Content uploaded by Elena Couce
Author content
All content in this area was uploaded by Elena Couce on Nov 30, 2017
Content may be subject to copyright.
Tropical coral reef habitat in a geoengineered, high-
CO2 world
E. Couce, P. J. Irvine, L. J. Gregorie, A. Ridgwell, and E. J. Hendy
Continued anthropogenic CO2 emissions are expected to impact tropical coral
reefs by further raising sea surface temperatures (SST) and intensifying ocean
acidification (OA). Although geoengineering by means of solar radiation
management (SRM) may mitigate temperature increases, OA will persist, raising
important questions regarding the impact of different stressor combinations. We
apply statistical Bioclimatic Envelope Models to project changes in shallow water
tropical coral reef habitat as a single niche (without resolving biodiversity or
community composition) under various representative concentration pathway and
SRM scenarios, until 2070. We predict substantial reductions in habitat suitability
centered on the Indo-Pacific Warm Pool under net anthropogenic radiative forcing
of ! 3.0 W/m2. The near-term dominant risk to coral reefs is increasing SSTs;
below 3 W/m2 reasonably favorable conditions are maintained, even when
achieved by SRM with persisting OA. “Optimal” mitigation occurs at 1.5 W/m2
because tropical SSTs overcool in a fully geoengineered (i.e., preindustrial global
mean temperature) world.
An edited version of this paper was published by AGU. Copyright 2013 American Geophysical
Union. Citation: Couce, E., P. J. Irvine, L. J. Gregorie, A. Ridgwell, and E. J. Hendy (2013), Tropical coral
reef habitat in a geoengineered, high-CO2 world, Geophys. Res. Lett., 40, doi:10.1002/ grl.50340.
1. Introduction
Tropical shallow water coral reefs cover 0.1% of the world’s oceans, yet rank
among the most productive and biodiverse ecosystems. Anthropogenic pressures
have been implicated in significant long-term reef decline as well as abrupt coral
mortality events associated with extreme temperatures and bleaching [Hoegh-
Guldberg et al., 2007]. Solar radiation management (SRM) — a form of
geoengineering achieved by adding reflective aerosols to the atmosphere [Crutzen,
2006], increasing cloud albedo [Latham and Smith, 1990], or increasing the albedo
of the Earth’s surface [Irvine et al., 2011], for example — has the potential to
mitigate surface warming and hence hypothetically help safeguard shallow water
coral reef habitat. But by only seeking to diminish downward radiation [Angel,
2006], SRM achieves no direct mitigation of atmospheric CO2 and resulting
“ocean acidification”. The latter undermines habitat construction that supports
coral reef ecosystems because higher pCO2 reduces carbonate ion concentration
and associated saturation ("Arag) levels, in turn lowering net carbonate production
by corals and calcareous algae [Kleypas et al., 1999].
Any implementation of SRM geoengineering would therefore produce a complex
pattern of marine environmental changes, overall characterized by relatively low
sea surface temperatures (SST) but with high levels of atmospheric pCO2 and
ocean acidification. This raises important questions about the primary global
environmental threat(s) to tropical coral reefs: whether it is increased SSTs,
reduced "Arag, or that both factors are equally significant. Our motivation in this
paper is hence not to make a case for or against SRM but to explore the spatial and
temporal consequences of different potential global temperature and ocean
acidification futures for shallow water coral reefs. Bioclimatic Envelope Modeling
can be applied to forecast effects of climate change on species’ distribution [e.g.,
Thuiller et al., 2005] and statistically analyze the environmental requirements of
coral reef ecosystems [Couce et al., 2012]. We use this approach to explore how
changing future environmental conditions with and without SRM geoengineering
could affect the potential suitability of global shallow water habitats for coral reef
ecosystems.
2. Methods
Bioclimatic Envelope Modeling analyzes the relationship between environmental
factors and the distribution of a species (or an ecosystem), using statistical
correlation to identify acceptable environmental ranges and the relative
significance of the different factors. We used two machine-learning techniques:
maximum entropy (MaxEnt) [Phillips et al., 2006] and boosted regression trees
(BRT) [Friedman, 2001]. The assumption behind MaxEnt is that a
species/ecosystem will occupy all suitable habitat in as random a way as possible;
MaxEnt then identifies which constraints maximize the entropy of the system.
