The Key Role of Atlantic Multidecadal Oscillation in Minimum Temperature Over North America During Global Warming Slowdown

Abstract

Daily minimum temperature (Tmin) is an important variable in both global and regional climate changes, and its variability can greatly affect the ecological system. In the early 21st century, warming slowdown is seen over the North Hemisphere and North America is one of the major cooling centers. In this study, we found that Tmin experienced an obvious decline in North America during warming slowdown period. Such Tmin decline is closely related to the Atlantic Multidecadal Oscillation (AMO); the correlation between the decadal components of Tmin and AMO reached 0.71 during 1950–2014. According to composite analysis, the AMO on the positive (negative) phase takes two low-pressure (high-pressure) systems in the northeastern Pacific and the North Atlantic at night, accompanied by cyclonic (anticyclonic) circulations and warm (cold) advection in North America. Therefore, the analyses conclude that the Tmin decline during warming slowdown period is a result of the synchronous decrease of the AMO. The results emphasize the key role of AMO on the decadal variation of Tmin in North America.

1 Introduction

The time series of global-mean temperature has shown a well-known rise since the early 20th century, most notably since the late-1970s (Foster & Rahmstorf, 2011). Its changes are characterized by significant increase in Tmin and insignificant change in maximum temperature (Tmax). Easterling et al. (1997) confirmed that Tmin increased about twice as fast as Tmax over global land areas since 1950, and this asymmetrical change led to an obvious decrease in the diurnal temperature range on both global scale (Dai et al., 1999) and regional scale (Zhang et al., 2017). Tmin is a sensitive variable for global warming and has an important impact on ecosystems. Information about the frequency and severity of frosts is valuable to the agricultural industries, especially in spring (Tabony, 1985). The increasing Tmin has a direct impact on crops and can lead to the reduction of rice yields. Increasing Tmin by 1 °C may reduce rice yields by about 10% (Peng et al., 2004), which will have a negative impact on the survival of mankind. Tutin and Fernandez (1993) found that a small permanent increase in Tmin resulting from global warming has dramatic consequences: not only would certain tree species cease to reproduce but also would drastically reduce the quantity of food available to frugivores. The above ground net primary productivity of grassland vegetation is negatively correlated with seasonal Tmin, and Tmin warming has dramatic effects on ecosystem-level carbon, nitrogen, water cycling, and carbon storage (Alward et al., 1999; Turnbull et al., 2002).

In North America (NA), the southwestern area has been observed the largest increase in Tmin over 1951–2000 (Morak et al., 2013). Global climate models show a large increase in warm nights in southwestern NA during 2081–2100 relative to 1980–1999 (Dupuis, 2014). It is thus important to gain a better understanding of observed decadal change of Tmin in NA. Alfaro et al. (2006) found that predictability of Tmin in NA is strongly influenced by large-scale patterns and remote climate conditions such as the sea surface temperature in the Pacific Ocean. The decadal change of Tmin in NA is sensitive to the effects of internal climate variability (ICV), especially oceanic ICV (Steinman et al., 2015). In the early 21st century, the warming rate of global-mean temperature showed a slowdown (Easterling & Wehner, 2009), which attracts great attention from the research community and the public. At present, most studies suggest that warming slowdown was mainly induced by the oceanic ICV (England et al., 2014; Meehl et al., 2014; Trenberth & Fasullo, 2013; Watanabe et al., 2014). For instance, the La-Niña-like decadal cooling favored the global warming slowdown through partially offsetting the greenhouse gas-induced warming (Kosaka & Xie, 2013). Many studies point out that this warming hiatus was a decadal signal (Guan et al., 2015; Huang et al., 2016), its effect was not limited to temperature, yet it has spread to different aspects of climate change, such as wetting signal over the Northern Hemisphere, increasing areas of evergreen conifer forest over Siberia, and more extreme climate events in China (Guan et al., 2017; He et al., 2017; Niu et al., 2017; Wang et al., 2016).

A previous result found that the decrease of Tmax has contributed most to the decrease in DTR during the warming slowdown period over China (Li et al., 2015). However, less is known about Tmax and Tmin in NA during the warming hiatus decade. NA is one of the major continental cooling centers during the warming slowdown period (Huang et al., 2016). Hence, we focus on the variation of Tmin in NA during the warming slowdown period in this study. We selected the area from 15 to 60°N of NA as our study region. This paper is arranged as follows. In section 2, we describe data and methodology. In section 3, we show the investigation on the variation of Tmin during the warming slowdown period and explore the key oceanic ICV signals affecting the decadal variability of Tmin. Section 4 presented the summary and discussion.

