Observational studies have revealed that changes in the stratosphere and troposphere are directly or indirectly associated with solar variability attributed to the decadal solar cycle (Balachandran et al., 1999; van Loon and Shea, 1999; van Loon and Labitzke, 2000; Gleisner and Thejll, 2003; Crooks and Gray, 2005; Gray et al., 2005; Haigh et al., 2005; Kodera and Shibata, 2006; Matthes et al., 2006; IPCC, 2007; Rind et al., 2008; Meehl and Arblaster, 2009; Meehl et al., 2009; Frame and Gray, 2010 and many others). However, great uncertainty remains concerning the actual atmospheric response and controversial conclusions in the various studies (Fröhlich and Lean, 1998; Haigh, 2003; Willson and Mordvinov, 2003; Hood, 2004, Keckhut et al., 2005, Scafetta and West, 2005, 2006; Lean, 2006). For example, the Intergovernmental Panel on Climate Change (IPCC, 2007) report pointed out that because of the uncertainties and sensitivity drifts in the measurements, irradiance increases of over 0.04% during the 27-year period of the Active Cavity Radiometer Irradiance Monitor (ACRIM) datasets (Willson and Mordvinov, 2003) have raised questions about the observations. However, such a trend is not present in the Physikalisch-Meteorologisches Observatorium Davos (PMOD) composite (Fröhlich and Lean, 2004). The controversies centre around a small solar radiation change that minimally affects a relatively massive atmospheric system and two aspects of the datasets themselves.
One key aspect associated with the controversy is the different data sources. First, due to various combinations of the observation record used to represent the solar variability (Willson and Mordvinov, 2003; Fröhlich and Lean, 2004; Dewitte et al., 2005), there is disagreement over the solar forcing results (Lean, 2006; Scafetta and West, 2006). On the basis of the total solar irradiance (TSI) composite of the PMOD between solar cycles 21–23, Lean (2006) pointed out that the solar contribution to global warming would be negligible. However, Scafetta and West (2006) relied on the ACRIM TSI composite concluding that the sun contributed at least 10–30% of the 0.40 ± 0.04 K global surface warming. Second, the different atmospheric datasets and techniques used to compile and integrate the atmospheric data lead to results with different characteristics that have been questioned. The main atmospheric datasets used for current community studies include conventional surface and rawinsonde observations along with rocketsonde data (Dunkerton et al., 1998), lidar data (Keckhut et al., 2005), satellite data from the Stratospheric Sounding Unit (SSU) and Microwave Sounding Unit (MSU) instruments (Scaife et al., 2000; Keckhut et al., 2001; Hood, 2004, Gray et al., 2009), and model assimilated datasets (ERA-40 and National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis). We note that both assimilated datasets include the SSU/MSU assimilated observations since November 1978. Also, we note the NCEP/NCAR reanalysis-assimilated derived temperatures from the satellite data (Kalnay et al., 1996), whereas the ERA-40 reanalysis assimilated the satellite measured radiance data directly (Uppala et al., 2005). These differences in approach could affect trend analyses. The differences in the reanalysis model physics could generate dynamic differences in the trends and anomaly comparisons (Mo et al., 1995). In addition to diverse data sources in the reanalyses, varying length data source periods can contribute to even more differences. Previous studies have employed a variety of data sources and observational periods (Pawson and Fiorino, 1999; van Loon and Shea, 2000; Labitzke et al., 2002; Haigh, 2003; Crooks and Gray, 2005; Keckhut et al., 2005; Xu and Powell, 2010) that may have impacted their conclusions. This analysis possibly suffers from the same issues. As the science becomes more complicated, it is difficult to understand the effects from a variety of data sources combined with models which may have many physical interactions contributing to a single outcome. Part of the purpose of this study is to resolve the similarities and differences between two recognized reanalyses that use different models and a direct satellite measurement to help resolve what is real and possibly what is not in terms of solar forcing impacts.
Another factor contributing to the diverse solar impact conclusions is the different analysis techniques. Many studies have made use of correlation analysis (van Loon and Labitzke, 2000; Gleisner and Thejll, 2003, and many other authors) and regression analysis (White et al., 1997; Haigh, 2003; Crooks and Gray, 2005). These studies only analyse for a linear relationship between solar variability and atmospheric response during the entire study period. However, the climate is not a linear system (Scafetta and West, 2005, 2006), and the relationship is likely to change with time (Labitzke and van Loon, 1988). Many results from zonal mean analyses (Tourpali et al., 2003; Coughlin and Tung, 2004 and the others) ignored important characteristics of longitudinal heterogeneity. In addition, some studies employed composite mean differences between solar maximum and solar minimum (van Loon and Shea, 2000; Labitzke, et al., 2002) to explore the response of temperature changes to solar variability, but they have not been able to establish a statistically significant difference.
In this work, both composite and multiple linear regression analysis results for the lower stratospheric and the middle tropospheric temperature changes associated with the solar variability are reported for the two reanalyses datasets (NCEP/NCAR and ERA-40) and the retrieved MSU measurements (only in the period 1979–2002) for the two periods of 1979–2002 and 1958–1978. The data and analysis techniques are described in the following section. The linear trend of the temperature and their differences in the two periods are offered in Section 3. Section 4 presents the composite analysis about the impacts of solar variability on the lower stratospheric and the middle tropospheric temperatures. The multiple regression analyses for the temperature and ozone modulation are described in Section 5. Section 6 gives the final conclusion and discussion.
