Big Jump of Record Warm Global Mean Surface Temperature in 2014–2016 Related to Unusually Large Oceanic Heat Releases
Abstract
A 0.24°C jump of record warm global mean surface temperature (GMST) over the past three consecutive record-breaking years (2014–2016) was highly unusual and largely a consequence of an El Niño that released unusually large amounts of ocean heat from the subsurface layer of the northwestern tropical Pacific. This heat had built up since the 1990s mainly due to greenhouse-gas (GHG) forcing and possible remote oceanic effects. Model simulations and projections suggest that the fundamental cause, and robust predictor of large record-breaking events of GMST in the 21st century, is GHG forcing rather than internal climate variability alone. Such events will increase in frequency, magnitude, and duration, as well as impact, in the future unless GHG forcing is reduced.
1 Introduction
The annual mean global mean surface temperature (GMST) broke the previous records 3 years in a row during 2014–2016 and by a large margin. The magnitude and duration of this record-breaking event were highly unusual (Mann et al., 2017) and exceeded previous events associated with strong El Niño. As a consequence, global warming has passed the 1°C milestone relative to the preindustrial period (1851–1880), with some monthly GMST values in 2015 and 2016 even close to the 1.5°C threshold. This jump of record warm GMST also represents a sizable fraction of the 2°C stabilization goal set by the 2015 Paris Agreement (Paris Agreements, 2015) to avoid dangerous climate change. The record high monthly and annual mean GMSTs in 2015 and 2016 were coincident with outbreaks of extreme weather worldwide (Herring et al., 2016, 2018), extensive melting of polar ice (Nicolas et al., 2017; Simpkins, 2017), and the third ever global coral bleaching event (NOAA, 2015). Given these significant impacts, it is extremely important and timely to understand the mechanism for this rapid warming and provide perspectives for future large record-breaking events of GMST. Here we examine various observational data and the simulations and projections by 40 climate models to address these issues.
2 Data and Models
Table 1 summarizes the observational data sets used in the present study. Six data sets are available for GMST including Hadley Center and Climate Research Unit analysis version 4 (HadCRUT4) (Morice et al., 2012), Goddard Institute for Space Studies (GISS) Surface Temperature Analysis (GISTEMP) (Hansen et al., 2010), NOAA Global Surface Temperature (NOAAGlobalTemp) (Vose et al., 2012), Japanese Meteorological Agency (JMA) (Ishihara, 2006), Berkeley (Rohde et al., 2013), and Cowtan&Way (Cowtan & Way, 2014). The detailed data descriptions and uncertainty quantification can be found in the references (Figure S1 in the supporting information). All six GMST data sets were downloaded as of 7 March 2017.
| Data set | Institute | Period | Reference | Website | |
|---|---|---|---|---|---|
| GMST | HadCRUT4 | UK Met Office Hadley Centre | 1850–2016 | (Morice et al., 2012) | http://www.metoffice.gov.uk/hadobs/hadcrut4/data |
| GISTEMP | NASA GISS | 1880–2016 | (Hansen et al., 2010) | http://data.giss.nasa.gov/gistemp/ | |
| NOAAGlobalTemp | NOAA NCEI | 1880–2016 | (Vose et al., 2012) | https://www.ncdc.noaa.gov/data-access/marineocean-data/noaa-global-surface-temperature-noaaglobaltemp | |
| JMA | Japanese Meteorol. Agency | 1891–2016 | (Ishihara, 2006) | http://ds.data.jma.go.jp/tcc/tcc/products/gwp/temp/ann_wld.html | |
| Berkeley | Berkeley Earth | 1850–2016 | (Rohde et al., 2013) | http://berkeleyearth.org/ | |
| Cowtan&Way | University of York | 1850–2016 | (Cowtan & Way, 2014) | http://www-users.york.ac.uk/~kdc3/papers/coverage2013/series.html | |
| OHC | Levitus | NOAA NCEI | 1955–2016 | (Levitus et al., 2012) | https://www.nodc.noaa.gov/OC5/3M_HEAT_CONTENT/ |
| Ishii | Japanese Meteorol. Agency | 1950–2016 | (Ishii & Kimoto, 2009) | http://www.data.jma.go.jp/gmd/kaiyou/english/ohc/ohc_data_en.html | |
| Argo | International Argo Program | 2004–2016 | (Roemmich & Gilson, 2009) | http://www.argo.ucsd.edu | |
| Sea Level | Tide gauge | PSMSL |
1950–2015 1948–2016 |
(Holgate et al., 2013) | http://www.psmsl.org/data/obtaining |
| Altimetry | AVISO | 1993-2016 | (Ducet et al., 2000) | http://www.aviso.altimetry.fr/en/data/data-access.html | |
| Global mean Reconstruction | CSIRO | 1880–2013 | (Church & White, 2011) | http://www.cmar.csiro.au/sealevel/sl_data_cmar.html | |
| ENSO | MEI | NOAA ESRL | 1950–2016 | (Wolter & Timlin, 2011) | https://www.esrl.noaa.gov/psd/enso/mei/ |
| Niño3.4 | NOAA ESRL | 1870–2016 | (Rayner et al., 2003) | https://www.esrl.noaa.gov/psd/gcos_wgsp/Timeseries/Nino34/ | |
| PDO | PDO | NOAA NCEI | 1854–2016 | (Mantua et al., 1997) | https://www.ncdc.noaa.gov/teleconnections/pdo/ |
- Notes. Abbreviations: CSIRO, Commonwealth Scientific and Industrial Research Organisation; ESRL, Earth System Research Laboratory; NCEI, National Centers for Environmental Information; PSMSL, Permanent Service for Mean Sea Level.
