干旱气象 ›› 2026, Vol. 44 ›› Issue (1): 103-114.DOI: 10.11755/j.issn.1006-7639-2026-01-0103

• 论文 • 上一篇    下一篇

基于机器学习订正ERA5的甘肃省地表太阳辐射时空分布

吴欣华1(), 王思晨2, 王菲菲3, 王天河2(), 杜源1, 陈涛1, 牛亮亮1, 赵怀宇1, 张昊天1   

  1. 1.甘肃省交通投资管理有限公司,甘肃 兰州 730000
    2.兰州大学大气科学学院,甘肃 兰州 730000
    3.甘肃省水利水电勘测设计研究院有限责任公司,甘肃 兰州 730000
  • 收稿日期:2025-09-01 修回日期:2025-11-19 出版日期:2026-02-28 发布日期:2026-03-25
  • 通讯作者: 王天河(1980—),男,甘肃静宁人,博士,教授,主要从事大气遥感、气溶胶-云-气候相互作用等研究。E-mail: wangth@lzu.edu.cn
  • 作者简介:吴欣华(1978—),男,甘肃武威人,高级工程师,主要从事交通运输与绿色能源研究。E-mail: 18551810@qq.com
  • 基金资助:
    甘肃省交通运输厅科技项目(2022-56);中国气象局青年创新团队(CMA2024QN13);天池英才引进计划(2024)

Spatiotemporal distribution of surface solar radiation in Gansu Province based on machine learning correction of ERA5

WU Xinhua1(), WANG Sichen2, WANG Feifei3, WANG Tianhe2(), DU Yuan1, CHEN Tao1, NIU Liangliang1, ZHAO Huaiyu1, ZHANG Haotian1   

  1. 1. Gansu Communication Investment Management Company Limited,Lanzhou 730000,China
    2. College of Atmospheric Sciences,Lanzhou University,Lanzhou 730000,China
    3. Gansu Water Resources and Hydropower Survey Design and Research Institute Company Limited,Lanzhou 730000,China
  • Received:2025-09-01 Revised:2025-11-19 Online:2026-02-28 Published:2026-03-25

摘要:

准确掌握地表太阳辐射的时空分布特征,对太阳能资源评估与区域新能源规划具有重要意义。以甘肃省地基辐射站点观测数据为基准,构建机器学习模型,对欧洲中期天气预报中心第五代再分析资料(ERA5)的逐小时地表下行太阳辐射进行偏差订正,在此基础上系统分析了2000—2024年甘肃省地表下行太阳辐射的时空变化特征,并统计了各地级行政区年总辐射量。结果表明:机器学习订正方法显著提升ERA5数据精度,订正后数据与地基观测值的相关系数(0.93)提高12.04%,均方根误差(106.2 W·m-2)降低36.45%;与中国科学院空天信息创新研究院发布的CARE(Cloud Remote Sensing,Atmospheric Radiation and Renewal Energy Application)卫星遥感产品对比,二者相关系数达0.87,偏差主要分布在青藏高原东北侧。研究期内,甘肃省地表下行太阳辐射年均值为206.73 W·m-2,折合年累计总辐射量为1 659.60 kWh·m-2,高于全国平均水平,空间上呈“西北高、东南低”的分布格局,其中酒泉地区可达1 828.44 kWh·m-2,具备优越的太阳能开发潜力,且全省未呈现明显的年际波动趋势。

关键词: 地表下行太阳辐射, 机器学习, 数据订正, 时空分布

Abstract:

Accurately characterizing the spatiotemporal distribution of surface solar radiation is crucial for solar energy resource assessment and regional renewable energy planning. In this study, ground-based radiation observations in Gansu Province were used as the reference to bias-correct the hourly surface downward solar radiation from the fifth-generation ECMWF (European Centre for Medium-Range Weather Forecasts) reanalysis (ERA5) using a machine learning approach. Based on the corrected data, the spatiotemporal variability of surface downward solar radiation in Gansu Province during 2000-2024 was systematically analyzed, and annual cumulative radiation totals are quantified for each prefecture-level administrative region. The results demonstrate that the machine learning-based method significantly improves the accuracy of the ERA5. The correlation coefficient between the corrected data and ground observations increases by 12.04%, while the root mean square error decreases by 36.45%. Compared with the CARE (Cloud Remote Sensing, Atmospheric Radiation and Renewal Energy Application) satellite remote sensing product released by the Aerospace Information Research Institute, Chinese Academy of Sciences, the correlation coefficient between them reaches 0.87, and the remaining biases are mainly concentrated along the northeastern margin of the Tibetan Plateau. Over the study period, the provincial mean surface downward solar radiation is 206.73 W·m-2, corresponding to an annual cumulative total of 1 659.60 kWh·m-2, which is higher than the national average. Spatially, the radiation exhibits a distinct pattern of being higher in the northwest and lower in the southeast. The radiation in Jiuquan area reached 1 828.44 kWh·m-2, indicating excellent solar energy development potential. Moreover, no significant interannual fluctuation trend was observed across the province.

Key words: surface downward solar radiation, machine learning, data correction, spatiotemporal distribution

中图分类号: