1. Introduction
Since the end of the twentieth century, the rapid warming of global surface temperature is reported to have slowed, which is often called the global warming “hiatus” (or “slowdown” to be more precise) (Easterling and Wehner 2009; Knight et al. 2009; Meehl et al. 2011; IPCC 2013). Two main lines of evidence are presented to support the slowdown (Trenberth 2015). First, unusually low growth trends during the early twenty-first century are successively observed in surface and tropospheric temperatures (Easterling and Wehner 2009; Knight et al. 2009; Solomon et al. 2010; Santer et al. 2014), upper ocean heat content (Katsman and van Oldenborgh 2011; Meehl et al. 2011), and sea level (Cazenave et al. 2014). The other line of evidence is the increasing disparity between the observed weak warming and the expected strong warming by models (Fyfe et al. 2013; Fyfe and Gillett 2014; Meehl et al. 2014; Schmidt et al. 2014). With rapidly increasing greenhouse gas concentrations, the unanticipated global warming slowdown triggers doubts about the acknowledged human-induced global warming, and the mechanisms behind it have been a focus of scientists over the last decade (e.g., Lean and Rind 2009; Solomon et al. 2010; Trenberth and Fasullo 2010; Meehl et al. 2011; Kosaka and Xie 2013; Tung and Zhou 2013; Chen and Tung 2014; England et al. 2014; Nieves et al. 2015).
Despite wide discussions about the potentially involved physical mechanisms, whether there is a warming slowdown in the early 2000s remains a controversial issue that attracts heated debate (e.g., Fyfe et al. 2016; Medhaug et al. 2017; Cahill 2018; Rypdal 2018). Some researchers argue that the so-called global warming hiatus does not really exist at all and instead is caused by measurement biases in some global surface temperature datasets or by the selection biases in target periods. For one thing, the warming trend of the surface temperature during the early twenty-first century is thought to be underestimated by some datasets such as HadCRUT4, which are subject to bias from incomplete spatial coverage, especially over the fast-warming Arctic region and the Southern Hemisphere (Hansen et al. 2010; Cowtan and Way 2014; Durack et al. 2014). For the other, the decreased warming reported in the early 2000s is attributed to sea surface temperature (SST) sampling bias from changes in instrumentation (Karl et al. 2015; Hausfather et al. 2017). For example, Karl et al. (2015) revised the recent warming trend by using a corrected ERSST4 dataset and concluded that the global surface temperature should have continued to warm during 2000–14 at the same rate as that during 1950–99, challenging the global warming hiatus. In addition to measurement biases, some studies ascribe the hiatus to the selection bias in the fitting intervals for estimating the temperature warming rate based on the least squares linear trend. In particular, three consecutive new record temperatures in the post-hiatus period 2014–16 further strengthen the doubt about the existence of the hiatus (Loeb et al. 2018; Zhang et al. 2019). Mudelsee (2019) claimed that the hiatus would disappear when the hiatus period is extended from 1998–2013 to 1998–2017. However, Trenberth (2015) and Fyfe et al. (2016) argued against the conclusion of no hiatus. They pointed out that the reference period 1950–99 is problematic because the warming rate during 1950–99 is unusually low with the “big hiatus” artificially included, thus impeding the identification of the recent slowdown. Instead, the period 1972–2001 is a more physically interpretable baseline than 1950–99, and the warming rate is significantly slower during 2001–14 than during 1972–2001 (Fyfe et al. 2016).
Overall, the controversy regarding the existence of the hiatus is closely associated with its short-term linear trend–dependent definition. Although the definitions vary greatly across different studies, a hiatus/slowdown event is usually identified by comparing the observed warming rate over a decadal-scale target period with a prescribed baseline. When the observed warming rate is lower than the baseline, a hiatus/slowdown event is recognized, and otherwise not. The controversy over the existence of the hiatus/slowdown may stem from three aspects. First, the warming rate severely suffers from the period selection bias, since it is commonly estimated by calculating the least squares linear trend of the global temperature, which is highly sensitive to fitting intervals with different start and end years (Wu et al. 2007; Trenberth and Fasullo 2013). Second, the differences in diverse temperature datasets, which arise from spatial coverage, observation biases, input data, techniques used to generate datasets, and even the version of the dataset, may lead to substantially different temperature trend estimates for the same period (Risbey et al. 2018). Moreover, the occurrence of a hiatus/slowdown also depends on which hiatus definition is adopted, as different definitions prescribe different reference values as baselines to evaluate whether the warming slows down during the target period. Various baselines can be grouped into three main types: 1) a near-zero warming rate (definition 1; e.g., Meehl et al. 2011); 2) the warming rate of a reference period, such as the rapidly warming decades over the last quarter of the twentieth century (definition 2a; e.g., Knight et al. 2009), or a long-term period, such as the second half or the whole twentieth century (definition 2b; e.g., IPCC 2013); and 3) the expected global warming rate projected by climate models (definition 3; e.g., Fyfe et al. 2013).
Therefore, it is understandable and even expected that apparently conflicting conclusions about the existence of the hiatus in the early 2000s would be drawn from different studies. Previous studies usually identify a hiatus event during a cherry-picked target period based on one or several arbitrary datasets by referring to a subjective baseline. As a result, the conclusion of one study may not hold for another study, which may take different period, dataset, or definition, degrading the robustness and comparability of the results. Although efforts have been made to reconcile the controversy about the status of the hiatus, this issue is not explicitly addressed. For example, Trenberth (2015) showed that “the perception of whether or not there was a hiatus depends on how the temperature record is partitioned,” emphasizing the decisive role of the period choice. Medhaug et al. (2017) examined the global surface temperature trends during periods spanning 1998–2012 and discussed the existence of global warming hiatus in the early 2000s by using six observational datasets (BEST, GISTEMP, HadCRUT4, MLOST, HadCRUTkrig, and HadCRUT3) based on abovementioned definitions 1, 2b, and 3. They found that different definitions of hiatus and the difference between observational datasets are responsible for the ongoing controversy. More recently, using four recent observational datasets (HadCRUT4, GISTEMP, ERA-Interim, and MLOST) based on definition 3, Tung and Chen (2018) revisited the existence of the slowdown during the period 1998–2012. However, they thought the difference between datasets would not influence the existence of the slowdown, as all the observed warming trends are significantly below the multimodel mean (MMM). A similar result was also given by Rypdal (2018), who adopted definition 2b and pointed out that the difference in observational datasets is too small to account for the warming slowdown during 1998–2014 relative to the long-term trend during 1970–2017.
In this work, using 32 datasets that cover almost all the available global surface temperature datasets, we reexamine the existence of the slowdown during the early twenty-first century under all existing definitions mentioned above by comprehensively investigating the influences of period, dataset, and definition choices on the recent temperature trend during all decadal-scale periods spanning 1990–2017. Particularly, besides the intuitive analysis made in previous studies, we further quantitatively estimate the pairwise temperature trend differences between various observational datasets, and test the statistical significance of these differences. More importantly, in addition to case studies focusing on a specific time period and fixed-length periods, we employ the probability of the occurrence of hiatus/slowdown events to quantitatively investigate the influence of period, dataset, and definition choices. We find the probability is significantly higher in the 2000s than in the 1990s, supporting a warming slowdown during the early twenty-first century. At last, to remove the discrepancy between the slowdown and the anthropogenic global warming, we present the evolution of the anthropogenic global warming signal by employing the ensemble empirical mode decomposition (EEMD) method.
2. Data and methods
Given that the arbitrary choices of period, dataset, and hiatus definition may be the main causes for the diversity of the recent warming rate changes and thus the controversy over the presence of a hiatus/slowdown event, we examined the influences of these three factors one by one. To investigate the influence of target period choice, we calculated the linear trends of the global mean surface temperature (GMST) for all decadal-scale periods spanning 1990–2017. To comprehensively check the impact of dataset choice and to guarantee comparable and general results, we collected 32 datasets that cover almost all the available routinely updated global surface temperature datasets widely used in the hiatus research (see Table S1 in the online supplemental material). Given that the warming slowdown is primarily recognized using the observed GMST, most analyses of this work (sections 3a, 3c, and 3d) were performed based on six current well-known observational combined land/marine surface temperature datasets: BEST (Berkeley Earth Surface Temperatures; Rohde and Hausfather 2020), GISTEMP [Goddard Institute for Space Studies (GISS) Surface Temperature Analysis (Hansen et al. 2010], HadCRUT4 (Hadley Centre/Climatic Research Unit, version 4; Morice et al. 2012), HadCRUT4krig (the same with kriging; Cowtan and Way 2014), JMA (the Japan Meteorological Agency dataset; Ishihara 2006), and MLOST (Merged Land–Ocean Surface Temperature; Vose et al. 2012). To explore the influence of hiatus definition choice, we considered all existing definitions with emphasis on the recent warming rate change relative to the preceding warming surge (definition 2a). This is because it is critical to determine whether the global warming slows down under the steady greenhouse gas radiative forcing, which is the essence of the warming slowdown and also the reason why the phenomenon attracts huge interest (e.g., Knight et al. 2009). Besides, a total of 233 historical simulations of climate models in phase 6 of the Coupled Model Intercomparison Project (CMIP6) were used to calculate the simulated warming rates in definition 3. In addition, both the statistical significances of individual trends and trend differences were tested according to Santer et al. (2000). Detailed descriptions of the datasets and statistical methodology can be found in the online supplemental material.
To explore the recent change of the nonlinear anthropogenic warming trend, which is difficult to be revealed by the linear trend, a time-varying intrinsic trend was employed. Wu et al. (2007) innovatively defined the intrinsic trend as the nonperiodic part of a time series, which should be monotonic or have at most one extremum within a given temporal span. After all the quasi-periodic variabilities on various time scales are removed from the data by using an adaptive and temporal local empirical mode decomposition (EMD) method (Huang et al. 1998), the remainder is naturally determined as the intrinsic trend. The intrinsic trend was then gradually proven to be robust and reliable and to have a significant improvement over the linear trend when applied to global temperature time series (Wu et al. 2007, 2011; Ji et al. 2014). Wu et al. (2011) and Ji et al. (2014) pointed out that the intrinsic trend of the centennial-scale instrumental global temperature time series suggests that the anthropogenic warming signal is induced by continuously increasing greenhouse gases. First, the temporal evolution of intrinsic trend reflects that the acceleration of warming corresponds to the accumulation of greenhouse gases in the atmosphere. Furthermore, the spatial pattern of widespread warming in the intrinsic trend agrees well with a response to the buildup of well-mixed greenhouse gases (Wu et al. 2011). Here we extracted the intrinsic trend of the global temperature by using EEMD (Wu and Huang 2009), which is an improved version of the original EMD and is particularly suitable for nonlinear and nonstationary climatic series. More details on EMD and EEMD can be found in the supplemental material.
3. Results
a. Influence of period choice
The arbitrarily chosen hiatus periods and datasets in the more than 10 years of hiatus research (Table 1) can directly lead to highly variable warming rates in recent decades and thus may result in considerable controversies over the existence of the recent warming slowdown. In this section, to focus on the influence of the target period choice, we temporarily set aside the effects of the dataset and definition. We just fixed the temperature time series as the mean of six observationally derived GMST time series, and fixed the baseline as the linear trend over 1975–97 (0.16°C decade−1), which represents the averaged state of the warming surge in the late twentieth century, according to definition 2a. Figure 1 shows the distribution of the hiatus periods used in previous studies. The hiatus periods usually start in the late 1990s and end in the early 2010s with a 10–16-yr duration, presenting a decadal scale (Figs. 1a,b). Generally, the warming rates during these periods are obviously lower than that of the previous warming surge (Fig. 1c). There is an evident warming slowdown during most periods in Table 1 (68 of 90 times) with the warming rate decreasing over half (Fig. 1d). The most popular hiatus periods in the scientific literature start from 1998–2001 (67 times) and end in 2012 or 2013 (62 times), lasting 14–16 years (46 times). The most frequent warming rate is approximately 0.08°C decade−1 (39 times), with a warming reduction of nearly half than that in 1975–97. Beyond the above common features, there are large discrepancies. The start years of the hiatus period range from 1993 to 2004, the end years range from 2008 to 2014, and the durations range from 7 to 20 years. Correspondingly, the warming rates show large diversity as well, although they are calculated based on the same data. The warming rates greatly vary between −0.07° and 0.17°C decade−1, representing weak cooling and strong warming, respectively. This may directly induce controversies over the authenticity of the slowdown.
The histogram of the hiatus periods in the scientific literature listed in Table 1. (a) Start and end years. (b) Lengths. (c) Observed warming rates during these periods and (d) the corresponding warming reductions relative to the warming surge during 1975–97. The purple line and spread corresponding to the right coordinate in (a) represent the mean of six observationally derived GMSTs and twice SDs apart from the mean, respectively. The warming rates are obtained by calculating the least squares linear trends of the mean GMST time series during all periods in Table 1.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The histogram of the hiatus periods in the scientific literature listed in Table 1. (a) Start and end years. (b) Lengths. (c) Observed warming rates during these periods and (d) the corresponding warming reductions relative to the warming surge during 1975–97. The purple line and spread corresponding to the right coordinate in (a) represent the mean of six observationally derived GMSTs and twice SDs apart from the mean, respectively. The warming rates are obtained by calculating the least squares linear trends of the mean GMST time series during all periods in Table 1.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The histogram of the hiatus periods in the scientific literature listed in Table 1. (a) Start and end years. (b) Lengths. (c) Observed warming rates during these periods and (d) the corresponding warming reductions relative to the warming surge during 1975–97. The purple line and spread corresponding to the right coordinate in (a) represent the mean of six observationally derived GMSTs and twice SDs apart from the mean, respectively. The warming rates are obtained by calculating the least squares linear trends of the mean GMST time series during all periods in Table 1.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The hiatus periods and surface temperature datasets used in the scientific literature over the last decade (2009–19) that explicitly defines a hiatus period. (Many expansions of acronyms are available at http://www.ametsoc.org/PubsAcronymList.)
The large variability of the recent warming rates, in turn, is primarily caused by the high sensitivity of the decadal-scale linear trend to the arbitrary target period choice. Figure 2 shows the warming rates during all decadal-scale periods spanning 1990–2017 including all the periods mentioned in previous studies. The warming rate is usually obtained by calculating the least squares linear trend of the GMST time series, neglecting that the climate system is inherently nonlinear and exhibits variabilities across a wide range of time scales. In particular, Tung and Chen (2018) noticed that the recent warming slowdown phenomenon is between two super El Niño events of 1997/98 and 2015/16, and they further pointed out that the strong “interannual variations prominently influence the determination of the linear trends.” As a result, the recent decadal-scale warming rates are severely affected by the period selection. Both the time and length of the period have great impacts. Some similar periods may show quite different warming rates. For one thing, the decadal warming rates (typically represented by the 15-yr running trend; purple line in Fig. 2) strikingly fluctuate over time. Taking the periods around 1998–2012 as examples, the warming rate during 1998–2012 (point 2005, 15, yellow square) is 0.08°C decade−1, with a half reduction relative to the 0.16°C decade−1 of the previous warming surge. When the 15-yr period moves back two years to 1996–2010 (point 2003, 15), the warming rate abruptly increases to 0.16°C decade−1, suggesting no slowdown. For another thing, the decadal warming rate dramatically changes over the length of period, even for periods centered on the same year. For example, fixing the center year as 2005, when the 15-yr period (1998–2012) is reduced by two years to 13 years (1999–2011) the warming rate would increase to 0.14°C decade−1, indicating no slowdown as well. The discrepancy in the warming rates is even greater in periods shorter than 10 years. Some periods show cooling, while neighboring periods present rapid warming, such as the periods 2001–08 (point 2004, 8) and 2000–07 (point 2003, 8), with warming rates of −0.01° and 0.23°C decade−1, respectively. The above examples illustrate that the period choice brings huge uncertainty to the decadal-scale warming rates and largely explains why there is so much debate about the reality of the recent warming slowdown in previous studies.
