Accurately quantifying agricultural water use is essential for protecting agricultural systems from the risk of water scarcity and promoting sustainable water management. While previous studies have innovatively provided spatially explicit analyses or datasets, they tend to have relatively coarse resolution (~8.3 km), and inadequately considered precise localization parameters. Here, we produced annual blue and green water use for 15 main crops with a resolution of 1 km for the years 1991–2019 in China. Firstly, we estimated the yearly crop blue and green water use at the site scale by incorporating more localized input parameters using a dynamic water balance model. Then, the random forest model was combined with site-scale simulation results to generate spatial predictions of blue and green water for each crop from 1991 to 2019. The resulting maps showed a high correlation with locally observed values at field stations (R
2
= 0.95), statistics (R
2
= 0.77), and exhibited some strengths compared with existing datasets that covered various scales. This dataset can play a key role in devising sustainable water management strategies.