The ability of a forest to buffer understory temperature extremes depends on the canopy structure, which is often measured from the ground. However, ground measurements provide only point estimates, which cannot be used for spatially explicit microclimate modeling. Canopy structures derived from airborne light detection and ranging (LiDAR) can overcome these limitations, but high point-density LiDAR is expensive and computationally challenging. Therefore, we explored whether unmanned aerial systems (UAS) processed with the structure-from-motion (SfM) algorithm could serve as an alternative source of canopy variables for forest microclimate modeling. Specifically, we compared the performance of the canopy cover and height derived from the ground measurements and passive (UAS-SfM) and active (UAS-LiDAR) remote sensing as predictors of air and soil temperature offsets (i.e. differences between the forest understory and treeless areas).

We found that the maximum air temperatures were substantially lower inside than outside the forest, with differences ranging from 9.0 to 12.5 °C. The soil temperatures under the canopy were also reduced, but the soil temperature offsets were lower and ranged from 1.1 to 2.8 °C. The air and soil temperature offsets both increased with increasing tree height and canopy cover. However, the prediction ability of tree height and canopy cover differed if they were ground-based or remotely sensed. The remotely sensed canopy indices explained air temperature offsets better (UAS-SfM: R² = 0.59, RMSE = 0.75 °C; UAS-LiDAR: R² = 0.57, RMSE = 0.76 °C) than ground measurements (R² = 0.51, RMSE = 0.80 °C). Ground-based metrics explained soil temperature offsets better (R² = 0.37, RMSE = 0.36 °C) than passive remote sensing approach (UAS-SfM: R² = 0.27, RMSE = 0.39 °C), but comparably to active one (UAS-LiDAR: R² = 0.35, RMSE = 0.37 °C).

Our results suggest that both UAS-SfM and UAS-LiDAR can substitute ground canopy measurements for air temperature modeling, but soil temperature modeling is more challenging. Overall, our results show that forest microclimate can be modelled at a very high spatial resolution using UAS equipped with inexpensive optical cameras. The increasingly available UAS-SfM approach can thus provide fine-resolution microclimatic data much needed for biologically relevant predictions of species responses to climate change.

森林缓冲林下极端温度的能力取决于树冠结构，这通常是从地面测量的。然而，地面测量仅提供点估计，不能用于空间明确的小气候建模。源自机载光探测和测距 (LiDAR) 的树冠结构可以克服这些限制，但高点密度 LiDAR 价格昂贵且计算量大。因此，我们探讨了使用结构自运动 (SfM) 算法处理的无人机系统 (UAS) 是否可以作为森林微气候建模的冠层变量的替代来源。具体来说，

我们发现森林内部的最高气温明显低于森林外部，差异范围从 9.0 到 12.5 °C。冠层下的土壤温度也降低了，但土壤温度偏移量较低，范围为 1.1 至 2.8 °C。空气和土壤温度偏移量都随着树高和树冠盖度的增加而增加。然而，无论是地面还是遥感，对树高和冠层盖度的预测能力都不同。遥感冠层指数 比地面测量更好地解释了空气温度偏移（UAS-SfM：R ²  = 0.59，RMSE = 0.75 °C；UAS-LiDAR：R ² = 0.57，RMSE = 0.76 °C）比地面测量（R ²  = 0.51， RMSE = 0.80 °C）。基于地面的指标更好地解释了土壤温度偏移（R²  = 0.37，RMSE = 0.36 °C）比被动遥感方法（UAS-SfM：R ²  = 0.27，RMSE = 0.39 °C），但与主动遥感方法（UAS-LiDAR：R ²  = 0.35，RMSE = 0.37 ℃）。

我们的结果表明 UAS-SfM 和 UAS-LiDAR 都可以替代地面冠层测量来进行气温建模，但土壤温度建模更具挑战性。总体而言，我们的结果表明，可以使用配备廉价光学相机的 UAS 以非常高的空间分辨率对森林小气候进行建模。因此，越来越可用的 UAS-SfM 方法可以提供精细分辨率的小气候数据，这些数据非常需要生物相关预测物种对气候变化的反应。