The recent decadal-scale warming rates. The warming rates are obtained by calculating the least squares linear trends of GMST during all decadal-scale periods spanning 1990–2017. The x axis and y axis indicate the center year and the length of the period, respectively. For example, the warming rate during the most selected period 1998–2012 is plotted on grid (2005, 15) (yellow box). The horizontal section across the grid (2005, 15) represents the 15-yr running trends (purple line corresponding to the right coordinate), which show how the 15-yr warming rate changes over time, while the vertical section from the bottom up successively displays the linear trends for periods centered on 2005 with lengths from 5 to 25 years, which denotes how warming rates around 2005 change over the period length. The GMST used here is the mean time series of six observationally derived GMSTs. The white boxes mark the hiatus periods listed in Table 1. The dots indicate the periods with significant trend (95% confidence level). The orange and green dashed boxes highlight the strong warming episode around the 1990s and the weak warming episode around the 2000s, respectively. The diamonds in two episodes denote the typical decadal-scale periods lasting from 10 to 16 years, which are frequently adopted scales in Table 1. The averaged warming rates of these periods are defined as the typical warming rates of these two episodes.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The recent decadal-scale warming rates. The warming rates are obtained by calculating the least squares linear trends of GMST during all decadal-scale periods spanning 1990–2017. The x axis and y axis indicate the center year and the length of the period, respectively. For example, the warming rate during the most selected period 1998–2012 is plotted on grid (2005, 15) (yellow box). The horizontal section across the grid (2005, 15) represents the 15-yr running trends (purple line corresponding to the right coordinate), which show how the 15-yr warming rate changes over time, while the vertical section from the bottom up successively displays the linear trends for periods centered on 2005 with lengths from 5 to 25 years, which denotes how warming rates around 2005 change over the period length. The GMST used here is the mean time series of six observationally derived GMSTs. The white boxes mark the hiatus periods listed in Table 1. The dots indicate the periods with significant trend (95% confidence level). The orange and green dashed boxes highlight the strong warming episode around the 1990s and the weak warming episode around the 2000s, respectively. The diamonds in two episodes denote the typical decadal-scale periods lasting from 10 to 16 years, which are frequently adopted scales in Table 1. The averaged warming rates of these periods are defined as the typical warming rates of these two episodes.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The recent decadal-scale warming rates. The warming rates are obtained by calculating the least squares linear trends of GMST during all decadal-scale periods spanning 1990–2017. The x axis and y axis indicate the center year and the length of the period, respectively. For example, the warming rate during the most selected period 1998–2012 is plotted on grid (2005, 15) (yellow box). The horizontal section across the grid (2005, 15) represents the 15-yr running trends (purple line corresponding to the right coordinate), which show how the 15-yr warming rate changes over time, while the vertical section from the bottom up successively displays the linear trends for periods centered on 2005 with lengths from 5 to 25 years, which denotes how warming rates around 2005 change over the period length. The GMST used here is the mean time series of six observationally derived GMSTs. The white boxes mark the hiatus periods listed in Table 1. The dots indicate the periods with significant trend (95% confidence level). The orange and green dashed boxes highlight the strong warming episode around the 1990s and the weak warming episode around the 2000s, respectively. The diamonds in two episodes denote the typical decadal-scale periods lasting from 10 to 16 years, which are frequently adopted scales in Table 1. The averaged warming rates of these periods are defined as the typical warming rates of these two episodes.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
Although the warming rates vary greatly with the hiatus period choice, the recent one or two decades are in the cooling episode (or weak warming age to be more exact) following the previous strong warming episode. As Fig. 2 shows, the decadal-scale warming rates represented by the 15-yr running trends in recent decades are vastly lower than those in previous decades. The typical warming rate of the strong warming episode centered in 1997 is 0.27° ± 0.10°C decade−1 [0.27 is the mean value, and 0.10 is two standard deviations (SDs)]. In contrast, the typical warming rate of the following weak warming episode centered in 2007 is only 0.05° ± 0.10°C decade−1. The warming rate is more than 4 times slower around the 2000s than around the 1990s. In addition, we marked the periods with significant linear trends at 95% confidence level. The percentage of periods with insignificant warming trend around the 2000s is significantly higher than around the 1990s (67% for 2000s vs 38% for 1990s). Therefore, from the perspective of the global warming rate change itself (definition 2a), the recent warming slowdown is real enough. Notably, the slowdown exists only in periods shorter than 20 years. And the magnitude of warming reduction decreases with lengthening of the time scale. Some periods shorter than 10 years show negative warming rates and thus present a real warming hiatus or pause. In contrast, for periods longer than 20 years, the warming slowdown no longer exists, and all periods show significant warming trends, indicating that the recent warming slowdown is only a decadal-scale phenomenon.
b. Influence of dataset choice
To focus on the influence of dataset and to avoid the interference of period and definition, here the hiatus period is temporarily fixed as the most selected period 1998–2012, and the reference period is fixed as 1975–98 according to definition 2a (more details can be found in the online supplemental material).We found that although the decadal-scale warming slowdown during the early 2000s widely existed in multiple global surface temperature datasets, the magnitude of warming reduction greatly varied across datasets, which intensified the conflict about its existence. The 32 global surface temperature time series derived from 12 merged land/marine surface temperature datasets (Fig. S1) and 20 SST datasets (Fig. S2) consistently show that the global surface temperature somewhat plateaus in 1998–2012 (blue interval) after the strong warming surge in 1975–97 (red interval) and present a warming slowdown. To quantitatively investigate the influence of the dataset choice, we estimated the observed warming rates during the periods 1975–97 and 1998–2012 based on 12 long observational datasets (Fig. 3). All the observed warming rates exhibit a decline during 1998–2012 relative to 1975–97. The observed warming rate of GMST is 0.16° ± 0.02°C decade−1 (0.16 is the mean value, and 0.02 is two SDs) during 1975–97 and then drops to 0.08° ± 0.07°C decade−1 during 1998–2012, with a reduction of 51% ± 39%. The observed SST warming rate reduces by 69% ± 48%, from 0.12° ± 0.03°C decade−1 during 1975–97 to 0.04° ± 0.06°C decade−1 during 1998–2012. Both GMST and SST warming rates decrease by over half. The warming slowdown is more significant in the ocean than on land, as there is no overlap between two periods in the dataset confidence intervals of the SST trends while for GMSTs there is a slight overlap, suggesting the important role of the ocean in the slowdown. In addition, there is a considerable overlap in the trend confidence intervals between 1975–97 and 1998–2012 (Fig. S6), as the confidence interval for 1998–2012 is quite large due to the small sample size and strong interannual variability. This indicates that the trends during the hiatus period and reference period are drawn from the same population since all the trends include the anthropogenic warming signal and the warming phase of the multidecadal variability.
Warming rates of (a) GMSTs and (b) global mean SSTs derived from observational datasets, and (c),(d) trend differences between various datasets. The warming rates in the upper panels are obtained by calculating the linear trends during the most selected hiatus period of 1998–2012 and warming surge period of 1975–97 based on long observational derived GMST and SST datasets covering 1975–2013. At the right edge, the circles at the center of the bar represent the mean trend of the six datasets on the left while the error bars indicate 2 times the SDs apart from the mean trends. The solid circle indicates the trend is significant with 95% confidence interval while the empty circle indicates the trend is insignificant. The lower panels show the trend differences between various datasets during 1998–2012. The plus sign (+) indicates a significant trend difference between the x-axis dataset and y-axis dataset (x-axis dataset minus y-axis dataset). The “mean” dataset is the average of the six observations and is treated as an individual dataset here. Matrices are symmetrical about the diagonal line but with opposite sign.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
Warming rates of (a) GMSTs and (b) global mean SSTs derived from observational datasets, and (c),(d) trend differences between various datasets. The warming rates in the upper panels are obtained by calculating the linear trends during the most selected hiatus period of 1998–2012 and warming surge period of 1975–97 based on long observational derived GMST and SST datasets covering 1975–2013. At the right edge, the circles at the center of the bar represent the mean trend of the six datasets on the left while the error bars indicate 2 times the SDs apart from the mean trends. The solid circle indicates the trend is significant with 95% confidence interval while the empty circle indicates the trend is insignificant. The lower panels show the trend differences between various datasets during 1998–2012. The plus sign (+) indicates a significant trend difference between the x-axis dataset and y-axis dataset (x-axis dataset minus y-axis dataset). The “mean” dataset is the average of the six observations and is treated as an individual dataset here. Matrices are symmetrical about the diagonal line but with opposite sign.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
Warming rates of (a) GMSTs and (b) global mean SSTs derived from observational datasets, and (c),(d) trend differences between various datasets. The warming rates in the upper panels are obtained by calculating the linear trends during the most selected hiatus period of 1998–2012 and warming surge period of 1975–97 based on long observational derived GMST and SST datasets covering 1975–2013. At the right edge, the circles at the center of the bar represent the mean trend of the six datasets on the left while the error bars indicate 2 times the SDs apart from the mean trends. The solid circle indicates the trend is significant with 95% confidence interval while the empty circle indicates the trend is insignificant. The lower panels show the trend differences between various datasets during 1998–2012. The plus sign (+) indicates a significant trend difference between the x-axis dataset and y-axis dataset (x-axis dataset minus y-axis dataset). The “mean” dataset is the average of the six observations and is treated as an individual dataset here. Matrices are symmetrical about the diagonal line but with opposite sign.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
Notably, the dataset choice introduces huge uncertainty into the warming reduction and thus contributes to the contradictory about the existence of the slowdown. Although the warming slowdown is widespread across all the datasets, the magnitude of the warming reduction varies greatly with datasets. The datasets with special considerations for the observational biases show smaller magnitudes of warming reduction. In some datasets, the SST sampling bias resulting from changes in instrumentation is emphatically correct. For example, ERSST4 has been carefully checked and corrected by Karl et al. (2015). It is blended in MLOST, and its successor ERSST5 is combined in GISTEMP. Besides, some datasets focus on special measures to infill the data-sparse regions, such as the fast-warming Arctic area. BEST, GISTEMP, HadCRUT4krig, and COBE-SST2 fill unsampled regions using a variety of statistical methods, such as by extrapolating from observed temperatures (Hansen et al. 2010; Cowtan and Way 2014; Rohde and Hausfather 2020), incorporating satellite observations (Cowtan and Way 2014) and producing SST data from observed sea ice concentrations based on the ice–SST relationship (Hirahara et al. 2014). In the corrected datasets, the cooling biases in the last decades due to observational biases are believed to have been eliminated. Consequently, compared with the uncorrected datasets, the corrected ones indeed exhibit a faster warming trend during 1998–2012, and thus smaller magnitude of slowdown relative to 1975–97. Besides, all the corrected GMST datasets exhibit significant warming trends during 1998–2012, while the uncorrected HadCRUT4 and JMA show no significant trend (Fig. 3a). As a result, the corrected datasets show milder warming reduction between 1998–2012 and 1975–97. Particularly, the warming reductions are only 32% for BEST and 35% for HadCRUT4krig, whereas they are 80% for JMA and 68% for HadCRUT4. For the SSTs, all datasets except COBE-SST2 show no significant warming trend during 1998–2012. And the magnitudes of warming reduction are generally larger than the GMSTs. Again, the corrected COBE-SST2 and ERSST5 exhibit smaller warming reductions of 38% and 39%, respectively, in contrast to large warming reductions ranging from 80% to 95% in other datasets.