BRT is based on decision trees. A single tree is built by repeatedly finding a
simple rule (whether one of the predictive variables is above or below a specific
threshold) that can split the data into groups providing the best separation of
presence and absence sites. A sequence of trees (typically >1000) is produced,
each grown on reweighted versions of the data, with final predictions obtained
from the weighted average across the tree sequence.
Couce et al. [2012] provides a detailed analysis and background to BRT and
MaxEnt in relation to establishing environmental controls on tropical coral reef
biogeography. In the current study 12 environmental fields were considered
including SST, "Arag, salinity, nutrients, and light availability. We chose "Arag
over pH because coral calcification is directly linked to saturation state, although
under rapid fossil fuel CO2 release changes in both variables will be closely
correlated [Hönisch et al., 2012]. In total, 27 predictive variables were used,
including mean annual and extreme monthly values for most fields in addition to
weekly extremes and standard deviation of SST (for complete list and relative
contribution to predictions see Appendix S1). Model training data sets were
generally observation-based except "Arag and SST, which were obtained from
1990 projections of the University of Victoria (UVic) Earth System Climate
Model [Weaver et al., 2001; Turley et al., 2010] of open ocean water in proximity
to reefs. All fields were mapped onto a 1°x1° global grid between 60°S and 60°N;
for cells outside the open-ocean mask, environmental data were extrapolated up to
1° by linear average of neighboring cells. The models were trained on a “shallow
water mask” defined by bathymetry within the euphotic zone and the area covered
by UVic projections (Figure S1.1). Locations of shallow water reef and coral
communities were provided by ReefBase (version 2000; http://www.reefbase.org
[Vergara et al., 2000]) and projected on the 1°x1° grid as binary presence/absence
data. See Appendix S1 and Couce et al. [2012] for further details on model
development and variables.
Figure 1. Simulated spatial anomalies, year 2070 minus preindustrial (P-I), of (top, a and b) sea
surface temperature (SST) and (bottom, c and d) aragonite saturation state ("Arag) under RCP 8.5
(a and c) and with SRM geoengineering returning total anthropogenic radiative forcing to P-I
values in the “RCP 8.5 & GEO 8.5” scenario (b and d). Change in shallow water tropical coral
1.5
1.5
2
2
2
2
2
2
2
2
2.5
3
0° 90°E 180° 90°W
60°S
30°S
0°
30°N
60°N
a) RCP 8.5
ï ï ï 0 1 2 3
SST anomaly (°C)
ï
ï
ï
ï
ï
0° 90°E 180° 90°W
b) RCP 8.5 & GEO 8.5
ï
ï
ï
ï
ï
ï
ï
0° 90°E 180° 90°W
60°S
30°S
0°
30°N
60°N
c) RCP 8.5
ï ï ï ï ï 0.0
1Arag anomaly
ï
ï
ï
ï
ï
ï
ï
ï
ï
0° 90°E 180° 90°W
d) RCP 8.5 & GEO 8.5
0° 90°E 180° 90°W
60°S
30°S
0°
30°N
60°N
e) RCP 8.5
ï ï ï 0.00 0.25 0.50
Average change in suitability
0° 90°E 180° 90°W
I5&3*(2
reef habitat suitability between 2070 and P-I, averaged from BRT and MaxEnt model outputs for
RCP 8.5 (e) and “RCP 8.5 & GEO 7,” with SRM geoengineering to reduce anthropogenic
radiative forcing to 1.5 W/m2 above P-I by 2100 (f). Green dotted line corresponds to 0 change;
black hatched pattern overlays area where projections move beyond training range with
significant influence on predictions. For other scenarios, see Appendix S2.