2. Data and methodology
2.1. NCEP/NCAR reanalysis
The monthly NCEP/NCAR reanalysis (Kalnay et al., 1996) with a 2.5° × 2.5° grid resolution is used for the two periods of 1979–2002 and 1958–1978. It should be noted that the reanalysis period 1958–1978 has no satellite data. The Television Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS) data, the MSU, the High Resolution Infrared Radiation Sounder (HIRS) and the SSU information were not available before the end of 1978. The Special Sensor Microwave/Imager (SSM/I) data was assimilated in this system from 1993. The reanalysis has 17 pressure levels that range from 10 hPa to the surface (1000 hPa).
The monthly ERA-40 reanalysis (Uppala, et al., 2005) is employed for the same periods as the NCEP/NCAR reanalysis datasets. The ERA-40 reanalysis data uses the integrated forecasting system (IFS) developed jointly by European Centre for Medium-Range Weather Forecasts (ECMWF) and Météo-France. Derived temperatures from the satellite data (Kalnay et al., 1996) were assimilated in the NCEP/NCAR reanalysis, whereas the satellite measured radiances were assimilated directly in the ERA-40 reanalysis. The reanalysis has 23 pressure levels that range from 1 hPa to the surface (1000 hPa).
2.3. Microwave sounding unit
The MSU monthly temperature dataset between the end of 1978 and 2006 was created from channels 2 and 4 brightness temperature measurements from the TIROS-N, National Oceanic and Atmospheric Administration (NOAA)-10, 11, 12, and 14 satellites (Zou, et al., 2008). The data were averaged over 2.5 × 2.5 latitude–longitude grids. In order to reduce the biases in the intersatellite MSU instruments, NESDIS scientists (Zou, et al., 2006, 2008) developed an intercalibration method based on the simultaneous nadir overpass (SNO) matchups. Due to orbital geometry, the SNO matchups are confined to the polar region where the brightness temperature range is slightly smaller than the global range. Nevertheless, the resulting calibration coefficients are applied globally to the entire life cycle of an MSU satellite.
Such intercalibration reduces intersatellite biases by an order of magnitude compared to pre-launch calibration and, thus, results in a well-merged time series for the MSU channels 2 and 4, which represent the deep layer temperature of the middle troposphere (∼600 hPa) and lower stratosphere (∼87 mb), respectively.
2.4. Ozone datasets
Two ozone datasets were used in this study; one is from the merged total ozone mapping spectrometer/solar backscatter ultraviolet (TOMS/SBUV) dataset from 1979 to 2008. Version 8 of the software merging TOMS/SBUV data was constructed by the TOMS science team at the NASA Goddard Space Flight Center and is available online at http://acdb-ext.gsfc.nasa.gov. The second ozone dataset is from the ERA-40 reanalysis dataset from 1958 through 2002.
2.5. Sunspot number and ultraviolet (UV) radio flux
The solar variability proxies used in this, and in most studies, are the sunspot number and the solar radio irradiance (the 10.7-cm radio flux). The sunspot number has historically been used to identify the solar activity level and observations extend back several hundred years. The solar radio irradiance spans the time period from 1947 to 2009 and can be found at http://www.ngdc.noaa.gov/stp/SOLAR. On the basis of previous studies that have tested different proxies (Keckhut et al., 1995), the 10.7-cm solar flux, which closely tracks the temporal behaviour of the ultraviolet (UV) changes on daily, monthly, and 11-year time scales (Gleisner and Thejll, 2003; Steinbrecht, et al., 2003; Crooks and Gray, 2005; Keckhut et al., 2005; Frame and Gray, 2010 and many others) is taken in our analysis to represent solar variability.
It is obvious that the two analysis periods have an important difference: the lack of satellite data before November 1978 and the use of assimilated satellite data after 1978. The MSU retrieved temperature data is available for the period 1978 through 2006. On the basis of the whole publicly available data from ERA-40 for 1958–2002, two data periods in this study were chosen: 1979–2002 and 1958–1978. All three datasets will be used in the 1979–2002 period; the two reanalysis datasets will be used in the earlier period without satellite data. The winter season was defined as December through February in this study. On the basis of these datasets, the composite analysis and multiple linear regression analysis will be used in this study.
2.6.1. Composite analysis
In order to select the solar maximum and the solar minimum years, we used the yearly variability of the sunspot number (Figure 1) which has a well-known correlation with solar irradiance, particularly in the UV portion of the spectrum. The solar variability (SV) is defined as follows: SV = (S − SC)/SC, where S is the sunspot number, SC is the sunspot mean during the period 1958–2002. We set a criterion to measure the intensity of the solar variation. The years with positive SV are defined as a solar maximum year and those with negative SV are defined as a solar minimum year for the purposes of this analysis.
The temperature anomaly (TA) is defined as TA = T − TC, where T is the original mean temperature in the December–February (DJF), TC is the climatological mean for the corresponding winter period over the entire period 1958–2002. Composites of the TAs were constructed with the linear trends (the term a1TRD in following Equation (1)) removed for various factors by applying a multiple linear regression analysis for the two data periods. The t-test analysis was employed to calculate the statistical significance of the temperature anomalies.
2.6.2. Multiple linear regression analysis
For a limited selection of atmospheric variables (Y), a multiple linear regression equation can be expressed as follows:
where a0 is the long-term mean for a certain variable, TRD is a linear trend and F10.7 represents the solar forcing quantified by the solar 10.7 cm radio flux. NINO3.4 is the El Nino index averaged sea-surface temperature in the equatorial Pacific (5°N–5°S, 170°W–120°W), quasi-biennial oscillation (QBO) is the equatorial zonal wind at 30 hpa from the Free University of Berlin (Labitzke, private communication, 2009), and AER is the global stratospheric aerosol optical depth (volcanoes), € is a residual error term. The coefficient a0, a1, a2, a3, a4 and a5 are determined by least squares regression. Note that each forcing term was normalized before the following calculation.