The large uncertainty of the reduction magnitudes is primarily attributed to the significant impacts of the differences between various datasets on the recent temperature trend. We tested whether the differences in datasets had a significant effect on the trends during 1998–2012 by examining whether there was a significant trend of the pairwise difference time series between various datasets (Figs. 3c,d). Given that the observational temperature time series derived from different datasets represent the same temperature change over the same period, the difference time series would remove the anthropogenic trend and natural variabilities common to both time series. And whether there is a significant trend of the difference time series should be related to differences in the datasets (Santer et al. 2000). From Figs. 3c and 3d, we found that there were significantly negative trend differences between uncorrected datasets, especially with incomplete coverage (HadCRUT4 and JMA), and corrected datasets. In contrast, the warming trend of the BEST and HadCRUT4krig with complete coverage is significantly higher than other datasets excepted for MLOST. For example, the trend difference between BEST and JMA is up to 0.08°C decade−1, which is comparable to the magnitude of the trend itself, suggesting the important influence of dataset choice. The trend differences between SST datasets have similar features. The filled-coverage COBE-SST2 and ERSST5 show significantly higher warming trends than other datasets during 1998–2012. Overall, the observational biases, particularly the spatial coverage, have a significant impact on the temperature trend during 1998–2012, and thus the magnitude of the warming reduction relative to 1975–98. It should be recalled that the impact is not just on definition 2a, but on all the definitions. Under all the definitions, the probability of observing a slowdown in the infilled and uncorrected datasets is higher than that in the filled and corrected datasets (Table 2). Consequently, the dataset choice further deteriorates the arguments about the existence of the slowdown.
Percentages (%) of hiatus/slowdown events occurring during 1990s and 2000s (the orange and green dashed boxes in Figs. 2 and 4). The “mean” denotes the average GMST of six observations, which is treated as an individual dataset here. The statistical significance of the difference between the percentages during two periods is assessed by the χ2 test, and all percentages during 2000s are significantly higher than that during 1990s (95% confidence level).
c. Influence of definition choice
In addition to period and dataset, the type of hiatus definition is another important reason for the controversy over its authenticity (Fig. 4; see also Figs. S7–S10). So far, the decade-long hiatus studies have not yet reached an agreement on the exact definition of the hiatus. In earlier studies, the global warming hiatus or pause is strictly defined as a period with no significant warming trend or even with a cooling trend, such as the period around 1998–2008 (e.g., Easterling and Wehner 2009; Knight et al. 2009; Meehl et al. 2011; Kaufmann et al. 2011). Further studies broaden the scope of the term “hiatus” and commonly use it to refer to the unforeseen interruption or slowdown of the ongoing rapid warming process under the background of continuously rising greenhouse gas emissions (e.g., Fyfe et al. 2013; IPCC 2013; Tung and Zhou 2013; Hawkins et al. 2014). The current existing definitions can be generalized as comparing the warming rate during the target period with a baseline, which can be near-zero warming rate (definition 1), the warming rate during a reference period such as the previous warming surge decades (definition 2a) or a long-term period like the second half or the whole twentieth century (definition 2b), and the expected warming rate in climate models (definition 3). The three definitions are associated with three distinct questions, respectively (Lewandowsky et al. 2015). Definition 1 employs near-zero warming rate to narrowly define an absolute hiatus or pause event without any heating. Definition 2 focuses on the global temperature evolution itself. Specifically, definition 2a uses the warming rate of the warming surge to evaluate whether the recent warming rate is slower than that in previous decades with an emphasis on the warming rate change under similar forcings, while definition 2b uses a long-term warming rate to estimate whether the recent warming is weaker than the average warming trend in response to the human-induced increasing greenhouse gas concentrations. In definition 3, the divergence from model-derived expectations verifies whether the current understanding of the temperature change mechanisms, which is included in the state-of-the-art models, is correct.
The warming rates of hiatus periods. As in Fig. 2, but only showing warming rates during hiatus periods identified by various baselines prescribed by definitions 1, 2a, and 2b, respectively.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The warming rates of hiatus periods. As in Fig. 2, but only showing warming rates during hiatus periods identified by various baselines prescribed by definitions 1, 2a, and 2b, respectively.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The warming rates of hiatus periods. As in Fig. 2, but only showing warming rates during hiatus periods identified by various baselines prescribed by definitions 1, 2a, and 2b, respectively.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
In a strict sense based on definition 1, a real warming hiatus or pause with a near-zero or negative warming rate usually exists in periods shorter than 10 years (Fig. 4a and Fig. S7). Combining the period and dataset choices, the probability of the occurrence of a real hiatus event in datasets with complete coverage (BEST, GISTEMP, HadCRUT4krig, and MLOST) is obviously lower than that in datasets with incomplete coverage (HadCRUT4 and JMA) (Table 2 and Fig. S7). When the time scale elongates to longer than 12 years, the real hiatus would disappear in all datasets. They all present positive warming rates during periods longer than 12 years, including the commonly selected period 1998–2012. This may be why the length of the real hiatus period is often chosen as 8 to 10 years when the hiatus event is defined based strictly on definition 1 (e.g., Meehl et al. 2011). In the context of escalating global warming, the real hiatus events lasting several decades, such as the so-called big hiatus during the 1940s–1970s, are expected to eventually disappear completely. The mostly positive decadal-scale warming rates suggest that the global temperature has continued to increase recently, although the recent warming rates are vastly slower than those in previous decades (i.e., a warming slowdown). Actually, the term “hiatus” or “pause” is inaccurate to refer to the warming slowdown because it may send a misleading message to the scientific community and public that human-induced global warming has stopped, which will be proven wrong in section 3d. But to avoid confusion and maintain consistency, we will continue using the term “hiatus period” in this paper.
In contrast to definition 1, which determines whether the recent warming rate is lower than a fixed near-zero value, definition 2 focuses on the global temperature change itself by comparing the recent warming rate with that in a reference period. However, the reference period choice also greatly affects the identification of a slowdown [e.g., Karl et al. (2015) vs Fyfe et al. (2016)]. Based on definition 2a, we selected the previous warming surge decades as reference period to determine whether recent global warming weakens under increasing greenhouse-gas concentrations. We found that the warming rates in recent decades indeed slowed down compared with the previous warming surge, even considering the influences of period and dataset. And all datasets show that there is a warming slowdown during nearly all periods in Table 1 relative to the previous warming surge (Fig. 4b and Fig. S8). However, if the reference period is extended to a long-term period (definition 2b), the slowdown of some periods and some datasets would disappear (Figs. 4c,d; see also Figs. S9 and S10). The second half of the whole twentieth century is frequently adopted to check whether the recent warming is slower than the overall average warming trend in response to greenhouse gases (e.g., IPCC 2013; Tung and Zhou 2013). Considering the uncertainty of datasets, the warming rate during the commonly selected hiatus period 1998–2012 (0.08° ± 0.07°C decade−1) is indistinguishable from the long-term warming rates during 1950–99 (0.09° ± 0.03°C decade−1) and during 1900–99 (0.07° ± 0.01°C decade−1). During 1998–2012, all filled and corrected datasets present no slower-than-average warming event (Figs. S9 and S10). Based on such kinds of reference periods and particular datasets, some studies claim that the recent global warming slowdown does not exist. Karl et al. (2015) argued that there is no slowdown recently, as the warming rate in 2000–14 is comparable to that in 1950–99 using corrected GISTEMP data. However, this result was questioned by Fyfe et al. (2016), who pointed out that the reference period of 1950–99 is inappropriate as the big hiatus period included. In fact, when the focus is shifted from a specific period to the change of the probability of the occurrence of slower-than-average warming events, in all datasets the probability is significantly higher in recent decade than in previous decade (Fig. 4 and Table 2).
