We developed a novel spatio-temporal fusion method to downscale the AVHRR NDVI products to the Moderate-resolution Imaging Spectroradiometer (MODIS) resolution. The algorithm effectively combines the high spatial variability of the MODIS NDVI data and the long-term temporal information of the AVHRR NDVI data. Finally, we successfully generated a monthly global long-term (since 1982) and high-resolution (250m) NDVI database. Here we provide the downscaled NDVI dataset of Idaho from 1982 to 2020. Datasets for other regions can be easily produced by the GEE platform with the code provided in the github (https://github.com/babyfoal/downsclaed_NDVI/tree/main). The spatial distribution and temporal variation of this dataset have been both well validated by the simulated and real-data experiments.