Future and preindustrial (P-I) projections of mean annual SST and "Arag were
determined using the UVic model [Weaver et al., 2001] version 2.9, which
comprises an atmosphere Energy Moisture Balance Model coupled to a 3D ocean
general circulation model, both at a spatial resolution of 1.8°x3.6°. Ocean
chemistry was calculated by the biogeochemical and carbon cycle model of
Schmittner et al. [2008]. The UVic model was forced with concentrations of CO2
and other greenhouse gases from the representative concentration pathways
(RCPs) [Moss et al., 2010] developed for the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change corresponding to a total
anthropogenic radiative forcing of 3, 4.5, 6, and 8.5 W/m2 above P-I at 2100,
respectively (labeled “RCP 3” to “RCP 8.5”). The extent of SRM geoengineering
considered for each RCP scenario either brought radiative forcing back to P-I
levels or to a particular forcing above P-I (the geoengineering forcing is labeled
“GEO” followed by the amount reduced; e.g., “GEO 1.5” refers to an equivalent
SRM geoengineering to bring anthropogenic forcing down by 1.5 W/m2 by 2100).
The SRM forcing was applied from 2020 with an e-folding time of 5 years and
following the equivalent RCP scenario when available (i.e., “RCP 6 & GEO 1.5”
will have the same total forcing as “RCP 4.5”). As for model training, the
maximum and minimum monthly and weekly SST values were computed by
adding observed present-day anomalies to UVic projected annual mean SST data
(i.e., assuming variability remains unchanged). Future irradiance levels under
SRM geoengineering were calculated by applying a -1% to -3% reduction to
present observed values depending on emission scenario and desired total level of
forcing. Additional variations in cloudiness patterns were not considered. All other
environmental fields were kept at present values. Predictions were generated at 10-
year intervals from 2010 to 2070 and for 1850 to establish the P-I baseline (for P-I
projection map, see Figure S1.4, Appendix S1). The 2070 cutoff for future
projections was chosen because 14% of coral reef cells are out of training range by
this date under RCP 8.5. Bioclimatic Envelope Models become less reliable for
forecasts that involve extrapolation to novel conditions because statistical
relationships observed in training may no longer hold.
0 1
0
0 1
0
0 1
0 1 0 1
0 1
0 1 0 1 0 1
0 1
0 1 0 1 0 1
0 1
0 1 0 1 0 1 0 1
sllec fo rebmuN
Modelled reef probability
-1.5
-2.5
-3.0
-4.0
-4.5
-5.5
-6.0
-7.0
-8.5
Level of geoengineering (W/m
2
)
Emission scenario (W/m2)
3.0 4.5 6.0 8.5
x0.62
0 0.4 0.8 1
0 50 150 250
3UHïLQGXVWULDO
x0.59
RCP 3
x0.52
RCP 4.5
x0.51
RCP 6
x0.49
RCP 8.5
x0.62 x 0.56 x 0.52
x0.51
x0.57
0 250
x0.58 x 0.56
x0.51
x0.54
0 250
x0.57
x0.56
0
x0.53
250
x0.56
x0.53
250
250
Year
+DELWDW6XLWDELOLW\,QGex
2010 2020 2030 2040 2050 2060 2070
0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00
a) RCP 8.5 (BRT)
Year
2010 2020 2030 2040 2050 2060 2070
b) RCP 8.5 (MaxEnt)
GEO 8.5
GEO 7
GEO 5.5
GEO 4
GEO 2.5
Unmitigated
c) Reef cells only, 2070 (BRT)
Figure 2. Habitat Suitability Index (defined as the average suitability for coral reefs within the
shallow water mask between 60°S and 60°N) for (a) BRT and (b) MaxEnt. Values are normalized
to preindustrial (P-I) predictions and show the evolution at 10-year intervals until 2070 under the
unmitigated RCP 8.5 scenario (black) and various level of SRM (lighter colors show
progressively higher degrees of SRM intervention). For all other scenarios, see Appendix S2 and
Figures S2.8 and S2.9. (c) Histograms showing the proportion of reef cells within binned BRT
modeled suitability values. The bottom left histogram is for P-I conditions; all remaining
histograms are for 2070 conditions and reflect potential changes in suitability under the four
unmitigated RCPs (bottom row, along x axis) and various levels of SRM geoengineering (y axis).