2 Materials and Methods

2.1 Reanalysis Data Sets and Oscillation Indices

In this study, we used the Climate Prediction Center global daily temperature, maximum temperature, and minimum temperature to analyze their variation in NA during the warming slowdown period. The data set covers the period from 1979 to 2016 with a spatial resolution of 0.5° by 0.5°. The monthly TS4.01 data set is obtained from the Climate Research Unit (CRU) at the University of East Anglia, which has a spatial resolution of 0.5° by 0.5° over 1901–2012 (Harris et al., 2014). The TS4.01 data set is used to verify our results and do the composite analysis. ERA-Interim daily data set is used to examine the geopotential height and wind fields and to identify the physical processes (Berrisford et al., 2011). The Atlantic Multidecadal Oscillation (AMO), the Pacific Decadal Oscillation (PDO), the North Atlantic Oscillation (NAO), the Nino 3.4 and the Arctic Oscillation (AO) indices are obtained from the Earth System Research Laboratory (ESRL), which covers the period from 1948 to 2016.

2.2 The Mann-Kendall Test Method

The Mann-Kendall (MK) test is a nonparametric statistical test for identifying the turning point in time series of long-term data (Kendall, 1975; Mann, 1945). Considering a sequence of observations x₁, x₂, …, x_n, the mathematical equations for calculating MK statistics S is as follows:

(1)

With

(2)

The UF and UB are defined as follows:

(3)

The E (S_k) and var (S_k) can be calculated by the following:

(4)

UF_i is the standard normal distribution, and UB_i is also calculated as above steps with inverse time sequence x_n, x_n − 1, …, x₁. If there is an intersection between the two curves of UF and UB, and this point is between the lines of ±1.96 (α = 0.05), then the year corresponding to the intersection could be considered as the time when the abrupt change starts.

2.3 Dynamic Adjustment Method

The dynamic adjustment method was proposed by Wallace and Johanson (2012) and described in detail by Smoliak et al. (2015). The core of the dynamic adjustment method is the partial least squares regression of surface air temperature to sea level pressure in a pointwise manner. It can divide the raw temperature into two parts, dynamically induced temperature and radiatively forced temperature. The dynamically induced part is associated with oceanic oscillation and atmospheric circulation. The radiatively forced part is linked to radiative factors such like greenhouse gases, volcanic eruptions, aerosol emissions, and local anthropogenic forcing. The method can only separate data until 2012 because the latest “NOAA-CIRES 20th Century Reanalysis version 2 Monthly Averages (20CR)” sea level pressure data set is updated to 2012.

2.4 Ensemble Empirical Mode Decomposition Method

The Ensemble Empirical Mode Decomposition (EEMD) method was presented in Wu and Huang (2009). It can divide climate variability into different oscillatory components with intrinsic timescales, including interannual, decadal, and multidecadal scales. The time series of variables can be decomposed into different intrinsic mode functions (IMFs). For time series X(t), we have

(5)

where C_j(t) represents the jth IMF, which is an oscillatory component with a specific frequency. R(t) is the residual of X(t). According to the frequencies of IMFs, we can extract IMF3, IMF4, and IMF5 that belong to the decadal scale and sum them up to obtain the temporal variation of decadal signals.

3 Results

Figure 1 shows the temporal evolution of annual-mean temperature (Temp) Tmax and Tmin in NA during 1982–2014. Tmax and Temp presented increase trends during 1982–1998 and then changed to warming slowdown during 1998–2014 (the warming slowdown period is referred to as the hiatus hereafter). Different from Tmax and Temp, Tmin exhibited an obvious cooling trend during the hiatus, with a rate of −0.047 °C/year (passed the significance test at α = 0.10). We further applied the MK test to the time series of Temp, Tmax, and Tmin to identify the turning point. The intersection between the solid and dashed curves represents the turning point of Temp, Tmax, and Tmin (Figure 2a). There was a turning point in Tmin but no abrupt changes in Tmax and Temp around year 1998, which also was the starting year of the hiatus. Given the particularly decrease trend in Tmin during the hiatus, we further show the 17-year running trend of Tmin beginning at 1982 in Figure 2b as a complementary. The 17-year running trend of Tmin shifted from positive to negative around year 1997, which is similar to the result of MK test. Such an abrupt change of Tmin in the hiatus was also found in space. Figures 2c and 2d show an upward trend of Tmin over most regions of NA during the global warming, but a downward trend during the hiatus, and the differences between the Tmin trend during the hiatus and the warming period were almost negative (Figure 2e).