As closely related fields, we use three data sets for ocean heat content (OHC) and temperatures including Levitus (Levitus et al., 2012), Ishii (Ishii & Kimoto, 2009), and Argo (Roemmich & Gilson, 2009) (Table 1). The OHC anomalies are relative to the WOA09, 1981–2010 and 2004–2015 climatology in the three data sets, respectively. The values are integrated in the upper 700 m and across each grid box. It should be noted that the three data sets are not independent. The Argo coverage is usually not considered adequate until 2005.
We choose two tide gauge data in the northwestern tropical Pacific (NWP) with relatively long and complete records (Table 1) (Holgate et al., 2013). The relative sea level data at Kwajalein (167.74°E, 8.73°N) and Guam (144.65°E, 13.44°N) span from 1950–2015 and 1948–2016, respectively. For satellite altimetry data, we use the Archiving, Validation, and Interpretation of Satellite Oceanographic data (AVISO) (Table 1) (Ducet et al., 2000). The data show dynamic sea level (i.e., departure of sea surface height from the geoid) at an eddy-resolving scale and over a global domain. We find that the tide gauge and altimetry data are highly correlated at Kwajalein and Guam during 1993–2016 (r = 0.98) and show similar interannual changes and decadal/interdecadal trends (Figure S2). The long-term global mean sea level reconstruction is from Church and White (2011).
The Multivariate El Niño–Southern Oscillation (ENSO) Index (MEI) is a comprehensive index for quantifying ENSO strength (Table 1) (Wolter & Timlin, 2011). According to the MEI time series (Figure S3), the 2015/2016 El Niño was strong but not as strong as the 1982/1983 and 1997/1998 El Niño. Although there were signs of El Niño development in 2014 (Hu & Fedorov, 2016), 2015 was the real onset of the recent strong El Niño. As another widely used index, Niño3.4 confirms the strong El Niño in 1982/1983, 1997/1998, and 2015/2016 (Figure S3) (Rayner et al., 2003). To identify decadal and longer time scale variability in the Pacific, we apply a 9 year low-pass filter to the annual mean index of the Pacific Decadal Oscillation (PDO) (Figure S4). We quantify the contribution of PDO to the ocean heat accumulation and sea level rise in the western tropical Pacific (WP) during 1993–2012. The use of the Interdecadal Pacific Oscillation (IPO) index gives similar results.
In terms of modeling data, we use 40 Coupled Model Intercomparison Project Phase 5 (CMIP5) models and their piControl and 20th century simulations, as well as 21st century projections under four Representative Concentration Pathway (RCP) emission scenarios (Table S1) (Flato et al., 2013; Taylor et al., 2012). Based on the 200 year piControl simulations, the CMIP5 models show a range of internal variability of GMST (as the standard deviation of the annual time series) from 0.06° to 0.16°C, with an ensemble mean of 0.11°C (Table S2). For observational estimates, we apply the Ensemble Empirical Mode Decomposition method (Wu et al., 2011) to remove the nonlinear secular trend from the observed GMST time series (Figure S5a). We also subtract the forced GMST response in the CMIP5 historical simulations from the GMST observations (Figure S5b). All of these methods consistently indicate a 0.11°–0.12°C standard deviation of internal GMST variability. Some observational data such as HadCRUT4 are spatially incomplete and blend surface air temperature and SST. This may slightly influence the estimated internal variability of GMST and therefore the data-model comparison (Cowtan et al., 2015). ENSO variability in the tropical Pacific contributes to GMST variability.