Reef cells are cells where reefs or non-reef coral communities are presently found (ReefBase
v2000). Novel environmental conditions, compared to the 1990 values used for model training,
are simulated by UVic Earth System Climate Model for SST and "Arag on some reef cells. The
solid colored histogram bars contain all cells either with environmental conditions within the
bioclimatic envelope used to train the models or where out-of-range variables do not significantly
affect predictions. The average suitability value (x
!) of reef cells for each scenario is calculated
from this sample set. Cells where predictions are less reliable (i.e., SST and/or "Arag values out of
training range and MaxEnt clamping value > 0.1; see Appendix S3) are indicated by hatched
pattern and have been excluded from the calculated average.
3. Results
Under the highest CO2 scenario considered (RCP 8.5), year 2070 tropical SSTs are
generally ~2°C–3°C higher than preindustrial (P-I), with the strongest warming
occurring in the western Pacific (Figure 1a). Associated with rising atmospheric
pCO2 and invasion of fossil fuel CO2 into the ocean, "Arag falls by 1.5–2 units,
with least change in upwelling areas (Figure 1c). Under these conditions, we
forecast a marked decline in environmental suitability for shallow coral reef
habitats across the central Indo-Pacific (Figure 1e; see also Appendix S2).
Elsewhere, conditions generally became less favorable, except for higher latitudes
and upwelling regions. Values of a Habitat Suitability Index (defined as the mean
probability of a coral reef being present within the shallow water mask,
normalized as a percentage relative to P-I) fell from 93%–97% in 2010 to 65%–
70% by 2070 (Figures 2a and 2b). As an alternative way to measure impact on
existing reefs, we also compared changes in suitability values across all 1° grid
cells with present-day coral communities and reefs (i.e., with entries from the
ReefBase v2000 database). These values declined substantially under all
unmitigated RCP scenarios and by 2070 had reached average values as low as 0.49
(RCP 8.5) compared to the P-I average of 0.62 for the BRT model output (Figure
2c, bottom row; MaxEnt values given in Figure S2.7). The pattern of impact does
not scale simply with increasing radiative forcing; instead an impact threshold is
apparent at ~3W/m2. When levels of anthropogenic forcing were below 3 W/m2,
the probabilities on cells currently associated with reefs remained high (Figures 2c
and S2.7), and the area of significantly reduced suitability was confined to within
the central Indo-Pacific Warm Pool (IPWP; Figure S2.1).
In the UVic simulations, application of SRM geoengineering sufficient to return
the average global temperature to P-I levels leaves the tropics on average ~1°C
cooler (Figure 1b), similar to previous findings using fully coupled GCM models
[e.g., Lunt et al., 2008; Irvine et al., 2010]. Because cooling increases CO2
solubility, a subsidiary consequence of this SRM-driven overcooling is that "Arag
is lower than under the unmitigated scenarios (Figure 1d). The net result of cooler
temperatures and further enhanced ocean acidification is that suitabilities for coral
reefs (averaged across cells associated with modern reef sites) are lower under a
geoengineering scenario of radiative forcing returned to 0 W/m2 compared to P-I
(i.e., 1:1 line in Figure 2c for BRT results). In fact, suitabilities for a fully
geoengineering climate are similar to those obtained for unmitigated RCP 4.5 and
RCP 3 scenarios, although this reduction was less significant for MaxEnt (Figure
S2.7). In contrast, application of SRM geoengineering equivalent to reducing the
forcing to 1.5 W/m2 above P-I not only forestalls the projected decline in shallow
water reef habitat suitability across the central Indo-Pacific but also leads to
improved conditions in the central Pacific due to the residual warming there
(Figure 1f; Appendix S2). The probability histograms calculated for currently
designated reef cells (Figure 2c) show that all SRM geoengineering scenarios
where forcing is reduced to 3 or 1.5 W/m2 maintained reasonably favorable
conditions and averages were preserved (0.56–0.62) near the P-I value (0.62).
