Abstract: Natural resource indicator data with high spatio-temporal resolution are essential for large-scale natural resource dynamic observation and trend assessment. The large amount of multi-source data under big data era could provide the possibility for efficient utilization and fusion of data. Taking the Normalized Difference Vegetation Index (NDVI) in Hanjiang Basin as an example, the authors in this paper have built a spatio-temporal big data processing underlying architecture for natural resources based on PostgreSQL, and integrated three types of methods, including data-level fusion, feature-level fusion and decision-level fusion. Besides, the intelligent fusion system of multi-source heterogeneous data has been constructed based on the machine learning algorithms to achieve efficient utilization of multi-source data and feature spatial preference. Meanwhile, the year-by-year NDVI 1 km dataset of Hanjiang Basin from 2000 to 2019 has been reconstructed to comprehensively reflect the dynamic changes of vegetation in Hanjiang Basin. These results could provide some scientific reference for the efficient extraction and simulation analysis of spatio-temporal big data in earth sciences, and provide a more accurate and convenient technical means for quantitatively accounting the scale of forest and grassland resources endowment and exploring the spatio-temporal evolution of ecosystem.