The Worldsoils project has developed a pre-operational Soil Organic Carbon (SOC) monitoring system in a cloud environment. The system predicts topsoil organic carbon content at regional and continental scales from Earth Observation (EO) satellite data with a continuous cover over Europe. The system utilizes spectral models for croplands and a digital soil mapping approach for permanently vegetated areas such as grasslands and forests. Models strongly rely on soil reflectance composites from the Sentinel 2 multispectral instrument providing the median reflectance for all valid pixels over a period of three years. The bare soil frequency, a proxy for the degree of crop cover, is clearly lower in a Mediterranean pilot region compared to croplands in temperate regions. This is due to the extensive crop cover in the Mediterranean with winter cereals and fodder crops. The graphical user interface provides SOC content and the prediction interval ratio (i.e. 90 % uncertainty interval divided by the median) for 50 m pixels in three pilot regions and 100 m pixels for the rest of Europe. The SOC prediction algorithms are reasonable compared to others at the continental scale (R2: 0.41 for croplands and 0.28 for permanently vegetated areas). Apart from tree crops in Macedonia (Greece) the soil reflectance composite attributes the correct model to validation sets of cropland and grassland in the pilot regions. The SOC prediction is satisfactory in Wallonia (Belgium; R2 0.51) but is less accurate in Greece and the Czech Republic. In particular in Greece, the poor performance is linked to the low bare soil frequency due to the abundance of tree crops, cereals and fodder crops. The monitoring system can reproduce spatial patterns in SOC content similar to the ones obtained from a detailed regional algorithm using the new generation of hyperspectral satellites. However, the very high values in kettle holes in a morainic landscape of Northern Germany are underestimated.

中文翻译：

---

利用 Sentinel 2 图像和 LUCAS 土壤数据库的欧洲土壤有机碳监测系统


Worldsoils 项目在云环境中开发了一个运行前的土壤有机碳 （SOC） 监测系统。该系统根据地球观测 （EO） 卫星数据预测区域和大陆尺度的表层土壤有机碳含量，并持续覆盖欧洲。该系统利用农田的光谱模型和永久植被区域（如草原和森林）的数字土壤测绘方法。模型在很大程度上依赖于 Sentinel 2 多光谱仪器的土壤反射复合材料，提供三年内所有有效像素的中位反射率。与温带地区的农田相比，地中海试点地区的裸土频率（代表作物覆盖程度）显然较低。这是由于地中海地区冬季谷物和饲料作物的广泛作物覆盖。图形用户界面提供了三个试点地区 50 m 像素和欧洲其他地区 100 m 像素的 SOC 含量和预测区间比率（即 90% 不确定性区间除以中位数）。与大陆尺度的其他算法相比，SOC 预测算法是合理的（R2：农田为 0.41，永久植被区域为 0.28）。除了马其顿（希腊）的树木作物外，土壤反射复合体还将正确的模型归因于试点地区的农田和草地验证集。瓦隆大区（比利时;R2 0.51），但在希腊和捷克共和国的准确率较低。特别是在希腊，由于树木作物、谷物和饲料作物丰富，裸土频率低，性能不佳。 该监测系统可以再现 SOC 内容的空间模式，类似于使用新一代高光谱卫星从详细的区域算法中获得的空间模式。然而，德国北部冰碛景观中壶穴的非常高的价值被低估了。