Increase of surface solar irradiance across East China related to changes in aerosol properties during the past decade

The surface solar radiation data come from three public sources, namely the Baseline Surface Radiation Network (BSRN, Ohmura et al 1998) at the World Radiation Monitoring Center (http://bsrn.awi.de/), the Global Energy Balance Archive (GEBA) at ETH Zurich (Gilgen and Ohmura 1999, www.geba.ethz.ch/), and the World Radiation Data Center (WRDC, http://wrdc.mgo.rssi.ru/wrdc_en_new.htm) at the Voeikov Main Geophysical Observatory. BSRN provides global, direct and diffuse radiation, WRDC reports global radiation and diffuse radiation for most sites, while for GEBA, only global radiation is available for the 2005–2015 period. With respect to BSRN measurements, direct radiation measurement uncertainty is estimated to be 1 W m⁻², whereas diffuse radiation has ~5 W m⁻² uncertainty (Ohmura et al 1998). The measurement accuracy of GEBA global radiation was estimated to be on the order of 2% (Gilgen and Ohmura 1999). Moreover, all three datasets have undergone rigorous quality control to ensure both high quality and homogeneity. Over the study period, there is one BSRN station (XiangHe), seven GEBA stations and nine WRDC stations that have more than ten years of continuous data records. Note there is a substantial overlap between GEBA and WRDC stations. The name and location of the stations used are listed in supplementary table S1 available at stacks.iop.org/ERL/13/034006/mmedia. The BSRN data is reported as 1 min observations, GEBA in monthly means and WRDC in daily sums. Both BSRN and GEBA report solar irradiance in W m⁻². WRDC collects integrated solar irradiance in J cm⁻², which has been converted to daily means in W m⁻² in this study.

We examine the two major factors affecting surface solar radiation—aerosols and clouds, and the data come from both satellites and ground based measurements for more robust analysis. Satellite aerosol products including aerosol optical depth (AOD) from the Moderate Resolution Imaging Spectroradiometer (MODIS, Collection 6, Levy et al 2013, https://ladsweb.modaps.eosdis.nasa.gov) onboard the EOS-Aqua platform, and from the Multi-angle Imaging Spectroradiometer (MISR, Martonchik et al 1998, https://eosweb.larc.nasa.gov/project/misr/misr_table) onboard the EOS-Terra platform. Level 3 gridded monthly means are used. The dataset names are 'MYD04_M3' and 'MIL3MAE' for MODIS and MISR respectively. The MODIS AOD accuracy over land is ±0.05 + 15% AERONET AOD (Levy et al 2013), and that for MISR AOD is estimated to be better than 0.05 (Kahn et al 2007). Ground based aerosol observations are provided by the Aerosol Robotic Network (AERONET, Holben et al 1998, available at https://aeronet.gsfc.nasa.gov/). This network has more than 400 stations worldwide, performing routine measurements and retrievals of major aerosol optical properties. In mainland China, only limited sites have a sufficient record for trend analysis and we only use data from the XiangHe site, which is collocated with the BSRN site. This study uses Level 2.0 direct beam AOD retrieval at 440 nm and SSA and asymmetry parameter from the inversion algorithm (Dubovik et al 2000), with accuracy estimates of ±0.02 for AOD (Eck et al 1999) and ±0.03 for SSA (Dubovik et al 2000)

The satellite cloud fraction and microphysics data are from the Aqua-MODIS Level 3 monthly joint atmosphere product (MYD04_M3), https://ladsweb.modaps.eosdis.nasa.gov/search/order/1/MYD08_M3--6. The MODIS cloud detection products have been validated against both ground based and other space-born sensors by Ackerman et al (2008). Ground based, gridded cloud cover observation from the Climate Research Unit (CRU, v4.00, Harris et al 2014, https://crudata.uea.ac.uk/cru/data/hrg/) is also used for corroboration. Comparison with other ground and satellite based datasets indicates reasonable agreement (Harris et al 2014).

Note that although only MODIS data on the Aqua platform is used, we compared the results with those from Terra and they are quite similar (comparing figure S1 and figures 2 (a) and (c), also see figure S5).

