Radiative transfer models (RTMs) designed to reproduce the anisotropy of surface brightness temperature (BT) are particularly useful for applications on Earth's energy budget when using remote sensing (RS) datasets. Despite the fact that several thermal infrared (TIR) RTMs have been developed, a quantitative analysis comparing the benefits and limits of these models remains necessary. Herein, three modeling frameworks (physical hybrid, analytical parameterization, and kernel driven) have been evaluated comparatively for homogeneous vegetation, a row-planted crop, and a sparse forest. Airborne measurements and the discrete anisotropy radiative transfer (DART) model simulations were retained as the benchmark. Forward modeling and inverse fitting schemes were proposed for the sake of comparison. Results reveal that: 1) in the forward modeling scheme, from airborne measurements, the hybrid model performs better with root-mean-squared errors (RMSEs) of 0.17 degrees C, 1.57 degrees C, and 0.38 degrees C for homogenous, row-planted vineyard, and sparse forest scenes, respectively; the analytical model appears similar performant (0.17 degrees C, 0.40 degrees C) for the homogeneous and sparse forest scenes, but less performant (2.39 degrees C) for the row-planted scene and 2) in the inverse fitting scheme, the uncertainties (95% of probability) of model coefficients and predicted directional anisotropies were considered. The kernel-driven model has fewer modeling constraints and statistically performs better for the homogeneous and sparse forest scenes with RMSEs of 0.07 degrees C and 0.19 degrees C, respectively, whereas it is less efficient for the row-planted scene with RMSE of 0.80 degrees C. This study highlights the differences in accuracy between models of different complexity and provides reference information for researchers to improve existing models and for users to choose their best modeling solution.