Based on definition 3, most recent decadal-scale periods are identified as having unexpected weak warming periods as their observed warming rates are generally lower than the simulations (Fig. 5). For example, during 1998–2012, the simulated warming rate of the 233 realizations in CMIP6 is 0.25° ± 0.25°C decade−1. The MMM is more than 4 times that of the mean observed trend (0.08°C decade−1), and 40% simulations (93 of 233) significantly overestimate the observed value. The divergence between simulations and observations raises questions about the simulation and prediction ability of sophisticated climate models. However, there are several problems with the comparison between simulation and observation. For one thing, the overestimation of the MMM does not mean that the models could not simulate the warming slowdown. As the natural variabilities in different simulations, which seriously affect the observed short-term trends, have canceled each other out in the MMM, the MMM should not be expected to reproduce the observed short-term trends. And when the uncertainty of simulations is taken into account, the observed warming rates (including those during 1998–2012) are well within the simulations (Fig. 5c). For another thing, it is unfair to directly compare the simulated and observed warming rates for a specific time period. When examining the simulated and observed probability distribution functions (PDFs) of the 15-yr trends over the whole historical period (Fig. 5d), we find that the PDF of the simulations is very similar to that of the observations, indicating the models could simulate the observed decadal-scale trends in a statistical sense. The apparent divergence over a specific period could be explained by climate natural variabilities, errors in radiative forcing and the response to forcings in the models (Fyfe et al. 2013; IPCC 2013; Schmidt et al. 2014; Medhaug et al. 2017). Particularly, as the historical forcings of CMIP6 are based on the observations (Eyring et al. 2016), most discrepancies should be attributed to natural variabilities, the cooling phases of which are the main cause of the observed warming slowdown (Wei et al. 2021). Since the historical simulations in CMIP6 are not initialized, the simulated natural variabilities are not necessarily in phase with the real world, and therefore the models do not necessarily have to reproduce the observed decadal-scale trend during a specific time period. Thus, the discrepancy actually has no bearing on the existence of the slowdown.
(a) The trend difference between simulation and observation, (b) the percentage of simulations overestimating the observed trend, and (c),(d) the PDFs of GMST trends. In (a) the trend differences are between the mean observation and the MMM of 233 simulations, and the significant trend differences (95% confidence level) are marked by dots. In (b) the percentage of simulations that significantly overestimates the observed trend for each period is shown. The bottom panels [(c) and (d)] show the simulated and observed PDFs of GMST trends during 1998–2012 and over 15-yr periods during the whole historical run period (1850–2014), respectively. All the six observations are used to calculate the observed PDFs. Note that (c) compares the observed and simulated trends for the same period (i.e., 1998–2012), which assesses the performance of models in simulating period-to-period trends; in contrast, (d) compares the observed and simulated trends for fitting intervals with the same length (i.e., 15 years) during the whole historical run period (1850–2014), which evaluates the capacity of models in reproducing the observed 15-yr trends in a statistical sense.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
(a) The trend difference between simulation and observation, (b) the percentage of simulations overestimating the observed trend, and (c),(d) the PDFs of GMST trends. In (a) the trend differences are between the mean observation and the MMM of 233 simulations, and the significant trend differences (95% confidence level) are marked by dots. In (b) the percentage of simulations that significantly overestimates the observed trend for each period is shown. The bottom panels [(c) and (d)] show the simulated and observed PDFs of GMST trends during 1998–2012 and over 15-yr periods during the whole historical run period (1850–2014), respectively. All the six observations are used to calculate the observed PDFs. Note that (c) compares the observed and simulated trends for the same period (i.e., 1998–2012), which assesses the performance of models in simulating period-to-period trends; in contrast, (d) compares the observed and simulated trends for fitting intervals with the same length (i.e., 15 years) during the whole historical run period (1850–2014), which evaluates the capacity of models in reproducing the observed 15-yr trends in a statistical sense.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
(a) The trend difference between simulation and observation, (b) the percentage of simulations overestimating the observed trend, and (c),(d) the PDFs of GMST trends. In (a) the trend differences are between the mean observation and the MMM of 233 simulations, and the significant trend differences (95% confidence level) are marked by dots. In (b) the percentage of simulations that significantly overestimates the observed trend for each period is shown. The bottom panels [(c) and (d)] show the simulated and observed PDFs of GMST trends during 1998–2012 and over 15-yr periods during the whole historical run period (1850–2014), respectively. All the six observations are used to calculate the observed PDFs. Note that (c) compares the observed and simulated trends for the same period (i.e., 1998–2012), which assesses the performance of models in simulating period-to-period trends; in contrast, (d) compares the observed and simulated trends for fitting intervals with the same length (i.e., 15 years) during the whole historical run period (1850–2014), which evaluates the capacity of models in reproducing the observed 15-yr trends in a statistical sense.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
In summary, by determining the baseline for measuring slowdown, the definition type greatly affects the identification of slowdown events. The probability of the occurrence of a slowdown event would decrease in turn, if definition 3, definition 2a, definition 2b, and definition 1 are used, respectively (Table 2). During all decadal-scale periods around 2000s, there is 97% ± 4% probability of an unexpected weak warming based on definition 3 referring to model-derived expectations, 77% ± 12% probability of a warming slowdown based on definition 2a referring to the warming rate of the previous warming surge, 45% ± 23% probability of a slower-than-average warming event based on definition 2b referring to the warming rate during 1950–99 (the probability is 34% ± 18% when referring to 1900–99), and 11% ± 13% probability of a real warming hiatus based on definition 1 referring to zero warming rate. Generally, the first one or two decades of the twenty-first century indeed experience a weak warming episode relative to the previous warming surge, although at higher greenhouse gas concentrations. Since the decadal-scale warming rates in 2000s are vastly lower than those in 1990s, a hiatus/slowdown event more likely occurs around 2000s than around 1990s. Therefore, from the perspective of the temperature change, the probability of the slowdown and hiatus events occurring in the 2000s is significantly higher than that in the 1990s, regardless of which definition and datasets are adopted. For a specific period, whether there is a warming slowdown depends on the magnitude of the warming reduction, which is in turn determined by definition and datasets choices. Overall, the choices of period, dataset, and definition almost fully account for the controversies over the existence of the slowdown, particularly the period choice.
d. Centennial-scale anthropogenic global warming trend presents no slowdown
The decadal-scale warming slowdown is not incompatible with the anthropogenic long-term warming trend. The slowdown is just a surface manifestation that the local cooling effects of natural variabilities temporarily superimpose onto the anthropogenic warming signal. Figure 6a displays the full range of the linear-trend-determined warming rates during the whole instrumental period, with the long-term and short-term warming rates in the upper and lower parts, respectively. The steady long-term warming rates correspond to the pure anthropogenic warming trend, while the highly variable short-term warming rates are primarily dominated by the natural variabilities. From bottom to top, the warming rates change less and less, indicating that the influence of period choice on the warming rate decreases with increasing period length. The short-term linear trend, with strong natural variabilities included, severely suffers from the period selection bias, thus making little sense in detecting the centennial-scale global warming change. In contrast, the long-term warming rates are quite steady and show little change over different periods. The significant anthropogenic warming signal could be clearly seen in the long-term linear trend, especially during periods longer than 80 years at which time scale most impacts of strong short-term variabilities are averaged out. However, limited by the length needed to eliminate the interference of short-term processes, the long-term linear trend of GMST has difficulty in reflecting the whole secular warming process completely and accurately, especially over the last several decades, which are closely watched.
