地表辐射收支是地球系统能量平衡的基础，同时还控制一系列天气，气候和海陆过程，影响着生物的繁衍生息，是一个极为重要的研究主题。地表辐射收支主要包含短波辐射收支和长波辐射收支两个部分，其中短波辐射直接来自于太阳辐射，目前对于它的认识已经较为成熟，而长波辐射主要来自地表物体自身辐射和大气层的辐射，牵涉变量极多，充满各种不确定性，存在很多亟待解决的问题。这其中，地表发射与大气下行辐射之差被称作地表有效辐射，它反映了地表损失能量的多少，与人类生产生活密切相关。随着遥感传感器的改进及遥感地表/大气产品的精度不断提高，为地表有效辐射的精确估算提供了支持，因此本文主要集中于地表有效辐射的遥感估算模型的建立与验证。本研究所提出的物理—统计混合模型主要基础是 MODTRAN4 对 TIGR 大规模辐射传输模拟所得到多个地气因子组成的模拟数据集。通过分析各个因子与地表有效辐射间的相关关系，并在因子间采用简单数学变换和组合后，筛选得到与地表有效辐射相关性最高的因子—地表温度和 955.1 hPa 下大气水汽压强，最终通过简单回归方法建立起三者间的回归关系。此外，研究还提出方法，通过分区域拟合来解决回归关系的区域性限制。采用美国 SURFRAD 通量观测站进行验证，结果表明，该方法对有效辐射估计有较高的精度，其R2 ，Bias 和 RMSE 分别为 0.628，8.699 W m-2 和 24.781 W m-2 。同时研究对 Bisht 等所提出算法进行改进，并应用到验证中，结果其 R2 ，Bias 和 RMSE 分别为 0.758，2.265 W m-2 和21.051 W m-2 ，夜间估计结果要好于日间估计。基于辐射传输模拟数据，研究采用机器学习方法 SVM 对地表有效辐射进行回归。采用六组不同参数组合分别进行拟合并选择结果最优一组。经过参数寻优，训练，测试和验证，选择 LAW-La 参数组合作为生成模型组合。同时，分析了每个参数的重要性及模型对于它们的敏感性。结果表明，地表温度最为重要，其次为大气水汽压和纬度，气温对模型贡献最小，敏感性方面，模型对于高估的温度和低估的大气水汽含量较为敏感，对于气温的误差几乎不敏感。验证表明其 R2 为0.4697，Bias 为 23.176 W m-2 ，RMSE 为 28.477 W m-2 。通过与混合方法交叉验证表明两者具有较好的一致性，R2 为 0.686，其差异主要集中在 4.0—7.0 W m-2 。最终，将模型应用于北京城市夜间热岛研究中，从辐射角度来认识夜间城市热岛的形成和发展。

Land surface radiation budget is the foundation of the energy balance in Earth system, controls a series of weather, climate and ocean-land processes, and influences the survival and propagation of living things, which is a crucial research topic. Two parts, short-wave radiation and long-wave radiation, make up the land surface radiation budget. The short-wave radiation is mainly from solar radiation, which has been known about very clearly, however, the long-wave radiation from land surface and atmospheric layers, being related with so many variables and full of many kinds of uncertainties, brings us many unknown problems which are needed to figure out. The term named effective radiation that the difference between the radiation emitted from land surface and emitted from atmosphere is introduced, which describes the loss of energy from land surface and is related with the human activities. As the remotely sensors and the accuracy of remotely sensed land/atmosphere products are being improved, accurate estimation of land surface effective radiation becomes possible. This paper concentrates on the development and validation of the remotely sensed estimation model of land surface effective radiation.This paper proposed a hybrid method integrating with physical and statistical processes. Based on the simulation from MODTRAN4 on TIGR atmospheric profiles, many variables were acquired in a dataset. Through the correlativity analysis between each variable and effective radiation, and some simple mathematical transformations, the most correlative variables-land surface temperature and atmospheric water vapor pressure at 955.1 hPa were selected as independent variables. In the end the non-linear regression equation was built. Additionally, based on latitude division, different coefficients corresponding different divisions were developed, solving the limitation from localization. Utilizing the observation results from SURFRAD in America for validation, the results indicate that this method has high accuracy, including R2, Bias and RMSE is 0.628, 8.699 W m-2 and 24.781 W m-2. Meanwhile, the method advanced by Bisht et al. was improved in this paper and the results become better than before, including R2, Bias and RMSE is 0.758, 2.265 W m-2 and 21.051 W m-2 respectively. Additionally, the estimation in nighttime is better than that in daytime.Based on the simulative data above, the machine leaning method-SVM was used for regression. Six different variables groups were used to select the best one. Through optimal parameters searching, training, testing and validation, LAW-La group was selected. Meanwhile, the importance of each variable in the group and their sensitivity for the model were analysized. The results indicate that land surface temperature is the first important variable, and the atmospheric water vapor pressure and latitude are the second and third. The air temperature has the fewest contributions. The model is more sensitive for the overestimated temperature and underestimated water vapor pressure. The validation indicates that its R2, Bias and RMSE are 0.4697, 23.176 W m-2 and 28.477 W m-2. The cross-validation with the method above indicates that both of two models have well consistency with 0.686 of R2 and the discrepancy concentrating at 4.0-7.0 W m-2. In the end the hybrid model is applied to the nocturnal urban heat island in Beijing research, from which we could know about the initiation and development mechanism of nocturnal urban heat island.