Climate sensitivity values in these models depend on representations of different feedback processes in the atmosphere. Different approaches have been used to estimate climate sensitivity in climate models (Yoshimori et al., 2016). CMIP5 used the approach of Gregory et al. (2004) to calculate effective climate sensitivity. The values range from 2.1° to 4.7°C across the CMIP5 models (Table S2), therefore covering a large uncertainty space. As the transient climate response values show a narrower spread across the CMIP5 models, we find that the use of effective climate sensitivity can enhance correlation analyses in this study.
3 Results
3.1 Observed Big Jump of Record Warm GMST Over 2014–2016
The mean of the six GMST data sets indicates that the past 3 years (2014–2016) taken together broke the previous record warm year set in 2010 by a large amount of 0.24°C (0.22°–0.27°C, mean and range of the six data sets) (Figure 1). Year 2014 alone was a record-breaking year in five of the six data sets, with a difference from the previous record that is within the observational uncertainty range (Figure S1). However, the 3 year jump of 0.24°C represents a large change compared with GMST variability. Such events with similar or greater record-breaking magnitudes over consecutive record-breaking years (0.24°C+ events hereinafter) are highly unusual in the instrumental records (Figure 1) (Mann et al., 2017). Most of the record increase in GMST over the past 3 years occurred in 2015 and 2016 during a strong El Niño. Years 2015 and 2016 surpassed 2014 by 0.15°C (0.11°–0.19°C) and 0.21°C (0.18°–0.24°C), respectively. In 2015–2016, forecasts were published (Peyser et al., 2016; UK Met Office, 2015) predicting this jump in GMST using either operational forecast systems (Folland et al., 2013) or Pacific sea level anomalies along with a dynamical-statistical method (Peyser et al., 2016). The subsequent observations confirm these GMST predictions.
In three longer data sets (HadCRUT4, Berkeley, and Cowtan&Way), the 1877–1878 GMST spike was the only other possible 0.24°C+ event in the instrumental period (Figure S1). This 19th century event was prior to large anthropogenic climate forcing and warming, but it is also associated with relatively large uncertainty. The large drop of GMST in the following year (1879) suggests that this GMST spike was likely caused by natural and transient processes, particularly a possible strong El Niño in the Pacific (Aceituno et al., 2009). By contrast, recent predictions of GMST based on operational systems suggest that 2017 will be among the warmest years and the GMST, although unlikely to exceed the 2016 record, will remain at a high level after the 2014–2016 jump (UK Met Office, 2016).
3.2 Unusually Large Oceanic Heat Releases From WP/NWP during 2014–2015
The 2014–2016 GMST jump is a combined effect of the long-term increase in the atmospheric GHG concentration and the short-term strong El Niño in the Pacific Ocean. The warming associated with the El Niño should not be considered as purely natural variability because, in addition to warming the SSTs along the eastern equatorial Pacific, the El Niño also released excess heat sequestered into the subsurface layer (50–300 m) of the WP, especially the NWP east of the Philippines (Figures 2, 3, and S6). This heat had built up over the previous two decades (Figures 2a and S7), leading to rapid sea level rise in the WP (up to four times faster than the global mean) since the 1990s due to the dominant thermosteric effect (Figures 3c and S2) (Griffies et al., 2014; Peyser et al., 2016).
The linear trends of the observed upper 700 m OHC indicate a total heat accumulation of 3.5–4.5 × 1022 J in the WP during 1993–2012 (Figure 3a). As shown in the low-pass filtered time series, the gradual buildup of ocean heat in the WP/NWP started from around 1993–1994 (Figure S7). The heat accumulation in the WP is an important component of global ocean heat uptake during the same period (Gleckler et al., 2016). Both internal variability and external forcing contributed to this unprecedented heat accumulation in the WP. On decadal/interdecadal time scales, the PDO or IPO is the dominant variability mode in the Pacific region that can influence OHC and GMST (England et al., 2014; Mantua et al., 1997; Medhaug et al., 2017; Meehl et al., 2016; Nieves et al., 2015; Steinman et al., 2015). It should be noted that the mechanisms of the PDO/IPO are still under investigation about whether it is an intrinsic variability mode, a mixture of tropical and higher latitude phenomena (Newman et al., 2016), or involves a forced component (Smith et al., 2016). The regression of the upper 700 m OHC onto the observed PDO or IPO indices since the 1950s suggests that the transition of PDO/IPO to their negative phase during 1993–2012 had caused heat accumulation primarily centered in the southwestern tropical Pacific (Figures 2c, S4, and S6). By contrast, the buildup of excess ocean heat in the NWP east of the Philippines is mainly attributable to factors external to the Pacific, including GHG forcing and possible remote effects of other ocean basins (Figures 2d, 3d, and S6) (Luo et al., 2012; McGregor et al., 2014).
