Merging ground-based sunshine duration 1 observations with satellite cloud and aerosol 2 retrievals to produce high resolution long- 3 term surface solar radiation over China

Although great progress has been made in estimating surface solar radiation ( R s ) 22 from meteorological observations, satellite retrieval and reanalysis, getting best 23 estimated long-term variations in R s are sorely needed for climate studies. It has been 24 shown that sunshine duration (SunDu)-derived R s data can provide reliable long-term 25 variability, but are avaliable at sparsely distributed weather stations. Here, we merge 26 SunDu-derived R s with satellite-derived cloud fraction and aerosol optical depth (AOD) 27 to generate high spatial resolution (0.1  ) R s over China from 2000 to 2017. The 28 geographically weighted regression (GWR) and ordinary least squares regression (OLS) 29 merging methods are compared, and GWR is found to perform better. Based on the 30 SunDu-derived R s from 97 meteorological observation stations, which are co-located 31 with those that direct R s measurement sites, the GWR incorporated with satellite cloud 32 fraction and AOD data produces monthly R s with R 2 = 0.97 and standard deviation = 33 11.14 W/m 2 , while GWR driven by only cloud fraction produces similar results with R 2 34 = 0.97 and standard deviation = 11.41 w/m 2 . This similarity is because SunDu-derived 35 R s has included the impact of aerosols. This finding can help to build long-term R s 36 variations based on cloud data, such as Advanced Very High Resolution Radiometer 37 (AVHRR) cloud retrievals, especially before 2000, when satellite AOD retrievals are 38 not unavailable. The merged R s product at a spatial resolution of 0.1  in this study can 39 be downloaded at https://doi.pangaea.de/10.1594/PANGAEA.921847 (Feng and Wang, 40 2020). 41 42 43 44 45

original CERES EBAF but not 0.1 degree cloud product." in Rs is closely related to the Earth's water cycle, the whole biosphere, and the amount 54 of available solar energy. This situation emphasizes the urgency to develop reliable Rs 55 products to obtain the variability in Rs. 56 Great progress has been made in the detection of variability in Rs by  Table 1 lists the current satellite-based Rs products, which have been widely  On the other hand, the spatial resolution of Rs data is crucial for regional 168 meteorology studies, as the minimum requirement of the spatial resolution of Rs data,   derived Rs are used as independent reference data to investigate the performances of the 220 fusion methods (Fig. 1). The whole area over China is further divided into nine zones  collected from approximately 2,400 meteorological stations (http://data/cma/cn/) from 242 the CMA, are used to calculate the SunDu-derived Rs (Fig. 1). Rs values are calculated    derived Rs data are randomly selected to validate these GWR parameters (Fig. 1). The 377 results show that R 2 increases and bias decreases when the number of NNPs decreases.

378
However, when the NNP is smaller than 30, the GWR-based fusion method produces 379 spatially incomplete Rs data due to the local collinearity problem with large spatial 380 variability. Therefore, 30 is selected as the NNP parameter (Table 3).

Site validation 390
Based on the independent SunDu validation sites, both the GWR and OLS 391 methods explain 97%~86% of Rs variability (Fig. 4). The GWR method generally  (Fig. 4). The comparative result shows that both 398 fusion methods show slightly reduced performances when using direct Rs observations 399 rather than the SunDu-derived Rs. Both the GWR and OLS methods explain 91%~82%

417
To analyse the impacts of AOD on the GWR fusion results, the GWR driven with

438
We also analyse the performances of fusion methods for different seasons at all 439 validation sites, as shown in Table 4. At seasonal scales, both the GWR-CF and GWR-  Rs and impact factors is not linearly stable and is closely related to spatial position. The 467 spatial distribution of the Rs trend derived from the GWR method is also consistent with 468 the SunDu-derived Rs trend, especially in western China (Fig. 8). In order to prove that 469 SunDu-derived Rs can add value to the 0.1 degree product, instead of cloud fraction 470 data alone. We perform a similar GWR analysis but using CERES EBAF interpolated 471 to 0.1 degree and 0.1 degree cloud, and compare the results with those using SunDu-472 derived Rs and 0.1 degree cloud (Fig .9). The results indicate that SunDu-derived Rs 473 can add value to the 0.1 degree product and the merged Rs by using interpolated 474 CERES EBAF and 0.1 degree cloud product are also similarly to original CERES 475 EBAF but not 0.1 degree cloud product.  (Figures 2, 3 and 10), indicating that the GWR-CF and GWR-CF-AOD methods can 500 produce reasonable annual Rs variations over China.

501
In zones I and II, located in northern arid/semiarid regions, the annual anomaly Rs    Comparison results from ) also indicate that the GWR method 553 is better than the multiple linear regression method and spline interpolation method for 554 near surface air temperature. By using spatial interpolation method, CERES EBAF Rs 555 can also be downscaled to 1km or 30m. These interpolated CERES Rs data cannot 556 represent the detailed Rs distributions at spatial resolution of 1km or 30m due to the 557 variability of Rs within a 1 degree box. Without additional high spatial resolution data, 558 interpolated cannot capture more detail variability of Rs.

565
Accurate estimation of Rs variability is crucially important for regional energy 566 budget, water cycle and climate change studies. Recent studies have shown that SunDu-567 derived Rs data can provide reliable long-term Rs series. In this study, we merged