针对静止轨道热红外单角度观测下观测天顶角影响导致的空间一致性和时序可比性降低的问题,提出1种面向林草植被覆盖区的地表温度角度归一化方法。以黑河流域地表过程综合观测网为研究区,基于风云四号B星(FY-4B)地表温度产品,观测几何参数及多维地表特征,结合方向性发射率模型与传感器光谱响应函数,将地表温度转换为热辐射,采用轻量级梯度提升机(LightGBM)模型建立热辐射与观测几何及地表属性之间的非线性映射关系;将观测天顶角设定为近垂直参考角度,其余地表特征保持不变,从而获得同一观测条件下的参考热辐射;为降低跨传感器差异对验证结果的影响,引入近垂直观测的Landsat-8/9热红外数据,并进行跨传感器偏差校正后作为独立参考。结果表明:1)模型能够较好地刻画热辐射对观测几何和地表属性的综合响应,训练集和验证集RMSE分别为0.241和0.263,R²分别为0.972和0.966,Bias均接近于0;2)跨传感器偏差校正后,FY-4B与Landsat参考热辐射的一致性明显提高,RMSE由0.421降至0.221,Bias由0.204降至0.020,R²由0.193提高至0.777;3)2023年8月27—29日多期实验中,角度归一化后FY-4B热辐射与校正后Landsat参考热辐射的散点分布整体向1∶1线收敛,RMSE均明显降低,且观测天顶角较大区域的修正幅度更为显著;4)国际地圈生物圈计划(IGBP)土地覆盖类型分组结果显示,稠密草原、森林等冠层结构复杂、空间异质性较强的地类归一化后精度改善更明显,而裸地等相对均一地表的改善幅度有限,表明该方法的校正效果与不同地类热辐射方向性强弱具有一致性。地表温度角度归一化方法无需多角度同步观测数据,能够在保持地表温度空间格局基本稳定的基础上有效削弱FY-4B地表温度产品中的角度效应,为静止轨道热红外数据的多时相分析和定量应用提供了可行技术途径。

To address the reduced spatial consistency and temporal comparability of land surface temperature(LST)products caused by view zenith angle(VZA)effects under single-angle geostationary thermal infrared observations,this study proposes an angular normalization method for LST over forest and grassland areas.Taking the integrated observatory network for land surface processes in the Heihe River Basin as the study area,this study utilized Fengyun-4 B(FY-4B)LST products,viewing geometry parameters,and multi-dimensional surface features.Based on a directional emissivity model and the spectral response function of the sensor,LST was first converted into thermal radiance.A Light Gradient Boosting Machine(LightGBM)model was then employed to establish the nonlinear relationship between thermal radiance,viewing geometry,and surface properties.The observed VZA was replaced with a near-nadir reference angle while the other surface features remained unchanged,to obtain reference thermal radiance under a unified viewing condition.To reduce the influence of cross-sensor differences on validation results,near-nadir Landsat-8/9 thermal infrared data were introduced and used as an independent reference after cross-sensor bias correction.The results showed that:1)The model effectively characterized the combined response of thermal radiance to viewing geometry and surface properties,with root-mean-square error(RMSE)values of 0.241 and 0.263 and coefficients of determination(R²)of 0.972 and 0.966 for the training and validation datasets,respectively,while the bias values were close to zero;2)After cross-sensor bias correction,the consistency between FY-4B and Landsat reference thermal radiance was substantially improved,with RMSE decreasing from 0.421 to 0.221,bias decreasing from 0.204 to 0.020,and R² increasing from 0.193 to 0.777;3)In the multi-date experiments from 27 to 29 August 2023,the scatter distribution of the angular-normalized FY-4B thermal radiance and corrected Landsat reference radiance converged towards the 1∶1 line,and RMSE values were markedly reduced,with larger correction magnitudes observed in areas with greater VZA;and 4)The International Geosphere-Biosphere Programme(IGBP)land-cover grouping analysis indicated that types with more complex canopy structures and stronger spatial heterogeneity,such as dense grassland and forest,showed more evident accuracy improvements after angular normalization,whereas relatively homogeneous surfaces such as bare land exhibited limited improvement,indicating that the correction effect of this method is consistent with the strength of the thermal radiance directionality of different land cover types.The LST angular normalization method does not require synchronous multi-angle observation data.It can effectively weaken the angular effects in FY-4B LST products while maintaining the basic stability of the LST spatial pattern,providing a feasible technical approach for multi-temporal analysis and quantitative applications of geostationary thermal infrared data.