The invention discloses a multi-year short-term climate prediction method based on a multi-factor regression set, a cross-check set, and a month-by-month rolling set, and belongs to the field of short-term climate prediction theory and methods in meteorology. The method of the present invention includes, firstly, according to the large-scale barotropic eddy equation, using the potential function field and the flow function field reflecting the effects of the advection term and the exogenous forcing term in the theoretical equation as the predictor field; then, considering the combination of different factor time effects and Its independence, comprehensive application of multi-factor regression set, cross-check set, month-by-month rolling set, to predict the target. The invention has improvements in the selection of the basic factor field, the determination of the final factor and the application of various sets. Its advantages lie in the characteristics of long prediction timeliness, simple calculation relative model, good prediction effect, strong regional adaptability, etc., and can be applied to prediction problems of other scales and different element variables.