Phytoplankton photosynthesis contributes approximately 50% of global primary productivity and plays a crucial role in the carbon cycle, forming the foundation of the oceanic food chain. Accurately estimating the magnitude of global ocean net primary productivity (NPP) and revealing its spatiotemporal patterns are essential for understanding the ocean's biological carbon pump and the carbon cycle in the atmosphere-ocean coupled system. However, the current quantification of global ocean primary productivity through satellite remote sensing shows poor correlation with observational data, and different remote sensing models yield significant variations in predictions across different ocean regions. To enhance our understanding of the spatiotemporal distribution of primary productivity, this study has established a more comprehensive global ocean NPP database, comprising 11,005 records of depth-integrated data based on 14C absorption methods, covering the period from 1958 to 2022. This represents an 82.53% increase compared to the most recent available global database. Using the updated global primary productivity database, we evaluated the predictive performance of current remote sensing models and found that several commonly used NPP models had poor correlations with measured primary productivity, with R² values ranging from 0.01 to 0.2, and root mean square errors (RMSE) reaching up to three orders of magnitude. To further reduce the gap between numerical simulations and in situ observations, this study employs machine learning techniques, incorporating changes in physicochemical parameters from 14C incubation experiments, to develop a more accurate global primary productivity remote sensing model.

remote sensing (RS), primary productivity