This global analysis will be completed in steps. First, a reference baseline will be established by reviewing the trends in the global temperatures and the resulting pattern. This baseline provides insight into the consistency of the datasets before performing any additional analysis. Next, a composite analysis is performed to establish the anomaly regions and the potential contributions of solar forcing on them. Finally, a multiple regression analysis that accounts for several forcing factors is performed and compared to the composite analysis. Once the impact regions have been established, an analysis of the ozone changes was performed as both a consistency check and to understand the role ozone may have contributed to the observed solar forcing.
3. Trend of global temperature
Before discussing the impacts of solar variability on temperature anomalies, it is important to first review the preliminary features of the DJF temperature trends in the two periods as a baseline. The trend was calculated by the term of a1TRD in Equation (1).
Because the measurement from MSU channel 4 represents the lower stratosphere layer temperature with peak at 87 hPa (Zou et al., 2008), the layer mean temperature between 70 and 100 hPa was chosen to represent the lower stratospheric temperature in the NCEP/NCAR and the ERA-40 reanalyses.
In 1979–2002, the temperature tended to decrease over most of the global areas except for the area north of 60°N latitude where warming consistently occurred (Figure 2(a)–(c)) in the three datasets. The basic pattern can be confirmed by visually comparing the analyses with each other. The largest warming with the rate of 2.1∼2.7 K/decade was identified over the high latitudes of the North American continent. However, it is worth noting that the temperatures tended to increase over the tropical eastern Pacific in the ERA-40 reanalysis, which is different from the other two datasets.
The temperature trend in the 1958–1978 period (Fig. 2(d) and (e)) differs from its counterparts in the 1979–2002 period in the two reanalyses. The largest warming is observed south of 60°S, the maximum rate was 1.5 K/decade. For the NCEP reanalysis (Figure 2(d)), the largest cooling rate of −1.8 K/decade is observed over the high latitudes of the Eurasian continents. In contrast, the cooling rate is only −0.9 K/decade over the same location in the ERA-40 reanalysis (Figure 2(e)). Note that a similar warming can be found over the tropical eastern Pacific in both reanalyses. The basic temperature patterns are confirmed in the polar regions of all three datasets and indicate the greatest positive change in the North Polar region during 1979–2002. In addition, the two reanalyses are similar for the 1958–1978 period showing warming in the southern hemisphere polar region—a reversal of the 1979–2002 pattern where the North Polar region showed warming.
Three of the four panels from the ERA-40 and NCEP-NCAR reanalyses indicate an equatorial warming over the eastern Pacific and is a region to compare during the analysis.
The retrieved temperature from the MSU channel 2 represents the layer temperature with a peak at 600 hPa (Zou et al., 2008); the mean temperature from 500 to 700 hPa is employed to represent the middle tropospheric temperature in the NCEP/NCAR and the ERA-40 reanalyses.
Compared to the stratospheric analysis, the tropospheric temperatures tended to increase in the two study periods. For the period 1979–2002 (Figure 3(a)–(c)), the main warming areas are found over the Eurasian continents, western Pacific, North American continents and the southern middle latitudes. The temperature over the tropical Indian Ocean and Pacific Ocean decreased in the two reanalyses (Figure 3(a) and (b)), but clear evidence is found that the MSU ch2 temperature tended to increase over these regions although its amplitude is small (Figure 3(c)). In addition, the temperature over the Antarctic shows a strong cooling trend in the NCEP/NCAR reanalysis, while the temperatures over some areas appear to have a warming trend in the ERA-40 reanalysis.
The main warming areas in 1958–1978 (Figure 3(d) and (e)) appeared over most of the southern hemisphere and the Arctic zone in the two reanalyses. The biggest temperature trend difference between 1979–2002 and 1958–1978 occurred in the northern middle-high latitudes, where the trend is dominated by negative values. The largest warming is observed over the southern high latitudes, but the pattern shows a significant difference in the NCEP and ERA-40 reanalyses. It is worth noting that a negative trend can be found over the northern high latitudes of the Eurasian continent, North Pacific Ocean and west coast of the North American continent in both periods.
Based on Table I, the global mean temperature tended to increase in the troposphere and decrease in the stratosphere between the two periods of 1958–1978 and 1979–2002. Both the stratospheric and tropospheric temperature trends in the two periods are similar in the two reanalysis datasets and can also be confirmed in the MSU measurements for 1979–2002 although there is a different temperature trend between the two reanalyses and the MSU measurements over the Antarctic zone and tropical eastern Pacific. However, the variability and temperature trend patterns also show significant differences between the two periods in the temperature structure observed over the tropical and middle-high latitudes.
Table 1. Trend of global mean temperature on DJF. Units: K/decade
It is clear from these analyses that the general global pattern trends have broad similarity with differences in the rates in both the stratosphere and the troposphere. In addition, there are clear regional discrepancies. To determine where the temperatures are most significant, and possibly correspond with solar forcing, the trends must be removed from the datasets.
4. Composite analysis
In order to identify the influence of solar forcing, the raw data was modified in two ways. First, because the satellite observations were introduced into the data assimilation system in late 1978, the reanalysis datasets have different trends between the periods as described in Section 3; hence, the linear trend was removed from the raw datasets based on the term a1TRA of Equation (1) in the separated period of 1958–1978 and 1979–2002, respectively. Second, three volcanic eruptions of Agung (March 1963), El Chichon (April 1982) and Mt. Pinatubo (June 1991) occurred during the study period. Salby and Shea (1991) pointed out that the volcanic eruptions could be problematic since any atmospheric behaviour attributed to solar variations may actually be due to volcanic eruptions. Hence, to avoid a significant influence from the volcanic eruptions, the data are omitted for 2 years following each eruption in the composite analysis. The ten solar maximum years (1979–1981, 1988–1990, 1999–2002) and nine solar minimum years (1984–1987, 1994–1998) are identified in 1979–2002; and the seven solar maximum years (1958–1960, 1967–1970) and 12 solar minimum years (1961–1962, 1965–1966, 1971–1977) are identified in 1958–1978.
