Journal of Geophysical Research: Atmospheres

Spatiotemporal changes in sunshine duration and cloud amount as well as their relationship in China during 1954–2005

Authors


Abstract

[1] Long-term trends in cloud amount and sunshine duration have been studied based upon surface observations at 618 meteorological stations across China. The degree of agreement between the two measures at interannual and decadal scales is analyzed, and a further understanding of the trends in sunshine duration is presented. A significant decreasing trend has been derived for sunshine duration (SSD) and total cloud cover (TCC); however, low-level cloud cover (LCC) shows an increasing trend, although it is not significant at the 95% level. Interannual variability of SSD is strongly inversely correlated to that of TCC and LCC, indicating short-term variability of SSD is dominantly determined by cloudiness. A positive correlation between decadal changes in SSD and TCC suggests long-term change in TCC cannot account for the decreasing trend in SSD. Long-term change in LCC appears to be one of important contributors to the trend in SSD in southern China, where long-term changes in SSD are inversely correlated to those of LCC. The decreasing trend in SSD is contributed by the declines in average SSDs under clear sky (13%), cloudy (51%), and overcast conditions (36%), 30% of which is offset by an increase in the frequency of clear sky.

1. Introduction

[2] The surface solar radiation (SSR), the sum of the direct and diffuse solar radiation incident on the surface, is an important factor in the context of climate change, the hydrological cycle, and agriculture. Observational and modeling studies suggest that SSR is not necessarily constant on decadal time scales. Long-term changes in the amount of solar radiation reflected and absorbed by the atmosphere have induced substantial decadal SSR changes that may be unnoticed over many years [Wild, 2009]. As revealed in a number of local and regional studies, SSR decreased (“dimming”) at the rate of 1 to 3 percent per decade between the 1950s and 1980s at continental sites [Liepert and Tegen, 2002; Stanhill and Cohen, 2001; Wild, 2009]. More recent observations show a recovery of SSR (“brightening”) in the last decade of the 20th century in some regions of the globe [Wild et al., 2009]. The causes of the decadal variation of SSR and its implication to climate change during the last fifty years are not fully understood. The most accepted cause is the changing atmospheric transparency. Cloudiness is the largest modulator of the transparency and therefore is a leading candidate. Indeed, some investigators have suggested that increases of the cloud optical depth and a shift from cloud-free to more cloudy skies were the major contributors to the decline in SSR [Liepert and Tegen, 2002]. The transparency is also impacted by changes in the mass and optical properties of aerosols, which is the most likely cause for clear-sky SSR trends. Hypotheses of an impact of aerosol loads on the SSR trends between the 1960s and 1980s are difficult to verify in a strict sense due to lack of direct information on historic changes in aerosol loads. However, analysis of high temporal resolution SSR data and aerosol optical depth data have shown that the switch from solar “dimming” to “brightening” during recent decades in Europe and the U.S. could be due to a reversal from increasing to decreasing anthropogenic aerosol emissions [Wild, 2005, 2009]. The emerging evidence for a widespread decadal change in SSR since the middle of the twentieth century leads to speculations that solar “dimming” and “brightening” have profoundly influenced global warming and the hydrological cycle [Liepert et al., 2004; Wild et al., 2007].

[3] The climate in China has changed at an unprecedented rate during the past few decades [Li et al., 2007]. One of most noted effects has been a significant decline in SSR over much of China, especially in eastern China where total SSR declined by more than 6% per decade [Che et al., 2005; Li and Zhou, 1998; Xia et al., 2006]. This is mostly because direct solar radiation decreased by 8.6% per decade from the 1960s to 1980s [Liang and Xia, 2005]. The decrease in SSR between the 1960s and 1980s is at odds with a general decreasing trend in annual mean cloud cover (1–3% per decade) [Kaiser, 2000] and rainy days (1–4% per decade) observed at most ground sites in central, eastern, and northeastern China [Liang and Xia, 2005] that is supported by an analysis using the more reliably observed frequencies of cloud-free sky and overcast sky [Qian et al., 2006]. Concurrent declines in total cloudiness and SSR before the 1990s indicate that total cloud cover changes should not be the major contributor to the SSR decline in China. Note that the dimming is consistent with the observed decline in visibility. Observations show that visibility was reduced by 35% from the 1960s to the 1980s in south China and frequencies of good visibility (visibility greater than 20 km) decreased by more than 20% per decade in eastern China [Che et al., 2007; Liang and Xia, 2005]. Using synoptic cloud and satellite retrievals of SSR under clear and cloudy skies, Norris et al. studied the impact of changing cloud cover on the SSR trends and pointed out that aerosols are major modulators of SSR in China [Norris and Wild, 2009].