As a good indicator of OHC, two long-term high-quality tide gauge data from the NWP confirm the unusual heat accumulation during 1993–2012 (Figure 3c). The relative sea level data at Kwajalein and Guam since 1950 are well correlated, suggesting common mechanisms. Sea level trends show a clear change during the 1990s at both locations. According to a clear linear trend, the total sea level rise over 1993–2012 was 194 and 186 mm at Kwajalein and Guam, respectively. The regression of the detrended tide gauge data onto the PDO index suggests that natural variability can explain 53 and 75 mm sea level rise at the two sites, respectively (Figures 3d and S8). Thus, other factors including GHG forcing and remote oceanic effects likely played a more important role in causing the rapid sea level rise in the NWP during 1993–2012 (Figures 3c and S2) (Hamlington et al., 2014; Han et al., 2014).
The excess subsurface ocean heat in the WP/NWP rapidly resurfaced and was then released to the atmosphere during 2014–2015 (Figures 2, 3, and S9). Here 2013 is used as the reference level for the heat release during 2014 and 2015. Compared to 2013, the WP OHC shows a significant drop by about 4.0–5.2 × 1022 J in 2015 (Figure 3a), with the maximum heat release at the NWP (Figures 2b and S6). Our results indicate that in addition to the natural and normal amount of ocean heat release, the strong 2015/2016 El Niño also completely released the ocean heat accumulated in the NWP during 1993–2012 (Figure 3d). The intense heat release resulted in a drastic sea level fall in the NWP by up to 300 mm during 2013–2015 (Figure S2). As a consequence, the once fastest sea level rise east of the Philippines as evidenced in altimetry data tends to subside after the 2014–2015 event (Figure S2).
EOF1 of the OHC in the tropical Pacific reveals an east-west seesaw pattern in the tropical Pacific (Figure S10). This seesaw also reflects vertical redistribution of ocean heat in the tropical Pacific associated with ENSO and PDO/IPO as well as induced by external forcing (Figure S9). Following a gradual decline during 1993–2012, PC1 shows a large jump during 2013–2015 exceeding the previous events associated with strong El Niño (Figure S10). It indicates an eastward movement and rapid upward emergence of massive ocean heat previously sequestered in deeper oceans (Figure S9), thereby directly boosting the Pacific SST and GMST (Figure 1).
During the large 1997–1998 El Niño, GMST also broke the previous record warm year set in 1995 by a large margin of 0.20°C (0.18°–0.22°C) (Figure 1). Conventional indices such as MEI and Niño3.4 suggest that the 1997–1998 El Niño was at least as strong as the 2015–2016 one (Figure S3). In terms of ocean heat release from the WP, and especially the NWP, however, the 2013–2015 event was clearly stronger and more pronounced than the previous ones during 1981–1982 and 1996–1997 (Figures 3 and S10). In fact, after two decades of unprecedented heat accumulation, the OHC in the WP dropped all the way back to the 1997 level in 2015. The ocean heat release started from 2014, coincident with the increase in GMST.
3.3 Model Simulations and Projections of Large Record-Breaking Events of GMST
The frequency, magnitude, and duration of record-breaking events of GMST are important statistics for characterizing the risk of future extreme warming and other climate events. Next we use 40 CMIP5 models to examine these record-breaking event statistics in the 20th century simulations and 21st century projections. We first evaluate these statistics for each model before averaging them to generate the multimodel ensemble mean. In the historical runs of the CMIP5 models, the GMST records are usually broken by about 0°–0.2°C each time (Figure 4a). According to the multimodel ensemble average, a 0.24°C+ event such as observed in 2014–2016 is unlikely to have occurred under the 20th century external forcing alone; the average frequency of such an event is much less than one during the period of 1901–2005, consistent with the observed lack of such an event during the 20th century (Figure 1).
In the context of the instrumental records of GMST, 13 (or 36%) of the 36 years from 1981 to 2016 were record-breaking years (Figure 1a). For the same period (1981–2016) of the CMIP5 model simulations and projections, the GMST records are broken by 10.5 ± 2.6, 10.0 ± 2.5, 9.9 ± 2.3, and 10.2 ± 2.5 (multimodel ensemble mean ± 1σ) times in the historical simulations (1981–2005) along with RCP2.6, RCP4.5, RCP6.0, and RCP8.5 projections (2006–2016), respectively. On centennial time scales, the record-breaking years account for 16%, 28%, 33%, and 49% of the total of 95 years (2006–2100) given RCP2.6, RCP4.5, RCP6.0, and RCP8.5 emission scenarios, respectively (Figure S11).
