An effort to distinguish the effects of cloud cover and aerosols on the decadal variations of surface solar radiation in the Northern Hemisphere

Surface solar radiation (SSR) serves as the primary energy source on Earth. However, a relative lack of research systematically quantifies long-term SSR variations and their driving factors based on complete and reliable baseline data. This paper presents a new assessment of the Northern Hemisphere/regional SSR variations and the influence of total cloud cover (TCC) on these variations, based on the latest reconstructed SSR gridded dataset. We also address multicollinearity among multiple aerosol types and quantify the effects of multiple aerosol/precursors on SSR variability using a partial least squares regression model. The results indicate that TCC is not the predominant driver of longer-term SSR variations, known as ‘dimming’ and ‘brightening’. The variations of NH3 and SO2 primarily drive inter-decadal SSR variations in North America, while the variations of SO2 and NO X mainly influence inter-decadal SSR variations in Europe.


Introduction
Surface solar radiation (SSR) is the primary energy source for all life on Earth.In terms of solar activity, it has been discovered that the variation of total solar irradiance at the top of the atmosphere (TSI) varies by 0.1% during the 11 year solar cycle [1].Some researchers argue that the variation in SSR is ten times higher than that in TSI [1][2][3][4][5][6].SSR drives most of the terrestrial processes, such as the carbon and hydrological cycles, and plays a vital effect in many fields, such as agriculture and solar energy production e.g.[7][8][9][10].The SSR variations are generally influenced by clouds and the cloud-free atmosphere (including the underlying atmospheric composition, aerosols, water vapor, etc.).Among these factors, water vapor and other active atmospheric have had a minor effect on SSR variations [5,11,12].It is generally accepted in the scientific community that clouds and aerosols are both significant modulators of SSR.Clouds primarily affect inter-annual SSR variations, while aerosols notably influence inter-decadal SSR variations [10,[13][14][15][16].For instance, pollution prevention and control measures in Beijing, China, have successfully transitioned SSR from 'dimming' to 'brightening' by an average magnitude of 3.5 W m −2 * decade since 2008 [17].Modelling studies have demonstrated that the 'dimming' in Europe and East Asia is only present in simulations that include variations in anthropogenic constituents aerosol and aerosol/precursor emissions [18][19][20][21].Aerosols are thus a substantial contributor to the 'dimming.'However, it has also been observed that a decrease in low cloud cover explains 70% of the global increasing trend in SSR from 1990-2000 [22].On a regional scale, variations in total cloud cover (TCC) even cause more than 80% of the 'dimming' in northern China during summer and winter [23].In no-cloud conditions, the trends of 'brightening' and 'dimming' in SSR are more pronounced in many regions (Europe, USA, China, India, and Japan) than in all-sky conditions [12,[24][25][26][27][28].These results demonstrate the intricate ways in which clouds and aerosols impact variations in SSR.
So far, there have been several challenges in quantifying and distinguishing the individual contributions of clouds and aerosols to SSR variations.Firstly, there are significant uncertainties in observations of SSR, clouds and aerosols [29][30][31][32][33]; Secondly, there is extensive multicollinearity (see table S1 in the supporting information) among clouds and aerosols, including their constituents.Additionally, the direct, semi-direct, and indirect interactions between atmospheric aerosols and clouds play a significant role in determining and influencing SSR variations [34].This paper proposes a systematic qualitative analysis and quantitative research of the driving factors (TCC and several aerosol/precursors) based on reprocessed SSR observation data in the Northern Hemisphere (NH).The paper is divided into five main sections: section 2 describes the materials and methods used.Section 3 isolates and examines the driving factors of SSR.Section 4 presents the conclusions.And section 4 provides the conclusions.

Data and regions
All data used in this paper (sections 2.1.1-2.1.3)are listed in table 1.