4.1. Period of 1979–2002
The solar cycle categorized stratospheric TAs with volcanic signals and temperature trend removed in the 70–100 hPa layer provide insights into possible solar influence. For solar maximum activity, the global TAs have very similar patterns in the two reanalyses and the MSU measurements. Two large temperature anomalies can be found in the Arctic, one positive and one negative. Negative TAs can be found over the Arctic zone, central Asian continent, northeastern Pacific and southeast of the North American continent. The strongest TAs occur over the high latitudes; the amplitude increases to −1.0 °C and + 1.0 °C in the respective anomaly regions in the NCEP/NCAR reanalysis (Figure 4(a)) and is similar to the magnitude in the MSU measurements (Figure 4(c)). The TAs in the ERA-40 reanalysis (Figure 4(b)) are in the same range but only reach 0.8 °C. Positive TAs appear over the European continent, the tropical eastern Pacific and the North Atlantic. The magnitude of the MSU TA is consistent with that in both the NCEP/NCAR reanalysis and ERA-40. In addition, negative TAs occur over the northwestern and southeastern sides of the TC warm centre (Figure 4(a)–(c) vs Figure 2(a)–(c)). On the basis of these anomalies, the response of the TAs to SV should enhance the temperature gradients over the northern middle latitudes. Over all tropical areas, the response of the TAs to SV shows very similar patterns of positive TAs in both the reanalyses (Figure 4(a) and (b)) and MSU measurements (Figure 4(c)); the generally weaker amplitude in the MSU measurements is noted. Poleward of 30°S, alternating negative and positive TAs in a wavelike pattern can be observed over the middle latitudes in both the reanalyses and the MSU measurements.
Compared to the solar maximum temperature anomalies, the TAs for solar minimum activity have very different TA patterns. The TA distributions at solar minimum are similar in each of the three datasets except for the tropical eastern Pacific (Figure 4(d)–(f)). Poleward of 60°N, a strong positive TA with values above 1.6 °C is observed around the Arctic Circle. On the basis of the analysis, it appears that during solar minimum a consistent negative relationship between the TAs and SV can be found over the tropical/subtropical areas and a positive relationship exists in the North Polar regions.
Finally, the shaded areas with t value of 1.33 denote statistically significant changes above the 90% level in some areas, but the large stratospheric changes in the three datasets are not statistically significant. A potential mechanism responsible for the relationship between the stratospheric temperature and solar variability will be discussed in Section 5.
During solar minimum, the North Polar region shows consistent strong positive anomalies (up to + 1.6 °C) and the south polar region generally shows weaker negative anomalies (up to −0.8 °C). The tropics generally show a weak negative anomaly during solar minimum. During solar maximum, the polar region over North America, the North Pacific and northeastern Asia is replaced by a large negative TA. In addition, the eastern Pacific shows stronger and more consistent regions of positive anomalies during solar maximum. These large changes suggest a dynamic change may have occurred to generate these remarkable differences between solar minimum and maximum periods.
Compared to the stratospheric TAs in DJF of 1979–2002, the tropospheric TAs (Figure 5) have complex features. For solar maximum activity, the global TAs show similar patterns in each of the three datasets except for the Antarctic zone (Figure 5(a)–(c)). Negative TAs tend to occur over the high latitudes (50–80°N). whereas the positive TAs occur over the middle latitudes (20–50°N). The wavelike pattern alternating between positive and negative TAs appears over both northern and southern middle latitudes. However, a large difference is found over the Antarctic zone in the three datasets. The MSU and NCEP data show weak positive anomalies, whereas the ERA-40 shows moderate negative anomalies over most of the Antarctic polar region.
For solar minimum activity, the TAs are largely positive over the northern high latitudes and negative over the middle latitudes (Figure 5(d)–(f)). This suggests a relationship between TA and SV over the northern middle and high latitudes (Figure 5(a)–(c)). The most significant region of positive anomalies is found over North America, the North Atlantic and the eastern North Pacific. However, the relationship disappears over the tropical areas. It is worth noting that the anomalous amplitudes are very weak in the MSU measurements.
Generally, the tropospheric anomalies are disjointed and irregular in shape. The anomalies are stronger during solar minimum and weaker during solar maximum suggesting a stronger temperature gradient during the solar minimum periods. The overall pattern tends to shift poleward during solar minimum and equatorward during solar maximum. Another interesting feature is the nearly consistent replacement of positive (negative) anomalies during solar minimum with negative (positive) anomalies at solar maximum and vice versa.
During the 1979–2002 period, the stratospheric TAs show a significant response to the SV over the northern middle latitudes, the Arctic zone in the stratosphere and the middle latitudes in the troposphere, but the stratospheric TAs have more uniform values than those in the troposphere. The strongest TA under a value of −1 °C occurs over the Arctic zone in the stratosphere during solar maximum activity and a positive anomaly dominated the entire Arctic zone with maximum up to 1.6 °C during solar minimum periods. The tropospheric TAs show a stronger longitudinal heterogeneity with a wavelike pattern of alternating positive and negative TAs over the middle latitudes. The TAs show a large difference over the tropical areas and most of the southern high latitudes in the three datasets. The TA relationship with SV is not significant over the tropical latitudes and Antarctic zone. Most of the central areas of TAs satisfy the statistical significance test at the 90% level.
4.2. Period of 1958–1978
In order to examine the reproducibility of the relationship between the atmospheric TA and SV, another composite for the period of 1958–1978 was investigated. There are no available MSU measurements in this period, only the NCEP/NCAR and the ERA-40 reanalysis datasets will be employed in this section.
Comparing the solar cycle categorized temperature anomalies in the 1979–2002 period with those in the 1958–1978 period in the NCEP and ERA-40 reanalysis (Figure 4), common features can be found in 1958–1978 (Figure 6(a) and (b) vs Figure 4(a) and (b)). The stronger response of TAs to SV always appears in or near the Arctic zone, but it is worth pointing out that the signs of the anomalous temperature response to the SV are mostly opposite in the two periods for solar maximum activity. During 1979–2002, the anomalies are negative over North America and the North Pacific but are positive across the entire Arctic in the 1958–1978 for solar maximum. The positive anomalies get up to 3.6 °C and the amplitude is much higher than that in 1979–2002 (Figure 6(a) and (b)). In addition, the patterns of TAs are very similar in the NCEP/NCAR and the ERA-40 reanalyses except for a portion of the southern high latitudes. Negative TAs emerge over most of the tropical oceanic areas during solar maximum activity.
For solar minimum activity in 1958–1978, the negative relationship between TAs and the SV can be identified over all the northern areas above 60°N and more weakly in the tropical–subtropical areas (Figure 6(a) vs (c), Figure 6(b) vs (d)) in the two reanalyses. Significant negative anomalies with maximum value of −1.4 °C are observed over the Arctic zone. However, the significant TAs disappear over the Antarctic zone (Figure 6(c)) in the NCEP/NCAR reanalysis, and the TAs are very weak over the tropical oceanic areas in both reanalyses (Figure 6(c) and (d)).
For the troposphere, the strongest TAs (Figure 7(a) and (b)) are found over the high latitudes in both reanalysis datasets, but the positive TA centres appear over the northern border of the two great land masses in the solar maximum period. The wavelike pattern of alternating positive and negative TAs can be observed over the northern middle latitudes (30–60°N). However, the pattern of the TAs over the southern middle and high latitudes exhibits a large difference in the two datasets. The TAs are very weak over tropical areas.
For solar minimum activity (Figure 7(c) and (d)), the magnitude of TAs is reduced substantially although the negative TAs emerge over the northern high latitudes, where the TAs do not satisfy the statistical significance test at the 90% level. Slightly stronger TAs exists over both middle latitudes.
It is clear that a strong positive connection between the TAs and SV exists over the Arctic zone for both reanalysis datasets in 1958–1978. However, both the stratospheric and tropospheric TAs do not reproduce the pattern in the 1979–2002 period. In addition, the magnitude of TAs decreases substantially for solar minimum activity.
To summarize, common features were observed in the two periods. The DJF TA has a significant response to the SV over the northern middle latitudes and the Arctic zone in the stratosphere and the middle latitudes in the troposphere. The stratospheric TAs have more uniform values than that in the troposphere. The tropospheric TAs show a stronger longitudinal heterogeneity with a wavelike pattern of alternating positive and negative TAs over the middle latitudes. The strongest TA occurs over the Arctic zone in the stratosphere during solar maximum activity. Large differences over the tropical areas and most of the southern high latitudes were observed in the three datasets. However, both the stratospheric and tropospheric TAs do not reproduce the pattern in the two periods: a strong positive relationship exists over the high latitudes in 1958–1978, whereas a negative relationship exists in 1979–2002.
5. Multiple linear regression analysis
On the basis of above composite analysis, the lower stratospheric and middle tropospheric temperatures have a significant response to solar variability and these anomalies can be largely confirmed in each of the three datasets. The strongest signal responding to solar variability was observed in the stratospheric Arctic zone; however, the sign of TAs was opposite in the two periods. This result poses a question: Can the TA response to solar variability be confirmed? This section will address the question.
Generally, three methods are used to assess the impact of solar changes on the Earth, modelling, composite analysis and statistical analysis. For the composite analysis in the preceding sections, Figures 4–7 likely indicate nonlinear relationships in the stratosphere and troposphere. However, for a complicated climate system, the composite analysis cannot identify which sources are contributing to the observed TA signal. The influence of trends and volcanic eruptions can easily be found. Fortunately, the multiple linear regression methodology is a good way to identify the signal from potential multiple sources (Haigh, 2003). Although the regression methodology can only address linear relationships, it is a step to improve our understanding of the relative source contributions. In other words, the regression method will overcome to some degree the limitations in the composite analysis. However, one must keep in mind that the regression approach also cannot make attributions for cause and effect but only point to possible areas for investigate.
5.1. Solar signal in the temperature anomalies
The method for determining the linear correlation between the TA data and each parameter is similar. For brevity, only the solar cycle forcing (a2) in Equation (1) from the TA data will be discussed in detail.
For the stratosphere, the multiple linear regression analysis of temperature with the normalized 10.7-cm solar flux in 1979–2002 shows (Figure 8(a)–(c)) a strong negative regression coefficient (cooling) in the Arctic zone with positive values (warming) in the tropical latitudes and a portion of the southern middle-high latitudes in the two reanalyses and MSU channel 4 measurements. In contrast, the regressed TA in 1958–1978 (Figure 8(d) and (e)) has a large positive regression coefficient in the Arctic zone and negative value in the tropical latitudes. The strongest correlations occur over part of the Arctic zone and the Pacific; in these areas the correlations exceed the statistical significance test at the 90% level. Compared to the composite analysis in Figures 4 and 6, the regression analysis is highly consistent for the amplitude and location of anomalies. The opposite sign of the temperature response to solar variation in the two periods has been significantly reproduced in most areas, especially for the stronger signal in the Arctic zone.
