CIML-TER v10 (2001-2020)

Accurate estimation of terrestrial ecosystem respiration (TER) is essential for refining global carbon budgets and informing climate change response policies. Current large-scale TER models predominantly rely on empirical structures derived from site-scale observations, often driven solely by hydrothermal factors. However, it is critical to incorporate ecosystem-scale information for more accurate large-scale TER modeling, such as the biotic factors (e.g., clumping index and non-photosynthetic vegetation cover) linked to ecosystem-scale vegetation structure and component, and the spatiotemporal variation factors that describing the continuous variations of land cover and phenology. These ecosystem-scale variables have not been well parameterized in existing models, because the mechanisms by which they affect TER remain unclear. To address this gap, we applied a causality constrained interpretable machine learning framework (PCMCI+, XGBoost, SHAP) to model the relationships between relevant variables and TER, and established a TER estimation model called “CIML-TER”. The CIML-TER model was trained with an integrated TER observations from two major flux networks (FLUXNET and ABCflux), and was applied to estimate global monthly TER at a spatial resolution of 0.05° during 2001-2020. Global annual TER estimated with CIML-TER ranged between 117 and 125 Pg C. Moreover, CIML-TER accurately depicted the naturally continuous spatial variations of TER, which were not well described in Fluxcom and LGS-Reco due to the limitations of using traditional discrete land cover data. The CIML-TER model revealed the underestimated contributions of some complex ecosystems (such as EBFs) to global TER in previous TER products and emphasized the need for future process models to account for ecosystem-scale variables’ effects.精准估算陆地生态系统呼吸（Terrestrial Ecosystem Respiration，以下简称TER），对于优化全球碳预算、为气候变化应对政策提供依据至关重要。当前的大规模TER模型主要依赖基于站点尺度观测得到的经验结构，且通常仅以水热因子作为驱动变量。然而，若要实现更精准的大规模TER建模，纳入生态系统尺度的信息至关重要——例如与生态系统尺度植被结构及组成相关的生物因子（如聚集指数（clumping index）和非光合植被覆盖度（non-photosynthetic vegetation cover）），以及表征土地覆盖与物候连续变化的时空变异因子。现有模型尚未对这些生态系统尺度变量进行充分参数化，原因在于其影响TER的作用机制仍未明确。为填补这一研究空白，本研究采用因果约束可解释机器学习框架（PCMCI+、XGBoost、SHAP）构建相关变量与TER间的关联模型，并搭建了名为"CIML-TER"的TER估算模型。CIML-TER模型以两大通量网络（FLUXNET与ABCflux）整合的TER观测数据进行训练，并被用于估算2001-2020年间全球0.05°空间分辨率下的逐月TER。通过CIML-TER估算得到的全球年度TER总量介于117至125 Pg C之间。此外，CIML-TER精准刻画了TER自然连续的空间分布特征，而此前的Fluxcom与LGS-Reco产品因依赖传统离散土地覆盖数据，无法很好地描述这一特征。CIML-TER模型揭示，此前的TER产品低估了部分复杂生态系统（如EBFs）对全球TER的贡献，并强调未来的过程模型需纳入生态系统尺度变量的影响。