Leveraging meteorological legacy effects and machine learning to improve daily gap-free normalized difference vegetation index reconstruction: a critical basis for fine-scale carbon sink capacity monitoring

Hui Li, Yue Cao, Jingfeng Xiao, Fei Yu, Rui Wang, Xinyi Hao, Zuoqiang Yuan

The daily Normalized Difference Vegetation Index (NDVI) is a critical indicator of terrestrial carbon sequestration capacity, essential for accurately quantifying vegetation carbon sink functions and their dynamic responses to climate change. We previously developed a long-term daily NDVI dataset across China, but the over smoothed polynomial fitting method constrained reconstruction accuracy in complex scenarios, hindering precise fine-scale carbon sink monitoring. To address this limitation, we improved the reconstruction framework by integrating meteorological legacy effects and the machine learning algorithm, with a focus on validating its application value for net primary productivity (NPP) estimation. The reconstructed daily gap-free NDVI (1982–2023) shows strong consistency with original valid NDVI, achieving a national average R 2 of 0.9, percentage bias ( PB ) of − 0.09%, and root mean square error (RMSE) of 0.03. Over 97% of China’s reconstructed NDVI reach high-quality ( R 2 > 0.8, |PB|<1%, RMSE < 0.08), representing a 13.9% R 2 improvement over our previous dataset. The national daily average NDVI exhibits a continuous increasing trend during the study period, primarily driven by Southwest, Central, Southeast, South, and Northeast China. Attribution analysis indicates that meteorological legacy effects serve as the primary drivers for daily NDVI reconstruction across China. Temperature-related factors including average temperatures over the past 10, 30 and 60 days dominate 68.94% of China, while 60-day cumulative precipitation contributes 12.16%. NPP estimation validation indicates that our daily NDVI-derived NPP shows high agreement with in-situ measurements ( R 2 = 0.82, PB = 5.18%, RMSE = 0.19 kg C m –2 yr –1 ), significantly outperforming coarse-temporal-resolution NDVI-based products ( R 2 < 0.28). This study provides a robust long-term daily gap-free NDVI dataset for China, serving as a critical basis for fine-scale terrestrial carbon sink capacity monitoring. Integrating meteorological legacy effects into NDVI reconstruction enhances the accuracy of vegetation dynamic simulation and improves the quantification of terrestrial carbon cycle processes, particularly for fine-scale NPP estimation. Our findings highlight the superiority of high-temporal-resolution NDVI in capturing fine-scale vegetation phenological dynamics, addressing the limitations of conventional coarse-resolution products and supporting more reliable ecological and climate change research.