Annick Bricaud,1
Carlos Mejia,3
David Blondeau-Patissier,2
Hervé Claustre,1
Michel Crepon,3
and Sylvie Thiria3
1When this research was performed A. Bricaud (bricaud@obs-vlfr.fr), D. Blondeau-Patissier, and H. Claustre were with the Laboratoire d'Océanographie de Villefranche, CNRS, Boîte Postale 08, 06238 Villefranche-sur-Mer Cedex, France, and the Laboratoire d'Océanographie de Villefranche, Université Pierre et Marie Curìe-Paris 6, Boîte Postale 08, 06238 Villefranche-sur-Mer, France.
2D. Blondeau-Patissier is now with the Environmental Remote Sensing Group, Commonwealth Scientific and Industrial Research Organization Land & Water, GPO Box 1666, Canberra, ACT 2601, Australia.
3C. Mejia, M. Crépon, and S. Thiria are with CNRS, Laboratoire d'Océanographie et du Climat: Expérimentation et Approches Numériques (LOCEAN∕IPSL), 4 place Jussieu, 75252 Paris Cedex 05, France, and Université Pierre et Marie Curie-Paris 6, LOCEAN, 4 place Jussieu, 75252 Paris Cedex 05, France.
Annick Bricaud, Carlos Mejia, David Blondeau-Patissier, Hervé Claustre, Michel Crepon, and Sylvie Thiria, "Retrieval of pigment concentrations and size structure of algal populations from their absorption spectra using multilayered perceptrons," Appl. Opt. 46, 1251-1260 (2007)
Spectral absorption coefficients of phytoplankton can now be derived, under some assumptions, from hyperspectral ocean color measurements and thus become accessible from space. In this study, multilayer perceptrons have been developed to retrieve information on the pigment composition and size structure of phytoplankton from these absorption spectra. The retrieved variables are the main pigment groups (chlorophylls
a, b, c, and photosynthetic and nonphotosynthetic carotenoids) and the relative contributions of three algal size classes (pico-, nano-, and microphytoplankton) to total chlorophyll
a. The networks have been trained, tested, and validated using more than 3700 simultaneous absorption and pigment measurements collected in the world ocean. Among pigment groups, chlorophyll a is the most accurately retrieved (average relative errors of 17% and 16%
for the test and validation data subsets, respectively), while the poorest performances are found for chlorophyll b (average relative errors of 51% and 40%). Relative contributions of algal size classes to total chlorophyll a are retrieved with average relative errors of 19% to 33%
for the test subset and of 18% to 47% for the validation subset. The performances obtained for the validation data, showing no strong degradation with respect to test data, suggest that these neural networks might be operated with similar performances for a large variety of marine areas.
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
The number of samples (N) and the dominant trophic state (TS; E = eutrophic, M = mesotrophic, O = oligotrophic) are indicated for each cruise. For the sake of simplification, the trophic state of waters is indicated as O when the surface Tchl a concentration is generally <0.2 mg m−3, M when it is in the range of 0.2–2 mg m−3, and E when it is >2 mg m−3.
Validation data subset.
Table 2
Performances of the MLPs Developed for Each Pigment Group for the Learning, Test, and Validation Subsetsa
Learning
Test
Validation
Pigment
N
Slope
Intercept
Error (%)
N
Slope
Intercept
Error (%)
N
Slope
Intercept
Error (%)
Tchl a
1653
0.977
−0.0177
0.972
15.44
413
0.950
−0.0389
0.880
17.05
1556
0.956
−0.0524
0.924
16.40
Tchl b
1164
1.236
0.135
0.623
45.25
287
1.227
0.120
0.439
51.05
1248
1.141
0.0163
0.438
40.17
Tchl c
1220
1.127
0.0461
0.831
39.08
298
1.132
0.0227
0.802
41.13
1257
1.189
0.185
0.833
29.59
TPSC
1603
0.975
−0.0362
0.965
28.57
396
0.988
−0.0440
0.844
30.32
1455
0.943
−0.0951
0.415
28.36
TPPC
1401
1.007
−0.0181
0.908
26.31
345
1.032
−0.0219
0.738
26.96
1353
0.897
−0.192
0.731
24.49
Slope, intercept, and determination coefficient corresponding to linear regression analyses between log transformed retrieved and measured values. For each case, the average relative error between retrieved and measured values, in absolute values [error, in percent; see Eq. (4)] is also indicated. For each subset, the samples for which the considered pigment concentration was <0.01 mg m−3 were discarded from computations (note they are displayed on scatterplots). N indicates the number of considered samples for each case.
