在过去的几十年中，化石燃料等不可再生能源的使用造成了一系列的生态环境问题，而全球的能源需求依旧在逐步增加。要满足全球能源供应并有效缓解生态环境问题，发展可再生能源来替代不可再生能源是最有效，也是最佳的选择。随着各国对可持续思想的贯彻和政策支持，可再生清洁能源逐步得到了开发和应用，而太阳能作为地球的主要能源，以其独特优势成为了可再生能源发展的增长极，太阳能光伏发电也随之飞速发展。对光伏发电潜力进行评估则有利于进行光伏发电系统的部署，对我国的能源规划与发展具有重要意义。

本研究运用站点数据、遥感产品数据及再分析数据进行中国光伏发电潜力的评估，以容量因子表示并进行时空变化特征的分析。首先基于站点辐射和气象观测数据，通过机器学习算法估算得到站点长时间序列的散射辐射；其次基于得到的站点散射辐射和地表入射太阳辐射计算得到散射辐射比，连同地表入射太阳辐射和提取的温度数据输入GSEE模型中估算得到中国长时间序列的光伏发电潜力，并进行相应的分析；最后基于中国高时空分辨率遥感辐射产品以及温度数据，运用GSEE模型估算得到空间连续的光伏发电潜力，分析其特征并与站点估算结果进行对比分析，以期更加合理地探索中国光伏发电潜力的时空变化特征。具体的研究内容与结论如下：

（1）收集中国气象局839个站点多年的每日气象观测数据和辐射观测数据，结合提取的MERRA2云和气溶胶数据，通过三种机器学习算法构建模型估算散射辐射，并基于站点实测数据对模型估算结果进行验证。综合在训练、验证和测试数据集上的结果，选择随机森林构建的模型作为最优模型，进而扩展估算得到长时间序列站点每日散射辐射。运用站点观测数据对该估算结果与地表五公里每日散射辐射进行验证对比，得出站点估算结果的R值为0.957，RMSE为12.625W/m²，MBE为-1.044W/m²，相比于地表五公里散射辐射遥感产品，其精度良好，用于后续研究。

（2）将筛选后数据连续的609个站点的散射辐射以及地表入射太阳辐射重采样为每小时分辨率并计算得到散射辐射比，将地表入射太阳辐射、散射辐射比以及提取的MERRA2温度数据输入GSEE模型中，设置参数估算得到中国1980-2019年基于站点的每小时光伏发电潜力，并在估算结果的基础上进行模型的敏感性分析。结果显示，在其余变量保持不变的条件下，地表入射太阳辐射越大，散射辐射比和温度越小，光伏发电潜力越大。基于得到的站点光伏发电潜力进行时空变化分析，并探究城市化对光伏发电潜力的影响，结果表明我国的光伏发电潜力大体呈现西北向东南递减的趋势，总体而言夏季最高，冬季最低。全国的发电潜力在1980-2019年呈现每十年6×10^-4的下降趋势，且超过半数的站点呈现下降趋势，而城市化会加速光伏发电潜力的下降，也会使得光伏发电潜力相对更低。

（3）基于较为完整的2012年每小时地表入射太阳辐射及散射辐射遥感产品数据，可以计算得到散射辐射比，并结合地表入射太阳数据和MERRA2温度数据输入GSEE模型中，估算得到中国每小时的高时空分辨率光伏发电潜力。分析可知，基于遥感产品估算得到的光伏发电潜力与基于站点得到的结果在空间上分布相似，都为西北高、东南低，最低值分布在四川、重庆及贵州地区。在季节上同样是夏季发电潜力高值更多，冬季低值分布更广。二者区别在于，基于遥感产品得到的光伏发电潜力高值相对更高（大于0.3）。通过站点估算得到的每日光伏发电潜力对基于遥感产品得到的结果进行验证对比也可以得出，2012年光伏发电潜力的R、RMSE和MBE分别为0.890、0.036及0.018，总体而言基于遥感产品得到的结果呈高估现象。

综上所述，本文分别基于站点数据和遥感产品数据评估了中国的光伏发电潜力并进行了时空变化特征的分析研究，得到了中国光伏发电潜力变化的基本规律，可以作为合理规划中国光伏发电产业的参考依据，也有助于能源的合理开发和应用。

Over the past few decades, the consumption of non-renewable resources, such as fossil fuels, have caused a series of ecological and environmental problems, while the demand of global energy continues to increase. To meet the supply of global energy and effectively alleviate the ecological and environmental issues, the most effective and optimal choice is to develop renewable energy to achieve the replacement of fossil resources. With the implementation of sustainable concepts and policy support, renewable and clean sources have been gradually developed and applied. As the main energy of the earth, the solar energy has become the growth pole of renewable energy development with its unique advantages, and the solar photovoltaic power generation has also developed rapidly. Evaluating the photovoltaic power generation potential is conducive to the deployment of photovoltaic systems and is of great significance to the energy planning and development of China.