4. Discussion and Conclusions
In our statistical models, unmitigated climate change leads to an SST-driven
collapse in environmental suitability for shallow water coral reefs, spreading from
the center of the IPWP and across the central Indo-Pacific as radiative forcing
increases beyond 3 W/m2. For a radiative forcing of > 4.5 W/m2, the affected area
encompasses the “Coral Triangle”, the richest region of biodiversity for corals and
reef-associated fauna [e.g., Tittensor et al., 2010]. In contrast, declines in shallow
water habitat suitable for coral reefs are averted in relatively aggressive SRM
geoengineering scenarios in which net radiative forcing is restricted to 3 W/m2
despite the existence of high pCO2. Due to residual warming, forecast
environmental conditions even improved slightly across the central Pacific, a
region sparsely populated in terms of shallow coral reefs, but critical in terms of
connectivity of reef-dependent species across the Pacific basin [e.g., Lessios and
Robertson, 2006; Mora et al., 2012]. Similarly, upwelling regions were generally
less impacted as a consequence of upwelled waters, previously isolated from the
atmosphere, providing some buffering against acidification (Figure 1c).
The difference in modeled response between unmitigated and geoengineered
scenarios reflects the importance placed on SST variables; both MaxEnt and BRT
use a combination of SST variables to explain 50%–60% of the variation in
models trained on present-day global shallow water coral reef distribution [Couce
et al., 2012]. As a result, simulated future SST changes dominate predictions.
Other environmental fields, in particular "Arag, light availability, and nutrients, are
used to reinforce the SST-derived pattern and to model coral reef presence at
regional scales where the correlation with temperature breaks down [Couce et al.,
2012]. Consequently, when global temperatures are controlled by SRM, the
strongest negative responses map onto regions identified as sensitive during model
development to reduced "Arag and light availability: the Coral Triangle, southwest
Pacific, and South China Sea [Couce et al., 2012]. This spatial impact pattern was
also observed in an empirically supported modeling study on the response of
global shallow water coral reefs to future "Arag reductions [Silverman et al., 2009].
The strongest decline in habitat suitability for shallow water coral reefs
corresponds to areas where maximum weekly SST increases above a threshold of
31.9°C and is centered on the IPWP. Shallow water coral reef ecosystems as a
whole are very sensitive to elevated SSTs as evident from the recent observations
of mass bleaching, mortality events, and subsequent reef deterioration associated
with SST anomalies [Hoegh-Guldberg et al., 2007]. However, the model focus on
the IPWP as a thermally sensitive region is supported by observations and
empirical studies of physiological tolerances to thermal stress in reef-forming
species of coral and coralline algae. Reduced thermal tolerance has been linked to
both low SST variability environments [e.g., Ateweberhan and McClanahan,
2010; Teneva et al., 2012] and synergistic stress from reduced "Arag [e.g., Anthony
et al., 2008]. The relative sensitivity of this region is further evident in recent
observations of declining coral cover [Bruno and Selig, 2007] and exceptionally
high susceptibility to mass bleaching events [Donner et al., 2005; Teneva et al.,
2012]. The amelioration of future SST warming is therefore of primary importance
for minimizing impacts in this key region.
The relative dominance of SST in our statistical models helps explain why, in
contrast to Silverman et al. [2009], our projections do not forecast a global
collapse of coral reefs by ca. 560ppm atmospheric CO2. Instead, the potential
presence of coral cover at high pCO2 values (up to 677ppm, by 2070 under RCP
8.5) is consistent with Fabricius et al. [2011], who observed massive Porites
colonies growing within this range of geochemical conditions with no significant
impact on calcification rates. Tropical coral reef ecosystems are treated as a single
entity in our models, so our results should be considered a simplified first order
approximation and cannot be directly compared to the substantial changes in coral
community composition and diversity versus environmental gradient observations
also reported by Fabricius et al. [2011]. The future loss of biodiversity is likely to
be significant under high pCO2, but the models cannot separate potentially
significant shifts in the distributions of individual reef-forming species and so the
modeled habitat suitability response is likely muted. Future use of correlative
models created at the species (of functional type) level may provide a means to
start addressing this question.
