Ecosystem assessment is vital important to carry out ecosystem management, and its results have become a basis to measure the status of regional ecosystem. Due to the insufficient consideration of the regional differences of ecosystem endowments and lack of effective measures to distinguish the impact of climate change and human activities, the practicability and regional comparability of ecosystem assessment results have been controversial. Under the background of ecological big data, this paper combined with long-term ground monitoring data, field investigation data, remote sensing data and other massive data, coupled with the ecosystem process model, and took the less disturbed, nearly native and integral zonal vegetation ecosystem (integral ecosystem) as references based on the ecological integrity theory. The composition, structure and function of the integral ecosystem were defined as the reference conditions. The deviation between the current and reference state of ecosystem was used to evaluate the ecosystem. An ecosystem evaluation method based on "Reference-State-Deviation (RSD)" was then constructed. The RSD evaluation method could transform the assessment of current state into deviation assessment and enhance the temporal and spatial comparability of the assessment results. Taking the self-developed model, CEVAS-RS model and CEVAS-CLI model as tools, the productivity of terrestrial ecosystem in China was evaluated, and the feasibility of the method was verified. This study is of great significance to improve China's ecosystem assessment system, promote the practicability of ecosystem assessment results, and support the national needs of China's key ecological function areas and counties assessment, ecological red line supervision, nature reserve system construction and so on.