A new look at solar dimming and brightening in China

1 Introduction

As the primary energy source of our planet, solar radiation received at the Earth's surface has not been stable over time but underwent significant decadal variations, constituting a decrease until the late 1980s and an increase thereafter, popularly known as “global dimming and brightening” [Wild et al., 2005; Wild, 2009, 2012]. Consistent with the global trend of surface solar radiation (SSR), there is abundant evidence of the “from dimming to brightening” transition in the SSR trend in China during the past half century [Che et al., 2005; Shi et al., 2008; Norris and Wild, 2009; Ohmura, 2009; Tang et al., 2011]. However, three serious issues remain in the previously analyzed SSR trends in China.

The first issue is the difference in the estimated magnitude of solar dimming and brightening in China, which varies from −2.5 to −12 W m⁻² decade⁻¹ and from 0.4 to 4 W m⁻² decade⁻¹, respectively [Wang and Yang, 2014]. The highest dimming rate was proposed by Ohmura [2009] by using 66 stations with long-term and continuous observations from the China Meteorological Administration (CMA) for the period of 1960 to early 1990s. By contrast, the lowest dimming rate was given by Tang et al. [2011] using hybrid model-derived solar radiation from sunshine duration data over 459 CMA stations for 1961–1989. With the same method, Tang et al. [2011] also suggested the lowest and negligible brightening rate of SSR in China after 1990. The highest SSR brightening rate in China was estimated in the study of Norris and Wild [2009], based on 23 stations mainly distributed in the eastern part of China for 1990–2002. The inconsistency in the analyzed SSR trends in China highlights the importance of spatial and temporal coverage of the data set and data quality control and calls for a reevaluation. Furthermore, differences in the methods of constructing the composite time series discussed in the literature may also contribute to the discrepancy in the published solar radiation trends.

The second issue is the inconsistency in the estimates of the seasonal trends of SSR. Inferred from 30 CMA stations across China with less than 1 month of continuous missing data during 1961–2003, Wang et al. [2009] noted that the dimming rate is the strongest in the period from March to August, which accounts for 55–85% of the annual reduction. Based on 72 stations with SSR data that passed the tests of a set of quality assessment algorithms, Shi et al. [2008] made an opposite conclusion that the largest decrease in SSR in China was in winter, followed by autumn, summer, and spring for 1957–2000. Xia [2010] analyzed the seasonal SSR trends for 1991–2005 in four subregions and noted that the most obvious brightening was in southeastern China in the spring season. The overall seasonal SSR trends for the brightening phase in China remain thus unclear.

The third issue is a sharp increase (“jump”) evident in composite time series of the raw SSR data set for the period of 1990–1993. This jump in SSR can also be found in the analyzed SSR time series using long-term records/first-class stations [Che et al., 2005; Liang and Xia, 2005; Qian et al., 2006] and even quality-controlled SSR data [Shi et al., 2008; Yang et al., 2013]. The unusual jump might influence the examined change point and the trend magnitude. Sunshine duration (SD; a widely used proxy for SSR) was therefore used to derive SSR trends in China [Tang et al., 2011; Wang et al., 2015] to circumvent this problem. A trend obtained from the “good” SSR stations unaffected by this jump is still needed.

Therefore, this study aims to revisit solar diming and brightening in China at different temporal and spatial scales by using only stations without abnormal SSR jump during 1990–1993. First, stations which contributed to the jump of SSR during 1990–1993 will be picked out, and the reasons will be clarified. Then, after excluding the stations contributing to the SSR jump due to artificial factors, such as changes in stations or instruments, new temporal and spatial SSR trends in China will be given.

