Urban forests are important for effectively mitigating urban heat island (UHI) effects. However, thorough investigations into how the three-dimensional (3D) structures of urban forests influences urban thermal conditions collectively and individually are limited. In this study, voxel-based landscape indices were innovatively extracted from UAV LiDAR data, and high-precision land surface temperature (LST) data were obtained using thermal infrared sensors mounted on a UAV. These were combined with a random forest (RF) model to analyze the relative influences and marginal effects of urban forest three-dimensional (3D) structure on LST. Our results showed the following: (1) The voxel-based landscape index exhibits a stronger capability to interpret LST than both the 2D landscape index and the gradient-based landscape index, with significant enhancements in model accuracy across all dimensions (an increase in R of 0.17–0.25 and a decrease in RMSE by 0.39–1.59°C). (2) Considering the vertical stratification of tree canopies, which voxel-based landscape index has the greatest LST fitting precision (R = 0.75, RMSE = 3.11°C). Including the canopy's vertical layers in analyses is pivotal, with the upper canopy layers exerting the most significant influence on reducing LSTs. (3) The scale of the grid impacts the accuracy of LST fitting, showing a trend where accuracy increases and then decreases with increasing grid scale; at the 40-m scale, the landscape indices demonstrate their highest explanatory capacity for LST (2D landscape index R=0.43, RMSE=4.65°C; gradient-based landscape index R=0.56, RMSE=4.07°C; voxel-based landscape index R=0.68, RMSE=3.94°C; vertical stratification (VS) voxel-based landscape index R= 0.75, RMSE= 3.30°C.). (4) Volume, proportion of volume, surface area, and diversity represent the parameters that most significantly influence variations in LST. Notably, volume, proportion of volume, and surface area exhibit a significant negative correlation with temperature, whereas diversity displays a distinct positive correlation. For the whole canopy at the optimal scale of 40 m, a volume within 4200 m3, proportion of volume within 0.8, and a surface area within 18000 m2 are associated with a cooling effect. For the upper canopy, volume within 1200 m3, proportion of volume within 0.22, and surface area within 2000 m2 are associated with a cooling effect. This study unequivocally confirms the feasibility of using drones with LiDAR and thermal infrared sensors to analyze small-scale UHI issues. This approach is beneficial for describing the 3D structure of a forest and fitting surface temperature. Urban planners can utilize these findings in practical applications by prioritizing forest configurations with optimal 3D structures in their planning efforts to effectively mitigate UHI effects. This research provides groundbreaking methods and highly reliable data to significantly deepen our understanding of the mechanisms behind the UHI effect.

中文翻译：

---

通过来自 UAV LiDAR 的新型基于体素的景观指数增强对地表温度的绿地表面的解释


城市森林对于有效缓解城市热岛 （UHI） 影响非常重要。然而，对城市森林的三维 （3D） 结构如何共同和单独影响城市热条件的深入研究是有限的。本研究创新性地从无人机 LiDAR 数据中提取基于体素的景观指数，并使用安装在无人机上的热红外传感器获得高精度地表温度 （LST） 数据。这些与随机森林 （RF） 模型相结合，以分析城市森林三维 （3D） 结构对 LST 的相对影响和边际效应。我们的结果表明如下：（1） 与二维景观指数和基于梯度的景观指数相比，基于体素的景观指数表现出更强的解释 LST 的能力，在所有维度上模型的准确性都有显著提高（R 增加 0.17-0.25，RMSE 降低 0.39-1.59°C）。（2） 考虑树冠的垂直分层，哪种基于体素的景观指数具有最大的 LST 拟合精度 （R = 0.75， RMSE = 3.11°C）。在分析中包括冠层的垂直层是关键的，上层冠层对减少 LST 的影响最为显着。（3） 网格的尺度影响 LST 拟合的精度，呈现精度随网格尺度增加后降低的趋势;在 40 m 尺度上，景观指数显示出它们对 LST 的最高解释能力（二维景观指数 R=0.43，RMSE=4.65°C;基于梯度的景观指数 R=0.56，RMSE=4.07°C;基于体素的景观指数 R=0.68，RMSE=3.94°C;垂直分层 （VS） 基于体素的景观指数 R= 0.75，RMSE= 3.30°C。 （4） 体积、体积比例、表面积和多样性代表了对 LST 变化最显着的参数。值得注意的是，体积、体积比例和表面积与温度呈显著的负相关，而多样性则表现出明显的正相关。对于40 m最佳尺度的整个冠层，体积在4200 m3以内，体积比例在0.8以内，表面积在18000 m2以内与冷却效果相关。对于上层树冠，体积在 1200 m3 以内，体积比例在 0.22 以内，表面积在 2000 m2 以内与冷却效果有关。这项研究明确证实了使用带有 LiDAR 和热红外传感器的无人机分析小规模 UHI 问题的可行性。这种方法有利于描述森林的 3D 结构和拟合表面温度。城市规划者可以在实际应用中利用这些发现，方法是在规划工作中优先考虑具有最佳 3D 结构的森林配置，以有效减轻 UHI 影响。这项研究提供了开创性的方法和高度可靠的数据，以显着加深我们对 UHI 效应背后机制的理解。