SSRIH 20CR
The SSR data utilized in this paper are sourced from the global land monthly surface (excluding Antarctica) SSR gridded 5 • * 2.5 • resolution dataset (SSRIH 20CR , 1961-2018) [31]).Initially, we developed the first homogenized monthly global SSR dataset by systematically integrating the latest results [25,[37][38][39] of SSR homogenization over the last ten years, based on the Global Energy Balance Archive (GEBA) [40] dataset.Subsequently, we trained a reasonable partial convolutional neural network model based on the 'state-of-art' global reanalysis dataset (20CRv3).We designed an observation constraint algorithm for the difference in SSR variations between sea and land regions.It enabled us to reconstruct a global terrestrially homogeneous, fullcoverage (excluding Antarctic), 5 • * 2.5 • resolution baseline SSR climate dataset with monthly resolution.Our dataset demonstrates high reliability in describing the multiscale variation and spatial pattern of global SSR, as evidenced by a comparative validation study [31].Specifically, the area and magnitude of the high and low centres of the SSRIH 20CR are the same as those of the SSRIH grid (gridded by homogenised SSR station data) (see their figures 7 and S6 in [31]).The time range of this dataset is 1955-2018.It is accessible and freely available to users on the Global Climate Observation and Modelling Data Platform (www.gwpu.net),which was established by our team.For this study, we analysed the years 1961-2014 to investigate the causes of SSR variation.

CRU TS v. 4.07 (driving factor)
The TCC data is sourced from the Climatic Research Unit gridded Time Series (CRU TS v. 4.07).CRU TS is a monthly observational dataset that encompasses land areas, except Antarctica, with a high resolution of 0.5 • × 0.5 • [35].The dataset contains historical time series of cloud cover and other parameters from 1901 to the present.The parameter for cloud cover is determined by direct station observations of cloud cover, as well as proxy measurements of related variables such as sunshine duration and daily temperature range.They use angular-distance weighting (ADW) as the interpolation method, and the generation of secondary variables has been adjusted to align with this approach.The ADW improves the traceability of each grid value to the input observations, providing a more insightful diagnostic for users to assess geographic differences in the quality of the dataset.
During the pre-satellite era, there were limited alternatives available to validate CRU cloud cover, and to the best of the authors' knowledge, there are no comparison analyses.However, the satellite measurements are too short for analysis in this study.Furthermore, the cloud data from the reanalysis is a model-derived product and is not suitable for observational analysis [19].Despite the uncertainty of the CRU TCC data in terms of long-term trends in some regions (such as China (1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002))) [41], the CRU dataset is undoubtedly irreplaceable due to its high spatial (0.5 • × 0.5 • ) and temporal (from 1901 and onward) resolution [19].The CRU dataset uses only observation for interpolation and does not include the uncertainty from the proxy data [35].To ensure coherence with the SSR gridded data, we resampled the TCC data to 5 * 2.5 • by bilinear interpolation [42].The interpolation method we use for all data is a bilinear interpolation.Bilinear interpolation involves performing linear interpolation in two directions.As a type of interpolation algorithm in numerical analysis, bilinear interpolation is widely used in signal processing, digital image, and video processing [42].For the analysis of the causes of SSR variation in this study, we focused on the years 1961-2014.

CEDS (driving factor)
We obtained aerosol/precursor data (black carbon (BC), organic carbon (OC), NH 3 , NO X , SO 2 , and non-methane volatile organic compounds (NMVOC)) from the Community Emissions Data System (CEDS) gridded monthly global emissions data [36].These species are cited as the main anthropogenically sourced aerosol/precursor emissions in the IPCC's sixth assessment report [43].The CEDS emission data, with a resolution of 50 km, is a widely utilized gridded global emission dataset that offers historical modelling of gases and aerosols, as well as future projections.This data is utilized in the Coupled Model Intercomparison Project phase 6, CMIP6 [44], and serves as input data for the participating Earth System Models.The CEDS dataset provides trends over recent decades by utilizing current energy consumption data and emission inventories specific to regions and countries.Emissions of all species are consistently estimated across all periods using the same activity data.The emissions are reported annually by country and sector and are also gridded with monthly seasonality.Although these estimated values are slightly higher than those in the existing global inventory, such as CMIP5 [45], EDGAR [46] and GAINS [47,48], the overall scenario is similar to the existing global inventory [49].The CEDS dataset covers the period from 1850 to 2014, with a high resolution of 0.5 • × 0.5 • .For this study, we focused our analysis on the years 1961-2014 to investigate the causes of variations in SSR.Additionally, we resampled the spatial resolution of all data to 5 • * 2.5 • by bilinear interpolation [42].

Interest regions
The solar elevation angle impacts the climatology and variations of the SSR, leading to differences in the SSR variations at various latitudes [30].The SSR variations are not necessarily identical, even at similar latitudinal ranges [50].Therefore, it is important to compare regional characteristics of SSR variations to understand inter-regional differences in climate change.In this paper, we selected the NH and two comparison regions (figure 1) to explore the driving factors of long-term SSR variations in the NH.We chose North America and Europe due to their similar latitudinal ranges, comparable economic levels, early industrialization, and early implementation of pollution controls [51,52].

