1
Auxiliary Material for
2
3
Detecting decadal-scale increases in anthropogenic CO2 in the
4
ocean
5
6
Shinya Kouketsu and Akihiko M. Murata
7
8
9
10
Research Institute for Global Change, Japan Agency for Marine-Earth Science and
Technology (JAMSTEC), 2-15 Natsushima-cho, Yokosuka, Japan
11
12
S1. Changes in CT and Anthropogenic CO2 in Vertical Sections
13
To obtain the changes in CT and anthropogenic CO2, we used data provided by the
14
Carbon Dioxide Information and Analysis Center (CDIAC, http://cdiac.ornl.gov) and the
15
CLIVAR Carbon Hydrographic Data Office (CCHDO, http://cchdo.ucsd.edu). We also
16
used data from observation lines P10 along 150°E (in 2011), S04I along 63°S (in 2012),
17
and P14 along 150° E (in 2012); we sampled along these observation lines recently.
18
We estimated the changes in anthropogenic CO2 in the same way as Murata et al. [2007]
19
by applying the ΔC* method [e.g., Gruber et al., 1996] to the two data collection periods
1
20
(WOCE and WOCE revisit). We simplified the calculations by making the following
21
assumptions. We defined the anthropogenic CO2 concentration in the ocean interior
22
ANT
(CT , μmol kg–1) as follows [e.g., Gruber et al., 1996, Sabine et al., 2002a; Sabine et al.,
23
2002b]:
24
ANT
CT =CTm−γC:O×AOU−{0.5×(ATm−AT0)+CT0+ΔCTdiseq},
25
where γC:O is the Redfield ratio for carbon and oxygen. Following Murata et al. [2007],
26
we assigned γC:O a value of 0.69 [Anderson and Sarmiento, 1994]. The apparent
27
oxygen utilization, AOU, is defined as the difference between the observed concentration
28
of dissolved oxygen (DO, μmol kg–1) and the saturated DO concentration at each potential
29
temperature and salinity. CTm (μmol kg–1) and ATm (μmol kg–1) are the measured CT and
30
measured total alkalinity (AT, μmol kg–1), respectively. AT0 (μmol kg–1) is the preformed
31
AT, which we assumed to have been constant from preindustrial times to the present. CT0
32
(μmol kg–1) is the theoretical CT of the water in equilibrium with an atmosphere without
33
anthropogenic CO2. ΔCTdiseq (μmol kg–1) is the difference between the CT in the mixed
34
layer in equilibrium with atmospheric CO2 and the CT at the time of water mass
35
formation.
36
When the changes in anthropogenic CO2 between the two data collection periods are
37
calculated, both CT0 and AT0 cancel out. Furthermore, we assumed that on a decadal time
38
scale changes in ATm and ΔCTdiseq were negligible and that these values could be ignored.
2
(1)
39
CAL
We calculated the change of anthropogenic CO2 (ΔnCT , μmol kg–1) over the decade as
40
follows:
41
CAL
CAL
CAL
ΔnCT =nCT (tr)−nCT (tw)
42
CAL
CAL
where nCT (tr) and nCT (tw) are the normalized preformed C T values for the two
43
sets of observations, WOCE revisit (tr) and WOCE (tw), respectively. The values were
44
normalized to a salinity of 35 to remove the influence of changes caused by the addition
45
or removal of freshwater; the normalization did not greatly affect the results.
46
We defined preformed natural CT (CAOU, μmol kg–1) to be equal to γC:O×AOU. We
47
then computed the changes in natural CO2 (ΔnCAOU, μmol kg–1) as follows:
48
ΔnCAOU=nCAOU(tr)−nCAOU(tw)
(2)
(3)
49
CAL
Both ΔnCT
and ΔnCAOU were calculated on given neutral density surfaces after
50
interpolating the individual properties (CT, AOU, and salinity) on the surfaces. Note that
51
ΔnCAOU reflects changes due to both circulation and biological activity, because below
52
the mixed layer, AOU is related to remineralization processes by the Redfield ratio. Thus,
53
AOU is considered to represent how long and how much water masses have been affected
54
by remineralization processes. Note that we did not distinguish the changes in ventilation
3
55
from the changes in circulation here, because the ventilation changes cause changes in
56
volume-averaged water mass ages (how long water masses have been affected) as well as
57
changes in transport, both of which contribute to changes of meridional overturning
58
circulation.