(a) The warming rates of GMST, (b) the intrinsic trend, and (c) natural variabilities. The purple curves in (a)–(c) corresponding to the right coordinate are the time series of GMST, intrinsic trend, and natural variabilities, respectively. The GMST time series is the mean of six observational derived GMST series. It is further divided into the monotonous intrinsic trend and the quasi-periodic natural variabilities by employing the EEMD method. The two parts roughly represent the anthropogenic warming signal and the climate internal variabilities, respectively. The circles on the intrinsic trend curve in (b) show the time derivatives of the intrinsic trend at 1850, 1860, …, 2010, and 2017, which represent the instantaneous long-term anthropogenic warming rates at those times. The color and size of the circle indicate the instantaneous long-term warming rate. The dots are marked on the periods with insignificant trend (95% confidence level).
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
(a) The warming rates of GMST, (b) the intrinsic trend, and (c) natural variabilities. The purple curves in (a)–(c) corresponding to the right coordinate are the time series of GMST, intrinsic trend, and natural variabilities, respectively. The GMST time series is the mean of six observational derived GMST series. It is further divided into the monotonous intrinsic trend and the quasi-periodic natural variabilities by employing the EEMD method. The two parts roughly represent the anthropogenic warming signal and the climate internal variabilities, respectively. The circles on the intrinsic trend curve in (b) show the time derivatives of the intrinsic trend at 1850, 1860, …, 2010, and 2017, which represent the instantaneous long-term anthropogenic warming rates at those times. The color and size of the circle indicate the instantaneous long-term warming rate. The dots are marked on the periods with insignificant trend (95% confidence level).
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
(a) The warming rates of GMST, (b) the intrinsic trend, and (c) natural variabilities. The purple curves in (a)–(c) corresponding to the right coordinate are the time series of GMST, intrinsic trend, and natural variabilities, respectively. The GMST time series is the mean of six observational derived GMST series. It is further divided into the monotonous intrinsic trend and the quasi-periodic natural variabilities by employing the EEMD method. The two parts roughly represent the anthropogenic warming signal and the climate internal variabilities, respectively. The circles on the intrinsic trend curve in (b) show the time derivatives of the intrinsic trend at 1850, 1860, …, 2010, and 2017, which represent the instantaneous long-term anthropogenic warming rates at those times. The color and size of the circle indicate the instantaneous long-term warming rate. The dots are marked on the periods with insignificant trend (95% confidence level).
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
To investigate the complete and accurate evolution of the centennial-scale global warming signal during the whole instrumental period, we employed the EEMD-determined intrinsic trend, which is more physically meaningful. The GMST time series (purple line in Fig. 6a) is divided into two parts. One part is the monotonically increasing intrinsic trend (purple line in Fig. 6b). It represents the long-term anthropogenic warming signal because its temporal-spatial evolution agrees well with the response to the buildup of the greenhouse gases as shown by Wu et al. (2011) and Tung and Chen (2018). The other part is the sum of various natural variabilities (purple line in Fig. 6c). The linear trends of the two parts are shown in Figs. 6b and 6c, respectively. The features of the linear trends of the intrinsic trend are very similar to the long-term GMST linear trends (upper part of Fig. 6a) but distinct from the short-term ones (lower part of Fig. 6a). As the interference of short-term natural variabilities is excluded in the intrinsic trend, its linear trends are totally independent of the period choice. All the linear trends of intrinsic trend during periods centered on the same year share almost the same value, regardless of whether the fitting length is 5 years or 100 years. The time derivative of the intrinsic trend at a specific time represents the instantaneous long-term warming rate at that time (Fig. 6b, the circles on the intrinsic trend curve; see also Fig. 7). For example, the instantaneous intrinsic warming rate in 1949 was 0.07°C decade−1, which is exactly equal to the 100-yr linear trend of GMST centered on 1949 (i.e., during 1900–99). Therefore, the intrinsic trend accurately represents the centennial-scale tendency underlying the complex GMST evolution. More importantly, it could instantaneously and completely capture the accurate evolution of anthropogenic centennial-scale warming over the whole data domain.
The instantaneous long-term warming rates. The intrinsic warming rates are defined as the time derivatives of the intrinsic trend, which is shown by the purple curve corresponding to the right coordinate (as in Fig. 6b). The circles show the warming rates in 1975, 1998, and 2012, respectively.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The instantaneous long-term warming rates. The intrinsic warming rates are defined as the time derivatives of the intrinsic trend, which is shown by the purple curve corresponding to the right coordinate (as in Fig. 6b). The circles show the warming rates in 1975, 1998, and 2012, respectively.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The instantaneous long-term warming rates. The intrinsic warming rates are defined as the time derivatives of the intrinsic trend, which is shown by the purple curve corresponding to the right coordinate (as in Fig. 6b). The circles show the warming rates in 1975, 1998, and 2012, respectively.
Citation: Journal of Climate 35, 6; 10.1175/JCLI-D-21-0373.1
The intrinsic trend shows that the centennial-scale anthropogenic global warming process keep accelerating during the instrumental period 1850–2017 without any hiatus or slowdown. The globally positive intrinsic warming rates indicate that the global temperature has been increasing continuously since 1850 (Figs. 6b and 7). The intrinsic warming rate gradually increases over time, which physically matches the sustained accumulation of greenhouse gases in the atmosphere (Fig. 7). The instantaneous intrinsic warming rate increased from 0.01°C decade−1 at 1850 to 0.15°C decade−1 at 2017. In contrast, the linear trends of natural variabilities exhibit quasi-periodically altering warming and cooling episodes (Fig. 6c). The two distinct signals are mixed in the linear trend of the GMST, especially for periods shorter than 80 years. Sometimes, the local cooling (or warming) linear trend from the natural variabilities temporarily masks (or enhances) the anthropogenic warming signal and directly leads to a local warming deceleration (or acceleration) event lasting one or several decades. The recent warming slowdown is such a case. During the period 1975–2012, the intrinsic trend steadily warmed with an evident acceleration (Figs. 6b and 7). The instantaneous intrinsic warming rates were 0.10°, 0.12°, and 0.14°C decade−1 during 1975, 1998 and 2012, respectively (Fig. 7), with a 46% increase from 1975 to 2012. The warming acceleration is physically reasonable since the greenhouse gas–induced radiative forcing rapidly increased during 1975–2012. However, the linear trend of GMST in 1998–2012 suddenly presented a warming slowdown relative to that in 1975–97. Wei et al. (2019) reported that the slowdown is primarily contributed by the interannual- and interdecadal-scale variabilities, as their negative local linear trends during 1998–2012 offset over half of the positive linear trends from the steady intrinsic trend and the warming phase of the multidecadal variability. Therefore, although the recent warming slowdown is real enough from the perspective of linear-trend-determined warming rate change, it is only a temporary and superficial artifact obtained by applying the short-term linear trend to nonlinear global temperature time series. The warming slowdown determined by the decadal-scale linear trend may lead the public to misunderstand the monotonic global warming process as it could never prevent the continuation of the global warming.
4. Summary and discussion
In this work, we partly reconciled the controversies over the existence of the global warming slowdown during the early twenty-first century and the discrepancy between the decadal-scale warming slowdown and the centennial-scale anthropogenic warming process. Ultimately, whether there is a hiatus or slowdown is evaluated by the linear warming rate change relative to a baseline prescribed by hiatus definition. The warming rate is usually determined by calculating the least squares linear trend of the global temperature estimates. Affected by the strong natural variabilities, the decadal-scale linear trend severely suffers from the period selection bias, which largely explains why there is so much debate about the reality of the recent warming slowdown. Besides, as the observational biases, particularly the spatial coverage, have a significant impact on the linear trend during the early 2000s, the dataset choice further deteriorates the arguments about the existence of the slowdown. The uncorrected datasets with incomplete coverage show significantly lower temperature trends over recent decades than the filled-coverage datasets. In addition, the type of hiatus definition is also an important reason for the controversies over its authenticity. By determining the baseline for slowdown, the definition greatly affects the identification of slowdown events. When the slowdown is identified based on definition 3 (unexpected weak warming), definition 2a (warming slowdown), definition 2b (slower-than-average warming), and definition 1 (real hiatus) respectively, the number of recognized slowdown events would decrease in each successive case. Overall, the choices of the period, dataset, and definition almost fully account for the controversies over the existence of the slowdown/hiatus, particularly the period choice. However, when the focus is shifted from specific periods to a wider perspective, the first one or two decades of the twenty-first century are in a weak warming episode following the strong warming surge in the late twentieth century. Thereby, the probability of a slowdown/hiatus event occurring in the 2000s is significantly higher than that in the 1990s, regardless of which definition and dataset are adopted. This supports a decadal-scale warming slowdown during the early twenty-first century relative to the warming surge in the late twentieth century, despite higher greenhouse gas concentrations.
