J. Meteor. Res.  2015, Vol. 29 Issue (1): 144-145   PDF    
http://dx.doi.org/10.1007/s13351-014-4934-1
The Chinese Meteorological Society
0

Article Information

Zhu Xiaochen, Qiu Xinfa, Zeng Yan, GAO Jiaqi, 2015.
A remote sensing model to estimate sunshine duration in the Ningxia Hui Autonomous Region, China
J. Meteor. Res., 29(1): 144-154
http://dx.doi.org/10.1007/s13351-015-4059-1.

Article History

Received April 30, 2014
in final form September 28, 2014
A Remote Sensing Model to Estimate Sunshine Duration in the Ningxia Hui Autonomous Region, China
ZHU Xiaochen1, QIU Xinfa1 , ZENG Yan2, GAO Jiaqi3,     
1 School of Remote Sensing, Nanjing University of Information Science & Technology, Nanjing210044;
2 Jiangsu Institute of Meteorological Sciences, Jiangsu Meteorological Bureau, Nanjing210008;
3 College of Atmospheric Science, Nanjing University of Information Science & Technology, Nanjing210044
Abstract:Sunshine duration (SD) is strongly correlated with solar radiation, and is most widely used to estimatethe latter. This study builds a remote sensing model on a 100 m ×100 m spatial resolution to estimateSD for the Ningxia Hui Autonomous Region, China. Digital elevation model (DEM) data are employedto reflect topography, and moderate-resolution imaging spectroradiometer (MODIS) cloud products (AquaMYD06−L2 and Terra MOD06−L2) are used to estimate sunshine percentage. Based on the terrain (e.g.,slope, aspect, and terrain shadowing degree) and the atmospheric conditions (e.g., air molecules, aerosols,moisture, cloud cover, and cloud types), observation data from weather stations are also incorporated intothe model. Verification results indicate that the model simulations match reasonably with the observations,with the average relative error of the total daily SD being 2.21%. Further data analysis reveals that thevariation of the estimated SD is consistent with that of the maximum possible SD; its spatial variation isso substantial that the estimated SD differs significantly between the south-facing and north-facing slopes,and its seasonal variation is also large throughout the year.
Key words: sunshine duration     digital elevation model data     -resolution imaging spectroradiometer (MODIS)     moderate     cloud cover     remote sensing estimation model    
1. Introduction

Sunshine duration(SD)is an important indicator of the amount of solar radiation received in a region. Adequate sunshine not only increases air temperature and effective accumulated temperature butalso aids photosynthesis in plants,thereby promotingtheir growth. SD is the most visible manifestation ofsolar radiation(Watson and Albritton, 2001), and itis strongly correlated with solar radiation(Takenaka,1988; Ertekin and Evrendilek, 2007). As a result,SDis the most widely used parameter to estimate solarradiation(Yin,1957; Weng,1964; Sen,2001; Sen and ¨Oztopal,2003). Many empirical models for the prediction of solar radiation from SD have been developed(Angstrom,1924; Ertekin and Evrendilek, 2007). SDis also an important parameter in the study of ecosystem process models,hydrological simulation models, and biophysical models(Weng,1997).

SD is affected by terrain factors(e.g.,slope,aspect, and terrain shadowing degree) and atmosphericfactors(e.g.,air molecules,aerosols,moisture,cloudcover, and cloud types). Here,we use digital elevation model(DEM)data to comprehensively reflectthe terrain. These data(with different resolutions)have unique abilities to reflect and characterize terrain. However,due to certain restrictions(e.g.,computation speed and model efficiency),the calculationmodel is rarely applied to DEM data on the 1:250000scale in large-scale distribution modeling research(Li et al., 2008). Additionally,we use the sunshine percentage estimated from the moderate-resolution imaging spectroradiometer(MODIS)cloud products(theAqua MYD06−L2 and the Terra MOD06−L2)to comprehensively reflect the atmosphere because cloudsstrongly affect the solar radiation on the earth’s surface(Suehrcke et al., 2013). The sunshine percentageis also known as the sunshine fraction,i.e.,the ratiobetween the actual sunshine duration and the possible sunshine duration. The possible sunshine durationmust be measured on a day that is entirely cloudless.