2 Station Selection Procedure

The raw CMA data set encompasses 130 SSR stations across latitudes 16°50′N–52°58′N, longitudes 75°59′E–130°18′E, and elevations 3–4507 m (Figure 1a). In the composite time series of the entire raw data set, there is a sudden jump of SSR at a rate of 5.6 W m⁻² yr⁻¹ for 1990–1993 (Figure 2a, situation A). This jump was not recorded by SD at the same 130 locations, which slightly decreased by 0.05 h d⁻¹ yr⁻¹ during the same period (Figure 2b, situation A). Two national-wide changes happened in SSR measurements in China during the jump period. The first is the replacement of the solar radiation instruments [Shi et al., 2008]. Before 1993, the Yanishevsky thermoelectric pyranometer was used to measure SSR. The SSR recorder was replaced by a Chinese-developed DFY-4 pyranometer after 1993. The second change is the update in the classification of solar radiation stations in China [Wang et al., 2015]. The measurements of net radiation, reflected radiation, and direct solar radiation on inclined surface were added after 1993; since then the CMA stations were reclassified into three classes according to the radiation variables they measured.

Besides the nation-wide changes in solar radiation measurements in China, there were 51 stations that were abandoned or established during the jump period (Figure 1a). Specifically, in 1990, five new stations were established. In 1991, 17 stations were abandoned while 1 new station was established. In 1992, 3 sites were removed while 15 new stations were installed. In 1993, 10 stations were established. After removing the 51 discontinuous stations, the rate of SSR increase during the “jump period” shrinks to 3.8 W m⁻² yr⁻¹ (Figure 2a, situation B). This indicates that the inclusion of abandoned/established stations in the composite time series can explain about one third of the SSR jump during the period of 1990–1993. As seen from Figure 1a, there are three pairs of collocated discontinuous stations in similar climatic and environmental conditions. The observational records in each pair of the collocated discontinuous stations are coherent and thus were combined to complete the data period.

In addition, there are six sites abandoned before 1990 and eight stations established after 1993. These 14 stations do not have available data for 1990–1993 (Figure 1a) and thus cannot contribute to the jump of SSR (Figure 2a, situation C). As shown in Figure 1a, 6 out of the 14 stations with null value for 1990–1993 collocate with the remaining stations. These six pairs of collocated stations were also combined in order to prolong the recording period of the six remaining stations.

After this, 68 solar radiation stations were left for analysis. The methods of the accumulated deviation curve and the Mann-Whitney U test were then used to identify the series with an abnormal jump during the period of 1990–1993 in the following two steps:

In the first step, the accumulated deviation curve was drawn for each of the remaining 68 stations to preliminarily detect a possible change point in the period of 1990–1993. The basic equation for the accumulated deviation curve [Mu et al., 2003] is determined as

(1)

where x_i is the observed annual time series of SSR (x₁, x₂, … x_n); is the mean of the sequence (x₁, x₂, … x_n); and X_i is the accumulated deviation for year i, i ∈ (1,2, …, n). n is the total length of the record (x₁, x₂, … x_n) and equals 57 for the studied sample period of 1957–2013. Possible change points x_t occur in the year when X_i reaches an extremum [Yu and Chen, 2009]. In this study, the stations with X_i reaching the minimum in the period of 1990–1993 were selected, which suggests a transition point from a decreasing to increasing trend.

In the second step, the significance of the change point detected in step 1 was examined by the Mann-Whitney U test (also known as the Wilcoxon-Mann-Whitney test or the Wilcoxon rank sum test). The change point x_t in the period of 1990–1993 detected in step 1 may divide the sequence (x₁, x₂, … x_n) into two subsequences of (x₁, x₂, … x_t) and (x_t + 1, x₂, … x_n). Suppose that subsequence (x₁, x₂, … x_t) has size n₁ and subsequence (x_t + 1, x₂, … x_n) has size n₂, then n₁ ∈ (34, 37) > n₂ ∈ (20, 23) ≥ 20. The sum of ranks for the smaller subsequence (x_t + 1, x₂, … x_n) can be obtained as W, which is approximately normally distributed. In this case, the standardized value Z is calculated as [DePuy et al., 2005]

(2)

When |Z| > Z_0.05/2 = 1.96, the change point x_t is regarded as significant [Zhou et al., 2011].

