Abstract When using the kernel-driven bidirectional reflectance distribution function (BRDF) model to process multi-angular measurements, the input multi-angular measurements must be corrected for atmospheric effects. However, in current databases, a significant number of ground-based multi-angular measurements contain either no corrections or only approximate corrections for atmospheric effects. Thus, the blended diffuse light in the total incident irradiance will result in considerable smoothing of the reflectance anisotropy retrieved by the kernel-driven model unless an atmospheric correction process is conducted. In this study, we propose a diffuse-light correction (DLC) form of the kernel-driven model that improves its ability to process multi-angular measurements blended with hemispherical diffuse light. The DLC form of the kernel-driven model can be used to retrieve the intrinsic reflectance anisotropy of the observed target from atmospheric-uncorrected multi-angular measurements. This study used multi-angular data simulated by the PROSAIL and Radiosity Applicable to Porous IndiviDual objects (RAPID) BRDF model, atmospheric-corrected Polarization and Directionality of the Earth's Reflectances (POLDER), Cloud Absorption Radiometer (CAR) multi-angular measurements and their simulated data based on the Second Simulation of the Satellite Signal in the Solar Spectrum (6S) tools to validate the effectiveness of the DLC form of the kernel-driven model. The results indicated that the reflectance factors directly retrieved by the kernel-driven model are considerably smoothed by the blended diffuse light, especially in hotspot regions. Even under clear and cloudless sky conditions, the retrieved hotspot reflectance in the red band is still underestimated by an average of 9.25%, 7.72%, 11.0% and 13.8% for the PROSAIL, RAPID, POLDER and CAR data, respectively. In contrast, the hotspot reflectance retrieved by the DLC form of the kernel-driven model is very close to the intrinsic reflectance anisotropy of the targets; the average relative error of the DLC form of the kernel-driven model is only 1.99%, 1.50%, 4.57% and 3.42%, respectively. Although the reflectance reconstructed by the DLC form of the kernel-driven model in the hotspot region represents a considerable improvement compared with the reflectance retrieved by the original kernel-driven model, its improvement on the root mean square error (RMSE) and the bias of the entire datasets is not very apparent. Using the DLC form of the kernel-driven model can significantly improve the ability of the kernel-driven model to process multi-angular measurements blended with hemispherical diffuse irradiance.

中文翻译：

---

用于校正地面多角度测量漫射光的内核驱动模型的修改版本

摘要 当使用核驱动的双向反射分布函数（BRDF）模型处理多角度测量时，必须对输入的多角度测量进行大气效应校正。然而，在当前的数据库中，大量基于地面的多角度测量结果要么不包含任何修正，要么仅包含对大气影响的近似修正。因此，除非进行大气校正过程，否则总入射辐照度中的混合漫射光将导致核驱动模型检索的反射各向异性相当平滑。在这项研究中，我们提出了内核驱动模型的漫射光校正 (DLC) 形式，以提高其处理与半球漫射光混合的多角度测量的能力。内核驱动模型的 DLC 形式可用于从大气未校正的多角度测量中检索观测目标的固有反射各向异性。本研究使用了 PROSAIL 模拟的多角度数据和适用于多孔单个物体 (RAPID) BRDF 模型的多角度数据、地球反射率的大气校正极化和方向性 (POLDER)、云吸收辐射计 (CAR) 多角度测量及其基于太阳光谱中卫星信号的第二次模拟 (6S) 工具的模拟数据，以验证内核驱动模型的 DLC 形式的有效性。结果表明，核驱动模型直接检索的反射系数被混合漫射光显着平滑，特别是在热点区域。即使在晴朗无云的天空条件下，PROSAIL、RAPID、POLDER和CAR数据在红波段反演的热点反射率仍然分别被平均低估了9.25%、7.72%、11.0%和13.8%。相比之下，内核驱动模型的DLC形式检索到的热点反射率非常接近目标的固有反射率各向异性；内核驱动模型的DLC形式的平均相对误差分别仅为1.99%、1.50%、4.57%和3.42%。虽然在热点区域由内核驱动模型的 DLC 形式重建的反射率与原始内核驱动模型检索的反射率相比有相当大的改进，但其在均方根误差 (RMSE) 和偏差上的改进整个数据集不是很明显。