Yang Yuhang, Hua Yongchun, Zhang Qiuliang, Wang Fei
College of Forestry, Inner Mongolia Agricultural University, Hohhot 010000, China.
Plants (Basel). 2025 Sep 4;14(17):2768. doi: 10.3390/plants14172768.
As critically important global carbon sinks, the net primary productivity (NPP) of boreal forests is crucial for understanding the terrestrial carbon cycle. However, a lack of long-term, high-resolution data has hindered progress in this field. In this study, we used a standardized tree ring chronology of to identify the dominant factors driving NPP changes in the Northern Greater Khingan Range, applying both Pearson correlation coefficients and SHAP importance values. We then integrated XGBoost and Extreme Random Forest (ERF) models to reconstruct interannual forest NPP across the region from 1968 to 2020. Our results reveal a significant correlation between NPP and tree radial growth, with both processes dominated by growing season drought. The combination of machine learning and tree ring methods proved to be a reliable approach, with the XGBoost model achieving higher reconstruction accuracy than the ERF model. The reconstructed NPP series showed strong regional correlation with MODIS NPP products (r > 0.6) and revealed interdecadal cycles of 10, 28, and 49 years, as well as shorter periodicities of 2-8 and 15-18 years. This study establishes a novel framework for high-resolution NPP reconstruction and clarifies the response mechanisms of the boreal forest carbon cycle to climate change.
作为至关重要的全球碳汇，北方森林的净初级生产力（NPP）对于理解陆地碳循环至关重要。然而，缺乏长期、高分辨率的数据阻碍了该领域的进展。在本研究中，我们使用标准化的树木年轮年表，通过皮尔逊相关系数和SHAP重要性值，来确定驱动大兴安岭北部NPP变化的主要因素。然后，我们整合了XGBoost和极端随机森林（ERF）模型，以重建1968年至2020年整个区域的年际森林NPP。我们的结果揭示了NPP与树木径向生长之间的显著相关性，这两个过程均受生长季干旱主导。机器学习和树木年轮方法的结合被证明是一种可靠的方法，XGBoost模型的重建精度高于ERF模型。重建的NPP序列与MODIS NPP产品显示出很强的区域相关性（r > 0.6），并揭示了10年、28年和49年的年代际周期，以及2 - 8年和15 - 18年的较短周期。本研究建立了一个用于高分辨率NPP重建的新框架，并阐明了北方森林碳循环对气候变化的响应机制。
