Improving maize yield prediction at the county level from 2002 to 2015 in China using a novel deep learning approach

•We estimated maize yields at the county level in China using a novel deep learning approach.•The over-complete dictionary matrix was proposed to enhance the representation of inter-province spatial diversity.•We used the margin-constraint optimizing formula to eliminate the distraction of unbalanced distributed records.•The proposed new approach reduced the large-scale maize yield estimation error.