SD is calculated by using the maximum possibleSD(MPSD) and sunshine percentage. Traditionally,the calculation of SD has been achieved by three major methods: the empirical equation method(Wang,1993),the analytical method(Fu,1958; Zhu,1988), and the graphical method(Weng et al., 1981; Sun and Fu, 1996). With the development of space technology,many researchers have tried to improve the simulationaccuracy with regard to the calculation of SD. Thesimulation accuracy of the portion of the MPSD usedin the calculation of SD was improved with the development and promotion of the DEM(Moore et al., 1991). Dozier and Outcalt(1979)proposed that usingthe DEM to simulate solar radiation enabled a digitalapproach to explore the effect of terrain on the MPSD.Hetrick et al.(1993),Dubayah et al.(1990),Kumaret al.(1997), and Bocquet(1984)studied the MPSDin mountains with regard to the influence of terrainon solar radiation by applying the geographic information system(GIS). Zeng et al.(2003)investigatedthe spatial distribution of MPSD in China using dataat a spatial resolution of 1 km×1 km by establishing a distribution model for MPSD in rugged terrains.Chen et al.(2002) and Yuan et al.(2008)used thesunshine percentage from the routine observation dataat weather stations to calculate the actual SD in specific research areas. Some researchers have attemptedto estimate SD from satellite data,which provide better spatial coverage and representation(Good,2010;Shamim et al., 2012; Bertrand et al., 2013). Shamimet al.(2012)presented an improved model for globalSD estimation. The methodology incorporated geostationary satellite images by including snow cover information,sun and satellite angles, and a trend correctionfactor for seasons for the determination of cloud coverindex.

Based on previous studies,we now consider theeffects of both terrain and atmospheric factors by using the DEM data for the Ningxia Hui AutonomousRegion(abbreviated as Ningxia hereafter)to establisha distribution model of MPSD and by using the 2003daily MODIS cloud products to construct a remotesensing estimation model for sunshine percentage. Asa result,we completed the estimation of SD in Ningxiabased on the MODIS cloud cover data. Section 2 describes the climate of Ningxia. Section 3 describes thedata used in the current study and the primary methods. Section 4 presents a systematic analysis of theSD estimation results. The discussion and conclusionsare presented in Sections 5 and 6,respectively.2. Climatic conditions in Ningxia,China

Ningxia is located in 35℃14'–39℃23'N,104℃17'–107℃39'E, and its capital city,Yinchuan,is located inthe center of the region,upstream of the Yellow Riverin western China. This region spans 456 km north tosouth and 250 km west to east,covering an area ofapproximately 66000 km2.

Two climates occur in Ningxia: the eastern monsoon climate and the northwest arid climate. Theregion is close to the Qinghai-Tibetan alpine area,placing it approximately at the transition zone ofthree natural regions: the eastern monsoon region,thenorthwest arid and semiarid region, and the QinghaiTibetan alpine region. According to historical recordsof st and ard weather stations,the annual average temperature of Ningxia is approximately 8.2℃, and theaverage annual rainfall is approximately 272.6 mm. 3. Data and methods3.1 Data

The data used for this study consist of:

(1)Observations obtained by weather stations inNingxia,including the monthly average total cloudcover and low cloud cover, and SD observed from st and ard weather stations in 2003,as well as the monthlyaverage SD at the intensive observation stations duringthe same year. The observational data at the st and ardweather stations are used to develop the remote sensing model for estimating the SD, and the data fromthe intensive observation weather stations are usedto assess the model error. These data are providedby the National Meteorological Information Center ofthe China Meteorological Administration and are subjected to strict quality control. Figure 1 presents adistribution map of st and ard and intensive observation weather stations in Ningxia.

