Decomposing multisource uncertainties in ecosystem service assessment: A quantitative framework and application to...

Ecosystem service (ES) assessments exhibit wide discrepancies in results that undermine decision-making credibility, yet systematic uncertainty attribution remains lacking. In this study, a quantitative framework for decomposing epistemic uncertainties across the full ES assessment chain was developed, using water conservation in China's Three-River-Source Region as a case study. By integrating single-factor controlled experiments with full-factorial variance decomposition, we quantified the contributions of service definition, model structure, input data, and parameters. The water conservation estimates varied 24-fold (11.21-283.49 mm). Within the SWAT framework, the service definition dominated the variance (70.30%), but its nonlinear interaction with the model parameters contributed an additional 16.04%, amplifying total uncertainty 2.8-fold beyond linear expectations. Model structure effects (257% discrepancy between SWAT and InVEST) were excluded from the full-factorial analysis because of cross-model incompatibility, indicating that these contribution rates are specific to physically based distributed models. Data resolution effects were negligible (CV < 0.03) at regional scales, although this may not hold for fine-scale process simulations. To bridge the science-policy gap, we propose a dual-path management strategy that combines short-term standardized accounting with long-term fundamental research. The proposed framework advances ES assessment from single-factor sensitivity analysis to systematic uncertainty decomposition, with transferable applicability to carbon sequestration, soil conservation, and biodiversity evaluation.