[4] Surface measurements of SSR using pyranometers are limited in space and time. Hence, there is still a need for more measurements with higher spatial resolution and longer temporal resolution. For this purpose, the analysis should be supported and extended with the use of proxy measures of SSR. Analysis of these quantities by independent measurements can provide, at least to some extent, information on decadal variations in SSR and allow for consistency checks. Sunshine duration (SSD) is defined by the World Meteorology Organization as the time during which the direct solar irradiance exceeds 120 W m−2 [World Meteorological Organization (WMO), 2003]. The SSD measurements are some of the oldest and most robust data for use as a proxy of SSR. The advantage of SSD, in relation to other variables, is that it is less subjective than visibility or cloudiness observations. Few types of sunshine duration recording instruments have been used since the first measurements were made. Analysis of SSD data from over 200 sites between the 1950s and 1990s revealed significant decreases in SSD over much of China, especially in the eastern half of the country [Kaiser and Qian, 2002; Liang and Xia, 2005; Yang et al., 2009; Zheng et al., 2008]. However, a systematic analysis of the temporal evolution of SSD and its relation to concurrent changes in clouds, based upon a larger amount of data available from the 1950s to 2000s, is still lacking. Thus, the objective of this study is to present spatiotemporal variations in SSD between 1955 and 2005 using SSD data at 618 sites over China. Correlations between SSD and cloudiness have been studied to determine the degree of agreement between both quantities. Implications of long-term changes in cloudiness on trends in SSD are studied.

[5] This study differs from previous studies in several ways. First, spatial and temporal coverage of data are significantly enhanced and subtle SSD changes are thereby derived. Second, homogeneous tests of SSD and cloud data are carried out and regionalization of data is accomplished by means of an objective analysis. Third, analysis of SSD and total cloud cover (TCC) and low cloud cover (LCC) data is applied to determine the degree of agreement between both variables at different temporal resolution. Fourth, contributions of changes in the frequency of different sky conditions and their average SSDs to the total SSD trend are studied.

2. Data and Methodology

2.1. Sunshine Duration and Cloud Data

[6] The measurements of SSD and clouds used in this study were supplied by the China Meteorological Administration (CMA). Monthly SSD and TCC/LCC at more than 700 meteorological sites are archived and available from the Climate Data Center, China Meteorological Administration (CDC/CMA). The number of sites increases from 340 in 1954 gradually to 680 in about 1960 and then remains stable. The number of sites analyzed was limited to the 618 sites having a minimum of 30 complete years of measurements during the past fifty years. Detailed information on the instruments measuring SSD is lacking; however, most of these were Jordan recorders and the Campbell-Stokes sunshine recorder was used in a few sites (J. Wang, personal communication, 2009). As for the Jordan recorder, SSD is the amount of time, expressed to the nearest 0.1 h, in which solar radiation entering one of the slits falls on light-sensitive paper that lines the curved side of the semicylinder. The sensitivity of the Jordan recording paper is variable, and this introduces an uncertainty in the evaluation of the record. Thus, the reading of the card may differ from one observer to another. The Campbell-Stokes sunshine recorder consists essentially of a glass sphere set into a bowl with the sun burning a trace on the bowl. The errors of this recorder are mainly generated by the dependence on the temperature and humidity of the burn card as well as by the overburning effect, especially in the case of scattered clouds. Another big problem is in the reading of the cards and the reading of the card may also differ from one observer to another [WMO, 2003]. It was reported that the Jordan recorder was about 10% more sensitive than the Campbell-Stokes recorder [Noguchi, 1981]. However, no correction is applied to the data because information on instruments is not available. In addition, monthly anomalies are used in the analysis, which likely allowed the combination of data from stations with different SSD recorders. Quality assurance checks, including gross errors (e.g., SSD exceeds the maximum SSD) and the consistency of calendar dates, were performed by CDC/CMA. Ground-based observations of TCC and LCC are based on subjective estimates by experienced individuals at stations. Observations are carried out according to the recommendations of the WMO (World Meteorological Organization), but naturally there is also a subjective influence of the observer on these measurements, as well.

