Nighttime light (NTL) is a foundational data source for studying human activities from a remote sensing perspective. This dataset from 1992 to 2021 is the first global long-term and fine-grained NTL observations. It represents a milestone in facilitating the study of human activities. It is created by a new super-resolution model DeepNTL which converts DMSP-OLS images into NPP-VIIRS images. Compared with baseline models, including RCAN, SwinIR and AutoEncoder, DeepNTL has the hightest accuracy and best generalization ability for untrained years. It provides a good extension of NPP-VIIRS to the early years, and the future annual NPP-VIIRS data can be directly appended to this dataset by users own. More information about the dataset can be found in the "read_me.txt" file. Technical details and evaluations are presented in our paper. Any questions are welcome to be sent to jinyuguo23@m.fudan.edu.cn.