Seen from Figure 1b, 23 stations were detected with a change point in the period of 1990–1993, and one third of them are significant (p < 0.05). The average rate of the SSR jump in the 23 stations is 10.4 W m⁻² yr⁻¹ for 1990–1993, which is almost about 2 times the increase in the raw data set (compare situations A and E in Figure 2a). It can be noticed from Figure 2b that the SD trend of the 23 jump stations also slightly increased after 1990, indicating that part of the detected SSR jump was due to natural factors. By the method of the Pearson's correlation analysis, SSR variations in 7 out of the 23 jump stations were identified to be similarly recorded by SD (significant correlation >0.6 for 1985–1998). Here we extended the period of 1990–1993 by 10 years to check the correlation between SSD and SD in order to reduce random effects.

In the other 45 stations with normal variations for 1990–1993, the jump does not exist anymore and SSR only slightly increases by 0.7 W m⁻² yr⁻¹ over this period (Figure 2a, situation D). SD at the 45 stations shows a similar trend as SSR (Figure 2b). The annual variations in SSR and SD are not totally consistent (Figure 2), because the SSR and SD records have different missing periods. The effect of missing data still exists in the obtained time series of SSR especially for the periods apart from 1990 to 1993. A more reliable overall trend will be given in the next step by selecting the long-term records and filling the data gaps to derive a new estimate of solar dimming and brightening in China.

In the end, the 45 stations with normal variations for 1990–1993 and the 7 stations with a change point during 1990–1993 due to natural factors were further qualified to give the temporal and spatial SSR trends in China. The CMA solar radiation monitoring network started from 25 stations in 1957. The number of stations exceeded 70 only after 1961. Therefore, this study chose the year of 1961 as the start of the analyzed trends. From the 52 stations with normal/natural SSR variations, 47 out of them have yearly SSR records of more than 45 years (85% of whole period). These 47 stations, covering all provinces across the mainland of China except Tibet (Figure 4), were used to reanalyze the global dimming and brightening phenomena in China. Forty-five percent of the selected 47 stations have complete yearly records for the period of 1961–2013. Besides for each month, more than 65% of the selected stations have complete monthly SSR records during 1961–2013. To minimize the effect of missing data on the overall trends, the yearly/monthly data gaps were filled by (1) the average of the two adjacent years, in case of 1 missing data and (2) based on the linear regression relationships established for the adjacent 10 years with available data, in case of ≥2 missing data.

3 Solar Dimming and Brightening in China

3.1 Annual Trends

Figure 3a shows the annual SSR trend in China for 1961–2013. No SSR jump was detected by the methods of the accumulated deviation curve and the Mann-Whitney U test in the composite time series of the 47 selected solar radiation stations during 1990–1993. To identify the transition year into the brightening phase in China, the method of Mann-Kendall [Sneyers, 1975] was further applied. Similar with the result obtained from the methods of the accumulated deviation curve and the Mann-Whitney U test, no significant change point for solar brightening was detected by the method of Mann-Kendall. However, there was a transition from the continuous decreasing trend to a slight increase in the forward sequence of the Mann-Kendall-Sneyers test from 1989, indicating that the year 1989 can be used as the cutoff point for solar dimming and brightening phases in China. Similarly, using quality-controlled SSR data, Shi et al. [2008] and Tang et al. [2011] also identified the year 1989 as the cutoff point to examine the solar dimming and brightening trends in China. Besides, the detected transition year of 1989 in the SSR trend is consistent with that noted in the SD trends in China [Wang et al., 2013; Wang and Yang, 2014]. Therefore, the solar dimming and brightening phases in China were defined as 1961–1989 and 1990–2013, respectively.

In the dimming phase, according to our analysis SSR significantly declined by a rate of −8.3 W m⁻² decade⁻¹ in China (Figure 3a). The dimming rate is thus near the middle of the range from −2.5 to −12 W m⁻² decade⁻¹ estimated in previous studies. As seen from Figure 4a, 44 out of the 47 stations show a decreasing SSR trend during 1961–1989, indicating a nationwide phenomenon of solar dimming in China. Around 82% of the decreasing SSR trends are significant at the 95% confidence level, and 43% of them are stronger than −10 W m⁻² decade⁻¹. The strongest dimming trends are mainly distributed in southeastern China, especially in the Yangtze River basin. Similar conclusions were drawn in studies of Liang and Xia [2005], Ohmura [2006], Xia [2010], and Wild and Schmucki [2011].

