Surface net radiation, representing surface radiation energy balance, is closely related to several land processes, such as evapotranspiration, photosynthesis, and turbulent and conductive heat fluxes. Reanalysis products can provide a long-term surface net radiation; however, their coarse spatial resolution and large uncertainties hinder us from well applicating the data at a regional scale. Satellite products also include surface net radiation retrievals with high accuracy. The short time span of satellite products (i.e., GLASS product) makes these satellite products not suitable for long-term climate change study. Therefore, we used a deep learning method to upscale in situ measurements collected from global-distributed sites to generate a daily surface net radiation product with 0.05° spatial resolution from AVHRR data (1981-2019). After comprehensive validation, the RMSE of AVHRR net radiation product was ~26 Wm-2, which is generally better than some current reanalysis and satellite products.