A General Framework of Kernel-driven Modeling in the Thermal Infrared Band for Land Surface Temperature Normalization

Land surface temperature (LST) is the direct driving force of turbulent heat fluxes at the surface and atmosphere interface and is widely used in the fields of evapotranspiration estimation (Su et al., 2002) and energy budget (Liang et al., 2019). Remote sensing products offer the only possibility for measuring LST with completely spatially averaged values. The thermal radiation directionality (TRD) effect has been widely concerned in the area of thermal infrared (TIR) remote sensing over 50 years which can lead to the directional brightness temperature (DBT) difference between different viewing directions up to 10 K (Cao et al., 2019). Many models have been proposed to simulate the DBT patterns over different underlying surfaces aimed to achieve the TRD effect correction for the satellite LST products. In practice, it is advised to handle only TRD models having a limited number of input parameters for operational normalization of LST products. The use of TIR kernel-driven models appears a good tradeoff between physical accuracy and operationality. It remains that the existing 4 TIR kernel-driven models (Ross-Li, LSF-Li, Vinnikov, RL) underestimate the hotspot effect, especially for continuous canopies. In this study, a new general framework of TIR kernel-driven modeling is proposed to overcome such issue. It is a linear combination of three kernels (including a base shape kernel, a hotspot kernel and an isotropic kernel) with the ability to simulate the bowl, dome and bell shapes in the solar principal plane. 4 specific models (Vinnikov-RL, LSF-RL, Vinnikov-Chen, LSF-Chen) within the new framework were further developed to assess their fitting abilities for both continuous and discrete vegetation canopies. To evaluate 4 existing models and 4 new models comprehensively, it was prepared 102 groups of 4SAIL/DART generated multi-angle datasets considering 6 different canopy architectures and 17 component temperatures. Results show that the 4 new models behave slightly better than the 4 existing models over discrete canopies (R2 increases from 0.791~0.989 to 0.976~0.996) whereas they significantly improved the fitting accuracy over continuous canopies (R2 increases from 0.661~0.970 to 0.940~0.997). The innovative new general framework with three kernels and four parameters improve the fitting ability significantly since the addition of one more degree of freedom. This new kernel-driven modeling framework is a potential tool to achieve angular correction of LST products.