In the past, studies have successfully identified climatic controls on the temporal variability of the land surface phenology (LSP). Yet we lack a deeper understanding of the spatial variability observed in LSP within a land cover type and the factors that control it. Here we make use of a high resolution LiDAR based dataset to study the effect of subpixel forest stand characteristics on the spatial variability of LSP metrics based on MODIS NDVI. Multiple linear regression techniques (MLR) were applied on forest stand information and topography derived from LiDAR as well as land cover information (i.e. CORINE and proprietary habitat maps for the year 2012) to predict average LSP metrics of the mountainous Bavarian Forest National Park, Germany. Six different LSP metrics, i.e. start of season (SOS), end of season (EOS), length of season (LOS), NDVI integrated over the growing season (NDVIsum), maximum NDVI value (NDVImax) and day of maximum NDVI (maxDOY) were modelled in this study. It was found that irrespective of the land cover, the mean SOS, LOS and NDVIsum were largely driven by elevation. However, inclusion of detailed forest stand information improved the models considerably. The EOS however was more complex to model, and the subpixel percentage of broadleaf forests and the slope of the terrain were found to be more strongly linked to EOS. The explained variance of the NDVImax improved from 0.45 to 0.71 when additionally considering land cover information, which further improved to 0.84 when including LiDAR based subpixelforest stand characteristics. Since completely homogenous pixels are rare in nature, our results suggest that incorporation of subpixel forest stand information along with land cover type leads to an improved performance of topography based LSP models. The novelty of this study lies in the use of topography, land cover and subpixel vegetation characteristics derived from LiDAR in a stepwise manner with increasing level of complexity, which demonstrates the importance of forest stand information on LSP at the pixel level.

中文翻译：

---

LiDAR 衍生的地形和林分特征在很大程度上解释了 MODIS 地表物候学中观察到的空间变异性

过去，研究已经成功地确定了对地表物候 (LSP) 时间变化的气候控制。然而，我们对在土地覆盖类型中观察到的 LSP 空间变异性及其控制因素缺乏更深入的了解。在这里，我们利用基于高分辨率 LiDAR 的数据集来研究亚像素林分特征对基于 MODIS NDVI 的 LSP 指标空间变异性的影响。多元线性回归技术 (MLR) 应用于来自 LiDAR 的林分信息和地形以及土地覆盖信息（即 2012 年的 CORINE 和专有栖息地地图），以预测德国巴伐利亚森林国家公园的平均 LSP 指标. 六个不同的 LSP 指标，即赛季开始 (SOS)、赛季结束 (EOS)、赛季长度 (LOS)、本研究模拟了生长季综合 NDVI (NDVIsum)、最大 NDVI 值 (NDVImax) 和最大 NDVI 天数 (maxDOY)。结果发现，无论土地覆盖如何，平均 SOS、LOS 和 NDVIsum 主要受海拔影响。然而，包含详细的林分信息大大改进了模型。然而，EOS 的建模更复杂，并且发现阔叶林的亚像素百分比和地形坡度与 EOS 的关联性更强。当额外考虑土地覆盖信息时，NDVImax 的解释方差从 0.45 提高到 0.71，当包括基于 LiDAR 的亚像素森林林分特征时，进一步提高到 0.84。由于完全同质的像素在自然界中很少见，我们的结果表明，亚像素林分信息与土地覆盖类型的结合可以提高基于地形的 LSP 模型的性能。这项研究的新颖之处在于，随着复杂程度的增加，逐步使用了从激光雷达获得的地形、土地覆盖和亚像素植被特征，这证明了林分信息在像素级别的 LSP 上的重要性。