Figure 1

Temporal variations of regional-mean maximum temperature (Tmax), regional-mean temperature (Temp), and regional-mean minimum temperature (Tmin) over North America (180–0°, 15–60°N) from 1982 to 2014. The light blue shading represents the period of global warming slowdown (the hiatus) from 1998 to 2014. The dashed lines represent linear regressions during 1982–2014 and 1998–2014.

Figure 2

(a) The Mann-Kendall mutation test results of Tmax (red curves), Temp (black curves), and Tmin (blue curves) over North America (180–0°, 15–60°N). The intersection between the solid and dashed curves represents the turning point if it is between the two black lines of ±1.96 (α = 0.05). The cross represents the turning point of Tmin. (b) The 17-year running trend of regional-mean Tmin, which is calculated by changing the initial year sequentially. The trend passed the significance test (α = 0.10) if the bar is fully filled. (c) The spatial distribution of Tmin trend from 1982 to 1998. Stippling indicates the 95% statistical confidence level. (d) Same as (c) but during the hiatus. (e) Differences between (d) and (c) (d minus c).

Previous studies (Guan et al., 2015; Guo et al., 2017) pointed out that temperature variability can be divided into dynamical and radiative parts, and the hiatus over the Northern Hemisphere is mainly dominated by the dynamic effect, which offsets the warming from the radiative effect. Therefore, we checked the effect from the dynamic part on the variability of Tmin. According to the dynamic adjustment method, the raw Tmin was divided into dynamically induced Tmin (DITmin) and radiatively forced Tmin (RFTmin). Figure 3a shows the temporal variations of DITmin and RFTmin. The DITmin is similar to Tmin and shows a downward trend during the hiatus with a linear trend of −0.045 °C/year (passed the significance test at α = 0.10), while little change (about −0.005 °C/year), much smaller than that of DITmin, is observed in the RFTmin (Figure 3b). Moreover, the abrupt change of DITmin tested by the MK method occurred around year 1998, which was the same year as Tmin changed abruptly (Figure 4a). The spatial distribution of DITmin trend turned to negative over most regions of NA during the hiatus, and the spatial distributions were consistent with those of Tmin but more homogeneous (Figures 4b and 4c). It is clear that the dynamic effect played an important role in Tmin decline.

Figure 3

(a) Temporal variations of regional-mean Tmin (180–0°, 15–60°N), dynamically induced Tmin (DITmin) and radiatively forced Tmin (RFTmin). (b) The linear trends of DITmin and RFTmin during 1998–2012.

Figure 4

(a) The Mann-Kendall mutation test results of DITmin over North America (180–0°, 15–60°N). The cross between the solid and dashed curves represents the turning point of DITmin (α = 0.05). (b) The spatial distribution of DITmin trend from 1982 to 1998. Stippling indicates the 95% statistical confidence level. (c) Same as (b) but from 1998 to 2012.

The dynamic effect is mainly affected by oceanic ICV, which affects large-scale atmospheric circulation (Deser et al., 2010; Feng et al., 2011; Solomon et al., 2012; Zhang & Cook, 2015). Previous research studies explored the influences of the PDO, AMO, NAO, AO, and Nino 3.4 on the warming slowdown (Huang et al., 2016; Wang et al., 2017; Yao et al., 2016). To detect the key factors for the Tmin decline, we calculated the correlation coefficients of these indices with regional-mean Tmin of NA during the hiatus (Figure 5). We first calculated the detrended Tmin by removing the linear growth before computing correlation coefficients. The comparison illustrates the strongest influence of the AMO on Tmin decline; the correlation coefficient with the AMO is 0.65 for detrended Tmin. The weakest negative correlation coefficients are with the PDO and Nino 3.4, which are −0.05 and −0.14, respectively. The stronger negative correlation coefficients are with the NAO and AO, which are −0.43 and −0.45, respectively. Except for the AMO, none of the indices passed the significance test at α = 0.05 or α = 0.01, and the NAO and AO passed the significance test at α = 0.10. Figure 6 shows the spatial distribution of correlation coefficient between detrended Tmin and AMO during the hiatus. It exhibits a strong positive pattern over most regions of NA except for a few southwestern areas, especially in the northwest and southeast regions with values greater than 0.47. These analyses revealed that the AMO may be the key ICV signal, which affects Tmin variation in NA.