Remove the effect of cloud cover
Considering the time scale differences in the effects of cloud cover and aerosol on SSR variations [10,[13][14][15][16], this paper initially proposes to separate the effects of TCC variations on SSR (SSR no-TCC ) from those of other radiatively effective components, such as aerosol, to ensure the accuracy of their respective effects.Based on physical experience, there is a pronounced linear relationship between variations in TCC and all-sky SSR (SSR all-sky ) (inverse correlation, table 2).Therefore, we used an algorithm that removes the effect of TCC from SSR all-sky variations through regression analysis.Specifically, we constructed a regression model with the SSR all-sky time series after detrending (denoted as Y_d) as the dependent variable and the TCC time series after detrending (represented as X_d) as the independent variable.The TCC/SSR regression coefficients are computed at each grid box and for the year.The coefficients of the regression are represented as A. Subsequently, input the time series of SSR (denoted as Y) and TCC (represented as X) into regression equation ( 2) to calculate the SSR no-TCC : It should be pointed out that while TCC effects are removed, changes in the cloud optical thickness (caused by changes in cloud microphysics/aerosol interaction, cloud water content, or cloud type) remain in the 'no-cloud' (TCC-cleared) dataset.So precisely what remains after eliminating the TCC effects are the effects from changes in the cloud-free atmosphere plus potential changes in cloud optical properties not removed with the TCC regression [53,54].

Driving factors analysis methods
Commonly utilized methods for separating driving factors include the random forest method [55] and partial least squares regression (PLSR) [56,57], among others.Likewise, the possible driving factors of the SSR all-sky in this paper exhibit multi-correlated (collinearity) with each other (see table S1 in the supporting information).Consequently, the ordinary multiple linear regression (MLR) equation is no longer the best linear unbiased estimator when using methods such as multiple regression directly [58,59].This paper employed the PLSR method, which has demonstrated strong performance in distinguishing the independent contributions of multiple factors that are significantly correlated with each other [56,57,[60][61][62][63].PLSR is a multivariate statistical data analysis method that was first proposed by Wold and Esbensen [62] in 1983.Over the last 30 years, PLSR has quickly expanded in theory and methodology, while its application areas have proliferated from the initial field of chemistry to more natural and social sciences [60,61,63].This method is also applicable to cases in which the number of samples is relatively small (even when they are less than the number of factors), and it is equivalent to implementing linear regression analysis, principal components analysis (PCA), and canonical correlation analysis at the same time.The PLSR, different from traditional regression analysis, standardizes multiple independent and dependent variables to form a standardized matrix of independent (E 0 ) and dependent variables (F 0 ).Compared with traditional regression analysis methods, it first standardizes the independent and dependent variables to obtain the standardization matrix.The next step involves performing PCA on the independent variables to extract the principal component and load vector corresponding to the eigenvalue that is most closely associated with the dependent variable.Subsequently, the principal component is regressed on the dependent variable, and the resulting residual matrices are calculated separately.These residual matrices are then processed similarly.
The explained variance of the predicted dependent variable, calculated from the residuals, is utilized to assess the stability of the equation estimated by the partial regression, thereby determining the number of selected principal components.[61].
Its model expression is: E 0 denotes the normalized matrix of independent variable E 0 = (E 01 ,…, E 0p ) (n×p) , F 0 denotes the normalized matrix of the dependent variable, F 0 = (F 01 ,…, F 0p ) (n×p) , S denotes the number of components extracted from the original variables, t i denotes the vector of principal components of the matrix of independent variables, extracted one at a time (i), p i denotes the load vector of the independent variable, r i denotes the projection vector of the dependent variable on the principal component axis, F s denotes the residual matrix, and the symbol (') represents the transpose.
Since t 1 …t s can all be expressed as a linear combination of E 01 , …, E 0p , equations ( 3) and ( 4) can be reduced to the form of the regression equation of the standardized dependent variable y k * = F 0k on the x j * = E 0j standardized independent variable, whose expression is: a kj X * j + F sk (5) k = 1, …, q, j = 1, …, p, α kj denotes the standardized coefficient of the j th independent variable concerning the kth dependent variable, F sk denotes the kth column of the residual matrix, and ( * ) denotes the standardization treatment.The advantages of PLSR analysis are as follows: (1) it can create regression models for multiple sets of dependent variables and independent variables.(2) It can effectively address issues found in the MLR model, such as multicollinearity among independent variables and the number of sample points being less than the number of variables in the modelling analysis [57,59,61].We used regional average SSR all-sky /SSR no-TCC as dependent variables, and the TCC and aerosol/precursors (BC, OC, NH 3 , NO X , SO 2 and NMVOC) in the corresponding as independent variables for the PLSR modeling.We removed significant volcanic eruption years (1963,1982,1991) from the PSLR analysis to remove the effect of volcanic aerosols on SSR variation.

