Realistic simulation and accurate prediction of El Niño-Southern Oscillation (ENSO) is still a challenge. One fundamental obstacle is the so-called spring predictability barrier (SPB), which features a low predictive skill of the ENSO with prediction across boreal spring. Our observational analysis shows that the leading empirical orthogonal function mode of the seasonal Niño3.4 index evolution (i.e., from May to the following April) explains nearly 90% of its total variance, and the principle component is almost identical to the Niño3.4 index in the mature phase. This means a good ENSO prediction for a year ranging May-next April can be achieved if the Niño3.4 index in the mature phase is accurately obtained in advance. In this work, by extracting physically oriented variables in the spring, a linear regression approach that can reproduce the mature ENSO phases in observation is firstly proposed. Further investigation indicates that the specific equation, however, is significantly modulated by an interdecadal regime shift in the air–sea coupled system in the tropical Pacific. During 1980–1999, ocean adjustment and vertical processes were dominant, and the recharge oscillator theory was effective to capture the ENSO evolutions. While, during 2000–2018, zonal advection and thermodynamics became important, and successful prediction essentially relies on the wind stress information and their controlled processes, both zonally and meridionally. These results imply that accounting for the interdecadal regime shift of the tropical Pacific coupled system and the dominant processes in spring in modulating the ENSO evolution could reduce the impact of SPB and improve ENSO prediction.