Land-use intensity quantification and management classifications in grasslands of Germany 2017/2018

In Lange et. al (2022) we quantified land-use intensity (LUI) and its key parameters - grazing intensity, mowing frequency and fertiliser application - across Germany. Key parameters were classified using Convolutional Neural Networks (CNN) and Copernicus Sentinel-2 satellite data with 20 m x 20 m spatial resolution. Predictions of LUI and its components were validated using comprehensive in situ grassland management data from the DFG Biodiversity Exploratories. A feature contribution analysis using Shapley values substantiated the applicability of the methodology by revealing a high relevance of springtime satellite observations and spectral bands related to vegetation health and structure. We achieved an overall classification accuracy of up to 66% for grazing intensity, 68% for mowing, 85% for fertilisation and an r^2 of 0.82 for the subsequent LUI depiction. We evaluated the methodology's robustness with a spatial 3-fold cross-validation by training and predicting on geographically distinctly separated regions. Spatial transferability was assessed by delineating the models' area of applicability (AOA). More information can be found in the related publication.

Data is provided in GeoTiff format (projection EPSG:32632). Grazing classification bases on grazing intensity (G), given as livestock units (depending on species and age) per ha and day (Class 0: G=0, Class 1: 0 < G <= 0.33, Class 2: 0.33 < G <=0.88, Class 3: G > 0.88). Mowing counts represent the number of moving events in the respective year. Fertilisation information were aggregated into two classes: fertilised and not fertilised. Each model's AOA is given in a separate GeoTiff file, with values of 0 for areas outside and 1 for areas inside the model's AOA.

Please note:
Grassland pixels were selected according to the digital landscape model 2015 of the official topographic-cartographic information system (© GeoBasis-DE / BKG 2015).

在Lange等人（2022）的研究中，研究者针对德国全境量化了土地利用强度（Land-use Intensity, LUI）及其核心参数——放牧强度、刈割频率与施肥量。核心参数的分类采用卷积神经网络（Convolutional Neural Networks, CNN）与空间分辨率为20 m×20 m的哥白尼哨兵-2（Copernicus Sentinel-2）卫星数据完成。土地利用强度及其各组分的预测结果，依托DFG生物多样性探索项目（DFG Biodiversity Exploratories）的全面原位草地管理数据完成验证。采用夏普利值（Shapley Values）开展的特征贡献分析，通过揭示春季卫星观测数据与植被健康、结构相关光谱波段的高度相关性，证实了该方法的适用性。本研究针对放牧强度实现了最高66%的总体分类精度，刈割分类精度达68%，施肥分类精度达85%；后续土地利用强度表征的决定系数r²为0.82。本研究通过在地理上显著分离的区域分别开展模型训练与预测，采用空间3折交叉验证评估了该方法的鲁棒性。通过划定模型的适用范围（Area of Applicability, AOA），评估了模型的空间可迁移性。更多详细信息可查阅相关研究论文。

数据集采用GeoTiff格式存储，投影坐标系为EPSG:32632。放牧强度分类以放牧强度（G）为依据，其单位为每公顷每日的牲畜单位（依据牲畜种类与龄期确定），共分为4个等级：0级（G=0）、1级（0 < G ≤ 0.33）、2级（0.33 < G ≤ 0.88）、3级（G > 0.88）。刈割次数代表对应年度内的刈割事件总数量。施肥信息被划分为两个类别：已施肥与未施肥。每个模型对应的适用范围（AOA）存储于独立的GeoTiff文件中，文件内取值为0代表该区域处于模型适用范围之外，取值为1则代表处于模型适用范围之内。

请注意：草地像元的选取依据官方地形制图信息系统的2015年数字景观模型完成（© GeoBasis-DE / BKG 2015）。