Li Yujie, Liu Xuanguang, Zhang Zhenchao, Lang Jun
Institute of Geospatial Information, Information Engineering University, Zhengzhou, 450001, China.
Lyuliang University, Lüliang, 033001, China.
Sci Rep. 2025 Jul 1;15(1):22346. doi: 10.1038/s41598-025-08927-x.
The gross primary productivity (GPP) of Shanxi Province, China, plays an important role in the carbon cycle of the Loess Plateau ecosystem. However, Shanxi Province lacks carbon flux stations, leading to imprecise GPP estimation results. Additionally, few studies have explored the drivers of long-term GPP change in Shanxi Province. Therefore, in this study, we aimed to estimate the GPP in Shanxi Province from 2001 to 2022 and determine the driving factors of long-term GPP trends. To this end, we proposed an improved GPP estimation method based on the CatBoost model. Our CatBoost GPP model reduces model overfitting in few-shot scenarios and effectively captures the time dependence in time-series data. In addition, it integrates the change characteristics of vegetation ecological indicators and topography constraints, improving GPP estimation accuracy. Subsequently, we explored the spatial and temporal variations driving force through methods such as Theil-Sen Median trend analysis and Geodetectors. Our results show that (1) Compared with existing methods, the proposed CatBoost GPP method achieved superior site-level accuracy, with an [Formula: see text] value of 0.890, root mean square error (RMSE) of 1.155 gC[Formula: see text], and mean absolute error (MAE) of 0.772 gC[Formula: see text]. Furthermore, we compared our results with previous GPP products to further assess the regional-level accuracy; (2) The GPP in Shanxi Province displayed a fluctuating increase, with a growth rate of 20.58 gC[Formula: see text] from 2001 to 2022. The overall spatial variation was characterized by low GPP in the northwest and high GPP in the southeast. The GPP change was mainly characterized by weak anti-persistence; thus, approximately 58.8% of the area may experience degradation in the future; and (3) Land use type significantly influenced GPP changes in Shanxi, with the restoration and improvement of grassland being the main contributor to the increase in GPP. The interaction between precipitation and temperature had the most complex and significant impact on GPP, affecting approximately 62.05% of the study area. The results of this study provide a theoretical basis for ecological protection and sustainable development in Shanxi Province.
中国山西省的总初级生产力（GPP）在黄土高原生态系统的碳循环中发挥着重要作用。然而，山西省缺乏碳通量观测站，导致GPP估算结果不准确。此外，很少有研究探讨山西省长期GPP变化的驱动因素。因此，在本研究中，我们旨在估算2001年至2022年山西省的GPP，并确定长期GPP趋势的驱动因素。为此，我们提出了一种基于CatBoost模型的改进GPP估算方法。我们的CatBoost GPP模型减少了少样本情况下的模型过拟合，并有效捕捉了时间序列数据中的时间依赖性。此外，它整合了植被生态指标的变化特征和地形约束，提高了GPP估算精度。随后，我们通过Theil-Sen中位数趋势分析和地理探测器等方法探索了时空变化驱动力。我们的结果表明：（1）与现有方法相比，所提出的CatBoost GPP方法在站点层面实现了更高的精度，决定系数（R²）值为0.890，均方根误差（RMSE）为1.155 gC·m⁻²·a⁻¹，平均绝对误差（MAE）为0.772 gC·m⁻²·a⁻¹。此外，我们将结果与之前的GPP产品进行比较，以进一步评估区域层面的精度；（2）山西省的GPP呈波动上升趋势，2001年至2022年的增长率为20.58 gC·m⁻²·a⁻¹。总体空间变化特征是西北部GPP低，东南部GPP高。GPP变化主要表现为弱反持续性；因此，未来约58.8%的区域可能会出现退化；（3）土地利用类型对山西省GPP变化有显著影响，草地的恢复和改善是GPP增加的主要贡献因素。降水和温度的相互作用对GPP的影响最为复杂且显著，影响了约62.05%的研究区域。本研究结果为山西省的生态保护和可持续发展提供了理论依据。