2.2. Regionalization

[7] Measurements must have 24–30 years worth of data to find the mean value of either cloud cover or insolation within 1% accuracy with 95% confidence [Hoyt, 1978]. These lengths of time are comparable to the periods of time on which climatic changes occur. One implication of these results is that one station alone is not adequate to determine climatic trends in cloud cover or in insolation. Rather, a network of stations of sufficient density to determine mean cloud cover or insolation is needed for climatic change studies. Therefore, the 618 stations are grouped by similar TCC variability and the regional averages of SSD and cloudiness are analyzed. The clustering of stations is achieved by means of a factor analysis with varimax rotation. The analysis is applied to the time series of monthly TCC normalized anomalies at sites where there are at least 40 years of observations. We select first nine factors which have eigenvalues greater than 1. Figure 1 presents the following nine identified subregions: (1) Northeast (NE), (2) North China (NC), (3) North China Plain (NP), (4) Middle/East (ME), (5) South China (SC), (6) Middle/South (MS), (7) Southwest (SW), (8) Tibetan Plateau (TP), and (9) Northwest (NW). Note that stations in NW are relatively sparse and the spatial distribution of stations is extremely asymmetrical. Therefore, result derived in this region is less representative than that obtained in eastern part of China. Among these, NE, NC, NP, and NW are further classified as northern China and the remaining regions are classified as southern China in sections 3, 6, and 7. Note that the regionalization using SSD and LCC data is similar as that based upon TCC time series. Therefore, regionalization results based upon TCC are used.

Figure 1.

Location of the 618 meteorological stations with sunshine duration and cloud observations and regionalization result based upon a factor analysis with varimax transformation.

[8] Prior to analysis, all data were converted to departures from normal (anomalies), with normals approximated by period means of cloud and SSD data. This allowed the combination of data from adjacent months into seasonal relationships, thus increasing the number of cases and the reliability of the statistically derived relationships. The monthly anomalies of TCC/LCC and SSD have been averaged by season (winter is defined as December, January, and February, spring as March, April, and May, summer as June, July, and August, and autumn as September, October, and November) and an annual average has been obtained from the average of the four seasons. Annual and seasonal deviations from the long-term mean have been determined for each station, and these deviations have been averaged for stations within each of the nine regions. The average deviation for mainland China was then obtained from an average of the regional deviations for the nine regions. Anomalies were used to avoid potential biases caused by few missing measurements.

[9] For compatibility so that the sunshine data can be plotted on the same diagram with the same ordinate, the cloudiness in tenths of sky cover has been changed to cloudiness in percent of sky cover.

2.3. Homogenization of SSD and Cloud Data

[10] Detection of climate change requires observational data sets that are not only of good quality, but also homogeneous through time. Homogeneity of the climate record means that observed climate variations are merely due to the behavior of the atmosphere. However, climate observations are often influenced by nonclimatic factors such as changes in station location, exposure of the observational site, instrumentation type, or measurement procedure, which could introduce abrupt or gradual changes in data records. These shifts should be corrected prior to data analysis. A great deal of effort has been made to develop methods to identify and remove nonclimatic inhomogeneities over the last decades. These methods are widely applied to create homogenous climate variables, such as temperature, precipitation, and pressure. However, homogeneity tests for SSD and cloud cover time series are still limited. These data series, in most cases, are generally considered to be homogenous based on general justifications that there have been no changes in instruments and that the different types of SSD recorders used have made no difference. Recent analysis of SSD over the Iberian Peninsula, however, has demonstrated that only 18 of 72 SSD series collected proved to be homogeneous [Sanchez-Lorenzo et al., 2007], which was likely due to changes in station location, observation environment, or observer.

[11] The RHtestV2 software package is used to detect, and adjust for, multiple shifts that could exist in a SSD or TCC/LCC series that may display first-order autoregressive errors. The penalized maximal t test [Wang, 2007] and the penalized maximal F test are embedded in a recursive testing algorithm [Wang, 2008a, 2008b]. The latest version of RHtestV2 has significant improvements, in that the lag 1 autocorrelation (if any) of the time series is empirically accounted for. RHtestV2 is first applied to each site's data to detect shifts (if possible) without using a reference series. The RHtestV2 package is then executed to detect and adjust shifts at those sites which have proved to be inhomogeneous by the above tests; however, at this point, a reference SSD series is used. The reference is the weighted average of those homogeneous series that are from the same subregion as that of the tested series, and the weights are the correlation coefficients between the series to be tested and those homogeneous series. The results show that 49, 11, and 28% of the total series proved homogeneous for TCC, LCC, and SSD data, respectively. Figure 2 presents the mean adjustment curve averaged over all stations. The national average of all adjustments to TCC and SSD raw series are within 1%. Note that the mean of adjustments applied to TCC and SSD series during 1980–2000 is characterized by values systematically lower than 1, but the mean adjustments during 1965–1980 is larger than 1 (for SSD) or close to 1 (for TCC). Note that the inhomogeneities of LCC are clearly not random. The inhomogeneities of LCC lead to a long-term trend of −1%/decade in the raw LCC series, half of which is contributed by inhomogeneities occurring during the first ten years. The inhomogeneities of SSD and TCC/LCC are more evident in regional series, which indicates that biased results are likely to be produced if the raw data series are used in estimating the long-term variability.

Figure 2.

Yearly ratios between the raw and the homogenized time series for TCC, LCC, and SSD.

3. Trends in Annual and Seasonal SSD and Cloudiness at Regional and National Scale

[12] A simple and robust estimator of trend is computed based on the nonparametric Kendall's rank correlation tau that is used to assess the temporal development of SSD and TCC/LCC. We have used the Kendall estimate instead of the least squares estimate because it is less sensitive to the nonnormality of the distribution and less affected by extreme values or outliers in the series. Both conditions can often be found in climatological time series. Figure 3 presents the annual and seasonal China series, together with their fits from the robust locally weighted regression algorithm “Lowess” for a better visualization of long-term and decadal variability [Makowski et al., 2009]. The overall trends of the series are shown in Table 1. Note that the scale for the cloudiness series has a reversed axis to show an agreement in interannual and decadal changes between cloudiness and SSD.

Figure 3.

Time series plots of average China anomalies of sunshine duration (SSD), total cloud cover (TCC), and low cloud cover (LCC), plotted together with the smoothed series using robust locally weighted regression algorithm “Lowess.”

Table 1. Trends for Annual and Seasonal Time Series of Total Cloud Cover, Low Cloud Cover, and Sunshine Duration Over 1955–2005 Perioda
 AnnualSpringSummerAutumnWinter
  • a

    TCC, total cloud cover; LCC, low cloud cover; SSD, sunshine duration. Bold values indicate trends with significance levels higher than 95%.

TCC1.21.41.11.2−1.1
LCC0.20.10.4−0.20.2
SSD1.7−0.72.41.71.9

[13] The majority of positive anomalies of TCCs and SSDs occur before 1980, and thereafter the annual TCCs and SSDs anomalies are rarely positive. Annual TCCs and SSDs have declined by −1.2% and −1.7% per decade, respectively. A strong negative trend of TCC and SSD has previously been reported for China from surface observations obtained from the CMA for the periods 1951–1996 [Kaiser, 2000] and 1955–2000 [Kaiser and Qian, 2002; Liang and Xia, 2005; Qian et al., 2006]. Our results confirm their finding for the period 1954–2005 based upon data from many more stations. LCC anomalies are much less than those for TCC, and so is the decadal trend of LCC. The annual LCC has increased by 0.2% per decade, which does support a decreasing trend in SSD to some extent. Analyses of changes in cloud cover and cloud types during 1971–1996 show that a previously reported decline in TCC over China appears to be largely attributable to changes in high and middle clouds and, in fact, stratocumulus (a cloud type substantially influencing sunshine recorder measurement) shows an overall positive trend [Endo and Yasunari, 2006; Warren and Eastman, 2007]. On a seasonal scale, the TCC time series also shows a decreasing trend, ranging from −1.4% per decade in spring to −1.1% per decade in the winter; however, seasonal LCC except in autumn tends to increase, although this is not significant at the 95% level. A significant decreasing trend is also observed for the seasonal SSD series, being 0.7% in spring (not significant), −2.4% in summer, −1.7% in autumn and −1.9% per decade in winter. Note that after a substantial decrease from the 1960s to 1980s, TCC seems to have leveled off since the mid-1990s, and furthermore LCC shows a significant increasing trend, which is in good agreement with the decrease in SSD.

[14] Figures 4 and 5a5d show the annual and seasonal TCC/LCC and SSD series for the nine regions. Spatial variations in the decadal SSD and cloud variability are evident. The overall trends of the TCC/LCC and SSD series are shown in Figure 6. The annual TCC time series in the nine regions presents a similar decreasing trend as in the national series. The strongest and most consistent evidence for a decreasing trend of TCC is seen in NE, NC, and NP regions, where the decreasing trends are −2.6%, 2.5%, and −1.6% per decade, respectively. The minimum trend of ∼−0.5% is observed in SC, MS, and SW regions and the trend is not significant at the 95% level. The regional LCC series, as compared to the TCC series, not only shows a smaller interannual variability but also shows a quite different decadal variability. The annual LCC series shows significant decreasing trends in NC, NP, and NW regions being −5.8%, −3.2%, and −5.7% per decade, respectively. In the remaining regions, however, an increasing trend of LCC is observed, although most trends are not significant at the 95% level. The annual regional SSD shows a significant and persistent decreasing trend except in the SW region, ranging from −0.7% per decade in TP to −3.7% per decade in the ME. In SW region, the SSD anomalies fluctuate and the overall trend is only −0.3% per decade. Note that the maximum decreasing SSD trends occur in ME, SC, and MS regions, where the trend exceeds −3.0% per decade, which is surprising as TCC shows the largest decreasing trend, but agrees with the fact that LCC shows the largest increasing trend there. Evidently, the TCC and LCC also show an increasing trend or level off after the mid-1990s in most regions, which is consistent with a decreasing trend of SSD.

Figure 4.

Time series plots of regional annual anomalies of sunshine duration (SSD), total cloud cover (TCC), and low cloud cover (LCC), plotted together with the smoothed series using robust locally weighted regression algorithm “Lowess.”

Figure 5a.

Same as Figure 4 but for seasonal anomalies of sunshine duration (SSD), total cloud cover (TCC), and low cloud cover (LCC) in spring.

Figure 5b.

Same as Figure 4 but for seasonal anomalies of sunshine duration (SSD), total cloud cover (TCC), and low cloud cover (LCC) in summer.

Figure 5c.

Same as Figure 4 but for seasonal anomalies of sunshine duration (SSD), total cloud cover (TCC), and low cloud cover (LCC) in autumn.

Figure 5d.

Same as Figure 4 but for seasonal anomalies of sunshine duration (SSD), total cloud cover (TCC), and low cloud cover (LCC) in winter.

Figure 6.

Trends in (left) total cloud cover (TCC), (middle) sunshine duration (SSD), and (right) low cloud cover (LCC) derived for nine regions. The solid circles represent that the trend is significant at the 95% level, and the open circles indicate trends that are not significant.

[15] The spring TCC series in northern China shows a significant and persistent decreasing trend (a decline of more than 2.0% per decade). In the ME, SC, and MS regions, TCC decreases during the first decade and then increases until the 1980s, and since then, TCC has shown a decreasing tendency. TCC anomalies fluctuate from 1955 to the mid-1990s in the SW and TP regions. TCC has increased in the SW since about 1990, leading to an overall positive trend of 0.2% per decade (not significant) there. Similar interannual and decadal variations of LCC as those in TCC are observed in ME, SC, MS, and SW regions; however, the decadal variations of LCC are quite different from those of TCC in other regions, especially in NE where LCC shows a significant increasing trend. All spring regional SSD series show a nonsignificant decreasing trend except for NE region, where there is a significantly decreasing SSD trend that is consistent with a significant increasing trend in LCC. The spring SSD series show an increasing trend since the mid-1990s in NP, ME, SC, and MS regions, which is in good agreement with a simultaneous decreasing trend of TCC and LCC.

[16] Summer TCC series show a decreasing trend, but with large fluctuations in the nine regions. The largest decreases (−2.1% and −1.7% per decade) occur in NE and NC regions. In the remaining regions, the trend of summer TCC is close to −1.0% per decade. LCC fluctuates following the pattern of TCC but shows an upward trend in such regions as ME, SC, SW, and MS regions. The summer SSD series show a decreasing trend in all regions. This is especially true in NP, ME, SC, and MS regions, where the trends are −3.3, −5.4, −2.5, and −4.9% per decade, respectively, in good agreement with the fact that LCC shows an increasing trend there.

[17] The largest and most significant decreases in autumn TCC occur in NE and NC regions where the trends exceed −2.0% per decade. In the other regions, autumn TCC decreases about −1.0% per decade, and these trends are not significant at the 95% level. LCC follows TCC in NE, NC, and NP regions, but LCC shows an increasing trend (about 1% per decade) in the ME, SC, SW regions. All SSD series show a decreasing trend and the largest decreases in SSD occur in ME and SC regions, where the trends exceed −3.0% per decade, which is likely due to an increase in LCC.

[18] The winter TCC series show a significant decreasing trend in NE, NC, and NP regions, and especially in NE where the trend is −3.5%, which is two to three times the rate of change in other regions. A decreasing but not significant trend is also observed in SW, TP, and NW regions where trends exceed −1.0% per decade. We do not see a clear TCC trend in ME, SC, and MS regions. Note that TCC shows an increasing trend since the end of the 1990s in all regions, especially in ME, SC, and MS. The LCC in ME, SC, and MS has increased by about 1%; however, in other regions, LCC also shows a decreasing trend. The largest decreasing trend of SSD (about −4.0% per decade) occurs in ME, SC, and MS, corresponding to an increasing trend of LCC there.