For the troposphere, the regression analysis for 1979–2002 generally captured the temperature anomalies in the composite analysis (Figure 9(a)–(c) vs Figure.5(a)–(c)). But the amplitude of the temperature anomalies over the tropical areas in the regression analysis is lower than its counterpart in the composite analysis. For example, the significant anomaly appears over the tropical eastern Pacific in the composite analysis (Figure 5(a)–(c)), while there is only a very small regression coefficient in the regression analysis (Figure 9(a)–(c)). In 1958–1978, the temperature regression analysis corresponding to solar variability is not opposite to its counterpart in 1979–2002 over most areas (Figure 9(a) vs(d) and Figure 9(b) vs (e)). There are significant differences over the tropical and southern hemisphere between the composite analysis (Figure 7(a) and (b)) and the regression analysis (Figure 9(d) and (e)).
To summarize, both composite and regression analyses indicate the temperature responds to the solar variation in the stratosphere and troposphere quite differently, and the regression analysis is in good agreement with the composite analysis in the stratosphere. The anomaly reversal with the solar variation can be identified over most areas in the two periods. However, the regression analysis cannot significantly reproduce the solar signal in the composite analysis in the troposphere; this result implies that solar variability is closely related to lower stratospheric temperature, but it only partially explains the TA in the troposphere which likely can be attributed to the contribution from the solar cycle forcing.
5.2. Ozone response
The IPCC (2007) fourth report pointed out that the most likely mechanism for the influence of the solar cycle on climate change is considered to be some combination of direct forcing by changes in TSI and indirect effects of UV radiation on the stratosphere. A measurement study (Lean et al., 1997) showed that 10 to 20% of solar cycle irradiance changes occur in UV radiation, which is largely absorbed by stratospheric ozone. Haigh (1996), using a model that simulated elevated UV irradiance through direct increases in stratospheric ozone, showed some climate changes. Shindell et al. (1999) found that realistic solar UV changes from solar minimum to solar maximum in their GCM leads to northern hemisphere surface temperature increases of as much as 0.5 K. Rozanov et al. (2002), using the University of Illinois at Urbana-Champaign (UIUC) General Circulation Model (GCM) that includes ozone-related processes and realistic solar UV irradiances, found increases in ozone and significant temperature changes in the stratosphere related in part to dynamical changes. Austin et al. (2008), using new simulations of coupled chemistry climate models, examined the 11-year solar cycles in ozone and temperature and pointed out a secondary maximum in stratospheric tropical ozone, in agreement with satellite observations. These previous studies show that ozone is a critical factor to modulate the impacts of solar cycle through stratospheric temperature change.
Unfortunately, the set of ozone observations does not entirely cover the two study periods of 1979–2002 and 1958–1978, so a proxy ozone dataset from the ERA-40 reanalysis will be used in our analysis. The quality of the data is critical in order to draw final conclusions. A detailed investigation of the ozone data quality in the ERA-40 reanalysis dataset is clearly beyond the scope of this article. However, a simple comparison between the ERA-40 data and other independent measurements is necessary to have confidence in the analysis. Figures 10 and 11 indicate that there are no major inconsistencies in the two datasets between the ERA-40 reanalysis and TOMS/SBUV satellite observation for the period of 1979–2002, the latter data is one of the best proxy ozone datasets. The height-latitude sections of the ozone mixing ratio for the DJF mean for 1979–2002 show (Figure 10) that the strongest centre was observed in the 5–10 hPa layer over the south tropical region in both the ERA-40 reanalysis and TOMS/SBUV ozone datasets. The central value of the ozone mixing ratio in the ERA-40 is 0.4 ppmv (equivalent to approximately 4% of the maximum value) lower than that in the TOMS/SBUV observation. The time series of the global mean (65°S–65°N) total ozone demonstrates (Figure 11) that the ozone change in the ERA-40 reanalysis is highly correlated with the TOMS/SBUV observation in terms of annually averaged values except for short periods in the intervals of 1989–1991 and 1995–1998, where the correlation coefficient between the two datasets for the period of 1979–2002 reaches a peak of 0.797. This result confirmed the conclusion of the original data creator (Dethof and Holm, 2004) that the quality of ozone in the ERA-40 reanalysis compared well with independent observations in most periods, even during the pre-satellite years. A comparison with the long-term series of total ozone from ground-based stations over Europe in the period 1960–2000 with ERA-40 total ozone changes shows similar temporal and pattern changes (Peters et al., 2008). More recent comparison studies further concluded similar results (Oikonomou and O'Neill, 2006; Shi, et al., 2008) although there exists some bias in tropical and high latitudes. These studies indicate that the data should be representative of the ‘true’ ozone measurement. Therefore, the consistency of the ERA-40 reanalysis data compared with the observed ozone data is not a major issue for this analysis.
Figure 12 shows the anomalous total ozone of the ERA-40 reanalysis in the two periods. For the period 1979–2002, the magnitude of ozone declined considerably over the tropical and northern hemisphere. In contrast, the positive anomaly dominated over the tropical and northern hemisphere in 1958–1978.
The results cause us to ask a question, what is the anomaly due to? To address this question, the multiple linear regression technique will be used as shown in Equation (1) to separate various forcing effects in the Ozone dataset.
For the regression for the total ozone variability which includes the linear trend and four separate forcing terms, the results are shown in Figure 13. First, the evidence shows that the total ozone tends to decrease in the northern middle-high latitudes and increase in the tropics and most areas of the southern hemisphere in 1979–2002 (Figure 13(a)), while the opposite distribution is observed in the 1958–1978 period (Figure 13(b)). It is very clear that the distribution of the total ozone trend during the two periods (Figure 12(a) and (b)) is significantly dominated by the linear trend of ozone variation except for the tropical areas. When compared to the temperature trend depicted in Figure. 2, it can be seen that the ozone increase (decrease) corresponds to the temperature cooling (warming) in the lower stratosphere (Figure 13(a) vs 2(a), Figure 13(b) vs 2(b) ) except for the tropical eastern Pacific.