The SSR variations in the NH and the TCC effect
Figure 2 illustrates the variations of annual and seasonal SSR all-sky /SSR no-TCC in the NH from 1961-2018.The trends of annual and seasonal SSR all-sky /SSR no-TCC in the NH are presented in table S2 (see in the supporting information).Over the entire time series, the annual and seasonal SSR all-sky in the NH generally exhibited a negative trend from 1961-2018 (except 1963, 1982, 1991), with the most significant negative trend occurring in autumn (−0.34 ± 0.19 W m −2 * decade) and the smallest in spring (−0.19 ± 0.23 W m −2 * decade).The minimum values of the annual and seasonal SSR all-sky occurred in 1991-1992, just around the time Mount Pinatubo erupted.It is worth mentioning the somewhat stronger trends in summer and spring than in autumn and winter, as we see it in several studies [16].Specifically, during the 'dimming' period from 1961-1992 (except 1963, 1982, 1991), SSR all-sky in the NH decreased significantly throughout the annual and seasonal periods, with no significant differences between seasons.Subsequently, during the 'brightening' period from 1992-2018, annual and seasonal SSR all-sky in the NH displayed a significant increase, with no significant differences in seasonal and annual SSR all-sky trends (see figure 2, and table S2 in the supporting information).The results are in support of the general 'global dimming/brightening' concept, suggested by Wild et al [5,64,65].
When we remove the effect of TCC on the SSR all-sky using the method described in section 2.2.1, the SSR no-TCC still exhibits long-term variations from 'dimming' to 'brightening' , similar to the performance of SSR all-sky .Furthermore, seasonal and annual variations in SSR no-TCC all demonstrate negative trends, although there are slight differences in the trend values and their significance.During the seasonal 'dimming' and 'brightening' periods, the trends of SSR no-TCC and SSR all-sky are not significantly different (see figures 2(a) and (b) and table S2 in the supporting information).This suggests that TCC does not appear to be the predominant driver of the interdecadal SSR variations in the NH.Overall, the interdecadal characteristics (transition from 'dimming' to 'brightening' trend) of the SSR no-TCC obtained using the method in this paper remain unchanged.

The TCC effect on the SSR at regional scale
We computed SSR no-TCC in the two regions using the method outlined in section 2.2.1.Figures 3(a) and (b) display the annual variations of SSR all-sky , SSR no-TCC and TCC in the two regions (depicted in figure 1) from 1961-2018.Additionally, figure S1(a) (see in the supporting information) illustrates the multi-year average (1961-2018) spatial distribution of TCC in the NH.Table S3 (see in the supporting information) presents the trends and their uncertainties for the two regions of SSR all-sky and SSR no-TCC .
It is evident that SSR and TCC in all two regions exhibit an inverse relationship in terms of interannual variations, indicating that the presence of clouds inevitably reduces the SSR.This finding is consistent with previous studies [16,19,21,23,28].Consequently, the annual average TCC is significantly (p < 0.01) negatively correlated with SSR all-sky in both regions.Specifically, from 1961 to the 1970s, TCC in NA demonstrated a monotonic increase corresponding to a decrease in SSR all-sky in the region (with an explained variance of about 32%).The variation in the anomalies of the TCC and SSR all-sky in Europe is consistently opposite (explained variance about 71%); Spatially (see figure S1(a) in the supporting information), TCC is high in Europe/Northern America.This result inversely corresponds to the spatial distribution of SSR climatology (small in Europe/Northern America) [30].The variations in SSR are comparable to those described earlier for the NH when the effect of clouds is removed: The SSR no-TCC also exhibits a similar inter-decadal/multidecadal variation to the SSR all-sky at the regional scale; In terms of magnitude of variations, the SSR no-TCC shows a similar or slightly smaller trend than the SSR all-sky across most of the region (see table S3 in the supporting information).This suggests that our method of removing the effect of TCC on SSR does  not significantly influence the 'dimming' and 'brightening' trends of SSR.Therefore, we further support the point that TCC is not the main driving factor of SSR variations on scales above the inter-decadal.