4. Correlation Between Cloud and SSD at Different Frequencies

[19] Given that clouds are one of most important factors modulating surface solar radiation and therefore SSD, SSD is expected to be inversely related to TCC and LCC. We can see this inverse relation between TCC/LCC and SSD in some cases, but as shown in Figures 4 and 5a5d, we can also see a quite different story emerges. In the first step of correlation calculations, the raw (i.e., nondetrended) anomaly time series of SSD and TCC/LCC are used and the results are presented in Figure 7 and Table 2 (for TCC/SSD) and Table 3 (for LCC/SSD). The correlation coefficients between SSD and TCC are unexpectedly small (spring: −0.45, summer: −0.29, autumn: −0.52, winter: −0.76), even 0.23 for the annual mean. The minimum correlation in summer is likely due to prevailing local convective clouds. These clouds contribute to the TCC observations but cannot influence SSD observations very effectively. We can explain why the maximum correlation occurs in winter using the lack of local convective clouds by the same token. These values are much less than those derived in the U.S. and Europe, where the correlation coefficients between SSD and TCC exceed 0.85 [Angell, 1990; Sanchez-Lorenzo et al., 2007]. Note that we observe good agreement of SSD variability with that of TCC on a short-term scale, and a very strong inverse relationship is expected on this time scale. However, this inverse relation is offset by a positive correlation on the long-term scale, leading to a relatively lower correlation between the raw SSD and TCC time series. In order to distinguish between the decadal and interannual agreement of the two measures, we detrended the raw time series and recalculated the correlations between the higher- (interannual variability) and the lower- (multiannual to decadal variability) frequency parts of each series. A fit was determined by the robust locally weighted regression algorithm “Lowess” to represent the long-term signal, which is subtracted from the raw series to derive the detrended residuals or high-frequency changes [Makowski et al., 2009]. The correlation coefficients for the detrended residuals in TCC and SSD are of the same magnitude as that obtained for other continents, namely spring: −0.85, summer: −0.88, autumn: −0.93, winter: −0.97, and −0.79 for the annual mean. In contrast, the correlations of the low-frequency (smoothed) time series are substantially different. Long-term behavior of SSD is positively related to that of TCC, namely: 0.25 in spring, 0.61 in summer, 0.18 in autumn, and 0.45 in winter. For the annual long-term SSD and TCC, the correlation coefficient even reaches 0.94. This is mostly due to a significantly positive correlation between low frequencies of both quantities during 1954–1990. We can see an inverse correlation during 1990–2005. Decadal changes have a substantial offsetting influence on high-frequency TCC and SSD series, causing a poor relationship between both raw measures.

Figure 7.

Correlation coefficients of (left) raw, (middle) low frequency, and (right) high frequency (top) between sunshine duration (SSD) and total cloud cover (TCC) and (bottom) between SSD and low cloud cover (LCC).

Table 2. Correlation Coefficients of Raw, High-Frequency, and Low-Frequency Time Series Between Sunshine Duration and Total Cloud Covera
 AnnualSpringSummerAutumnWinter
  • a

    Values are expressed in percentage per decade. Bold values indicate trends with significance levels higher than 95%.

Raw0.230.450.290.520.76
Long term0.940.250.610.180.45
Short term0.790.850.880.930.97
Table 3. Same as Table 2 but for Correlations Between Sunshine Duration and Low Cloud Cover
 AnnualSpringSummerAutumnWinter
Raw0.520.760.780.750.85
Long term0.420.830.530.44−0.23
Short term0.780.770.860.920.95

[20] Interannual variability of LCC is negatively correlated to that of SSD. However, unlike TCC, annual and seasonal low-frequency variations in the LCC series are also inversely correlated to those of SSD, which meets our expectations. The inverse correlation between low frequencies of both quantities is more pronounced during 1990–2005. The correlation coefficients are −0.42 for annual, −0.83 for spring, −0.53 for summer, −0.44 for autumn, and −0.23 for winter. The inverse correlation between the high- and low-frequency partitions of LCC and SSD results in an overall negative correlation for the raw series of LCC and SSD. The correlation coefficients are −0.52, −0.76, −0.78, −0.75 for seasonal series, and −0.85 for annual series.

[21] Figure 7 presents correlation coefficients obtained in the nine regions for raw, short-term, and long-term time series of TCC/LCC and SSD. The interannual variation of SSD is closely related to that of TCC and LCC, especially in ME, SC, and MS regions where the correlation coefficients exceed −0.80. This implies that short-term variability of SSD is dominantly determined by that of TCC/LCC. The low-frequency variability of the annual TCC series is positively correlated to that of SSD except in SW region. The correlation between the low-frequency signal of seasonal TCC and the SSD series is very complex. The seasonal low-frequency variability of TCC is always positively correlated to that of SSD in NC. We can see positive or negative correlations between seasonal low-frequency variability of TCC and SSD in other regions. The correlation coefficients between seasonal low-frequency variability of TCC and SSD, especially for the spring time series, are generally negative or close to zero in ME, SC, MS, and SW regions where SSD shows a larger decreasing trend. This indicates that decadal changes in TCC should exert some effects on SSD trends in these regions.