59
To make vertical sections (e.g., section S4) of the decadal changes of CT (ΔC T),
60
CAL
ΔnCT , and ΔnCAOU, we interpolated the observed CT data onto 0.05 kg m–3 × 0.2°
61
neutral density (γn) surfaces. The shortcomings of this method and comparisons of this
62
method with other methods are discussed in detail by Kouketsu et al. [2013].
63
64
Figure S1. Observation lines (black lines) used for the analysis along transect lines and the isopycnal
65
method. Green, blue, yellow, and red regions denote the Atlantic, Pacific, Indian, and Southern Oceans
66
defined in this study. The years when the observation lines were occupied are listed in Table S1.
4
67
68
Table S1.
List of sections used in this study with dates of occupation. The superscripts
69
a–E correspond to the letters a–E in Figures 1 and S2.
Section name
Initial survey
Revisit
P01a
1999
P02b
1994
2007
2004
P06c
1992
2003
P06d
2003
2009
P10e
1993
2005
P10f
2005
P14g
1992 and 1993
2011
2007
P14Sh
1992 and 1993
2012
P16i
1991
2005 and 2006
P17j
1993
2001
P18k
1994
2008
P21l
1994
2009
A01Em
1991
2000
A05n
1992
1998
AR07o
1994
2006
A09p
1991
2009
A10q
A12r
A13s
A16At
A16Su
A16Nv
1992
1992
1995
1989
1998
1993
20031998
1998
2010
2004
2004
2004
5
A20w
A22x
I03y
I05z
I06A
I08B
I09C
S04ID
S04PE
1997
2003
1995
1995
1996
1995
1995
1995
1992
2003
2012
2003
2009
2008
2007
2007
2012
2011
70
71
CAL
S2. Water Column Inventories of ΔnCT
72
CAL
We obtained the distributions of water column inventories of ΔnCT
by the same
73
CAL
method used by Kouketsu et al. [2013]. At each station, we interpolated nCT on
74
neutral density surfaces at 0.1 kg m–3 intervals. We then calculated the longitudinal or
75
CAL
latitudinal interval averages of nCT
and their standard deviations on the neutral
76
density surfaces. The intervals were set to 20° and 10° for the zonal and meridional
77
sections, respectively. Because the station locations were different between the original
78
CAL
WOCE cruises and the revisits, the interval averages and standard deviations of ΔnCT
79
were calculated by using equation (2), with weights based on the station intervals. We
80
CAL
CAL
then obtained annual ΔnCT
rates (μmol kg–1 a–1) by dividing the average ΔnCT
81
by the number of elapsed years.
6
82
CAL
We estimated the specific water column inventories of ΔnCT
(mol m–2 a–1) by
83
CAL
integrating the ΔnCT
rates from the neutral density surface at the bottom of the
84
winter mixed layer (WML) to the neutral density surface reached by detectable positive
85
CAL
ΔnCT
(shown in the next section). The WML, defined as the depth where the density
86
was 0.03 kg m–3 greater than the density at the surface, was calculated by using data from
87
the World Ocean Atlas 2005 [Locarnini et al., 2006; Antonov et al., 2006]. Then,
88
CAL
following Murata et al. [2007], we calculated ΔnCT
values within the WML by
89
CAL
assuming that the ΔnCT
rates were the same as those just below the WML. We added
90
CAL
the calculated ΔnCT
values to the values below the WML to obtain the final water
91
CAL
column inventory values. We did not compute ΔnCT
within the WML from actual
92
measurements, because the measurements were made in different seasons and were
93
therefore influenced by seasonal variations.
7
94
95
CAL
Figure S2. Specific water column inventories of ΔnCT . Black lines denote
96
observation sections. The values and characters denote the estimation errors and the pair
97
of cruises used for the comparison, respectively. The locations are shown in Figure S1,
98
and the details are provided in Table S1.
