Century to multi-century sea level rise projections from CMIP5 models

Authors


Abstract

[1] Long-term projections of global-ocean thermal expansion (GTE) and the dynamic sea level (DSL) change are analyzed with 34 new CMIP5 models and under three greenhouse-gas emission scenarios. Multi-model ensemble mean (MEM) and ensemble standard deviation are calculated to identify robust features and quantify uncertainty. While the MEM of GTE shows moderate difference by 2100, with magnitudes of 13, 18 and 28 cm in RCP2.6, RCP4.5 and RCP8.5, respectively, it increases and diverges significantly by 2300, with magnitudes of 21, 52 and 119 cm in the three scenarios. Model-to-model spread seems reduced in CMIP5 compared to CMIP3. The MEM changes of the DSL show similar patterns between different RCPs, but with progressively larger magnitudes in RCP2.6, RCP4.5 and RCP8.5. Notable features identified previously in the CMIP3 projections also occur in CMIP5, indicating their robustness across generations of climate model and emission scenario. The CMIP5 models still show disagreement in projecting the DSL changes, even under the same external forcing.

1. Introduction

[2] Thermal expansion and ocean dynamics are important factors in understanding and projecting the magnitude and spatial variability of sea level rise (SLR) [Bindoff et al., 2007; Milne et al., 2009]. Global-ocean thermal expansion (GTE) is the largest contribution and accounts for 30–40% of the total global SLR since the 1970s [Church et al., 2011]. The significant and non-uniform pattern of SLR observed by satellites since the early-1990s is mainly attributable to the local thermosteric effect and ocean dynamics such as those related to the Pacific Decadal Oscillation [Becker et al., 2012; Bromirski et al., 2011]. Unlike land ice melt and ice sheet dynamics that are inadequately represented, GTE and the change of the dynamic sea level (DSL, i.e., sea surface height relative to the geoid) are explicitly simulated or can be accurately diagnosed in current coupled general circulation models [Griffies and Greatbatch, 2012]. Nonetheless, previous studies show uncertainties in projecting GTE and the DSL change over the 21st century [Gregory et al., 2001; Pardaens et al., 2011; Slangen et al., 2012; Yin et al., 2010]. The GTE projected by the Coupled Model Intercomparison Project Phase 3 (CMIP3) models diverges rapidly with time in the 21st century [Meehl et al., 2007, Figure 10.31]. For the DSL change, qualitatively robust features can be identified only in some particular regions, such as the faster and larger dynamic SLR north of the Gulf Stream and along the northeast coast of North America [Yin et al., 2009], and the pronounced belt-like pattern in the Southern Ocean [Yin et al., 2010].

[3] Since CMIP3, new generations of climate and Earth system models have been developed at modeling centers worldwide, many of which have been participating in CMIP5 [Taylor et al., 2012]. The CMIP5 archive has matured recently to allow a systematic investigation of future GTE and the DSL change. It is of great interest to examine whether the 21st century projections of GTE and the DSL from CMIP5 are consistent with those of CMIP3 and whether uncertainties are reduced. In addition, GTE and the DSL change reflect heat storage and density change in all ocean layers including the deep ocean. So they could continue for a very long time until the coupled climate system reaches a new equilibrium state and the radiative imbalance at the top of the atmosphere diminishes. The new Representative Concentration Pathways (RCPs) used in CMIP5 specify the greenhouse-gas concentrations until 2500 [Moss et al., 2010]. This facilitates the SLR studies on longer time scale and with larger model ensemble sizes in CMIP5 compared to CMIP3. This paper focuses on century to multi-century projections of GTE and the DSL change from 34 CMIP5 models.

2. Models and Method

[4] Table 1 lists the CMIP5 models with the global thermosteric (i.e., GTE) and DSL data available at the CMIP5 archive (http://cmip-pcmdi.llnl.gov/cmip5/) and used by the present study. For some models a set of realizations have been performed with slightly different initial conditions (Table 1). To generate the multi-model ensemble mean (MEM), the mean of all realizations for a particular model is calculated first.

display math

i and j are model and realization index, respectively. Δhij is the global thermosteric SLR relative to 2006. ηi,jHIS and ηi,jRCP are the DSLs in 1986–2005 of the historical runs and in 2091–2100 or 2291–2300 of the projection runs, respectively. math formula and math formula are the mean of all realizations. The MEM of GTE ( math formula) and the DSL change ( math formula) is then obtained by

display math

For simplicity all available models are used and carry equal weight in the MEM calculation. It should be noted that the DSL by definition is the sea level deviation from the global mean in models. It always has a zero global mean. The sum of GTE and the DSL change gives the total SLR in the CMIP5 models. The contributions from land ice melt and the related global halosteric SLR are not considered due to model limitations. Other factors besides ocean dynamics that can influence local sea level (e.g., vertical land motion and the geoid change induced by land ice melt) are also not taken into account here [Milne et al., 2009].