To what degree can the statistical model projections be treated as robust in the face
of potential future changes in both variable correlation and spatial patterns? Under
SRM scenarios, the first-order inverse correlation that exists between SST
variables and "Arag in the modern surface ocean no longer holds. As a result, the
two Bioclimatic Envelope Model class types used in our study might have yielded
divergent projections because of their different internal use of correlated variables
[Couce et al., 2012] (Appendix S1). Instead, the strong agreement between the
MaxEnt and BRT predictions (Appendix S2) suggests the models are not over-
relying on present-day correlations between variables, thus increasing confidence
in the projections. There is also an implicit decoupling between specific local
and/or hourly conditions occurring at a reef site and the relatively large spatial
(1°x1° scale) and weekly-to-annual average data employed in our models.
However, as long as local reef environments change in tandem with large-scale
“open ocean” changes, our results should not be substantially biased.
It is important to note that it becomes necessary to extrapolate when variables
exceed the range of present-day environmental values used for model calibration
(e.g., when mean annual SST increases over 31.4°C). Both BRT and MaxEnt
techniques deal with such situations by setting the response outside of training
range at the level set for the nearest most extreme within-training value. A detailed
discussion of the effect of the chosen extrapolation method on the results is given
in Appendix S3. The net result is a constant positive response in the case of
increasing "Arag (e.g., experienced under P-I conditions) and a conservative
assessment of the negative impacts of warming by setting a constant negative
response in the case of higher SSTs. Grid cells with novel conditions for which the
extrapolation method strongly impacts predictions are explicitly shown in the
results (hatched areas in Figures 1e and 1f and in the histograms in Figures 2c and
S2.7). By 2070, these areas of problematic extrapolation affect a minority of cells
where shallow water coral communities and reefs are currently found (0–14%; on
average 2.5%), and conclusions remain unaltered by excluding these areas (e.g.,
the general reduction in shallow reef habitat suitability under all unmitigated RCP
scenarios in Figures 2c and S2.7 is a robust finding). In fact, the extrapolation of a
negative response onto extreme SSTs imposed by both models would be a logical
decision from empirically driven evidence (e.g., thermal damage limits of coral
reviewed in Brown and Cossins [2011]). Significantly, this response implies that
the data set used to calibrate our statistical models contains sites where present-
day shallow water coral reef distribution is already limited by thermal thresholds.
The data set does not, however, include coral reefs from the Red Sea and Arabian
Gulf, which tolerate similar extreme maximum SSTs but are potentially
conditioned by very high SST variability [Ateweberhan and McClanahan, 2010],
because it was not possible to simulate conditions using the UVic model in these
enclosed seas. While assessment of habitat beyond 2070 and under CO2
concentrations higher than the maximum we consider here (677 ppm at year 2070
under RCP 8.5) may be desirable for a fuller and longer-term picture, the utility of
the Bioclimatic Envelope Modeling approach becomes increasing limited as more
of the ocean exceeds training limits.
Overall, our work highlights the complex patterns of global change induced by
even simple (and spatially uniform) geoengineering scenarios, with consequences
that can be non-obvious. Specifically, we find that tropical overcooling by full
geoengineering, together with a relatively low comparative sensitivity to "Arag in
our models, creates an apparent “optimum” for shallow coral reef habitat (this is
particularly evident in the BRT model output; Figures 2a, 2c, and S2.8). This
optimum occurs under environmental conditions corresponding to a partially, but
not fully, mitigated high CO2 climate (i.e., SRM geoengineering of radiative
forcing to 1.5 W/m2 above P-I). A high degree of geoengineering with a global net
residual warming acts to even out surface meridional temperature gradients while
preventing tropical overcooling to the net advantage of tropical corals. This
outcome is possibly exaggerated because terrestrial carbon storage feedback
cannot be explicitly accounted for under the fixed atmospheric CO2 concentrations
of the RCP-based approach. For example, Matthews et al. [2009] found that SRM
could slightly mitigate ocean acidification, although "Arag would still decrease,
due to a simulated increase in terrestrial CO2 uptake and hence atmospheric pCO2
drawdown.