Linear trends of the radiation, aerosol and cloud time series and their significance levels are estimated using Sen's slope method (Sen 1968) and the Mann–Kendall test (Mann 1945, Kendall 1975) respectively. Sen's slope trend is calculated as the median of the slope between all data pairs in the time series. The Mann–Kendall test is a non-parametric trend test method. Before calculating the trend, the multi-year averaged seasonal cycle is removed and the time series is pre-whitened, in order to reduce autocorrelation. The pre-whitening process follows that suggested by Yue et al (2002), which first removes the linear trend from the time series as:

$$\begin{equation} X_t^\prime = {X_t} - {bt} \end{equation}$$

where X_t is the original time series and b is the linear trend slope. Then the AR(1) process is removed from X_t' as:

$$\begin{equation} Y_t^\prime = X_t^\prime - {r_1}X_{t - 1}^\prime \end{equation}$$

where r₁ is the lag-1 autocorrelation. Finally, the linear trend is added back using:

$$\begin{equation} {Y_t} = Y_t^\prime + {bt} \end{equation}$$

A more detailed description and analysis procedure can be found in Li et al (2014, 2016).

In order to verify that the changes in surface radiation are associated with changes in aerosol properties, we use the RRTM (Mlawer et al 1997, Clough et al 2005) by Atmospheric and Environmental Research. Only the shortwave component (RRTM_SW) is used. This model calculates shortwave fluxes and heating rates in 14 spectral bands using the correlated k-distribution method and the discrete ordinate radiative transfer integration of the radiative transfer equation (Stamnes et al 1988). The model agrees well with line by line calculations, with 1 W m⁻² accuracy for direct irradiance and 2 W m⁻² for the diffuse. This model has both high efficient and accuracy and has been widely used for aerosol forcing calculations (e.g. Guan et al 2010, Lin et al 2016).

Maps showing decadal trends of surface global and diffuse solar irradiance are presented in figure 1, and the time series and linear trends for all stations are displayed in supplementary figures S2 and S3. The GEBA and WRDC global solar irradiance results are essentially the same (figure not shown) as they come from the same data sources, only with slightly different quality assurance. Therefore only WRDC trends are shown in the maps. From figure 1(a), it is clearly seen that except for the northwest site of Urumqi, almost all East China sites exhibit significant (>90%, P-value < 0.1) increasing trends in global solar irradiance over the past decade. The strongest trends are found over Northeast China (Harbin and Shenyang) and South China (Kunming and Guangzhou), reaching 10 W m⁻²/decade. Trends at XiangHe, Wuhan and Wenjiang are also above 5 W m⁻²/decade. The Shanghai site shows slightly negative but insignificant trends.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For the diffuse radiation (figure 1(b)), the results appear more variable, with sites showing both positive and negative trends. Diffuse radiation is not only affected by the amount of scattering agents, but also by their optical properties such as SSA and phase function. Its changes can be more related to aerosol compositional changes and will be discussed in the next sections.

The trends for each season are also examined and displayed in supplementary figure S4. Overall, the seasonal trends are consistent with annual trends, but some variability can still be seen. For the North China sites (Harbin, Shenyang, XiangHe and Beijing), the largest trends are found during the spring (March, April, May) and summer (June, July, August) seasons, whereas winter (December, January, February) and fall (September, October, November) show weak positive or insignificant trends. For South and central China sites, the seasonal differences are minimal.

Among the various factors influence the solar radiation incident on the Earth's surface, aerosols and clouds are two important and highly variable components. To help understand the observed radiation trends, we further examine the changes of AOD and cloud amount.

Figure 2 shows the decadal trends of cloud fraction from MODIS (figure 2(a)) and CRU datasets (figure 2(b)). Although there are some regional signals and differences, both datasets indicate insignificant trends over East China where solar radiation appears to have significant changes. Additionally, cloud effective radius and cloud optical depth also show mostly insignificant trends (figure S5). These results indicate that clouds should not be the main reason for the surface solar radiation anomalies over East China.

The trends for aerosols, however, appear different and more promising. The linear decadal trends from both MODIS and MISR AOD are estimated in figures 2(c) and (d). These two instruments have differences in terms of viewing geometry, sampling frequency and retrieving algorithms, which would inevitably result in differences in the magnitudes and the trends. Nonetheless, they consistently indicate a downward trend over most of East China, with stronger decreases over the southern parts. Because aerosols reduce incoming solar radiation, its reduction will then result in an increase of surface solar radiation, which agrees well with the radiation trends observed in the section 3.1. The radiation and AOD trends are most coherent over South China. Over the very northeast region (i.e. Heilongjiang province), the MODIS and MISR trends are inconsistent. This region is typically covered by snow for ~5 months per year, which results in fewer sampling and higher AOD retrieval uncertainty due to the high surface albedo. However, Li et al (2016) indicated a decline of surface extinction there deduced from visibility, which appears consistent with the radiation trends. Moreover, seasonal AOD trends (supplementary figure S6) indicate the strongest negative trends during the spring and summer over North China, and weaker or even positive trends during the fall and winter, both in line with the seasonal solar radiation trends.