Table 1. 34 CMIP5 Models Used in the Present Studya
ModelRCP2.6RCP4.5RCP8.5
210023002100230021002300
  • a

    The numbers indicate the realizations used for the calculation of the DSL change. Crosses indicate available data for the GTE calculation. Detailed description of the models can be found at http://pcmdi3.llnl.gov/esgcet.

ACCESS1.0  1 1 
ACCESS1.3  1 1 
BCC-CSM1.11 ×1 ×  1 ×1 ×
BCC-CSM1.1-M  1 × 1 × 
CanESM25 ×1 ×5 ×1 ×5 × 
CCSM45 ×   5 × 
CESM1-BGC  1 1 
CESM1-WACCM  1 1 
CMCC-CM  1 × 1 × 
CMCC-CMS  1 × 1 × 
CNRM-CM5  1 ×15 ×1 ×
CSIRO-Mk3.610 × 10 × 10 × 
EC-EARTH1 5 5 
FGOALS-g21 1 1 
FGOALS-s21 313 
FIO-ESM3 3 3 
GFDL-ESM2M1 × 1 × 1 × 
GFDL-ESM2G1 × 1 × 1 × 
GFDL-CM31 × 1 ×1 ×1 × 
GISS-E2-R3 ×3 ×3 ×3 ×3 ×3 ×
HadGEM2-CC  1 × 3 × 
HadGEM2-ES4 ×1 ×4 ×1 ×4 ×1 ×
INM-CM4  1 × 1 × 
IPSL-CM5A-LR3 ×1 ×4 ×1 ×4 ×1 ×
IPSL-CM5A-MR1 × 1 × 1 × 
IPSL-CM5B-LR  1 1 
MIROC53 × 3 × 3 × 
MIROC-ESM1 × 1 ×1 ×1 × 
MIROC-ESM-CHEM1 × 1 × 1 × 
MPI-ESM-LR3 ×1 ×3 ×1 ×3 ×1 ×
MPI-ESM-MR1 × 1 × 1 × 
MRI-CGCM31 × 1 × 1 × 
NorESM1-M1 × 1 ×1 ×1 × 
NorESM1-ME1 1 × 1 × 

[5] Several issues related to the DSL need attention. Many CMIP5 models employ curvilinear grids for the ocean components. To facilitate the MEM calculation, all modeling data are first converted and interpolated to a common longitude-latitude grid of 0.5° × 0.5° resolution. In addition, some models show clear DSL and GTE drifts in the pre-industrial control runs even after long spin-up periods. The statistically significant drifts have been corrected before generating the MEM. Finally, the DSL in the MIROC5 and GISS-E2-R models has been converted to the effective DSL which include the equivalent water of sea ice [Griffies et al., 2009; Yin et al., 2010].

[6] Three concentration-driven scenarios are considered - RCP2.6 (low), RCP4.5 (medium) and RCP8.5 (high) [Moss et al., 2010] - to explore the full range of GTE and the DSL change. 34 CMIP5 models have made the 21st century projections, while only a subset of them (less than 10) extended the experiments to 2300 (Table 1 and auxiliary material). Due to the different ensemble sizes, it is necessary to properly interpret different levels of uncertainty during the two time periods.

3. Results

[7] Due to the very different strength of greenhouse-gas emissions, the 21st century GTE shows deceleration, roughly linear trend and acceleration in RCP2.6, RCP4.5 and RCP8.5, respectively (Figure 1). By the end of this century, GTE can reach 13 ± 3, 18 ± 3 and 28 ± 3 cm (ensemble mean ± ensemble standard deviation) in RCP2.6, RCP4.5 and RCP8.5, respectively. These magnitudes are comparable with the previous estimates from CMIP3 [Meehl et al., 2007, Table 10.7]. In addition, model-to-model spread seems reduced in CMIP5 compared to CMIP3 [Meehl et al., 2007]. The GTE curves show less divergence and tend to group together over the 21st century, especially in the strong forcing case of RCP8.5 (Figure 1).

Figure 1.

Individual and MEM projections of GTE (m) under (a) RCP2.6, (b) RCP4.5 and (c) RCP8.5. The curves show the GTE relative to 2006. Thick black lines indicate the MEM. The discontinuity at 2100 is due to the change of ensemble size.