Figure 5

Correlation coefficients of regional-mean detrended Tmin (180–0°, 15–60°N) with the Atlantic Multidecadal Oscillation (AMO), North Atlantic Oscillation (NAO), Pacific Decadal Oscillation (PDO), Arctic Oscillation (AO), and Nino 3.4 during the hiatus. The dashed lines represent the 90%, 95%, and 99% statistical confidence level, respectively.

Figure 6

Correlation coefficients between detrended Tmin and the AMO during the hiatus. Stippling indicates the 95% statistical confidence level.

To further distinguish the effects of the NAO, AO and AMO on Tmin in NA, we applied a multiple linear regression for the regional-mean detrended Tmin with the NAO, AO, and AMO indices and a simple linear regression (SLR) with the AMO as shown in Figure 7a.We find that the time series of detrended Tmin during the hiatus can be approximately described by SLR or multiple linear regression. However, compared with SLR, the addition of NAO and AO indices just slightly improved the goodness of fit from 0.42 to 0.44, and the relative contribution rates of the AMO, NAO, and AO to the detrended Tmin are 73%, 10%, and 17%, respectively (Figure 7b). Thus, the AMO played a key role in Tmin variation. According to composite analysis, the annual-mean of detrended Tmin anomaly based on the positive and negative phases of the AMO was investigated by the long-term CRU data set. The years of positive and negative phases were selected by the values of standardized AMO larger than 1 or smaller than −1 (Figure 8a). Figures 8b–8d show the annual-mean Tmin anomalies according to the AMO phases, together with the differences between negative and positive phase combinations. The spatial distribution of detrended Tmin anomalies over the entire NA has a positive correlation with the AMO. The Tmin anomalies are positive and larger than 0.5 °C over most regions of NA when the AMO is in its positive phase, particularly along the northeastern side of the continent (Figure 8b). Figure 8c shows the spatial distribution of Tmin anomalies when the AMO is in its negative phase, which is totally the inverse of Figure 8b. The differences of Tmin anomalies between the negative and positive phases are negative values. The result indicates that the variation of Tmin is closely related with AMO phase change.

Figure 7

(a) The time series of regional-mean detrended Tmin (black curve) over North America (180–0°, 15–60°N). The blue curve is the multiple linear regression (MLR) for the detrended Tmin with the Atlantic Multidecadal Oscillation (AMO), North Atlantic Oscillation (NAO), and Arctic Oscillation (AO) during the hiatus, and the red curve is the simple linear regression (SLR) for the detrended Tmin with the AMO. (b) Relative contribution rates of AMO, AO, and NAO for the MLR result.

Figure 8

(a) The standardized time series of Atlantic Multidecadal Oscillation (AMO) from 1950 to 2014. The red bar represents the year when the AMO is in positive phase (AMO > 1), the blue bar represents the year of negative phase (AMO < −1), and the gray bar represents the year when the AMO is greater than −1 and less than 1. (b and c) Composite analysis results of Tmin anomaly. Annual-mean detrended Tmin anomalies for the years when the AMO is in positive or negative phase, respectively, during 1950–2014. (d) Differences between (c) and (b) (c minus b). Stippling indicates the 95% statistical confidence level.