Notably, the decadal-scale warming slowdown is not incompatible with the centennial-scale anthropogenic warming trend. In the long run, the global temperature has undoubtedly been warming over the whole instrumental period, responding to the continuously increasing greenhouse gas concentrations in the atmosphere. The secular anthropogenic warming has been smoothly accelerating, corresponding to the sustained accumulation of greenhouse gases, without any hiatus or slowdown. However, cooling (or warming) episodes accompanied by locally strong natural variabilities sometimes mask (or enhance) the steady anthropogenic warming signal, leading to a temporary global warming deceleration (or acceleration) event. Both the warming slowdown in the early twenty-first century and the warming surge in the late twentieth century are such cases. Nevertheless, the short-term cooling effect could never prevent the continuation of the anthropogenic warming, since natural variabilities fluctuate in cycles and the corresponding cooling episodes will eventually end. The recent warming slowdown is only a superficial manifestation resulting from arbitrarily applying the short-term linear trend to nonlinear global temperature time series, without distinguishing the centennial-scale anthropogenic warming signal corresponding to increasing greenhouse gases concentrations from the short-term noises from unforced climate natural variabilities.
The warming slowdown determined by the decadal-scale linear trend may make the monotonic global warming process misunderstood. Some careful considerations about the limitations of the linear trend should be recalled before using it to detect or attribute the global temperature change, although it is broadly accepted and frequently adopted. First, it makes little sense to use a straight line to fit the global temperature time series, which is significantly nonlinear and nonstationary. Second, the short-term linear trend lacks robustness with high sensitivity to the period choice. It may give completely different trend values for similar periods, which is the most important reason for the current controversy over the reality of the slowdown. Besides, it can also easily change when new data are added, which disagrees with the physical reality that the future should not affect the present (Fig. S5). Third, the linear trend could not accurately reflect the overall tendency and instead it represents only an average state, such as a 50-yr trend, with mixed anthropogenic warming signal and unforced natural variabilities. In contrast, the ensemble empirical mode decomposition (EEMD)-determined intrinsic trend has been proven to be a more physically meaningful tool to accurately reflect the overall tendency hidden within the intricate global temperature evolution. The intrinsic trend successfully captures the centennial-scale anthropogenic warming signal over the whole data domain as all the natural variabilities are removed. Furthermore, the intrinsic trend could provide the instantaneous state of the overall tendency at each moment. In this work, it is clearly shown that anthropogenic global warming has recently accelerated rather than paused or slowed down. Most importantly, the intrinsic trend is quite robust, as it is insensitive to the period choices and hardly changes with the addition of new data (Fig. S5). With these advantages, the nonlinear intrinsic trend should be a preferable choice for detecting and attributing nonlinear anthropogenic climate change.
Acknowledgments.
We thank Professor N. E. Huang of the First Institute of Oceanography, State Oceanic Administration and Professor R. Hu of Ocean University of China for the encouragement and helpful discussions. We thank Dr. Z. Wu of Florida State University for kindly offering the EEMD code. We thank all the data providers. M. Wei is jointly supported by the National Natural Science Foundation of China (NSFC) (41806043) and the Basic Scientific Fund for National Public Research Institutes of China (2019Q08). F. Qiao is supported by the NSFC (41821004). Z. Song is jointly supported by the National Key R&D Program of China (2016YFA0602200), the National Program on Global Change and Air–Sea Interaction (GASI-IPOVAI-06), and the China–Korea Cooperation Project on Northwestern Pacific Climate Change and its Prediction. Q. Shu is supported by the Basic Scientific Fund for National Public Research Institute of China (Shu Xingbei Young Talent Program 2019S06).
Data availability statement.
Data analyzed in this study are openly available at locations cited in Table S1 in the online supplemental material.
REFERENCES
Allan, R. P., C. Liu, N. G. Loeb, M. D. Palmer, M. Roberts, D. Smith, and P. Vidale, 2014: Changes in global net radiative imbalance 1985–2012. Geophys. Res. Lett., 41, 5588–5597, https://doi.org/10.1002/2014GL060962.
Arora, A., S. A. Rao, R. Chattopadhyay, T. Goswami, G. George, and C. T. Sabeerali, 2016: Role of Indian Ocean SST variability on the recent global warming hiatus. Global Planet. Change, 143, 21–30, https://doi.org/10.1016/j.gloplacha.2016.05.009.
Banholzer, S., and S. Donner, 2014: The influence of different El Niño types on global average temperature. Geophys. Res. Lett., 41, 2093–2099, https://doi.org/10.1002/2014GL059520.
Cahill, N., 2018: No “slowdown” in the story of global warming. Significance, 15, 24–27, https://doi.org/10.1111/j.1740-9713.2018.01107.x.
Cazenave, A., H. Dieng, B. Meyssignac, K. von Schuckmann, B. Decharme, and E. Berthier, 2014: The rate of sea-level rise. Nat. Climate Change, 4, 358–361, https://doi.org/10.1038/nclimate2159.
Chen, X., and K. Tung, 2014: Varying planetary heat sink led to global-warming slowdown and acceleration. Science, 345, 897–903, https://doi.org/10.1126/science.1254937.
Cheng, L., G. Wang, J. Abraham, and G. Huang, 2018: Decadal ocean heat redistribution since the late 1990s and its association with key climate modes. Climate, 6, 91, https://doi.org/10.3390/cli6040091.
Clement, A., and P. DiNezio, 2014: The tropical Pacific Ocean—Back in the driver’s seat? Science, 343, 976–978, https://doi.org/10.1126/science.1248115.
Cowtan, K., and R. G. Way, 2014: Coverage bias in the HadCRUT4 temperature series and its impact on recent temperature trends. Quart. J. Roy. Meteor. Soc., 140, 1935–1944, https://doi.org/10.1002/qj.2297.
Dai, A., J. C. Fyfe, S. Xie, and X. Dai, 2015: Decadal modulation of global surface temperature by internal climate variability. Nat. Climate Change, 5, 555–559, https://doi.org/10.1038/nclimate2605.
Dai, X., and P. Wang, 2018: Identifying the early 2000s hiatus associated with internal climate variability. Sci. Rep., 8, 13602, https://doi.org/10.1038/s41598-018-31862-z.
Deser, C., R. Guo, and F. Lehner, 2017: The relative contributions of tropical Pacific sea surface temperatures and atmospheric internal variability to the recent global warming hiatus. Geophys. Res. Lett., 44, 7945–7954, https://doi.org/10.1002/2017GL074273.
Dieng, H. B., A. Cazenave, B. Meyssignac, K. von Schuckmann, and H. Palanisamy, 2017: Sea and land surface temperatures, ocean heat content, Earth’s energy imbalance and net radiative forcing over the recent years. Int. J. Climatol., 37, 218–229, https://doi.org/10.1002/joc.4996.
Dong, L., and M. J. Mcphaden, 2017: The role of external forcing and internal variability in regulating global mean surface temperatures on decadal timescales. Environ. Res. Lett., 12, 034011, https://doi.org/10.1088/1748-9326/aa5dd8.
Douville, H., A. Voldoire, and O. Geoffroy, 2015: The recent global warming hiatus: What is the role of Pacific variability? Geophys. Res. Lett., 42, 880–888, https://doi.org/10.1002/2014GL062775.
Drijfhout, S. S., A. T. Blaker, S. A. Josey, A. J. G. Nurser, B. Sinha, and M. A. Balmaseda, 2014: Surface warming hiatus caused by increased heat uptake across multiple ocean basins. Geophys. Res. Lett., 41, 7868–7874, https://doi.org/10.1002/2014GL061456.
Durack, P. J., P. J. Gleckler, F. W. Landerer, and K. E. Taylor, 2014: Quantifying underestimates of long-term upper-ocean warming. Nat. Climate Change, 4, 999–1005, https://doi.org/10.1038/nclimate2389.
Easterling, D. R., and M. F. Wehner, 2009: Is the climate warming or cooling? Geophys. Res. Lett., 36, L08706, https://doi.org/10.1029/2009GL037810.
England, M. H., and Coauthors, 2014: Recent intensification of wind-driven circulation in the Pacific and the ongoing warming hiatus. Nat. Climate Change, 4, 222–227, https://doi.org/10.1038/nclimate2106.
England, M. H., J. B. Kajtar, and N. Maher, 2015: Robust warming projections despite the recent hiatus. Nat. Climate Change, 5, 394–396, https://doi.org/10.1038/nclimate2575.
Eyring, V., S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor, 2016: Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev., 9, 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016.