[8] The divergence of the GTE curves continue beyond 2100 (Figure 1). By 2300 the MEM GTE increases to 21 ± 5, 52 ± 10 and 119 ± 15 cm in RCP2.6, RCP4.5 and RCP8.5, respectively. It should be noted that different model ensembles are used in different RCPs for this period (Table 1). There is a considerable difference (1 m) between RCP2.6 and RCP8.5, highlighting that GTE on the multi-century time scale is highly dependent on emission scenarios. The GTE difference between RCPs is moderate in the 21st century when the ocean climate system makes the initial response. By 2300 GTE tends to level off in RCP2.6 as global oceans approach a quasi-equilibrium state more quickly in response to a small forcing. By contrast, it shows little sign of slowdown in RCP4.5 and RCP8.5, indicating the ocean is still taking up a large amount of heat.

[9] While GTE reflects the globally-averaged density change of the ocean, this change is not spatially uniform. The related DSL displays pronounced anomaly patterns in the CMIP5 projections. In some ocean regions, the DSL change can be comparable in magnitude to GTE, thereby either offsetting or doubling the local SLR rate. In general, the MEM patterns of the DSL change from CMIP5 (Figures 2a, 2c, and 2e) are similar to those of CMIP3 by the end of the 21st century [Pardaens et al., 2011; Yin et al., 2010]. Some outstanding features repeatedly appear in the high-latitude and polar regions, such as the notable dynamic SLR north of the Gulf Stream, in the Beaufort gyre and North Pacific subtropical gyre, as well as the belt-like pattern in the Southern Ocean. The significant decline and strengthening of the DSL gradient across the Gulf Stream and Antarctic Circumpolar Current, respectively, are very pronounced. In addition, the Arctic and northern North Atlantic undergo an overall dynamic SLR, resulting from ocean freshening, the projected slowdown of the Atlantic meridional overturning circulation and rapid ocean warming in the intermediate and deep layers [Yin et al., 2011]. Despite similar anomaly patterns of the DSL, the magnitude differs significantly in different RCPs (Figure 2). The DSL shows minimum changes in RCP2.6 while the magnitude reaches maximum in RCP8.5. Unlike the high latitudes the DSL change in the broad tropical oceans is insignificant, so that local SLRs basically follow the global mean on centennial time scales.

Figure 2.

MEM projections of the DSL change (m). Shown are the projected anomalies in (a, c, e) 2091–2100 and (b, d, f) 2291–2300 relative to 1986–2005 for (top) RCP2.6, (middle) RCP4.5 and (bottom) RCP8.5. See auxiliary figures for individual model projections.

[10] By the end of the 23rd century, the notable features of the DSL anomalies weaken and even reverse in RCP2.6, but continue to strengthen in RCP4.5 and RCP8.5 (Figures 2b, 2d, and 2f and auxiliary figures). RCP2.6 assumes near-zero emission by 2100 and therefore represents a strong mitigation and very low emission scenario. The CO2concentration peaks and then declines in the 21st century, and eventually stabilizes at a level close to or below the present-day value after 2100. Under this diminishing external forcing, the DSL tends to recover to the original pattern in the 20th century, except that the belt-like anomalies persist in the Southern Ocean (Figure 2b). By contrast, RCP8.5 is a non-mitigation and very high emission scenario as the CO2 concentration increases to approximately 2000 ppm by 2300. So the DSL displays the strongest response in this case (Figure 2f). For example, the dynamic SLR along the Arctic coast and densely populated northeastern North America and northern Europe can be greater than 35 cm by 2300, compared to 15–20 cm by 2100. These additional rises superimposed on the global mean SLR would lead to greater vulnerability of these coastal regions [Sallenger et al., 2012; Yin et al., 2009]. For the medium emission scenario of RCP4.5, the magnitude of the DSL change is between those of RCP2.6 and RCP8.5.

[11] Another striking feature in the MEM is the DSL depression south of 50°S (Figure 2f), related to the strengthening of the westerlies and Antarctic Circumpolar Current [Yin et al., 2010]. In RCP8.5 the DSL depression can reach 60–80 cm by 2300, thereby largely offsetting GTE in this region (Figure 1c). In addition to potential impact on the Antarctic ice sheet and floating ice shelves, this great DSL depression aggravates the dynamic SLR elsewhere through ocean mass redistribution (Figure 2f). Other notable features include the dynamic SLR in the North Pacific subtropical gyre [Suzuki et al., 2005] and the slight DSL depression in the southeastern Pacific [Timmermann et al., 2010].