99
S3. Estimations Based on Gridded Data
100
To make gridded data, we used CARINA and PACIFICA datasets as well as the World
101
Ocean Database (WOD). In addition, temperature and salinity in the WOD were used to
102
improve the estimates of water mass distributions. Although observations in the databases
103
of CARINA and PACIFICA were much more numerous than those along the sections, it
104
was unclear whether the observations were enough to estimate statistical parameters (e.g.,
105
degree of freedom and correlation length scales). To avoid assuming constant statistical
106
parameters, we used a Markov chain Monte Carlo method (MCMC), which is a method
107
used to sample from a probability distribution with a Markov chain, the equilibrium
108
distribution of which is the distribution used for sampling. We used the MCMC method to
8
109
make gridded data based on the datasets. Recently, MCMCs have been used to estimate
110
parameters based on sparse or heterogeneous observations [e.g., Majkut et al., 2014]. We
111
used the MCMC to search for the grid value distribution that was most likely to reproduce
112
the observations, and in this way we were able to obtain many statistical parameters as
113
well as gridded values. In applying this method, we did not assume spatial and temporal
114
length scales to avoid strong constraints on spatial structures, which could not be easily
115
known in this experiment. We evaluated the uncertainty of the estimates without
116
assuming strong spatial correlation functions. For the MCMC, we assumed the following
117
probability distribution:
118
p(yo|yg,τo) ∝ p(τo)
p(yg|mg,cg,τg,αg) p(τg)p(αg)
p(mg|mmg,τmg,αmg) p(τmg)p(αmg)p(mmg)
(4)
p(cg|ccg,τcg,αcg) p(τcg)p(αcg)p(ccg)
119
These parameters and probability functions are summarized in Table S2. Here, p(X|Y)
120
denotes the conditional probability of X with a priori Y. For p(yg|mg,cg,τg,αg),
121
p(mg|mmg,τmg,αmg), and p(cg|mcg,τcg,αcg), we used a Generalized Multivariate
122
Conditional Auto-Regressive model (GMCAR) [e.g., Jin et al., 2005]. The GMCAR used
123
a Gaussian with a mean calculated from neighbors as a conditional probability function
124
for a grid point. Because a sample of a grid value is generated from a Gaussian with a
125
mean equal to the mean of neighbors, the obtained grid value is also indirectly related to
126
observations far from the grid, and the resulting gridded data were smoothed with a
9
127
precision (τg), but without assuming spatial correlation structures. The parameters mg
128
and cg are the mean and average trend at each level for temperature (T) and salinity (S)
129
and for each of the temperature and salinity segments on the T–S relationships for
130
chemical values. By defining mg and cg as the values on the T–S relationships, we
131
could implement potentially strong relationships between water masses and chemical
132
properties to improve the estimation. We divided the data into two periods, 1980–1995
133
and 1996–2013, and generated 2° × 2° gridded data for the two periods. To avoid
134
excessive computational costs and to take account of regional changes in water masses,
135
we divided the world ocean into 13 sub-regions (Figure S3) and 3 vertical layers (0–550
136
m, 350–2250 m, and 1625–6500 m) and sampled with the MCMC. Note that the
137
boundaries between sub-regions overlapped each other to facilitate connecting the
138
solutions in the sub-regions smoothly. We used the last 1000 samples of 4000 MCMC
139
iterations in this study. To obtain a smooth map, we have shown a 10° × 6° areal average
140
map.
10
141
142
Figure S3. Sub-regions for the MCMC. Each sub-region is shown by a different color.
143
The boundaries between regions overlap each other by 4°.