Fig. 1. Distribution of the st and ard and intensive observation weather stations in Ningxia.

(2)Satellite remote sensing data,includingthe daily TERRA/MODIS(MOD06−L2) and theAqua/MODIS(MYD06−L2)cloud data for Ningxia and its surrounding regions in 2003,with a spatial resolution of 5 km×5 km. These data are obtained fromthe TERRA/MODIS and Aqua/MODIS image totaldaily cloud cover data by using a geometric correction →image mosaic →total daily cloud cover image superposition procedure. Due to movement of thesensing platform,undulating terrain,curvature of theearth’s surface,atmospheric refraction, and the earth’srotation,the remote sensing image distortion occursfor a given geometric position. Through geometric correction,such problems can be solved. Due to changesin the scanning position,the cloud cover of the studyarea is distributed over multiple sets of images. Animage mosaic is constructed by splicing a series of flatimages together to create a continuous panoramic image. The cloud cover for 15 April 2003 was obtained byusing geometric correction,image mosaic, and cloudcover image superposition,as shown in Fig. 2.

Fig. 2. The cloud cover over Ningxia on 15 April 2003,obtained by using geometric correction,image mosaic, and a cloud cover image superposition procedure

(3)National basic geographic data,including theDEM(100 m×100 m)data for Ningxia,an administrative map of Ningxia, and basic geographic information collected from the st and ard and intensive observation weather stations.The data used for this study are subjected to rigorous quality control. The software used for data analysis includes SPSS(Statistical Package for the SocialSciences),Access database,ArcGIS(the Environmental Systems Research Institute,Inc.’s geographic information software), and ENVI(Environment for Visualizing Images)remote sensing image processing software. In this study,ENVI is used for remote sensingimage analysis and display,SPSS is used for statistical analysis of the data, and ArcGIS is used for theprojection and its secondary development.3.2 Methods 3.2.1 The sunshine percentage estimation model

Previous studies(e.g.,Liu et al., 2009)havedemonstrated that there is always some deviation between the satellite-retrieved cloud data and the totalcloud data observed at weather stations; thus,it isnecessary to correct the satellite-retrieved cloud databefore they are used. Cao et al.(2012)proposed fivecorrection methods for the MODIS cloud product withground observation data,of which the ratio methodshowed the best correction results. A linear relationship exists between the sunshine percentage and thecloud cover, and the latter is the primary factor thataffects the variation of the former(Ding et al., 2005).Recently,Matuszko(2012)also analyzed the influenceof cloudiness on SD. The major factors that affect sunshine percentage include low and total cloud cover,which are corrected by the ratio method. Taking theseeffects on sunshine percentage into consideration,weestablished a remote sensing estimation model for sunshine percentage as follows.

where SRg represents the sunshine percentage observed at the weather station; CLp is the total cloudcover at the pixel corresponding to the weather station location; Lowp is the low cloud cover data at thepixel corresponding to the weather station location;a,b, and c are regression coefficients; interpolate()represents the interpolation result ofa,b, and c;SRisthesunshine percentage image data; CL is the total set ofcloud image data after correction; and Low representsthe low cloud cover image data after correction.

The sunshine percentage estimation process includes the following steps. First,calculate the monthlymean MODIS cloud cover based on the daily data.Second,revise the MODIS total cloud cover imagebased on the monthly total cloud cover reported byground observations and the monthly mean MODIScloud cover by using the ratio correction method.Third,establish the ratio relationship between totalcloud cover and low cloud cover reported by groundobservations, and calculate the MODIS low cloudcover image and MODIS total cloud cover image aftercorrection. Fourth,extract the monthly MODIS total and low cloud cover for each meteorological station after correction,establish the linear regression equationbetween the sunshine percentage reported by groundobservations and total and low MODIS cloud cover using Eq.(1), and calculate the monthly coefficientsa,b, and cin Eq.(1). Fifth,interpolate the monthly coefficientsa,b, and cto obtain the corresponding spatialdistribution. Sixth,calculate the sunshine percentagefrom the image data by raster calculation,accordingto Eq.(2).3.2.2 The SD model

