Abstract. We present a machine learning dataset for tree species classification in Sentinel-2 satellite image time series of bottom of atmosphere reflectance. The dataset is based on the German national forest inventory of 2012, as well as analysis ready satellite imagery computed using the FORCE processing pipeline. From the national forest inventory data, we extracted the tree positions, filtered 387 775 trees in the upper canopy layer and automatically extracted the corresponding bottom of atmosphere reflectance time series from Sentinel-2 L2A images. These time series are labeled with the corresponding tree species, which allows pixel-wise classification tasks. Furthermore, we provide auxiliary information such as the approximate tree position, the year of possible disturbance events or the diameter at breast height. Temporally, the dataset spans the years from July 2015 to end of October 2022 with ca. 75.3 million data points for trees of 51 species and species groups, as well as 13.8 million observations for non-tree background. Spatially, it covers entire Germany. The dataset is available under following DOI (Freudenberg et al., 2024): https://doi.org/10.3220/DATA20240402122351-0

中文翻译：

---

德国用于树种分类的 Sentinel-2 机器学习数据集


摘要。我们提出了一个机器学习数据集，用于 Sentinel-2 大气底部反射率卫星图像时间序列中的树种分类。该数据集基于 2012 年德国国家森林清查，以及使用 FORCE 处理管道计算的可供分析的卫星图像。从国家森林清查数据中，我们提取了树木位置，过滤了上层树冠层的387 775棵树木，并从Sentinel-2 L2A图像中自动提取了相应的底部大气反射率时间序列。这些时间序列标有相应的树种，这允许执行像素级分类任务。此外，我们还提供辅助信息，例如树木的大致位置、可能发生干扰事件的年份或胸高直径。从时间上看，该数据集的时间跨度为 2015 年 7 月至 2022 年 10 月底，大约为 2022 年 10 月。 51 个物种和物种组的树木的 7530 万个数据点，以及非树木背景的 1380 万个观测值。从空间上看，它涵盖了整个德国。该数据集可通过以下 DOI 获取（Freudenberg et al., 2024）：https://doi.org/10.3220/DATA20240402122351-0