Random effects and environmental sensitivity improve the compatible biomass model systems of moso bamboo forests in Southern China

• Developed an individual compatible biomass model for moso bamboo in southern China. • Inclusion of growth regional-level random effects has improved the accuracy of models. • Consideration of correlations between component models in model development. Accurate estimates of bamboo forest biomass can help assess the impact of climate change and help achieve the vision of carbon neutrality. Estimate of component-biomass and total-biomass of the moso bamboo ( Phyllostachys edulis ), which is one of the major bamboo species, is also crucial for assessing the productivity of bamboo forest ecosystem. This study is based on the measured biomass data from 368 bamboo individuals, collected across the nine provinces (growth regions) of China (Jiangsu, Anhui, Zhejiang, Jiangxi, Fujian, Sichuan, Guangxi, Hunan, and Hubei). This study applied three important modeling approaches, namely seemingly unrelated regression (SUR), two-stage error-in-variable modeling (TSEM) and seemingly unrelated mixed-effects modeling (SURM) to develop a compatible biomass model system, which is an aggregate form of a set of component-biomass models. Only the important factors describing effects of the individual bamboo characteristics, soil properties, and climate features were considered as predictor variables in a model system. The introduction of bamboo individuals (height to crown base), soil properties (soil bulk density and soil organic carbon), and climate factors (de Martonne aridity index) increased the R 2 of each component-biomass model by 0.7% − 4.6%. Introduction of the growth regional-level (province-level) random effects significantly improved the component models in a model system (R 2 increased by 1.2% − 12.4%). The model validation against the partitioned data showed a better prediction performance of SURM compared to TSUM and SUR models. A response calibration (localization) of SURM showed an increased the prediction accuracy with increased number of samples used in calibration of the random effects. Considering the investigation time and cost, we recommend using six randomly selected bamboo individuals per growth region for optimal prediction accuracy. Our models can provide a promising basis for estimation and evaluation of carbon storage in moso bamboo forests and formulating effective management strategies in southern China.