Table 3
Average Relative Errors on Retrieved Values for Each Pigment Groupa
Learning
Test
Validation
N
Error (%)
N
Error (%)
N
Error (%)
Tchl a
301
15.84
76
16.93
270
16.38
Tchl b
161
45.32
36
59.41
210
39.45
Tchl c
181
31.53
42
33.22
216
27.49
TPSC
284
38.68
73
22.13
253
26.09
TPPC
299
22.17
74
29.10
238
25.45
Errors in percent [see Eq. (4)]. Subsets are restricted to samples collected within the first optical depth.
Table 4
Average Relative Errors for Each Pigment Group and for the Various Concentration Intervalsa
Concentration Interval (mg m−3)
0.01–0.03
0.03–0.07
0.07–0.10
0.1–0.3
0.3–0.7
0.7–1
>1
Tchl a
22.4 (22.2)
19.2 (24.8)
24.3 (17.3)
17.2 (15.9)
11.7 (14.5)
8.9 (13.8)
21.7 (15.3)
Tchl b
73.7 (81.0)
59.4 (36.2)
32.2 (22.4)
39.9 (34.2)
77.8 (60.3)
— (—)
— (—)
Tchl c
65.8 (51.9)
29.8 (28.8)
24.4 (18.7)
28.9 (19.0)
18.8 (18.0)
10.2 (—)
30.1 (—)
TPSC
26.9 (59.5)
72.5 (33.0)
18.9 (26.6)
22.4 (22.6)
14.9 (20.4)
35.2 (12.9)
26.2 (10.7)
TPPC
39.6 (29.5)
23.9 (23.7)
15.7 (19.9)
29.6 (26.9)
37.2 (62.1)
78.0 (—)
38.5 (—)
Errors in percent [see Eq. (4)]. Values are given for the test subset and, in parentheses, for the validation subset. Concentration values lower than 0.01 mg m−3 were not considered.
Table 5
Performances of the MLPs Developed for Simultaneously Retrieving the Biomass Contributions of the Three Phytoplankton Size Classes for the Learning, Test, and Validation Subsetsa
Learning
Test
Validation
Class
N
Slope
Intercept
Error (%)
N
Slope
Intercept
Error (%)
N
Slope
Intercept
Error (%)
Pico
1728
0.832
0.0528
0.840
22.78
435
0.861
0.0546
0.834
22.88
1571
0.777
0.0971
0.767
26.56
Nano
1728
0.719
0.121
0.680
18.39
435
0.575
0.183
0.423
18.78
1571
0.643
0.159
0.508
17.83
Micro
1728
0.877
0.0429
0.833
33.08
435
0.872
0.0418
0.823
33.34
1571
0.544
0.0739
0.362
46.96
Slope, intercept, and determination coefficient corresponding to linear regression analyses between retrieved and measured values. For each case, the average relative error between retrieved and measured values [errors in percent; see Eq. (4)] is also indicated. N indicates the number of considered samples for each case.
Table 6
Average Relative Errors between Retrieved and Measured Valuesa
Learning
Test
Validation
N
Error (%)
N
Error (%)
N
Error (%)
Pico
1728
23.94
435
23.82
1571
26.21
Nano
1728
23.13
435
22.60
1571
19.60
Micro
1728
33.76
435
35.80
1571
49.55
Errors in percent [see Eq. (4)]. Biomass contributions of the three size classes are estimated using three independent MLPs. N indicates the number of considered samples for each case.
Tables (6)
Table 1
Information Concerning the Cruises Where Absorption and HPLC Data Were Simultaneously Collecteda
Cruise
Location
Period
N
TS
TOMOFRONT
Northwestern Mediterranean
April 1990
28
M
EUMELI 3
Tropical North Atlantic
October 1991
49
O, M
FLUPAC
Equatorial and subequatorial Pacific
September–October 1994
80
O
OLIPAC
Equatorial and subequatorial Pacific
November 1994
183
O
MINOS
Eastern and Western Mediterranean
May 1996
115
O
ALMOFRONT 2
Alboran Sea
December 1997–January 1998
477
M
PROSOPE
Morocco upwelling, Eastern and Western Mediterranean
The number of samples (N) and the dominant trophic state (TS; E = eutrophic, M = mesotrophic, O = oligotrophic) are indicated for each cruise. For the sake of simplification, the trophic state of waters is indicated as O when the surface Tchl a concentration is generally <0.2 mg m−3, M when it is in the range of 0.2–2 mg m−3, and E when it is >2 mg m−3.
Validation data subset.