In conclusion, while SRM geoengineering fails to tackle the causes or
consequences of ocean acidification, the detrimental effect of higher SSTs appears
to strongly outweigh the impacts of reduced "Arag for tropical shallow water coral
reefs when treated as a single entity. Further studies are needed to resolve potential
changes in coral reef community composition and biodiversity; however, severe
reductions in the area of suitable shallow water coral reef habitat might be averted
if anthropogenic forcing is limited # 3 W/m2 or returned below this level via SRM.
Overall, our work highlights the need for a multistressor and spatially explicit
framework in assessing ecological implications of future global change, whether
mitigated or not, so that the complex patterns of induced change and the nonlinear
combinations of environmental pressures can be adequately evaluated.
Acknowledgments. This work was supported by a U. Bristol postgraduate scholarship to E.C., a
UK NERC postgraduate studentship to P.J.I., a Royal Society Advanced Fellowship to A.R., and
an RCUK Academic Fellowship to E.J.H. L.J.G. was funded by the UK Ocean Acidification
Research Program (NE/H017453/1) and EPSRC grant EP/I014721/1. We thank H. Russell and A.
Wilmot-Sitwell for additional funding to support E.C.
References
Angel, R. (2006), Feasibility of cooling the Earth with a cloud of small spacecraft near the inner
Lagrange point (L1), Proc. Natl. Acad. Sci. U. S. A., 103(46), 17184–17189,
doi:10.1073/pnas.0608163103.
Anthony, K. R. N., D. I. Kline, G. Diaz-Pulido, S. Dove, and O. Hoegh-Guldberg (2008), Ocean
acidification causes bleaching and productivity loss in coral reef builders, Proc. Natl. Acad. Sci.
U. S. A., 105(45), 17442–17446, doi:10.1073/pnas.0804478105.
Ateweberhan, M., and T. R. McClanahan (2010), Relationship between historical sea-surface
temperature variability and climate change-induced coral mortality in the western Indian Ocean,
Mar. Pollut. Bull., 60, 964–970.
Brown, B. E., and A. R. Cossins (2011), The Potential for Temperature Acclimatisation of Reef
Corals in the Face of Climate Change, in coral Reefs: An Ecosystem in Transition, edited by Z.
Dubinsky, and N. Stambler, pp. 421–433, Springer Dordrecht.
Bruno, J. F., and E. R. Selig (2007), Regional decline of coral cover in the Indo-Pacific: Timing,
extent, and subregional comparisons, PLoS ONE, 2, e711, doi:10.1371/journal.pone.0000711.
Couce, E., A. Ridgwell, and E. J. Hendy (2012), Environmental controls on the global
distribution of shallow water coral reefs, J. Biogeogr., 39(8), 1508–1523, doi:10.1111/j.1365-
2699.2012.02706.X.
Crutzen, P. J. (2006), Albedo enhancement by stratospheric sulfur injections: A contribution to
resolve a policy dilemma?, Climatic Change, 77(3-4), 211–219, doi:10.1007/s10584-006-9101-y.
Donner, S. D., W. J. Skirving, C. M. Little, M. Oppenheimer, and O. Hoegh-Guldberg (2005),
Global assessment of coral bleaching and required rates of adaptation under climate change,
Global Change Biol., 11(12), 2251–2265, doi:10.1111/j.1365-2486.2005.01073.X.
Fabricius, K. E., C. Langdon, S. Uthicke, C. Humphrey, S. Noonan, G. De’ath, R. Okazaki, N.
Muehllehner, M. S. Glas, and J. M. Lough (2011), Losers and winners in coral reefs acclimatized
to elevated carbon dioxide concentrations, Nat. Clim. Change, 1(3), 165–169.
Friedman, J. H. (2001), Greedy function approximation: A gradient boosting machine, Ann. Stat.,
29(5), 1189–1232.
Hoegh-Guldberg, O., et al. (2007), Coral reefs under rapid climate change and ocean
acidification, Science, 318(5857), 1737–1742, doi:10.1126/ science.1152509.
Hönisch, B., et al. (2012), The geological record of ocean acidification, Science, 335(6072),
1058–1063.