The likelihood of the large 0.24°C+ events increases with the increase of GHG forcing. Over 2006–2100, the 0.24°C+ event occurs 0.7 ± 0.7, 1.1 ± 1.1, 2.1 ± 1.7, and 6.0 ± 2.7 times (multimodel ensemble mean ± 1σ) given RCP2.6, RCP4.5, RCP6.0, and RCP8.5 forcings, respectively (Figure 4a). The corresponding return periods are 136, 86, 45, and 16 years. In RCP8.5, the frequency of the 0.24°C+ event more than doubles after 2050. In addition to GHG forcing, different effective climate sensitivity between the CMIP5 models (Table S2) is also a critical factor (Figure 4c). Under RCP8.5, for example, the 0.24°C+ event occurs 4.2 times over 2006–2100 given a low simulated climate sensitivity of 2.5°C. In contrast, a high climate sensitivity of 4.5°C can significantly increase the frequency to 9.3 times. These values are inevitably associated with uncertainty as shown by the scatter around the linear regression line in Figure 4c. Under the low RCP2.6 and RCP4.5 emission scenarios, climate sensitivity is less influential in modulating the frequency of the 0.24°C+ event. These distinct differences between different RCP projections highlight importance of present-day mitigation efforts in controlling the frequency and magnitude of future extreme GMST events.
Surprisingly, different internal variability of GMST in different models (Table S2) has a much weaker influence on the projected frequency of 0.24°C+ events in the 21st century, as shown by their low correlation (Figure 4d). Thus, the 0.24°C+ events in model projections are fundamentally modulated by GHG forcing and climate sensitivity, which jointly determine the overall upward slope of GMST (Figure S11). However, internal variability does influence the duration of the 0.24°C+ event. Stronger internal GMST variability tends to shorten the length of large record-breaking events of GMST and vice versa (Figure S12). This is because the cooling phases associated with large internal variability are more likely to compensate the GHG-induced warming and interrupt possible record-breaking streak of GMST. Given a 0.11°–0.12°C internal variability of the observed GMST (Figure S5), the 0.24°C+ event typically occurs over a course of 3.3–3.7 years (Figure S12), consistent with the observation during 2014–2016. In the historical simulations of the CMIP5 models, most record-breaking events are single-year events along with a couple of 2 year events (Figure 4b). The average frequency of 3 year and longer events is less than 1. During the 21st century, these long-lasting events occur 1.1 ± 1.0, 2.2 ± 1.5, 3.3 ± 2.3, and 6.2 ± 3.0 times on average in the RCP2.6, RCP4.5, RCP6.0, and RCP8.5 projections, respectively.
4 Discussion and Conclusions
The observed 0.24°C jump of record high GMST over three consecutive record-breaking years is highly unusual in the perspective of historical climate variability and change (Mann et al., 2017). It was mainly induced by a rapid release through the recent strong El Niño of the excess ocean heat previously accumulated in the NWP. According to model projections, large record-breaking events of GMST could become more routine by 2100, particularly if GHG emissions continue at high rates. Our analyses suggest that these large events, although often realized during El Niño, are fundamentally caused by the background warming due to GHG forcing. As caveats, we note that heat storage and rapid release from the WP/NWP may not be the only possible process that can cause 0.24°C+ events during the 21st century. Nonetheless, large GMST events similar to the observed 2014–2016 occur in the CMIP5 model projections in terms of both magnitude and mechanism (Figure S13).
Future large jumps in GMST will continue to be amplified in some regions, particularly over land areas and at high latitudes, and will continue to be associated with other climate extremes and impacts related to rapid warming. The increase in frequency, magnitude, and duration of rapid global warming events has the potential to make adaptation more difficult. As shown by our RCP2.6 and RCP4.5 results, climate change mitigation would be effective at reducing or even eliminating such events in the long term.
Acknowledgments
The authors thank many observation, modeling, data service, and climate research centers for making their data and scripts available. We appreciate the comments from the anonymous reviewers. Part of the work was done during J.Y.'s sabbatical at Princeton University and GFDL/NOAA. J.Y. thanks the GFDL host S. Griffies and the Visiting Scientist Program of Princeton University. The research was supported by NOAA grant NA13OAR4310128 and NSF grants 1304083 and 1513411. The observational and model data used in this study can be accessed from the URLs found in the tables. For all other data inquiries, please contact Jianjun Yin (yin@email.arizona.edu).