Table 2
Performances of the MLPs Developed for Each Pigment Group for the Learning, Test, and Validation Subsetsa
Learning
Test
Validation
Pigment
N
Slope
Intercept
Error (%)
N
Slope
Intercept
Error (%)
N
Slope
Intercept
Error (%)
Tchl a
1653
0.977
−0.0177
0.972
15.44
413
0.950
−0.0389
0.880
17.05
1556
0.956
−0.0524
0.924
16.40
Tchl b
1164
1.236
0.135
0.623
45.25
287
1.227
0.120
0.439
51.05
1248
1.141
0.0163
0.438
40.17
Tchl c
1220
1.127
0.0461
0.831
39.08
298
1.132
0.0227
0.802
41.13
1257
1.189
0.185
0.833
29.59
TPSC
1603
0.975
−0.0362
0.965
28.57
396
0.988
−0.0440
0.844
30.32
1455
0.943
−0.0951
0.415
28.36
TPPC
1401
1.007
−0.0181
0.908
26.31
345
1.032
−0.0219
0.738
26.96
1353
0.897
−0.192
0.731
24.49
Slope, intercept, and determination coefficient corresponding to linear regression analyses between log transformed retrieved and measured values. For each case, the average relative error between retrieved and measured values, in absolute values [error, in percent; see Eq. (4)] is also indicated. For each subset, the samples for which the considered pigment concentration was <0.01 mg m−3 were discarded from computations (note they are displayed on scatterplots). N indicates the number of considered samples for each case.
Table 3
Average Relative Errors on Retrieved Values for Each Pigment Groupa
Learning
Test
Validation
N
Error (%)
N
Error (%)
N
Error (%)
Tchl a
301
15.84
76
16.93
270
16.38
Tchl b
161
45.32
36
59.41
210
39.45
Tchl c
181
31.53
42
33.22
216
27.49
TPSC
284
38.68
73
22.13
253
26.09
TPPC
299
22.17
74
29.10
238
25.45
Errors in percent [see Eq. (4)]. Subsets are restricted to samples collected within the first optical depth.
Table 4
Average Relative Errors for Each Pigment Group and for the Various Concentration Intervalsa
Concentration Interval (mg m−3)
0.01–0.03
0.03–0.07
0.07–0.10
0.1–0.3
0.3–0.7
0.7–1
>1
Tchl a
22.4 (22.2)
19.2 (24.8)
24.3 (17.3)
17.2 (15.9)
11.7 (14.5)
8.9 (13.8)
21.7 (15.3)
Tchl b
73.7 (81.0)
59.4 (36.2)
32.2 (22.4)
39.9 (34.2)
77.8 (60.3)
— (—)
— (—)
Tchl c
65.8 (51.9)
29.8 (28.8)
24.4 (18.7)
28.9 (19.0)
18.8 (18.0)
10.2 (—)
30.1 (—)
TPSC
26.9 (59.5)
72.5 (33.0)
18.9 (26.6)
22.4 (22.6)
14.9 (20.4)
35.2 (12.9)
26.2 (10.7)
TPPC
39.6 (29.5)
23.9 (23.7)
15.7 (19.9)
29.6 (26.9)
37.2 (62.1)
78.0 (—)
38.5 (—)
Errors in percent [see Eq. (4)]. Values are given for the test subset and, in parentheses, for the validation subset. Concentration values lower than 0.01 mg m−3 were not considered.
Table 5
Performances of the MLPs Developed for Simultaneously Retrieving the Biomass Contributions of the Three Phytoplankton Size Classes for the Learning, Test, and Validation Subsetsa
Learning
Test
Validation
Class
N
Slope
Intercept
Error (%)
N
Slope
Intercept
Error (%)
N
Slope
Intercept
Error (%)
Pico
1728
0.832
0.0528
0.840
22.78
435
0.861
0.0546
0.834
22.88
1571
0.777
0.0971
0.767
26.56
Nano
1728
0.719
0.121
0.680
18.39
435
0.575
0.183
0.423
18.78
1571
0.643
0.159
0.508
17.83
Micro
1728
0.877
0.0429
0.833
33.08
435
0.872
0.0418
0.823
33.34
1571
0.544
0.0739
0.362
46.96
Slope, intercept, and determination coefficient corresponding to linear regression analyses between retrieved and measured values. For each case, the average relative error between retrieved and measured values [errors in percent; see Eq. (4)] is also indicated. N indicates the number of considered samples for each case.
Table 6
Average Relative Errors between Retrieved and Measured Valuesa
Learning
Test
Validation
N
Error (%)
N
Error (%)
N
Error (%)
Pico
1728
23.94
435
23.82
1571
26.21
Nano
1728
23.13
435
22.60
1571
19.60
Micro
1728
33.76
435
35.80
1571
49.55
Errors in percent [see Eq. (4)]. Biomass contributions of the three size classes are estimated using three independent MLPs. N indicates the number of considered samples for each case.