ARIMA-SVR Ensemble Model for Forecasting the Grain Yields in China with the CEEMDAN Components

China’s grain security remains a worldwide concern due to its massive population and rapid urbanization. Although the China’s total grain yield has steadily increased in recent years, this growth exhibits temporal and spatial variations. This study used the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) method to analyze the variations in China’s total grain yield of three staple grains at different temporal and spatial scales from 1990 to 2022, and examined the impact of different grains and producing regions on total grain yield fluctuations. At the temporal scale, the CEEMDAN decomposition reveals that the total grain yield of China’s three staple grains comprises two fluctuation components and one trend component, exhibiting quasi-triennial short-period fluctuations and quasi-quindecennial long-period fluctuations. On the spatial scale, it is found that the primary grain-producing zones wield the greatest impact on total grain yield fluctuations, followed by the primary sales territories and balanced production–consumption regions. Subsequently, an ARIMA-SVR ensemble forecasting model is proposed based on the CEEMDAN components. The forecasting results indicate that the total yield of China’s three staple grains will increase from 2023 to 2025, but the annual growth rate is expected to be lower than that in the past three years. Our results highlight the necessity of maintaining vigilance in policy-making to ensure food security.