In general,two different definitions exist forMPSD: 1)the astronomical MPSD(without taking atmospheric and terrain shading factors into account) and 2)the geographical MPSD(which considers terrain-shading factors without atmospheric effects). The MPSD in this study refers to the geographical MPSD(Zuo,1990). Studies have shown that theSD at any point in actual rugged terrains for a givenday is the product that combines the MPSD for ruggedterrains and the sunshine percentage. This model iswritten as

where Lαβ is the SD of rugged terrains, and L0αβ isthe MPSD of rugged terrains. L0αβ is calculated byusing the MPSD distribution model developed by Zenget al.(2003)for rugged terrains, and SR is derived from the sunshine percentage estimate modeldescribed in Section 3.2.1. Figure 3 illustrates the calculation procedure for the SD model.
Fig. 3. Flow chart of the calculation procedure for the SD model
4. Results4.1 Estimation of sunshine percentage

Based on the computation flow chart in Fig. 3 and Eqs.(1)–(3),we derived the corrected total cloudcover image data,the estimated low cloud cover imagedata(Cao et al., 2012), and the estimated sunshinepercentage image data combined with its associatedground data. The sunshine percentage estimates obtained by using remote sensing are presented in Fig. 4.The overall sunshine percentage in January and October was greater than that in April and July,primarilybecause sunny days predominated during the fall and winter seasons of 2003 in Ningxia. The sunshine percentage magnitudes are higher in the north than inthe south in this case. An error analysis for sunshinepercentage is conducted by using 13 intensive observation stations in Ningxia, and the results are presentedin Table 1. The maximum monthly average absoluteerror(MAE)is 8.23% and it occurred in April, and the average absolute error for the entire year of 2003is relatively small(5.30%).

Fig. 4. Sunshine percentage estimates over Ningxia in(a)January,(b)April,(c)July, and (d)October 2003.

Table 1. Error analysis for sunshine percentage estimation at the intensive observation stations in Ningxiaduring 2003
4.2 Estimation of SD and its seasonal variation

We estimated SD using the aforementionedmethod. Its estimation results and magnified regionalmaps are presented in Fig. 5,which clearly demonstrates that the annual average total daily SD inNingxia in 2003 was distributed with higher values inthe north and west than in the south and east. Ingeneral,the average total daily SD in 2003 was 7.18 hin Ningxia. The duration was lowest over the southern mountainous area,where the annual average totalSD was only 4.3–5.3 h. The second lowest SD was inthe southeastern and southwestern mountainous areas,where the annual average daily total SD ranged from5.2 to 6.4 h. The highest SD was in the northernmost and mid-west of Ningxia,which had an annual averagedaily total SD of approximately 8.6 h. It can be easily inferred that the regional topography significantlyaffected the SD,especially in mountainous areas withrugged terrains. The shading effects of slope,aspect, and the surrounding terrains on SD are fully demonstrated here.

Fig. 5. Annual average distributions of the total daily SDin Ningxia in 2003.

January,April,July, and October were chosento represent winter,spring,summer, and fall,respectively. Figure 6 illustrates spatial distributions of themonthly averaged total daily SD over Ningxia duringthe four seasons of 2003. In general,the spatial distribution of SD during the four seasons is consistentwith that of the annual SD,with obvious distributioncharacteristics,e.g.,SD is higher in the mid-west thanthe south, and higher in high-latitude areas than lowlatitude areas. The average SDs in Ningxia were 7.25,7.77,7.25, and 7.03 h in spring,summer,fall, and winter,respectively. Moreover,SD values decreasedin the order: summer >spring >fall >winter; thisorder reveals an asymmetric SD distribution for thefour seasons. The maximum SD appeared in the order as follows: spring>summer>fall >winter. Theminimum SD occurred in the following order: summer>spring>fall >winter.