144
145
146
147
Table S2. Parameter list for the MCMC
yo
yg
Observation
Grid point value
τo
Precision for probability function for observations
τg
Precision for probability function for grid point value
τmg
Precision for probability function for mean value
τtg
Precision for probability function for mean trend
11
mg
Mean value for each level (TS) / within each TS
cg
segment
Mean
trend for each level (TS) / within each TS
mmg
segment
Mean
value for all levels / all segments
mcg
Mean trend for all levels / all segments
p(yo|yg,τo)
Gaussian with mean of yg and precision of τo
p(α), p(αmg), p(αtg)
Metropolis-Hastings algorithm sampling ([0, 1])
p(τo), p(τg), p(τmg)
Gamma distribution
p(m
p(τtgmg
) ), p(mtg)
Gaussian distribution
p(yg|mg,cg,τg,αg),
p(mg|mmg,τmg,αmg),
GMCAR
p(cg|mcg,τcg,αcg)
148
149
S4. Dissolved Inorganic Carbon Changes in Vertical Sections
150
CAL
Here, we show the vertical sections of ΔC T, ΔnCT , and ΔnCAOU. Whereas only
151
CAL
ΔnCT
values were used in our primary manuscript, we here show ΔCT and ΔnCAOU,
152
because they were used in the calculation and clearly reflect the raw observations of CT
153
and DO. In the Atlantic Ocean, a ΔCT of more than 5 μmol kg–1 was reached around γn
154
= 28.0 kg m–3 (Figures S4, S5, and S6). CT increases in the deep layer (around γn = 28.0
155
kg m–3) were detected along multiple sections, especially in the North Atlantic. In the
12
156
Pacific Ocean, detectable ΔCT (>3 μmol kg–1) were observed above γn = 27.5 kg m–3, as
157
reported by Kouketsu et al. [2013] (Figure S7). The detectable ΔCT in the South Pacific
158
appeared to reach deeper layers than in the North Pacific. Positive ΔCT values were also
159
observed above γn = 27.5 kg m–3 from 40°S to 20°S in the Indian Ocean, whereas
160
positive ΔCT values were detected above γn = 27.0 kg m–3 to the north of 10°S
161
(Figure S8 and S9). ΔCT values greater than 5 μmol kg–1 were detected in the deepest
162
layer (around 28.3 kg m–3) in all basins.
163
Based on the ΔCT in the sections, we identified the bottom layers for the calculation of
164
water column inventories. The bottom layers were defined by a γn of 27.6 kg m–3 north
165
of 20°S and 27.8 kg m–3 in the area bounded by 20–40°S in the Pacific and Indian Oceans.
166
The bottom layer was defined by a γn of 28.0 kg m–3 north of 40°S in the Atlantic Ocean.
167
The bottom layer was defined by a γn of 28.3 kg m–3 south of 40°S.
168
References
169
Anderson, L., and J. Sarmiento (1994), Redfield ratios of remineralization determined by
170
171
nutrient data analysis, Global Biogeochemical Cycles, 8(1), 65–80.
Antonov, J. I., R. A. Locarnini, T. P. Boyer, A. V. Mishonov, and H. E. Garcia (2006),
172
World Ocean Atlas 2005, Volume 2: Salinity, U.S. Government Printing Office,
173
Washington, D.C.
174
Gruber, N., J. L. Sarmiento, and T. F. Stocker (1996), An improved method for detecting
13
175
176
anthropogenic CO2 in the oceans, Global Biogeochem. Cycles, 10, 809–837.
Jin, X., B. P. Carlin, and S. Banerjee (2005), Generalized hierarchical multivariate CAR
177
models for areal data, Biometrics, 61(4), 950–961,
178
10.1111/j.1541-0420.2005.00359.x.
179
Kouketsu, S., A. Murata, et al. (2013), Decadal changes in dissolved inorganic carbon in
180
the pacific ocean, Global Biogeochemical Cycles, 27, 65–76,
181
doi:10.1029/2012GB004413.
182
Locarnini, R. A., A. V. Mishonov, J. I. Antonov, T. P. Boyer, and H. E. Garcia (2006),
183
World Ocean Atlas 2005, Volume 1: Temperature, U.S. Government Printing Office,
184
Washington, D.C.
185
Majkut, J. D., J. L. Sarmiento, and K. B. Rodgers (2014), A growing oceanic carbon
186
uptake: Results from an inversion study of surface pCO2 data, Global Biogeochem.
