Future habitat suitability for coral reef ecosystems under global warming and ocean acidification

Abstract

Rising atmospheric CO2 concentrations are placing spatially divergent stresses on the world's tropical coral reefs through increasing ocean surface temperatures and ocean acidification. We show how these two stressors combine to alter the global habitat suitability for shallow coral reef ecosystems, using statistical Bioclimatic Envelope Models rather than basing projections on any a priori assumptions of physiological tolerances or fixed thresholds. We apply two different modeling approaches (Maximum Entropy and Boosted Regression Trees) with two levels of complexity (one a simplified and reduced environmental variable version of the other). Our models project a marked temperature-driven decline in habitat suitability for many of the most significant and bio-diverse tropical coral regions, particularly in the central Indo-Pacific. This is accompanied by a temperature-driven poleward range expansion of favorable conditions accelerating up to 40-70 km per decade by 2070. We find that ocean acidification is less influential for determining future habitat suitability than warming, and its deleterious effects are centered evenly in both hemispheres between 5° and 20° latitude. Contrary to expectations, the combined impact of ocean surface temperature rise and acidification leads to little, if any, degradation in future habitat suitability across much of the Atlantic and areas currently considered 'marginal' for tropical corals, such as the eastern Equatorial Pacific. These results are consistent with fossil evidence of range expansions during past warm periods. In addition, the simplified models are particularly sensitive to short-term temperature variations and their projections correlate well with reported locations of bleaching events. Our approach offers new insights into the relative impact of two global environmental pressures associated with rising atmospheric CO2 on potential future habitats, but greater understanding of past and current controls on coral reef ecosystems is essential to their conservation and management under a changing climate.

Keywords: Bioclimatic Envelope Modeling; MaxEnt; boosted regression trees; coral reef ecosystems; global warming; maximum entropy; ocean acidification.

Figure 1

Current global distribution of coral reefs (red) according to ReefBase v2000 data set. The 1990 UVic-predicted fields for aragonite saturation (yellow–green color bar) and mean annual sea surface temperature (gray/black contour lines) used to train the Bioclimatic Envelope Models are also shown.

Figure 2

Predicted changes in suitability for coral reef ecosystems under the A2 scenario for future change in sea surface temperature (SST) variables only; 2010 (a), 2040 (b) and 2070 (c). Histograms show predicted suitability values for the 1124 presence sites of the study grid (i.e., cells currently with reefs according to the ReefBase v2000 data). All environmental fields apart from SST variables were kept constant at their present values (used in model training). Areas where projections are less reliable (i.e., SST out of training range and MaxEnt clamping value >0.1; see Data S2) are indicated by the hatched pattern, in both maps and histograms. The figure shows BRT_OPT model's results (see figure S3.1 in Data S3 for MaxEnt_OPT's projections).

Figure 3

Predicted changes in suitability for coral reef ecosystems under the A2 scenario for future change in aragonite saturation only; 2010 (a), 2040 (b) and 2070 (c). Histograms show predicted suitability values for the 1124 presence sites of the study grid (i.e., cells currently with reefs according to the ReefBase v2000 data). All environmental fields apart from Ω_Arag were kept constant at their present values (used in model training). Unlike the SST results (Fig. 2), changes in Ω_Arag do not give rise to extrapolation uncertainties by producing novel conditions. The figure shows BRT_OPT results (see figure S3.2 in Data S3 for MaxEnt_OPT projections).

Figure 4

Predicted suitability of coral reef presence as a function of latitude, when changing sea surface temperature variables only (a), aragonite saturation only (b) and both factors simultaneously, for the entire ocean (c) and restricted to shallow waters (d). Different colors show variations over time under the A2 scenario. All figures have been made averaging the modeled suitability values in 5° latitudinal bands, and show MaxEnt_OPT model's results (see figure S3.3 in Data S3 for the equivalent BRT_OPT projections).

Figure 5

Projected change in suitability for coral reef ecosystems between 1990 and 2070 under the A2 scenario, considering simultaneous changes in aragonite saturation and surface ocean temperature. The projected response has been averaged across both ‘OPT’ models and the MaxEnt_SIM model (models defined in Table 1; BRT_SIM's results were excluded from the average, as discussed in the text and Data S3). Green dashed line indicates null projected change, whereas the hatched pattern identifies areas with novel conditions where projections are less reliable, due to a significant impact of the chosen extrapolation method (as explained in Methods). The sites of past (green triangles) and present (orange diamonds) range expansion of coral are also indicated, based on data from Lighty et al. (1978), Veron (1992), Marsh (1993), Vargas-Ángel et al. (2003), Greenstein & Pandolfi (2008), Woodroffe et al. (2010), and Yamano et al. (2011). Black crosses indicate coral reef distribution during the last interglacial period (data from Kiessling et al., 2012).

Figure 6

Average global suitability for coral reef presence in shallow water areas over time for boosted regression trees (a) and MaxEnt (b) models. Color of the line refers to the CO₂ emission scenario, with red, blue, and green for A2, A1B, and B1, respectively, while the pattern of the line represents the model, with continuous lines for ‘OPT’, and dashed for ‘SIM’ models. Suitability values have been normalized to 2010 average.

Figure 7

Changes in suitability for coral reef ecosystems between 1990 and 2010 predicted by BRT_SIM (left) and MaxEnt_SIM (right) and the point location of over 3500 bleaching events taking place between 2008 and 2012 according to the ReefBase database. The changes in suitability assigned to grid cells currently with reefs (c, f) provide the baseline to which those corresponding to the bleaching events’ locations (d, g) should be compared (in figures c, d, f and g the mean and standard deviation are indicated). The p-values from a Welsh's t-test comparing the suitability distribution of bleaching events and that of all reef cells is also shown in d and g. Figures e and h illustrate the number of bleaching events within a cell as a function of the change in suitability assigned to that cell.