Fig. 6. Monthly average distributions of the total dailySD in Ningxia in(a)January,(b)April,(c)July, and (d)October 2003.
4.3 Model verification

To estimate the accuracy of SD simulation forrugged terrains,we compared the observation dataat the 13 intensive observation weather stations inNingxia with the simulated SD results. This verification mainly focused on two aspects:(1)we verifiedthe error of the interpolation results at the st and ardstations using the SD data from the intensive observation stations in Ningxia;(2)we verified the error ofthe SD estimation using the remote sensing method(Fig. 3)at the intensive observation stations. To obtain the value at the grid point corresponding to thelatitude and longitude of each weather station,the effect on the errors of the geographic and topographicparameters was considered. The major errors includeprecision error in the DEM data and weather stationlocation error. Therefore,we adopted a previously described pixel method(Qiu et al., 2009)to extract thedata at the grid points corresponding to the weatherstations. As Table 2 shows,MAE of SD was 0.39–0.78h,MRE(monthly average relative error)of SD was5.41%–11.82%,the annual average absolute error was0.57 h, and the annual average relative error was approximately 7.9% for the 2003 data in Ningxia,basedon the interpolation method at the st and ard stations and the verification with the observational data at theintensive observation stations. Table 3 lists the SDsimulation errors derived by using the intensive station data and the MODIS cloud data. Table 3 alsoshows that the MAE of SD was 0.10–0.51 h,MRE ofSD was 1.47%–8.80%,the annual average absolute error was approximately 0.16 h, and the annual averagerelative error was approximately 2.21% for the 2003data in Ningxia. The error in Table 3 was generallymuch smaller than in Table 2,indicating that the introduction of the MODIS cloud data has increased theaccuracy of the SD estimation.

Table 2. Error analysis for the 2003 SD results derived by using the interpolation method based on the stationdata

Table 3. Error analysis for the 2003 SD estimates based on the MODIS cloud data
4.4 Sub-regional distributions of SD

To further analyze the effect of the local topography on SD and to highlight the variation of MPSD and SD with terrain,we examined the MPSD and SD anomalies,which are the differences between the gridpoint values for certain terrain aspects and the averageof all the grid points for all terrain aspects. The MPSDanomaly variations with terrain aspects for differentmonths are illustrated in Figs. 7a and 7b. The figures show that the MPSD anomalies for the 90°–180°aspect decreased in the order of January>October>April>July, and the order of MPSD anomalies forthe 0°–90° and 180°–360°aspects were July>April>October>January. Moreover,the MPSD variation onthe 15° southern terrain slope was larger than the 10°slope in January. The SD variation trend illustratedin Figs. 7c and 7d is similar to the trend in Figs. 7a and 7b. The cloud cover anomaly variations with aspects for different months are illustrated in Figs. 7e and 7f. These figures show that in January,the cloudvariation is as strong as in other months, and the cloudcover and the terrain have no obvious regularity. InJanuary,the direct sunshine point is located near theTropic of Capricorn,the maximum solar elevation angle is relatively small, and a general decreasing trendof SD was observed from south to north. The spatialdifference for SD in the mountainous areas was obvious, and a large deviation in SD was observed betweenthe south- and north-facing slopes.

Fig. 7. Anomaly variation curves of(a,b)MPSD,(c,d)SD, and (e,f)cloud cover,associated with terrain slope aspectof(a,c,e)10° and (b,d,f)15°. The curve in black is for January,in purple for April,in blue for July, and in green for October.
5. Discussion

Several studies have used satellite data to estimate solar radiation(Tarpley,1979; Gautier et al., 1980; Cano and Monget, 1986; Rigollier et al., 2004;Liu et al., 2008; Zhang and Wen, 2014). SD is stronglycorrelated with solar radiation, and it has been themost widely used variable to estimate solar radiation(Hu et al., 2010). Sahin et al.(2013)compared artificial neural network(ANN) and multiple linear regression(MLR)solar radiation estimation models inTurkey using NOAA/AVHRR data. Their results indicated that the ANN model achieved satisfactory performance compared to the MLR model. Moreover,satellite data improve the accuracy of solar radiationestimation results. Lu et al.(2011)proposed a simple and efficient algorithm to estimate the daily global solar radiation from geostationary satellite data. Theirresults demonstrated that the ANN model using geostationary satellite data showed acceptable accuracywith regard to both space and time.

We developed a remote sensing model to estimateSD using MODIS data. The annual average relativeerror was approximately 2.21%, and the relative interpolation error was approximately 7.90%. It is foundthat the use of satellite data to improve the model estimation accuracy led to significant improvement. Ourmodel only requires the use of three commonly available data(i.e.,DEM data,MODIS cloud cover data, and ground station observation data), and the acquisition of these data is generally feasible. Nonetheless,the current approach must be employed in more placesto provide a more generalized SD estimation model.

SD is observed at meteorological stations and used to estimate solar radiation. The distribution ofSD is significantly different in regions without meteorological stations or rugged mountains. Robaa(2008)developed three empirical formulae to estimate SD using readily available observational cloud cover data,which can only reflect the SD of these stations. Ourestimation clearly reflects the distribution of SD,especially the SD estimates for rugged terrains. Hu et al.(2010)developed a calculation method for SD at anygiven point within natural canopy gaps. They primarily studied SD among natural canopy gaps,whereas wefocused on SD estimation in rugged terrain areas. Sen and öztopal(2003)presented a method to group solar irradiation/SD data into convenient seasonal subgroups, and then made quantitative predictions withineach group. They sought to identify the biases causedin parameter estimates and to eliminate them withadditional consideration in solar energy calculation.6. Conclusions

Given the effects of terrain and atmospheric factors on SD,we used the Ningxia DEM data to providecomprehensive information about the local terrains, and used the MODIS cloud products for 2003 to simulate sunshine percentage and SD in this region. Byapplying this comprehensive SD model,we estimatedthe SD distributions in Ningxia and analyzed its seasonal and sub-regional variations. The major conclusions of this study are as follows:

(1)SD estimation is more accurate with the introduction of MODIS cloud data.

(2)The SD variation trend in Ningxia is consistent with that of the MPSD. The SD fluctuation islargely due to the effect of cloudy weather in April,July, and October; however,sunny days predominatein January, and the SD fluctuations in that month arerelatively mild.

(3)The SD spatial variation is so substantialthat the differences in the estimated SD between thesouth- and north-facing slopes are quite significant.Furthermore,the seasonal variation is also significantthroughout the year.

The SD estimation method applied in this studyonly requires the use of three commonly available data(i.e.,the DEM data,the MODIS cloud cover data, and the ground station observation data), and the acquisition of these data is generally feasible. Moreover,this estimation method could be of great significancefor other areas of research,such as solar-resource assessment,surface-radiation numerical simulation, and surface energy balance research. This method mightimprove the accuracy and reliability of SD computation, and may be of great potential value in varioussolar energy related applications.