187
Cycles, 28, doi:10.1002/2013GB004585.
188
Murata, A., Y. Kumamoto, S. Watanabe, and M. Fukasawa (2007), Decadal increases of
189
anthropogenic CO2 in the South Pacific subtropical ocean along 32°S, J. Geophys.
190
Res., 112, C05033, doi:10.1029/2005JC003405.
191
192
193
Sabine, C., et al. (2002a), Distribution of anthropogenic CO2 in the Pacific Ocean, Global
Biogeochem. Cycles, 16(4), 1083.
Sabine, C., R. Key, R. Feely, and D. Greeley (2002b), Inorganic carbon in the Indian
194
Ocean: Distribution and dissolution processes, Global Biogeochem. Cycles, 16(4),
195
1067.
14
196
197
15
198
CAL
ΔC T, ΔnCT , and ΔnCAOU on the A16N line (2004–1993) along 20°W.
199
Figure S4.
200
Black lines denote the winter mixed layer density based on the World Ocean Atlas 2009
201
(WOA2009).
16
202
203
Figure S5. Same as Figure S4, but on the A16S line (2004–1998) along 25°W.
17
204
205
206
207
Figure S6. Same as Figure S4, but on the A16A line (2004–1989) along 30°W.
18
208
209
Figure S7. Same as Figure S4, but on the P16 line (2005/6–1991) along 150°W.
19
210
211
Figure S8. Same as Figure S4, but on the I08 line (2007–1995) along 90°E.
20
212
213
Figure S9. Same as Figure S4, but on the I09 line (2007–1995) along 90°E.
21
214
215
Figure S10. Same as Figure S4, but on the A01E line (2000–1991) along 57°N.
22
216
217
Figure S11. Same as Figure S4, but on the A05 line (1998–1992) along 25°N.
23
218
219
Figure S12. Same as Figure S4, but on the AR07 line (2006–1994) along 57°N.
24
220
221
Figure S13. Same as Figure S4, but on the A09 line (2009–1991) along 20°S.
25
222
223
Figure S14. Same as Figure S4, but on the A10 line (2003–1992) along 30°S.
26
224
225
Figure S15. Same as Figure S4, but on the A13 line (2010–1995) along 10°E.
27
226
227
Figure S16. Same as Figure S4, but on the A20 line (2003–1997) along 50°W.
28
228
229
Figure S17. Same as Figure S4, but on the A22 line (2012–2003) along 65°W.
29
230
231
Figure S18. Same as Figure S4, but on the I03 line (2003–1995) along 20°S.
30
232
233
Figure S19. Same as Figure S4, but on the I05 line (2009–1995) along 30°S.
31
234
235
Figure S20. Same as Figure S4, but on the I06 line (2008–1996) along 30°E.
32
236
237
Figure S21. Same as Figure S4, but on the P01 line (2007–1999) along 47°N.
33
238
239
Figure S22. Same as Figure S4, but on the P02 line (2004–1994) along 30°N.
34
240
241
Figure S23. Same as Figure S4, but on the P06 line (2003–1992) along 30°S.
35
242
243
Figure S24. Same as Figure S4, but on the P06 line (2009–2003) along 30°S.
36
244
245
Figure S25. Same as Figure S4, but on the P10 line (2005–1993) along 150°E.
37
246
247
Figure S26. Same as Figure S4, but on the P10 line (2011–2005) along 150°E.
38
248
249
Figure S27. Same as Figure S4, but on the P14 line (2007–1992/3) along 180°E.
39
250
251
Figure S28. Same as Figure S4, but on the P17 line (2001–1993) along 140°W.
40
252
253
Figure S29. Same as Figure S4, but on the P18 line (2008–1994) along 110°W.
41
254
255
Figure S30. Same as Figure S4, but on the P21 line (2009–1994) along 17°S.
42
256
257
Figure S31. Same as Figure S4, but on the S04P line (2012–1995) along 65°S.
43
258
259
Figure S32. Same as Figure S4, but on the S04I line (2011–1992) along 63°S.
44