While surface solar radiation and AOD show largely coherent trends, there is still a need to examine simultaneous radiation and aerosol measurements in order to clarify this effect. Unfortunately, there is only one site—the XiangHe site—where collocated, high temporal resolution radiation and aerosol measurements are available in China. This site has 1 min global, direct and diffuse solar radiation (by BSRN) as well as routine aerosol optical property retrievals (by AERONET). Situated in a suburb of the highly polluted city of Beijing, the aerosol properties at the site were shown to be representative of the North China Plain (Fan et al 2015).

Compared to GEBA and WRDC which only include monthly or daily data, the one-minute BSRN data allows the separation of clear from cloudy skies and the isolation of aerosol effects. For this purpose we follow the procedure introduced by Xia et al (2007). Briefly, the clear sky global surface radiation can be modeled by a power law function of the cosine of the solar zenith angle. Then the observed global surface radiation is compared with the modeled value and cloud conditions are screened using a two-step process: (1) if the absolute relative deviation between observation and model calculation is greater than 20%; (2) if the difference of the running standard deviation during each 30 min period between normalized observation and model is greater than 0.02. Figure 3 shows the surface global irradiance for all sky (figure 3(a)), clear sky (figure 3(b)) and cloudy sky (figure 3(c)). All sky and clear sky irradiances both show significant positive trends, with the clear sky trend being stronger. In contrast, cloudy sky shows a negative trend. This is a clear indication that clear sky trends are the driver of all sky trends and that aerosols are likely the primary factor.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We then match BSRN and AERONET observations to further examine the aerosol effect. Because the effect of aerosols on surface radiation highly depends on the solar zenith angle that varies significantly during the course of a day, it is not fair to average over all collocated observations since AERONET only samples under certain conditions, i.e. the sampling for different sun angles may be different and results in a bias in the averaged result. We therefore respectively calculate the decadal trends for each hour of the day to roughly overcome this sun angle dependency. Although this is not a precise solution, the interannual variability of solar zenith angles during the study period is found to be within 0.08°. Figure 4 shows the trends for global (a), direct (b) and diffuse (c) irradiances for each hour during 8:00 am to 6:00 pm local time. We can see that overall global radiation shows an upward trend for all hours, with an average increase of ~25 W m⁻²/decade. Similarly, direct radiation exhibits a slightly stronger positive trend, with a ~29 W m⁻²/decade increase on average. By contrast, the trends for diffuse radiation are all negative and the average change is only ~−6 W m⁻²/decade. This appears somewhat inconsistent with what would be expected if AOD were the only changing quantity, as a decrease of AOD would substantially decrease diffuse radiation. Therefore, we also examine another two parameters, namely SSA and asymmetry parameter, which also affect downward solar radiation (Yang et al 2016a, 2016b). A higher SSA means more radiation can be scattered thus increasing diffuse radiation, and a larger asymmetry parameter indicates more forward scattering which also increases solar radiation reaching the surface. It can be seen (figure 5) that while AOD has been decreasing over the past ten years, SSA has been increasing, whereas no significant trend shows up in the asymmetry parameter. In total AOD decreased by ~0.2 and SSA increased by ~0.02 during 2005–2015. This SSA increase will to some extent compensate for the reduction of diffuse radiation caused by the AOD decrease. We use the RRTM_SW radiative transfer model for a simple calculation of this effect. The model setting and environmental parameters are explained in supplementary section 2, in which the initial AODs and SSAs are taken as the AERONET January and July means of 2005 and the final values are the initial minus the linear trend. Results from tables S2–S5 indicate that decreasing AOD by ~0.2 alone would result in a ~23 W m⁻² (January and July mean) increase of the global irradiation and ~11 W m⁻² decrease of diffuse radiation. However, simultaneously increasing SSA by ~0.02 results in only a ~5 W m⁻² decrease of the diffuse radiation. Although the absolute numbers are different from observation due to model assumptions, the results qualitatively explain the weak diffuse reduction. They also suggest that surface solar irradiation is very sensitive to SSA, which is related to aerosol composition (Xia et al 2016).