Irvine, P. J., A. Ridgwell, and D. J. Lunt (2011), Climatic effects of surface albedo
geoengineering, J. Geophys. Res., 116, D24112, doi:10.1029/ 2011JD016281.
Kleypas, J. A., R. W. Buddemeier, D. Archer, J. P. Gattuso, C. Langdon, and B. N. Opdyke
(1999), Geochemical consequences of increased atmospheric carbon dioxide on coral reefs,
Science, 284(5411), 118–120, doi:10.1126/science.284.5411.118.
Latham, J., and M. H. Smith (1990), Effect on global warming of wind-dependent aerosol
generation at the ocean surface, Nature, 347(6291), 372–373, doi:10.1038/347372a0.
Lessios, H. A., and D. R. Robertson (2006), Crossing the impassable: genetic connections in 20
reef fishes across the eastern Pacific barrier, Proc. R. Soc. B., 273(1598), 2201–8,
doi:10.1098/rspb.2006.3543.
Lunt, D. J., A. Ridgwell, P. J. Valdes, and A. Seale (2008), “Sunshade world”: A fully coupled
GCM evaluation of the climatic impacts of geoengineering, Geophys. Res. Lett., 35(12), L12710,
doi:10.1029/ 2008GL033674.
Matthews, H. D., L. Cao, and K. Caldeira (2009), Sensitivity of ocean acidification to
geoengineered climate stabilization, Geophys. Res. Lett., 36, L10706,
doi:10.1029/2009GL037488.
Mora, C., E. A. Treml, J. Roberts, K. Crosby, D. Roy, and D. P. Tittensor (2012), High
connectivity among habitats precludes the relationship between dispersal and range size in
tropical reef fishes, Ecography, 35(1), 89–96.
Moss, R. H., et al. (2010), The next generation of scenarios for climate change research and
assessment, Nature, 463(7282), 747–756, doi:10.1038/nature08823.
Phillips, S. J., R. P. Anderson, and R. E. Schapire (2006), Maximum entropy modeling of species
geographic distributions, Ecol. Modell., 190(3–4), 231–259,
doi:10.1016/j.ecolmodel.2005.03.026.
Schmittner, A., A. Oschlies, H. D. Matthews, and E. D. Galbraith (2008), Global
Biogeochem.Cycles, 22(1), GB1013, doi:10.1029/2007GB002953.
Silverman, J., B. Lazar, L. Cao, K. Caldeira, and J. Erez (2009) Coral reefs may start dissolving
when atmospheric CO2 doubles. Geophys. Res. Lett.,36, L05606, doi:10.1029/2008GL036282.
Teneva, L., M. Karnauskas, C. A. Logan, L. Bianucci, J. C. Currie, and J. A. Kleypas (2012),
Predicting coral bleaching hotspots: The role of regional variability in thermal stress and potential
adaptation rates, Coral Reefs, 31(1), 1–12, doi:10.1007/s00338-011-0812-9.
Thuiller, W., S. Lavorel, M. B. Araújo, M. T. Sykes, and I. C. Prentice (2005), Climate change
threats to plant diversity in Europe, Proc. Natl. Acad. Sci. U. S. A., 102, 8245–8250,
doi:10.1073/pnas.0409902102.
Tittensor, D. P., et al. (2010), Global patterns and predictors of marine biodiversity across taxa,
Nature, 466, 1098–1011, doi:10.1038/nature09329.
Turley, C., M. Eby, A. J. Ridgwell, D. N. Schmidt, H. S. Findlay, C. Brownlee, U. Riebesell, V.
J. Fabry, R. A. Feely, and J. P. Gattuso (2010), The societal challenge of ocean acidification, Mar.
Pollut. Bull., 60(6), 787–792, doi:10.1016/j.marpolbul.2010.05.006.
Vergara, S., et al. (2000), ReefBase 2000: improving policies for sustainable management of
coral reefs. Ver. 2000. 164p. + CD-ROM. ICLARM, Manila, Philippines.
Weaver, A. J., et al. (2001), The UVic Earth System Climate Model: Model description,
climatology, and applications to past, present and future climates, Atmos.-Ocean, 39(4), 361–428.