The aerosols effect on the SSR at regional scale
Aerosols can reduce the amount of SSR reaching the surface by scattering or absorbing SSR.However, due to the multi-collinearity among aerosol/precursors (see table S1 in the supporting information) and the different spatial distribution and temporal variation characteristics of different aerosol/precursors in different regions (see figures S1(b)-(h) and S2 in the supporting information), it becomes very challenging to conduct quantitative research on the driving factors of SSR variations.For our PLSR modeling, we used annual SSR no-TCC from the two regions of the NH as dependent variables, and annual TCC and aerosol/precursors (BC, OC, NH 3 , NO X , SO 2 and NMVOC) as independent variables.Figure 4(a) depict the standard coefficient and its actual contribution ratios (the PLSR coefficient (the scaled coefficients from the normalized data by the ratios between the standard deviations of SSR itself and of each factors) × accumulate change) for each driving factor in the PLSR model for the SSR no-TCC variations.The results show that PLSR modelled the SSR no-TCC in the two regions (the explained variance ranged from 53 to 76%, see table S4 in the supporting information), and there were some differences in the significance of the driving factors across the different regions.
In NA, the maximum contributions come from NH 3 and SO 2 , with their contribution weights accounting for 41% and 39%.In Europe, SO 2 (35%) and NO X (16%) are the main contributors to SSR no-TCC variation.These finding supports the idea that SO 2 is by far the largest source of anthropogenic aerosol forcing globally [19,20,43,66].
Based on the significant negative correlation between annual mean TCC and SSR variations (see table 2 in the supporting information), we can remove the effect of TCC on SSR variations through linear regression.The removal of TCC has minimal impact on the inter-decadal variations of the average SSR in the NH and the regions, as depicted in figures 3 and 4. Essentially, TCC only significantly affects the SSR interannual variations, not the interdecadal SSR variations, which aligns with previous studies [10,12,19,21,28,67,68].
Furthermore, we replicated the experiments outlined in section 3.3.using the regional SSR all-sky as dependent variables, TCC and the aforementioned six aerosol/precursors as independent variables.The results indicate that, compared to SSR no-TCC , the total variance explained by the driving factors of SSR all-sky in Europe increases to some extent when considering the influence of TCC, but the change is not significant in the NA (see figures 4(a), (b), supporting information table S4).Meanwhile, The contribution of TCC to the SSR no-TCC is negligible in all regions (<5%), while the impact of the most crucial aerosol/precursor on the SSR no-TCC variations becomes more significant than on SSR all-sky .It is worth noting that even before removing the effect of TCC on SSR, the contribution of TCC to the SSR variations was only in the range of 5%-10%.This suggests that the use of linear regression to remove the cloud cover effect on SSR is reasonable and consistent with the limited effect of TCC in inter-decadal SSR variations.However, the SSR no-TCC calculated in this paper only eliminates the portion of SSR all-sky that TCC associated with SSR all-sky , and does not mitigate the interactive effects between clouds and aerosols or any other process that can affect cloud optical depth, e.g.changes in cloud types, cloud liquid water content etc.Because the drivers of SSR variations are complex, this study does not exhaust all of them, such as desert dust, wildfire smoke, sea salt and volcanic aerosols [69] water vapour and aerosol-cloud interactions, as can be seen from the modest variance explained by PLSR (about 54%-76%) and the discrepancy with the results of previous studies.In addition, the PLSR ignores the long-range transport of aerosols.This is a significant limitation of this method.Moreover, there is some uncertainty in the TCC dataset [29,35,41], which may also introduce biases in the results of SSR no-TCC and their driving factors.i.e. for individual regions, the driving factors (TCC) remain significant, albeit with smaller contribution ratios (see figure 4(a) in the supporting information).
The effect of aerosols on SSR is intricate (figure 5).Certain factors can result in reduced SSR reaching the surface through scattering or absorbing solar radiation (direct effect) [13,20,70,71](e.g.NH 3 and SO 2 in NA, SO 2 and NO X in Europe, figure 4); Some factors (e.g.BC in Europe) can increase SSR by heating the air, leading to cloud inhibition or burn-off (semi-direct effect) [72,73].Additionally, the majority of the factors (e.g.NO X , SO 2 , and OC) can also act as cloud condensation nuclei (CCN)/ice nuclei (IN) (first indirect effect), resulting in less or more solar radiation reaching the surface [20,74,75].They can also indirectly affect SSR by reducing the radius of cloud particles, thereby inhibiting precipitation or increasing cloud lifetime/height (second indirect effect) [76].From this perspective, this study has not yet been able to clarify such indirect and interactive effects as described above.

Conclusions
Accurate quantification of respective contributions of aerosol/precursor factors to SSR is difficult due to the extensive multiple correlations among aerosol/precursor factors (see table S1 in the supporting information).PLSR overcomes some of the limitations of ordinary linear regression equations.It is relatively effective in separating the specific contributions of individual factors to SSR variations across the two regions.Some preliminary conclusions can be drawn: In terms of long-term variations, with or without the cloud effect, the annual and seasonal SSR in the NH undergoes a process of 'dimming'(60 s-90 s) to 'brightening'(90 s-), with insignificant seasonal differences.There was no significant effect on the 'dimming' and 'brightening' trends when the effect of TCC on SSR was removed, indicating that the TCC is not the predominant driver of the inter-decadal variations ('brightening' and 'dimming') of the SSR.Most of the aerosol/precursors are deterministic driving factors for SSR in all two regions.The contribution of aerosols to SSR no-TCC is more apparent than SSR all-sky when we remove the contribution of TCC to SSR.Variations in NH 3 and SO 2 mainly drove the inter-decadal SSR variations in NA and Europe.Overall, the conclusions of this paper that clouds and aerosols deterministically influence SSR variations in specific regions of the NH are in good agreement with existing results e.g.[10,12,19,21,25,28,50].
However, it is worth noting that this paper does not take into account cloud and aerosol interactions, nor does it incorporate into the modelling all the driving factors that may affect regional SSR variations, e.g.different types of clouds [22,23,77], water vapour [12], etc.Furthermore, the reasons for these discrepancies [50,66,68,78] are due to data uncertainty [41,79], different methods, different areas and periods.It is worth mentioning that the PLSR method used in this paper has high requirements for the quality of modelling data (inhomogenities, inter-annual biases).Inaccurate data will result in a certain sensitivity of the coefficients.The researchers are not all in agreement about the types of aerosols that influence SSR variations in Europe.For example, some scholars believe that the SSR in Europe can be linearly fitted by SO 2 and BC [66].Others indicated the emissions of aerosols (SO 2 , NO X and NMVOC) are contributing to the SSR dimming and subsequent brightening in Europe [19].The differences in our results may be due to the use of different methods and data.Leibensperger et al found that the main contribution to the radiative forcing of US anthropogenic aerosols comes from sulfate (−2.0 W m −2 ), but the role of NH 3 was not mentioned [80].However, North America is the high-value area in terms of the spatial distribution of NH 3 (see figure S1(e) in the supporting information).In the future, we plan to consider combining model data from different experimental and forcing conditions to explore the effects of aerosol-cloud interactions on long-term SSR variations.Meanwhile, we will conduct further targeted optimization of the driving factors and models, as well as perform an in-depth mechanistic analysis of the statistical results above.

Figure 1 .
Figure 1.The study area of this paper.

Figure 2 .
Figure 2. Long-term anomaly variations of all-sky SSR (SSR all-sky ) and no-TCC SSR (SSR no-TCC ) during 1961-2018.(a)-(d) Long-term SSR anomaly variations (relative to 1971-2000) in annual and seasonal (a)-(b) SSR all-sky and (c)-(d) SSR no-TCC in the Northern Hemisphere (NH) from 1961-2018.The black solid line represents the annual SSR anomalies in the NH, and the coloured solid lines represent the seasonal SSR anomalies; the bar charts show the annual and seasonal trends (except for 1963, 1982, 1991) and their significance (at the 5% level) in the NH (1961-1992 (a, c, insert map), 1992-2018 (a, c, insert map) and 1961-2018 (b), (d)).

Figure 4 .
Figure 4. Influence of aerosol/precursors on (a) SSR all-sky (b) SSR no-TCC (Aerosol/precursors & TCC) anomaly variations in Northern America and Europe, 1961-2018.Left: standard coefficient and its significance (at the 5% level), as well as the actual contribution for each driving factor (right).

Figure 5 .
Figure 5.The radiative effects of cloud and aerosols.

Table 1 .
List of information on the various types of data used in this paper.

Table 2 .
Correlation analysis between annual mean SSR and TCC for the period 1961-2018 in the Northern Hemisphere (NH) land areas and the two regions Northern America (NA) and Eur (Europe).