Second, the ozone response to the solar cycle forcing term represented by the F10.7 cm shows (Figure. 13(c) and (d)) an opposite pattern over most of the northern hemisphere during the two periods, and the solar cycle forcing makes a positive contribution except for the northern high latitudes in 19581978. Third, there is a stronger response to the El-Niño Southern Oscillation (ENSO) forcing over the northern high latitudes, but the pattern is almost opposite in the two periods (Figure 13(e) and (f)) except for the northeastern Pacific. In addition, an interesting result worth noting is that the ENSO forcing contributes negatively to the ozone variation over most of the tropical ocean areas in 1979–2002 and positively in 1958–1978 (Figures 13(e) vs 12(a), Figures 13(f) vs. 12(b)) although the response amplitude is very weak. Fourth, the ozone response to the QBO forcing in the two periods (Figure 13(g) and (h)) is positive in the tropical areas and negative in both middle-high latitudes except for a narrow Arctic zone in 1979–2002.
Fifth, the response to the volcanic eruptions represented by the stratospheric aerosol optical depth (Figure 13(i) and (j)) has a similar pattern in the two periods with positive contributions in the northern middle-high latitudes and negative contributions in the southern hemisphere except for the high latitudes of the southern Pacific in 1979–2002. Similar to the ENSO forcing, the volcanic eruptions make a weak negative contribution to the ozone variation over most of the tropical ocean areas in 1979–2002 and a positive contribution in 1958–1978 (Figures 13(i) vs 12(a), Figures 13(j) vs 12(b)). The aerosol forcing response over the tropical areas shows a pattern similar to the ENSO forcing with negative contributions in 1979–2002 and positive contributions in 1958–1978.
To summarize, the response of temperature to solar variability showed the strongest temperature anomalies occurring repeatedly over the Arctic, but its sign is negative in 1979–2002 and positive in 1958–1978 for solar maximum activity. The opposite sign observed in the two periods using composite analysis was generally confirmed by the linear regression analysis—although the regression analysis shows a somewhat different distribution from the composite analysis in the troposphere. However, it is worth noting that the solar response in the regression analysis is not statistically significant in the Arctic region (Figures 8 and 9). The different temperature responses to the solar variability in the two periods can be reasonably explained by the modulation of the ozone amount. Because of the declining ozone amount over the Arctic in 1979–2002 (Figures 12(a) and 13(a)), more UV heat led to atmospheric warming (Figure 2(a)–(c)). In contrast, the increased amount of ozone (Figures 12(b) and 13(b)) in 1958–1978 less UV heat led to atmospheric cooling (Figure 2(d) and (e)).
The linear trend of the total ozone amount (Figure 13(a) and (b)) reasonably explains the opposite distribution of the anomalous ozone over both middle-high latitudes in the two periods of 1979–2002 and 1958–1978 (Figure 12(a) and (b)). This corresponds to the linear trend of the lower stratospheric temperature (Figure 2(a) and (b)). The opposite response to solar cycle forcing (Figure 13(c) and (d)) and ENSO forcing (Figure 13(e) and (f)) is only observed over the northern middle-high latitudes. The response to QBO forcing shows a similar pattern in the two periods (Figure 13(g) and (h)) and the volcanic eruption forcing (Figure 13(i) and (j)) always produces a positive anomaly over the northern middle-high latitudes and negative anomaly over the southern middle latitudes in the two periods. In addition, ENSO volcanic eruption contributes positively to the opposing ozone anomalies over the tropical ocean areas in the two periods.
6. Conclusion and discussion
On the basis of three TA composites and multiple linear regression computed from two reanalysis (NCAR/NCEP and ERA40) datasets and one satellite-retrieved temperature (MSU) dataset, the temperature responses to solar forcing in winter (DJF) is compared between the pre- (1958–1978) and post (1979–2002) periods of satellite data assimilation. The results show the following:
1.The global mean temperature tended to increase in troposphere and decrease in stratosphere in the two periods of 1958–1978 and 1979–2002 (Figures 2 and 3). However, the change rate and patterns of the temperature trends show a significant difference between the two periods. A heterogeneous temperature structure is observed over the tropical and middle-high latitudes. Both stratospheric and tropospheric temperature trends in the two periods are similar in the two reanalysis datasets and were also confirmed in the MSU measurements for 1979–2002, although there are different temperature trend rates between the two reanalyses and the MSU measurements over the Antarctic zone and tropical eastern Pacific.
2.During the two periods of 1979–2002 and 1958–1978, common features were observed showing the most sensitive areas of DJF TAs to SV emerge over the Arctic, the northern high and middle latitudes and the tropical–subtropical eastern Pacific. The patterns of DJF TAs associated with the SV in the lower stratosphere (Figure 8) and the middle troposphere (Figure 9) was similar in each of the three datasets; the exception areas are located over the tropical oceans and most of the middle-high latitudes in the southern hemisphere. The response of TAs in DJF to SV has a substantial spatial heterogeneity. The stratospheric TAs have more uniform values, whereas the tropospheric TAs have a wavelike pattern alternating between positive and negative values over both northern and southern middle latitudes.
3.The sign of TAs in the stratosphere responding to solar variability was opposite in the two periods. The strongest TA occurs over the Arctic zone in the stratosphere with a value of under −1 °C in 1979–2002 (Figure. 4(a) and (c)), whereas the TA is over 3.6 °C in 1958–1978 (Figure 6(a) and (b)) for solar maximum activity. In contrast, the tropospheric temperature did not show opposite responses in the two periods (Figures 5 and 7).