References
[1] Angstrom, A., 1924: Solar and terrestrial radiation. Re-port to the International Commission for Solar Research on actinometric investigations of solar and atmospheric radiation. Quart. J. Roy. Meteor. Soc., 50, 121-126.
[2] Bertrand, C., C. Demain, and M. Journee, 2013: Es-timating daily sunshine duration over Belgium by combination of station and satellite data. Remote Sens. Lett., 4, 735-744.
[3] Bocquet, G., 1984: Method of study and cartography of the potential sunny periods in mountainous areas. J. Climatol., 4, 587-596.
[4] Cano, D., J. M. Monget, M. Albuisson, et al., 1986: A method for the determination of the global solar radiation from meteorological satellite data. Sol. Energy, 37, 31-39.
[5] Cao Yun, He Yongjian, Qiu Xinfa, et al., 2012: Correction methods of MODIS cloud product based on ground observation data. J. Remote Sens., 16, 325-342.
[6] Chen Hua, Sun Danfeng, Duan Zengqiang, et al., 2002: Sunshine hours modeling and spa-temporal variation analysis based on DEM in mountainous area鈥擳ake Mentougou mountain in West Beijing as the exam-ple. J. Mountain Res., 20, 559-563. (in Chinese)
[7] Ding Shouguo, Zhao Chunsheng, Shi Guangyu, et al., 2005: Analysis of global total cloud amount variation over the past 20 years. J. Appl. Meteor. Sci.,16, 670-676. (in Chinese)
[8] Dozier, J., and S. I. Outcalt, 1979: An approach toward energy balance simulation over rugged terrain. Geographical Analysis, 11, 65-85.
[9] Dubayah, R., J. Dozier, and F. W. Davis, 1990: Topographic distribution of clear-sky radiation over the Konza Prairie, Kansas. Water Resour. Res., 26,679-690.
[10] Ertekin, C., and F. Evrendilek, 2007: Spatio-temporal modeling of global solar radiation dynamics as a function of sunshine duration for Turkey. Agr. For-est Meteor., 145, 36-47.
[11] Fu Baopu, 1958: Effects of sloping fields on sunshine and solar radiation. J. Nanjing Univ. (Natural Science Edition), 2, 23-46. (in Chinese)
[12] Gautier, C., G. Diak, and S. Masse, 1980: A simple phys-ical model to estimate incident solar radiation at the surface from GOES satellite data. J. Appl. Meteor.,19, 1005-1012.
[13] Good, E., 2010: Estimating daily sunshine duration over the UK from geostationary satellite data. Weather,65, 324-328.
[14] Hetrick, W. A., P. M. Rich, F. J. Barnes, et al., 1993: GIS-based solar radiation flux models. ACSM-ASPRS Annual Convention, 3, 132-132.
[15] Hu, L., B. Yan, X. Wu, et al., 2010: Calculation method for sunshine duration in canopy gaps and its application in analyzing gap light regimes. For. Ecol. Manag., 259, 350-359.
[16] Kumar, L., A. K. Skidmore, and E. Knowles, 1997: Modeling topographic variation in solar radiation in a GIS environment. Int. J. Geogr. Inf. Sci., 11,475-497.
[17] Li Mengjie, Zheng Jianfei, Zeng Yan, et al., 2008: Dis-tributed modeling of direct solar radiation over rugged terrains of Zhejiang Province. Adv. Earth Sci., 23, 299-305. (in Chinese)
[18] Liu Ruixia, Chen Hongbin, Zheng Zhaojun, et al., 2009: Analysis and validation of total cloud amount data in China. J. Appl. Meteor. Sci., 20, 571-578. (in Chinese)
[19] Liu Yonghong, Quan Weijun, Xia Xiangao, et al., 2008: Net surface solar radiation in the clear sky based on MODTRAN model and satellite data. Plateau Meteor., 6, 1410-1415.(in Chinese)
[20] Lu, N., J. Qin, K. Yang, et al., 2011: A simple and efficient algorithm to estimate daily global solar radiation from geostationary satellite data. Energy,36, 3179-3188.
[21] Matuszko, D., 2012: Influence of cloudiness on sunshine duration. Int. J. Climatol., 32, 1527-1536.