In this research, the photovoltaic power generation potential (represented by capacity factor (CF)) was evaluated using datasets at stations, remote sensing products and reanalysis products, and its spatial and temporal variation characteristics in China were also analyzed. Firstly, based on the radiation and meteorological measurements at stations, a machine learning algorithm was used to estimate the long-term diffuse radiation. Secondly, the ratio of diffuse radiation at stations were calculated and input into the global solar energy estimator (GSEE), and the obtained radiation and extracted temperature at stations were also input into GSEE to estimate the long-term CF in China. The corresponding analysis was carried out. Finally, the CF with spatial continuity was also estimated by GSEE, using high spatial and temporal resolution remote sensing products and temperature dataset, and the results were compared and analyzed to explore the characteristics of CF in China. The detailed contents and conclusions of this research are as follows:

(1) The daily meteorological observations and radiation data from 839 stations of the Climate Data Center of the Chinese Meteorological Administration (CDC/CMA) were collected. Combined with the cloud and aerosol data extracted from MERRA2, these collected datasets were applied to build the model for estimating the surface diffuse incident solar radiation (DISR) using three machine learning algorithms, and the estimations were validated against the ground measurements. Considering the results from training, validating and test datasets, the random forest was selected as the best performing model to estimate the long-term daily DISR at stations. Then, ground observations were employed to validate the DISR we estimated and the DISR from remote sensing product with five-kilometer resolution. The R, RMSE and MBE of DISR at stations were 0.957, 12.625W/m² and -1.044W/m², respectively. Compared to remote sensing products, the precision of the estimated results at stations is better and can be used for subsequent research.

(2) The surface incident solar radiation (ISR) and the DISR with complete records of 609 stations were resampled to hourly resolution and then the ratio of DISR was calculated. The ISR, ratio of DISR and the extracted MERRA2 temperature were input into GSEE, then we set the parameters to calculate the hourly CF of each station in China from 1980 to 2019. The sensitivity of the GSEE was also analyzed using these input variables. The results indicated that the CF was greater with higher ISR, smaller ratio of DISR and lower temperature, with other variables being constant. Based on the estimated results at stations, the variation of spatial and temporal characteristics and the urbanization impacts on CF were analyzed. It was found that the CF generally decreased from northwest to southeast of China, with the highest CF in summer and the lowest in winter. During the period of 1980-2019, the average CF showed a downward trend of 6×10^-4 per decade, and over half of the stations were in decreasing trends. The urbanization speeded up the decreasing of CF, and resulted in relatively lower CF.

(3) Based on the hourly ISR and DISR which were relatively complete in 2012 from remote sensing products, the ratio of DISR can be calculated. The ISR, ratio of DISR, and the corresponding MERRA2 temperature were input into GSEE to estimate the CF with high spatial-temporal resolution. The spatial distribution of estimated CF based on remote sensing products was similar to that based on datasets at stations, which were both higher in northwest China and lower in the southeast (the lowest CF values were distributed in Sichuan, Chongqing and Guizhou). In terms of seasonality, high values of CF were more common in summer, while the low values were more widespread in winter. The difference was that the higher value of CF based on remote sensing products was relatively higher (greater than 0.3). We evaluated the CF with spatial continuity using the CF at stations and found that the R, RMSE and MBE of CF in 2012 were 0.890, 0.036 and 0.018, respectively. Overall, the estimated results based on remote sensing products were overestimation.

In summary, this study evaluated the photovoltaic power generation potential in China based on data at stations and remote sensing products, and conducted spatial and temporal analysis of its characteristics. The basic rules of photovoltaic power generation potential variation characteristics in China were obtained, which can serve as a reference for the rational planning of photovoltaic power generation industry and the development and application of energy in China.