Parametric kernel-driven models are crucial for operationally adjusting satellite-derived urban land surface temperatures (LSTs) obtained at slant angles to hemispherically-representative values. Various parametric models have been proposed to simulate urban thermal radiation directionality, but a comprehensive comparison of the performances of the published parametric models, especially over a variety of urban surfaces under different solar radiation conditions, remains lacking. It is also unknown whether the combination of the available hotspot and base shape kernels can be used to derive new parametric models with even better performances compared with existing models. Based on both forward-modelling and satellite datasets, here we systematically evaluate three single-kernel and eight dual-kernel parametric models. The main findings are as follows: (1) Amongst the three single-kernel models, the VIN model has the best overall performance, with an average root-mean-square error (RMSE) of 0.79 and 1.35 K, based on forward-modelling and satellite data, respectively. However, the ROU and RL models outperform the VIN model when the solar zenith angle is less than 30°, and in particular it has a higher accuracy for hotspot description. (2) The dual-kernel models usually perform better than the single-kernel models. Amongst the eight dual-kernel models, those with the hotspot kernel K_{Hotspot_rou} (used by the ROU model) are more competent than those using K_{Hotspot_vin} (obtained from the Vinnikov model) as the hotspot kernel. The RVI model, in general, has the highest accuracy, with average RMSEs of 0.49 and 0.77 K based on forward-modelling and satellite data, respectively. (3) Compared with the single- and dual-kernel models, the multi-kernel models sometimes have better accuracies but the performance improvements are relatively limited. We also provide recommendations for model selection under various scenarios. Our systematic assessment improves our understanding of urban thermal radiation directionality regimes and potentially enables the improved correction of remotely-sensed urban LSTs, thus helping to advance thermal remote sensing of the urban environment.

参数内核驱动模型对于将卫星衍生的城市地表温度 (LST) 以倾斜角获得的操作调整为半球代表性值至关重要。已经提出了各种参数模型来模拟城市热辐射方向性，但仍然缺乏对已发表参数模型的性能的综合比较，尤其是在不同太阳辐射条件下的各种城市表面。与现有模型相比，是否可以使用可用热点和基本形状内核的组合来推导出具有更好性能的新参数模型，这也是未知的。基于前向建模和卫星数据集，我们系统地评估了三个单核和八个双核参数模型。主要发现如下： (1) 在三个单核模型中，VIN 模型的综合性能最好，平均均方根误差 (RMSE) 为 0.79 和 1.35 K，基于前向建模和卫星数据。然而，当太阳天顶角小于 30°时，ROU 和 RL 模型优于 VIN 模型，特别是它具有更高的热点描述精度。(2) 双内核模型通常比单内核模型性能更好。在八种双内核模型中，带有热点内核的模型 当太阳天顶角小于 30°时，ROU 和 RL 模型优于 VIN 模型，特别是它具有更高的热点描述精度。(2) 双内核模型通常比单内核模型性能更好。在八种双内核模型中，带有热点内核的模型 当太阳天顶角小于 30°时，ROU 和 RL 模型优于 VIN 模型，特别是它具有更高的热点描述精度。(2) 双内核模型通常比单内核模型性能更好。在八种双内核模型中，带有热点内核的模型K _{Hotspot_rou}（由 ROU 模型使用）比使用K _{Hotspot_vin 的}人更胜任（从 Vinnikov 模型获得）作为热点内核。RVI 模型通常具有最高的精度，基于前向建模和卫星数据的平均 RMSE 分别为 0.49 和 0.77 K。(3) 与单核和双核模型相比，多核模型有时具有更好的精度，但性能提升相对有限。我们还提供了各种场景下模型选择的建议。我们的系统评估提高了我们对城市热辐射方向性状况的理解，并有可能改进遥感城市 LST 的校正，从而有助于推进城市环境的热遥感。