[22] Moore, I. D., R. Grayson, and A. Ladson, 1991: Digital terrain modelling: A review of hydrological, geomorphological, and biological applications. Hydrol. Processes, 5, 3-30.
[23] Qiu Xinfa, Qiu Yueping, and Zeng Yan, 2009: Distributed modeling of monthly mean air temperature of rugged terrain of Chongqing. Adv. Earth Sci., 24, 621-628. (in Chinese)
[24] Rigollier, C., M. Lef`evre, and L. Wald, 2004: The method Heliosat-2 for deriving shortwave solar radiation from satellite images. Sol. Energy, 77, 159-169.
[25] Robaa, S., 2008: Evaluation of sunshine duration from cloud data in Egypt. Energy, 33, 785-795.
[26] Şahin, M., Y. Kaya, and M. Uyar, 2013: Comparison of ANN and MLR models for estimating solar radi-ation in Turkey using NOAA/AVHRR data. Adv. Space Res., 51, 891-904.
[27] Şen, Z., 2001: Angstrom equation parameter estimation by unrestricted method. Sol. Energy, 71, 95-107.
[28] Şen, Z., and A. Öztopal, 2003: Terrestrial irradiation sunshine duration clustering and prediction. Energ. Convers. Manage., 44, 2159-2174.
[29] Shamim, M. A., R. Remesan, D. W. Han, et al., 2012: An improved technique for global daily sunshine duration estimation using satellite imagery. J. Zhejiang Univ. Sci. A, 13, 717-722.
[30] Suehrcke, H., R. S. Bowden, and K. Hollands, 2013: Relationship between sunshine duration and solar radiation. Sol. Energy, 92, 160-171.
[31] Sun Hanqun and Fu Baopu, 1996: The elliptical integral model for computing the extra terrestrial solar ra-diation on mountain slopes. Acta Geogr. Sin., 51,559-566. (in Chinese)
[32] Takenaka, A., 1988: An analysis of solar beam penetra-tion through circular gaps in canopies of uniform thickness. Agr. For. Meteor., 42, 307-320.
[33] Tarpley, J. D., 1979: Estimating incident solar radiation at the surface from geostationary satellite data. J. Appl. Meteor., 18, 1172-1181.
[34] Wang Yu, 1993: Vertical distribution of sunshine duration in mountainous areas of Yunnan Province. J. Mountain Res., 11, 1-8. (in Chinese)
[35] Watson, R. T., and D. L. Albritton, 2001: Climate Change 2001: Synthesis Report. Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge,408 pp.
[36] Weng Duming, 1964: Overview of the climatological calculation methods of the total radiation. Acta Meteor. Sinica, 34, 304-315. (in Chinese)
[37] Weng Duming, Chen Wanglong, and Shen Juecheng, 1981: Microclimate and Farmland Microclimate. China Agriculture Press, Beijing, 357 pp. (in Chi-nese)
[38] Weng Duming, 1997: Studies on Radiation Climate of China. China Meteorological Press, Beijing, 363 pp. (in Chinese)
[39] Yin Hong, 1957: The use of empirical formula to calculate the total radiation distribution from the sunshine record in eastern China. Acta Meteor. Sinica, 28,101-107. (in Chinese)
[40] Yuan Shujie, Miao Qilong, Qiu Xinfa, et al., 2008: The spatial and temporal distributions of insolation du-ration over rugged terrains in the Guizhou Plateau. J. Appl. Meteor. Sci., 19, 233-237. (in Chinese)
[41] Zeng Yan, Qiu Xinfa, Miao Qilong, et al., 2003: Distri-bution of possible sunshine durations over rugged terrains of China. Prog. Nat. Sci., 13, 761-764.
[42] Zhang Chungui and Wen Mingzhang, 2014: Using satel-lite data to estimate solar radiation of clear sky over Fujian. J. Nat. Resour., 9, 1496-1507. (in Chinese)
[43] Zhu Zhihui, 1988: The global distribution of astronom-ical solar radiation on nonhorizontal surfaces. Sci. China (Ser. B), 10, 1100-1110.
[44] Zuo Dakang, 1990: Dictionary of Contemporary Geogra-phy. The Commercial Press, Beijing, 863 pp. (in Chinese)