【Objective】In the context of China’s “Dual Carbon” strategy and the digital transformation of forest resource monitoring, traditional forest stock volume monitoring methods still face challenges such as high cost, low timeliness, and insufficient accuracy under small-sample conditions. To address these issues and meet the municipal demand for “Annual data release”, this study proposes an annual forest stock volume updating method based on a data assimilation algorithm, integrating optimal generalized differential stand growth equations for different forest types with sampling techniques.【Method】Using five consecutive forest inventory datasets from Heihe City collected between 1995 and 2015, we selected 434 permanent sample plots. A non-replacement sampling scheme was employed to simulate the operational rotation mechanism of surveying 1/5 of the plots each year. A hybrid estimator model was then constructed, and its performance was compared with that of simple random sampling (SRS) in terms of estimation accuracy and variance control. In addition, a 1/7 sampling experiment was designed to further evaluate the robustness of the method under reduced sampling intensity. 【Result】The hybrid estimator model outperformed the traditional method under both sampling scenarios. Under the 1/5 sampling scheme, the estimated total stock volume obtained by the mixed estimator was closer to the true value, with a standard error of 0.8-4.2 m³/hm², which was substantially lower than that of SRS (1.8-4.7 m³/hm²). Even when the sampling proportion was reduced to 1/7, corresponding to an approximately 28.6% decrease in sampling intensity, the standard error of the hybrid estimation model remained lower (1.9-4.5 m³/hm²) than that of simple random sampling (2.7-5.6 m³/hm²). These findings indicate that the hybrid estimation model has good adaptability and convergence performance under small-sample conditions. 【Conclusion】By integrating theoretical growth equations with statistical sampling, the proposed model effectively controls estimation error while substantially reducing sampling intensity, thereby maintaining high precision and strong robustness. It provides a reliable solution for annual dynamic monitoring of forest resources based on small samples and offers a practical technical pathway for municipal-level annual monitoring. Moreover, its data assimilation framework avoids sensitivity to initial parameter values and is well suited for time-series modeling of complex ecosystems. The method can provide an important data foundation for carbon sink accounting and international compliance, and it also has the potential to be extended to the dynamic monitoring of other natural resources, such as wetlands and grasslands.