Based on the above result of composite analysis, then we explored the possible cause of Tmin decline during the hiatus. Since the Tmin decline was a typical decadal-scale change, we examined the decadal components of the standardized AMO and detrended Tmin (Figure 9). We extracted decadal components of the temporal evolution of variables using the EEMD method (Ji et al., 2014), and the decadal component is thought to be induced mainly by the ICV (Wu et al., 2011; Wallace & Johanson, 2012). We find that the temporal evolutions of the decadal components of detrended Tmin and AMO were coherent during 1950–2014, with a significant correlation coefficient of 0.71 (p < 0.01). The statistical relationship is credible as the long-term trends of Tmin and the AMO have been removed from the original data sets by the EEMD. The rising and falling stages of Tmin corresponded well to the phase changes of the AMO. Especially, the decadal component of detrended Tmin after 1990 increased obviously to a high value around year 2000, and the AMO shifted toward its positive phase during the same period. Then, the high Tmin anomaly began to decrease to around zero during the hiatus, accompanied by the synchronous phase shift of the AMO. The statistical relationship between the AMO and Tmin indicates the AMO may have affected the decadal variation of Tmin and caused the decline during the hiatus.

Figure 9

The decadal components of standardized AMO and regional-mean (180–0°, 15–60°N) detrended Tmin during 1950–2014 extracted by the Ensemble Empirical Mode Decomposition (EEMD) method.

To further understand how the AMO affects Tmin of NA, and identify the physical processes involved, we examine the daily geopotential height and wind fields from the ERA-Interim data set. Because temperature reaches a minimum value at night, we first calculated the average of geopotential height and wind at 00:00 and 06:00 American Central Standard Time. Figure 10 shows the composite analysis of effects of the AMO on the geopotential height and wind vector anomalies at 500 hPa. There are two low-pressure systems in the northeastern Pacific and the North Atlantic when the AMO is in its positive phase (Figure 10a). Oceanic regions are warmer than land areas at night. Thus, cyclonic circulations from oceans and southerly warm advection transport warm air to land, causing an increase in Tmin. When the AMO is in its negative phase, low-pressure systems change to high-pressure systems and airflows bring cold air from the north, and Tmin decreases (Figure 10b). Therefore, the decadal component of AMO changed from the positive to negative phase during the hiatus, causing a decline on Tmin.

Figure 10

(a) Composite analysis results of geopotential height and wind vector anomalies at 500 hPa. Annual-mean geopotential height and wind vector anomalies for the years when the Atlantic Multidecadal Oscillation (AMO) is in the positive phase during 1982–2014. The red wind vector represents warm advection. (b) Same as (a) except the AMO is in the negative phase. The blue wind vector represents cold advection.

4 Summary and Discussion

In this study, we revealed that Tmin in NA experienced an obvious decline during the hiatus, while Tmax barely changed. We found that the decline of Tmin started in 1998, the starting year of the hiatus, and the turning point of Tmin tested by the MK method was also this year. The Tmin trends over most regions of NA convert into negative during the hiatus. The AMO played a key role in the natural decadal variation of Tmin in NA. According to composite analysis, the spatiotemporal variations of Tmin are closely related to the AMO. At night, there are two low-pressure (high-pressure) systems in the northeastern Pacific and the North Atlantic when the AMO is in positive (negative) phase, and warm (cold) advection in NA is majorly from south (north). The Tmin decline happened during the hiatus because the synchronous decrease of the AMO phase.

Our results not only showed that Tmin experienced an obvious decline in NA but also identified that the AMO is the key oceanic ICV signal that affected Tmin variation. Many authors have emphasized the importance of the tropical Pacific [El Niño–Southern Oscillation, PDO] in explaining the current warming slowdown in NA (England et al., 2014; Maher et al., 2014; Meehl et al., 2014). However, we found that the AMO influenced not just the natural decadal change of Tmin during the hiatus but also during a longer period from 1950 to 2014. This support the studies that have emphasized the key role of North Atlantic change on the global warming slowdown (Chen & Tung, 2014; Pasini et al., 2016), and the results presented this paper are to be considered as a suggestion for new studies in decadal variations of Tmin.

Acknowledgments

This work was jointly supported by the National Science Foundation of China (41722502, 41575006, 41521004, and 91637312), the China 111 project (B13045). CPC global daily temperature data provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their website at https://www.esrl.noaa.gov/psd/. We thank the Climate Research Unit at the University of East Anglia for producing and making available the long-term CRU TS V 4.01 temperature data set, and the data set is available at https://crudata.uea.ac.uk/cru/data/hrg/. The AMO, PDO, NAO, AO, and Nino 3.4 data sets are provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, from their website at https://www.esrl.noaa.gov/psd/data/climateindices/list/. The ERA-Interim daily geopotential height and wind data sets are available at http://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=pl/.