[22] The annual LCC series is positively correlated to annual SSD in NC, NP, and NW regions, and in other regions the decadal changes in LCC support the long-term decline in SSD. In ME, SC, MS, and SW regions, seasonal LCC is negatively correlated to that of SSD and the correlation coefficients exceed −0.7, indicating that SSD decline is closely associated with the LCC decadal change.

[23] The raw TCC and LCC are generally negatively correlated to SSD. This is mostly due to a highly negative correlation in the high-frequency portion of the time series, and furthermore this is also attributed to, at least partly, by a strong inverse correlation between decadal changes in TCC/LCC and SSD in some regions, such as ME, SC, MS, and SC.

5. Contributions of Changes in SSDs Under Different Sky Conditions to Overall Secular Trends in Total SSD Under All-Sky Conditions

[24] A brief picture of decadal change in SSD under all sky conditions has been presented in section 3; however, a rather fundamental question still needs further analysis. Specifically, how much of any SSD change under all-sky conditions is attributable to changes in the frequency of different sky conditions and how much contributed by their average SSDs. For example, increased SSD could be derived from simply having more days during the year with clear sky, as the average SSD of clear sky is generally larger than that of cloudy skies. Alternatively, one could also envision a situation where the number of clear sky days does not change, but the SSD increases as a result of an increase of the average SSD under some sky condition.

[25] Following the method introduced by Karl and Knight [Karl and Knight, 1998] to study contributions of precipitation frequency and intensity to secular trends of precipitation amount, the contribution of frequency of different sky conditions and their associated average SSD to the secular trends of total SSD under all-sky conditions has been studied. The days are classified into clear, cloudy, and overcast cases based upon daily TCC observations. If this daily TCC is between 0 and 2/10, the day is clear; between 8/10 and 10/10 overcast; and cloudy otherwise. The proportion of any trend in total SSD that is attributable to changes in sky category frequency versus changes in its average SSD is estimated. This is calculated for the frequency component by determining the average SSD per sky category (equation image) and the trend in the frequency of sky category (Tf). Then the change in SSD due to the trend in the frequency of sky category is simply defined by the product (equation image × Tf). For the average SSD component, the trend is directly calculated as a residual between the trend in total SSD and Tf. Since this analysis is based on daily data, we only selected the completely homogenous series in both SSD and TCC variables, and just those containing at least 30 years of data at daily resolution during the 1971–2000 period. These restrictions resulted in a selection of only 96 stations, which are well distributed in southeastern China. As shown in Figure 8, on an annual basis, total SSD under clear sky shows an increasing trend that is mainly due to an increasing trend in clear sky frequency [Qian et al., 2006]. Total SSD under all-sky conditions shows a decreasing trend of −2.2% per decade that appears to be largely attributable to a significant decreasing trend in total SSD under cloudy (−1.4% per decade) and overcast conditions (−1.4% per decade). Virtually the average SSDs under clear sky, cloudy, and overcast conditions show a statistically significant decreasing trend, namely: −0.4% for clear sky, −1.8% for cloudy, and −1.0% per decade for overcast, which in turn contribute to 13%, 51%, and 36%, respectively, of the observed decreasing trend in SSD. Nearly 30% of the decreasing trend in SSD due to the decline in average SSDs under different sky conditions is offset by an increase in the frequency of clear sky. On a seasonal basis, every season has a statistically significant decrease in the average SSD under clear sky; however, the frequency of days with clear sky shows an increasing trend, and therefore the contribution of SSD under clear skies to the overall secular trend in total SSD under all-sky conditions is marginal. The secular decreasing trend in seasonal SSD under all-sky conditions is mostly contributed to by a decreasing SSD trend under cloudy and overcast conditions. The latter is primarily due to a strong decrease in the average SSDs under these conditions and contributions by the frequency changes in cloudy and overcast skies are generally less than 10%.

Figure 8.

(top) SSD trends (unit is percent per decade) under all sky (cloud category 1), clear sky (cloud category 2, TCC < 2), cloudy sky (cloud category 3, 2 ≤ TCC < 8), and overcast sky (cloud category 4, TCC ≥ 8). Contribution to SSD trends (middle) by frequency and (bottom) by the average SSD under different sky conditions (unit is % per decade).