[12] In terms of uncertainty, the significant model-to-model disagreement in projecting the DSL change is not reduced in CMIP5 (Figure 3) compared to CMIP3 [Yin et al., 2010]. Large MEM DSL changes are usually associated with large model spread, such as in the Arctic, Southern Ocean and northern North Atlantic (auxiliary figures). Model spread becomes larger with time and the increase in external forcing (Figure 3).

Figure 3.

Ensemble standard deviations (m) in projecting the DSL changes by the end of the (a, c, e) 21st and (b, d, f) 23rd century for (top) RCP2.6, (middle) RCP4.5 and (bottom) RCP8.5.

4. Discussion and Conclusions

[13] The new set of experiments under the CMIP5 protocol allows systematic studies on the long-term GTE and DSL change. With the CMIP5 data, it is critical to estimate the possible range of SLR and its components in addition to the most likely values, thereby providing sufficient information for coastal planning. Based on the projections from 34 CMIP5 models, it is found that the 21st century projections of GTE and the DSL change are consistent with those of CMIP3 when similar scenarios are compared. On multi-century time scales, the results are highly scenario-dependent. For the best-case scenario of RCP2.6 in which the global warming magnitude is kept below 2°C, the GTE projection is about 13 cm by 2100 and does not exceed 21 cm by 2300. The DSL shows minimum changes and local SLRs at most ocean regions just follow the global mean. These results suggest that strong mitigation is effective at reducing the risk of future large SLR [Washington et al., 2009], provided that polar ice sheets do not show nonlinear and tipping behavior in the next century. Over a period of several centuries, coastal communities may be able to adapt to this minimum SLR.

[14] Under the highest emission scenario of RCP8.5, GTE reaches 28 cm by 2100 and 119 cm by 2300. Three quarters of the 119 cm SLR occur after the 21st century. These magnitudes are comparable to but slightly smaller than the projected values from simplified climate models [Schewe et al., 2011; Solomon et al., 2009]. The DSL also shows the strongest response. For example, the northeast coast of North America can experience 15–20 cm and more than 35 cm dynamic SLR by 2100 and 2300, respectively. So even without considering the contribution from the ice sheet melt, the total SLR in this region is already close to half meter by 2100 and one and a half meters by 2300. The consequence would be significant for the coastal communities [Strauss et al., 2012]. For the mid-range scenario of RCP4.5, GTE is 18 and 52 cm by 2100 and 2300, respectively. The DSL shows medium changes between RCP2.6 and RCP8.5.

[15] Under a particular scenario, the CMIP5 models display less spread in projecting GTE. One possible reason is that the range of the projected global surface warming is narrowed in a RCP compared to the similar scenario used in CMIP3 [Rogelj et al., 2012], as equilibrium GTE is proportional to global surface warming. In addition to climate sensitivity, GTE and ocean heat uptake are also influenced by the sensitivity of the North Atlantic Deep Water formation [Knutti and Stocker, 2000], ocean mixing and subduction processes [Church et al., 1991; Griffies and Greatbatch, 2012], and the control ocean climate states (R. Hallberg, Sensitivity of 21st century global-mean steric sea level rise to ocean model formulation, submitted toJournal of Climate, 2012). The role of these factors in the MEM and ensemble standard deviation of the GTE projections is beyond the scope of this paper. As a globally-integrated variable, the GTE projected by the CMIP5 models (Figure 1) is consistent with those from simple 1-D, 2-D or intermediate climate models [e.g.,Schewe et al., 2011], suggesting that GTE may not critically depend on small scale features resolved in 3-D general circulation models.

[16] The CMIP5 models still show disagreement in projecting the DSL change (auxiliary figures). The DSL projections may be critically influenced by model resolution (e.g., coarse-resolution models tend to smooth the anomalies), numerical schemes and parameterizations, sensitivity of ocean circulation to external forcing, wind stress changes [Han et al., 2010] and detailed patterns of temperature and salinity anomalies. Nonlinear processes may occur associated with the DSL changes. Therefore, it is more challenging to reduce the uncertainty in the DSL projections. Given that temporal and spatial variability of SLR is significant, a better understanding of the causes of model spread and the role of natural variability and external forcing in the DSL predictions and projections should be a high priority in the climate and sea level rise modeling community.

Acknowledgements

[17] The work is supported by the author's start-up funds at the University of Arizona. The author thanks J. Gregory, S. Griffies, R. Hallberg, W. Hurlin, J. Krasting, S. Malyshev, J. Overpeck, R. Stouffer and an anonymous reviewer for discussions and comments. I acknowledge the World Climate Research Programme, the U.S. Department of Energy, and the climate modeling groups for producing and making their model output available.

[18] The Editor thanks two anonymous reviewers for their assistance evaluating this paper.

Ancillary