4.Except for the possible impacts from the reanalysis datasets with and without the satellite data assimilation, the temperature responses to the solar variability in the two periods can be reasonably explained by the modulation of the ozone amount. Due to a decline of the total ozone amount over the Arctic in 1979–2002 (Figures 12(a) and 13(a)), the increased UV heat absorption by ozone supports warming (Figure 2(a)–(c)). When the amount of ozone increases in the stratosphere in 1958–1978 (Figures 12(b) and 13(b)), the decreased l UV heat absorption will reduce the temperature over the Arctic (Figure 2(d) and (c)). The tropospheric temperature anomalies are not significantly sensitive to the ozone change but may be influenced through dynamic interactions with the stratosphere.
5.The linear trend of the total ozone amount essentially reverses its global pattern between the two time periods of 1979–2002 and 1958–1978. In the period 1958–1978 (Figure 13(b)), the northern hemisphere is positively impacted and the southern hemisphere is negatively impacted; vice versa in the 1979–2002 period (Figure 13(a)). The opposing response to solar cycle forcing (Figure 13(c) and (d)) and ENSO forcing (Figure 13(e) and (f)) is observed over the northern middle-high latitudes. The response to the QBO forcing (Figure 13(g) and (h)) shows a similar pattern in the two periods and the volcanic eruption forcing (Figure 13(i) and (j)) always produces a positive anomaly over the northern middle-high latitudes and negative anomaly over the southern middle latitudes in the two periods. In addition, ENSO volcanic eruptions make a positive contribution to the opposite of ozone anomaly over the tropical ocean areas in the two periods.
There are a lot of studies using correlation, regression or composite mean difference (solar maximum - solar minimum) analysis to investigate the relationship between atmospheric temperature and solar variation. van Loon and Shea (2000) used the NCEP/NCAR reanalysis data for 1958–1998 and found the whole northern hemisphere has a positive correlation. Coughlin and Tung (2004) pointed out that the zonal mean warming was positively correlated with the solar cycle over most of the troposphere. Labitzke et al. (2002) showed a positive correlation between the tropospheric TAs and SV. An earlier correlation analysis (Labitzke and van Loon, 1988) claimed that the correlation between atmospheric changes and the solar cycle depends on the phase of the equatorial QBO in the stratosphere. In this study, the dataset was divided into two periods (1979–2002 and 1958–1978) based on the available satellite data assimilated in the two reanalysis datasets before and after 1979. Our study shows that the relationship between the temperature changes and the solar variation in 1979–2002 cannot be completely reproduced in 1958–1978 period. Also, the sign of stratospheric TAs responding to solar variability is opposite in most areas, especially in the polar zones. However, it is worth noting that the most sensitive areas of temperature change because of solar variability always appear over the polar zones and most of the tropical–subtropical eastern Pacific Ocean in the two separate periods. The different responses of temperature to the solar forcing in the two periods probably come from two aspects: (1) the introduction of satellite data and (2) the possibility of a climate regime shift. The introduction of satellite observations into the reanalysis data assimilation system in late 1978 may result in the discontinuities observed in these reanalysis dataset around 1978–1979 (Santer et al., 1999; Trenberth et al., 2001). Another possibility is the 1977/1978 climate regime shift, which has been identified in many previous studies (Nitta and Yamada 1989; Trenberth, 1990; Meehl et al., 2009 and many others), may have changed the global dynamics. Just due to the change of climate background, the relationship between the solar forcing and temperature variation has been changed (Powell and Xu, 2010).
In addition, a separate statistical analysis method, e.g. multiple linear regression analysis, confirmed that the stratospheric temperature anomalies are mainly attributed to solar variability, but it is significantly modulated by the change of ozone amount in the lower stratosphere. For the period 1979–2002, the magnitude of ozone declined considerably over the tropical and northern hemisphere. In contrast, a positive anomaly dominated the tropical and northern hemisphere in 1958–1978. The results are consistent with previous model and observational studies (Haigh, 1994, 1996; Shindell et al., 1999; Rozanov et al., 2002; Shine et al., 2003; Hood, 2004 and many others). Since absorption by ozone is the main heat source in the stratosphere, a decline of the ozone mixing ratio in the lower stratosphere in 1979–2002 will lead to decreased absorption of energy and cooling, especially over the north high latitudes. When the ozone amount is increased over the northern middle-high latitudes of the stratosphere in 1958–1978, the greater absorption by ozone will elevate the temperatures there. It is worth noting that the tropospheric temperature anomalies are not sensitive to the ozone change; this result confirms the results of previous studies that conclude that the tropospheric temperature is not considered to be associated with solar radiative direct forcing and that the changes are mainly from the dynamics of stratospheric circulation change (Ramaswamy et al., 2001; Gray et al., 2005; IPCC, 2007).
The NCEP/NCAR monthly reanalysis data were obtained from NOAA/CDC web site. The ERA-40 reanalysis data were obtained from the ECMWF web site and the solar sunspot number from the NOAA/NGDC web site. The authors thank these agencies for providing the data. Special thanks to Dr C. Zou from NOAA/NESDIS/STAR for many excellent discussions and the MSU temperature datasets that were provided. We also thank Dr Gray and the other reviewers for their valuable comments and suggestions.
This work was supported by the National Oceanic and Atmospheric Administration (NOAA), National Environmental Satellite, Data and Information Service (NESDIS) and Center for Satellite Applications and Research (STAR). The views, opinions and findings contained in this publication are those of the authors and should not be considered an official NOAA or U.S. Government position, policy or decision.