6. Discussion

[26] A significant decreasing trend of SSD in China is surprising if only a decreasing trend of TCC is considered. For example, SSD has decreased by more than 3% per decade in summer, autumn, and winter in southern China where TCC has also decreased by about 1% per decade. It is widely suggested, based upon the fact that both SSD and TCC show a decreasing trend, that global “dimming” occurring in China is likely attributable to aerosol loading rather than cloud cover. However, analysis of LCC in this paper has shown that LCC shows an increasing trend in southern China and the low-frequency variability of the LCC series is strongly inversely correlated to that of SSD, which implies that part of decline in SSD in southern China is likely associated with LCC changes. Therefore, a distinction between low, middle, and high cloudiness is at least required to discover the role of clouds in global “dimming.” Actually, cloud properties such as cloud cover, cloud type, cloud position relative to sun, cloud optical depth, etc., should be fully considered in order to get a better understanding of clouds' role in the secular trend in SSD.

[27] Another surprising fact is that global “dimming” in China is accompanied by more frequent cloud-free skies. This fact has led to us to speculate that increased air pollution may have produced a fog-like haze that reflected/absorbed radiation from the sun and resulted in less solar radiation reaching the surface, despite concurrent increasing trends in cloud-free sky over China [Qian et al., 2007]. Increases in cloud-free skies, as expected, contribute to increases in SSD but this is partly offset by a decrease in the average SSD under cloud-free sky conditions. Declines in SSD over China are primarily attributable to a significant decreasing trend in the average SSDs under cloudy and overcast conditions and changes in the frequency of cloudy and overcast skies contributes little. This leads us to speculate that an increase of aerosol loading does result in a decline in SSD not only under cloud-free conditions but also under cloudy conditions. Under cloudy conditions, increases in aerosol loading may lead to declines in surface solar radiation and therefore SSD due to direct scattering and absorption effects. Moreover, surface solar radiation and SSD may be influenced by aerosols via their indirect effect, i.e., aerosols, by acting as cloud condensation nuclei, may influence cloud microphysical and macrophysical properties. Further study of aerosol effects under cloudy conditions will not only deepen our understanding of aerosol indirect effects but also provides clues regarding global dimming in China. SSD measurements are also impacted by precipitation, dust, fog. Further studies on the long-term variability of these factors and their effects on SSD are also required.

7. Conclusions

[28] The spatial and temporal variability of SSD and TCC/LCC, as well as their correlation, based upon monthly homogenized data sets covering the period 1955–2005 from 618 stations across China has been examined in this paper. Contributions of the frequency of different sky cover scenes and their average SSDs to the overall SSD changes are revealed by an analysis of daily SSD and TCC data using a statistical method. The main conclusions are as follows.

[29] A significant decreasing trend exceeding −1.0% per decade has been derived for the annual and seasonal TCC series. However, LCC shows a slightly increasing trend. Annual SSD decreases of −1.7% per decade and seasonal decadal SSD trends were found (−0.7% in spring, −2.4% in summer, −1.7% in autumn, and −1.9% per decade in winter). After a substantial decrease from the 1960s to the 1980s, TCC seems to have leveled off since the mid-1990s, when LCC shows a significant increasing trend and SSD shows a decreasing tendency.

[30] The annual and seasonal TCC series show significant decreases of 2–3% sky cover per decade in northern China, and the TCC trend is within −1.0–0.2% in southern China. SSD in northern China shows decreases of up to 2% per decade; however, much larger decreases of SSD (2–5% per decade) have been found in southern China, which is in accordance with increases of 0.5–2% of LCC per decade there.

[31] A strong inverse correlation between SSD and TCC is derived for the shorter-term variability, indicating interannual variability of SSD is attributable to changes of TCC. However, for long-term variability of SSD is generally positively correlated to TCC. On the contrary, a strong inverse correlation between SSD and LCC, on both short-term and also long-term time scales, is derived in southern China. As a matter of fact, we observe an inverse correlation between low frequencies of SSD and TCC during some periods and seasons in southern China. These facts suggest that decadal changes in SSD in southern China are likely associated with changes in cloudiness.

[32] The secular decreasing trend in total SSD under all-sky conditions is mostly contributed to by the decreasing SSD trend under cloudy and overcast conditions. The latter is primarily due to a strong decrease in the average SSDs under these conditions, and contributions by the frequency changes in cloudy and overcast skies are generally less than 10%.

Acknowledgments

[33] The surface observation data of cloudiness and sunshine duration are obtained from the Climate Data Center, Chinese Meteorological Administration. The author appreciates Xiaolan L. Wang at the Climate Research Division, Atmospheric Science and Technology Directorate, Science and Technology Branch, Environment Canada, for providing the RHtestV2 algorithm. The research was supported by the National Basic Research Program of China (2009CB723904), the Knowledge Innovation Program of the Chinese Academy of Sciences (grant KZCX2-YW-QN201), and the National Science Foundation of China (40775009 and 40875084).