The noted abnormal SSR jump mainly influences the brightening trends. To avoid its effect, previous studies used the methods of ignoring the jump period of 1990–1993 [Shen and Wang, 2011] or using the proxy of SD [Tang et al., 2011] to derive the SSR trend in China for the brightening phase. By identifying and excluding the stations with abnormal jumps in the SSR trends during 1990–1993, the present study found that SSR still shows a significant increasing trend for 1990–2013 with a rate of 2.1 W m⁻² decade⁻¹ (Figure 3a). Our estimated brightening rate is also near the middle of the range from 0.4 to 4 W m⁻² decade⁻¹ estimated in the previous studies. Therefore, rather than a transition “from dimming to leveling off” as indicated in SD trends [Wang and Yang, 2014], it is further verified that the SSR trend in China shows a transition “from dimming to brightening” during the past decades, consistent with the global SSR trend. It can be noted from Figure 4b that increasing SSR trends cover about 68% of the studied stations, and 41% of the increasing SSR trends are significant (p < 0.05). Solar brightening mainly occurs in south China, while dimming continues in north China. This is consistent with the findings of Xia [2010] and Wang et al. [2011]. A similar spatial pattern for the brightening phase was also noted in SD [Wang and Yang, 2014].

A new picture of solar dimming and brightening in China was shown in this study by eliminating stations containing discontinuous records and artifacts. However, it is still limited by the spatial representation of solar radiation measurements in China. Besides, most of the long-term SSR records are from city-scale stations. Taking into account potential urbanization effects, an overestimation of solar dimming in China might exist. In the brightening phase when pollution regulations become effective, urbanization effects on SSR and SD become insignificant [Wang et al., 2016]. There remain also potential issues with long-term drifts in the diffuse component of the SSR measurements [Wang et al., 2015].

4 Conclusions

Focusing on the main issue of a sudden upward jump in the published composite time series of SSR during 1990–1993, this study clarified that about one third of the magnitude of the SSR jump were accidentally caused by 51 stations (~39% of total) abandoned or newly established during 1990–1993. The remaining two thirds of the SSR jump were only caused by 23 stations detected by the methods of the accumulated deviation curve and the Mann-Whitney U test. SSR variations in 7 out of the 23 stations, which account for about 30% of the remaining jump, were similarly recorded by SD and thus should be due to natural factors. The other 70% were caused by the other 16 stations as a result of instrument replacement, changes in the classification or location of stations, or potential operational errors.

By excluding the above-mentioned stations that contributed to the SSR jump due to artificial factors, as well as the stations with SSR records containing less than 45 years during 1961–2013, a new estimate for the magnitude of solar dimming and brightening has been derived for China. A transition from dimming to brightening still exists in the SSR trend in China, with a significant decrease by a rate of −8.3 W m⁻² decade⁻¹ for 1961–1989 and thereafter a significant increase by 2.1 W m⁻² decade⁻¹ for 1989–2013. In the dimming phase, SSR declines occurred all over China, especially in the southeastern part, while in the brightening phase, increasing SSR trends cover about two thirds of China especially in south China. Continued decreasing trends can still be noted in north China.

The seasonal trends of solar dimming and brightening in China were further clarified. The strongest solar dimming occurred in summer (−11.4 W m⁻² decade⁻¹), followed by spring (−8.6 W m⁻² decade⁻¹), winter (−7.2 W m⁻² decade⁻¹), and autumn (−6.4 W m⁻² decade⁻¹) during 1961–1989. A nationwide solar dimming can be noted for all seasons, while solar brightening is only significant in spring, with a rate of 5.5 W m⁻² decade⁻¹ for 1989–2013. In the other seasons, SSR slightly increased by 1.9 W m⁻² decade⁻¹ and 1.5 W m⁻² decade⁻¹, respectively, in winter and summer and leveled off by 0.1 W m⁻² decade⁻¹ in autumn. Significant brightening trends were mainly found in the Yangtze River basin in spring and the southwestern and southeastern China in the other seasons.