SACCF
1 3 5 7 9 11 12
0 20 40 60 80 100
STD PP AS % MEAN
SAZ
NAPFZ
PF
SAPFZ
SACCF
1 3 5 7 9 11 12
MLD� SST� and PAR decrease with latitude�
Chl maxima in SAZ� PF� and SACCF�
Max PP in Jan and Nov�Dec� local minimum in SAPFZ
�Jan and Dec�� Minima in May�July�
Model spread at best ���� increasing moving south
and in winter months�
SUBANTARCTIC ZONE
70 77 64−51 68 54
114 −58 59 −52 51 80 64−55 168 −51 79 97
173 −52 68 9470 57 −52 51
72 65 535 10 15 20
13579
1112
NORTH POLAR FRONT ZONE
60 −59 −55 −54 70 6090 56 −65 −54 −58 74 60 55198 −51 −78 −62 −52 −55 −57 61 157 125
−64 252 −85 −75 −62 90 −57 −63 347 290276 79 −59 52 −52 −62 97 153165 81 −54 −56 −56 9886 70 −54 −59 −52
5 10 15 20
13579
1112
POLAR FRONT
Mon
th o
f 199
8
86 −58 −55 59 5094 84 −56 −63 71 59 87216 63 −60 −57 −62 57 244 170
−62 272 −55 −68 97 −60 −66 668 489277 122 −55 −64 178 149163 111 55 −59 −57 12388 97 68 −61 −52
5 10 15 20
13579
1112
SOUTH POLAR FRONT
104 75 −57 −58 55 11896 113 −57 −64 64 52 134226 71 −64 −69 −57 −65 71 57 51 823 555
−51 286 −71 64 −62 −67 2.68e+031.54e+03−53 265 −54 −58 169 −59 −57 −57 326 283
142 150 85 −56 −59 −54 22891 133 95 −53 −60 −56 164
5 10 15 20
13579
1112
% o
f int
egra
ted
PP
−100
−80
−60
−40
−20
0
20
40
60
80
100
SOUTH OF ACC FRONT
MODEL NUMBER
Spectral | VGPM | DINT | Ecosystem
233 99.8 −57 −51 −68 193117 169 −64 −52 187242 88 −63 −96 −56 −64 67 53 61 1.58e+03835369 −50 −61 75 −70 −55 −56 −68 6.11e+032.03e+03
−58 264 −50 −56 −56 −75 −67 206 58 60 −98 −59 840 532132 −58 −58 −63 −66 344 210 −86 −67 1.44e+0315661 −58 325 62 219 −73 −72 689
5 10 15 20
13579
1112
Disagreement increases moving south�
Uniformly high ���� or low ������
m0 50 100 150 200
MLD JPLCONDITIONS ALONG 170oW IN 1998
1 3 5 7 9 11 12
−70
−65
−60
−55
einstein m−2 d−10 15 30 45 60
PAR
1 3 5 7 9 11 12
−70
−65
−60
−55
oC
−2 2 6 10 14
SST
1 3 5 7 9 11 12
−70
−65
−60
−55
mg m−30.1 0.18 0.32 0.56 1
Chlorophyll
1 3 5 7 9 11 12
−70
−65
−60
−55
mg m−30.06 0.1 0.18 0.32 0.56 1 1.8
MEAN PP
1 3 5 7 9 11 12
−70
−65
−60
−55
%0 20 40 60 80 100
STD PP AS % MEAN
1 3 5 7 9 11 12
−70
−65
−60
−55
Progression as in Paci�c sector�
Maximum chlorophyll and PP in Jan south of ���Sand
in Dec in SAZ to ���S � Higher standard deviation of
models south