Folland, C. K., O. Boucher, A. Colman, and D. E. Parker, 2018: Causes of irregularities in trends of global mean surface temperature since the late 19th century. Sci. Adv., 4, eaao5297, https://doi.org/10.1126/sciadv.aao5297.
Fyfe, J. C., and N. P. Gillett, 2014: Recent observed and simulated warming. Nat. Climate Change, 4, 150–151, https://doi.org/10.1038/nclimate2111.
Fyfe, J. C., N. P. Gillett, and F. W. Zwiers, 2013: Overestimated global warming over the past 20 years. Nat. Climate Change, 3, 767–769, https://doi.org/10.1038/nclimate1972.
Fyfe, J. C., and Coauthors, 2016: Making sense of the early-2000s warming slowdown. Nat. Climate Change, 6, 224–228, https://doi.org/10.1038/nclimate2938.
Gastineau, G., A. R. Friedman, M. Khodri, and J. Vialard, 2019: Global ocean heat content redistribution during the 1998–2012 Interdecadal Pacific Oscillation negative phase. Climate Dyn., 53, 1187–1208, https://doi.org/10.1007/s00382-018-4387-9.
Gleisner, H., P. Thejll, B. Christiansen, and J. K. Nielsen, 2015: Recent global warming hiatus dominated by low-latitude temperature trends in surface and troposphere data. Geophys. Res. Lett., 42, 510–517, https://doi.org/10.1002/2014GL062596.
Guan, X., J. Huang, R. Guo, and P. Lin, 2015: The role of dynamically induced variability in the recent warming trend slowdown over the Northern Hemisphere. Sci. Rep., 5, 12669, https://doi.org/10.1038/srep12669.
Guemas, V., F. J. Doblas-Reyes, I. Andreu-Burillo, and M. Asif, 2013: Retrospective prediction of the global warming slowdown in the past decade. Nat. Climate Change, 3, 649–653, https://doi.org/10.1038/nclimate1863.
Hansen, J., R. Ruedy, M. Sato, and K. Lo, 2010: Global surface temperature change. Rev. Geophys., 48, RG4004, https://doi.org/10.1029/2010RG000345.
Hausfather, Z., K. Cowtan, D. C. Clarke, P. Jacobs, M. Richardson, and R. Rohde, 2017: Assessing recent warming using instrumentally homogeneous sea surface temperature records. Sci. Adv., 3, e1601207, https://doi.org/10.1126/sciadv.1601207.
Hawkins, E., T. Edwards, and D. McNeall, 2014: Pause for thought. Nat. Climate Change, 4, 154–156, https://doi.org/10.1038/nclimate2150.
Haywood, J. M., A. Jones, and G. S. Jones, 2014: The impact of volcanic eruptions in the period 2000–2013 on global mean temperature trends evaluated in the HadGEM2-ES climate model. Atmos. Sci. Lett., 15, 92–96, https://doi.org/10.1002/asl2.471.
Held, I. M., 2013: The cause of the pause. Nature, 501, 318–319, https://doi.org/10.1038/501318a.
Hirahara, S., M. Ishii, and Y. Fukuda, 2014: Centennial-scale sea surface temperature analysis and its uncertainty. J. Climate, 27, 57–75, https://doi.org/10.1175/JCLI-D-12-00837.1.
Huang, N. E. , and Coauthors, 1998: The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proc. Roy. Soc. London, A454, 903–995, https://doi.org/10.1098/rspa.1998.0193.
Huber, M., and R. Knutti, 2014: Natural variability, radiative forcing and climate response in the recent hiatus reconciled. Nat. Geosci., 7, 651–656, https://doi.org/10.1038/ngeo2228.
Hunt, B. G., 2011: The role of natural climatic variation in perturbing the observed global mean temperature trend. Climate Dyn., 36, 509–521, https://doi.org/10.1007/s00382-010-0799-x.
IPCC, 2013: Climate Change 2013: The Physical Science Basis. Cambridge University Press, 1535 pp., https://www.ipcc.ch/report/ar5/wg1/.
Ishihara, K., 2006: Calculation of global surface temperature anomalies with COBE-SST (in Japanese). Wea. Serv. Bull., 73, S19–S25.
Ji, F., Z. Wu, J. Huang, and E. P. Chassignet, 2014: Evolution of land surface air temperature trend. Nat. Climate Change, 4, 462–466, https://doi.org/10.1038/nclimate2223.
Johnson, N. C., S.-P. Xie, Y. Kosaka, and X. Li, 2018: Increasing occurrence of cold and warm extremes during the recent global warming slowdown. Nat. Commun., 9, 1724, https://doi.org/10.1038/s41467-018-04040-y.
Kamae, Y., H. Shiogama, M. Watanabe, and M. Kimoto, 2014: Attributing the increase in Northern Hemisphere hot summers since the late 20th century. Geophys. Res. Lett., 41, 5192–5199, https://doi.org/10.1002/2014GL061062.
Karl, T. R., and Coauthors, 2015: Possible artifacts of data biases in the recent global surface warming hiatus. Science, 348, 1469–1472, https://doi.org/10.1126/science.aaa5632.
Katsman, C. A., and G. J. van Oldenborgh, 2011: Tracing the upper ocean’s “missing heat”. Geophys. Res. Lett., 38, L14610, https://doi.org/10.1029/2011GL048417.
Kaufmann, R. K., H. Kauppi, M. L. Mann, and J. H. Stock, 2011: Reconciling anthropogenic climate change with observed temperature 1998–2008. Proc. Natl. Acad. Sci. USA, 108, 11 790–11 793, https://doi.org/10.1073/pnas.1102467108.
Kerr, R. A., 2009: What happened to global warming? Scientists say just wait a bit. Science, 326, 28–29, https://doi.org/10.1126/science.326_28a.
Knight, J. R. , and Coauthors, 2009: Do global temperature trends over the last decade falsify climate predictions? [in “State of the Climate in 2008”]. Bull. Amer. Meteor. Soc., 90, 22–23.
Kosaka, Y., 2014: Increasing wind sinks heat. Nat. Climate Change, 4, 172–173, https://doi.org/10.1038/nclimate2138.
Kosaka, Y., and S.-P. Xie, 2013: Recent global-warming hiatus tied to equatorial Pacific surface cooling. Nature, 501, 403–407, https://doi.org/10.1038/nature12534.
Kosaka, Y., and S.-P. Xie, 2016: The tropical Pacific as a key pacemaker of the variable rates of global warming. Nat. Geosci., 9, 669–673, https://doi.org/10.1038/ngeo2770.
Lean, J. L., 2018: Observation-based detection and attribution of 21st century climate change. Wiley Interdiscip. Rev.: Climate Change, 9, e511, https://doi.org/10.1002/wcc.511.
Lean, J. L., and D. H. Rind, 2009: How will Earth’s surface temperature change in future decades? Geophys. Res. Lett., 36, L15708, https://doi.org/10.1029/2009GL038932.
Lee, S., W. Park, M. O. Baringer, A. L. Gordon, B. Huber, and Y. Liu, 2015: Pacific origin of the abrupt increase in Indian Ocean heat content during the warming hiatus. Nat. Geosci., 8, 445–449, https://doi.org/10.1038/ngeo2438.
Lewandowsky, S., J. S. Risbey, and N. Oreskes, 2015: On the definition and identifiability of the alleged “hiatus” in global warming. Sci. Rep., 5, 16784, https://doi.org/10.1038/srep16784.
Lewandowsky, S., J. S. Risbey, and N. Oreskes, 2016: The “pause” in global warming: Turning a routine fluctuation into a problem for science. Bull. Amer. Meteor. Soc., 97, 723–733, https://doi.org/10.1175/BAMS-D-14-00106.1.
Li, Q., and Coauthors, 2015: China experiencing the recent warming hiatus. Geophys. Res. Lett., 42, 889–898, https://doi.org/10.1002/2014GL062773.
Liang, X., and L. Yu, 2016: Variations of the global net air–sea heat flux during the “hiatus” period (2001–10). J. Climate, 29, 3647–3660, https://doi.org/10.1175/JCLI-D-15-0626.1.
Liao, E., W. Lu, X. Yan, Y. Jiang, and A. Kidwell, 2015: The coastal ocean response to the global warming acceleration and hiatus. Sci. Rep., 5, 16630, https://doi.org/10.1038/srep16630.
Liao, E., X. Yan, Y. Jiang, and A. N. Kidwell, 2019: New findings on the route of heat transport between the Indo-Pacific and Southern Ocean. Climate Dyn., 52, 5145–5151, https://doi.org/10.1007/s00382-018-4436-4.
Liu, W., and S.-P. Xie, 2018: An ocean view of the global surface warming