The quantification and spatially explicit mapping of carbon stocks in terrestrial ecosystems is important to better understand the global carbon cycle and to monitor and report change processes, especially in the context of international policy mechanisms such as REDD+ or the implementation of Nationally Determined Contributions (NDCs) and the UN Sustainable Development Goals (SDGs). Especially in heterogeneous ecosystems, such as Savannas, accurate carbon quantifications are still lacking, where highly variable vegetation densities occur and a strong seasonality hinders consistent data acquisition. In order to account for these challenges we analyzed the potential of land surface phenological metrics derived from gap-filled 8-day Landsat time series for carbon mapping. We selected three areas located in different subregions in the central Brazil region, which is a prominent example of a Savanna with significant carbon stocks that has been undergoing extensive land cover conversions. Here phenological metrics from the season 2014/2015 were combined with aboveground carbon field samples of cerrado sensu stricto vegetation using Random Forest regression models to map the regional carbon distribution and to analyze the relation between phenological metrics and aboveground carbon. The gap filling approach enabled to accurately approximate the original Landsat ETM+ and OLI EVI values and the subsequent derivation of annual phenological metrics. Random Forest model performances varied between the three study areas with RMSE values of 1.64 t/ha (mean relative RMSE 30%), 2.35 t/ha (46%) and 2.18 t/ha (45%). Comparable relationships between remote sensing based land surface phenological metrics and aboveground carbon were observed in all study areas. Aboveground carbon distributions could be mapped and revealed comprehensible spatial patterns. Phenological metrics were derived from 8-day Landsat time series with a spatial resolution that is sufficient to capture gradual changes in carbon stocks of heterogeneous Savanna ecosystems. These metrics revealed the relationship between aboveground carbon and the phenology of the observed vegetation. Our results suggest that metrics relating to the seasonal minimum and maximum values were the most influential variables and bear potential to improve spatially explicit mapping approaches in heterogeneous ecosystems, where both spatial and temporal resolutions are critical.

中文翻译：

---

巴西大草原的Landsat物候指标及其与地上碳的关系。

对陆地生态系统中的碳储量进行量化和空间明晰的绘制对于更好地理解全球碳循环以及监测和报告变化过程非常重要，尤其是在诸如REDD +等国际政策机制或实施国家自主贡献（NDC）的情况下以及联合国可持续发展目标（SDG）。尤其是在热带稀树草原等异类生态系统中，仍然缺乏准确的碳定量，因为那里的植被密度变化很大，季节性强，阻碍了数据的连续采集。为了解决这些挑战，我们分析了从填空的8天Landsat时间序列进行碳测绘得出的地表物候指标的潜力。我们选择了位于巴西中部地区不同分区域的三个地区，这是萨凡纳（Savanna）的一个典型例子，该地区拥有大量的碳储量，并且已经进行了广泛的土地覆盖转换。在此，使用Random Forest回归模型将2014/2015年季节的物候指标与塞拉多森严密植被的地上碳田样本相结合，以绘制区域碳分布图，并分析物候指标与地上碳之间的关系。填补空白的方法能够准确估算原始的Landsat ETM +和OLI EVI值，以及随后得出的年度物候指标。三个研究区域之间的随机森林模型性能各不相同，RMSE值分别为1.64吨/公顷（平均相对RMSE 30％），2.35吨/公顷（46％）和2.18吨/公顷（45％）。在所有研究区域都观察到了基于遥感的土地表面物候指标与地上碳之间的可比关系。可以绘制地上碳的分布图，并揭示可理解的空间格局。物候指标来自8天的Landsat时间序列，其空间分辨率足以捕获异种萨凡纳生态系统碳储量的逐渐变化。这些指标揭示了地上碳与观测到的植被物候之间的关系。我们的研究结果表明，与季节最小值和最大值有关的度量标准是影响最大的变量，并且具有改进异质生态系统中空间明确的制图方法的潜力，在异质生态系统中，空间和时间分辨率都至关重要。