Uncertainty analysis of modeled carbon and water fluxes in a subtropical coniferous plantation

Abstract

Estimating the exchanges of carbon and water between vegetation and the atmosphere requires process‐based ecosystem models; however, uncertainty in model predictions is inevitable due to the uncertainties in model structure, model parameters, and driving variables. This paper proposes a methodological framework for analyzing prediction uncertainty of ecosystem models caused by parameters and applies it in Qianyanzhou subtropical coniferous plantation using the Simplified Photosynthesis and Evapotranspiration model. We selected 20 key parameters from 42 parameters of the model using one‐at‐a‐time sensitivity analysis method and estimated their posterior distributions using Markov Chain Monte Carlo technique. Prediction uncertainty was quantified through Monte Carlo method and partitioned using Sobol' method by decomposing the total variance of model predictions into different components. The uncertainty in predicted net ecosystem CO2 exchange (NEE), gross primary production (GPP), ecosystem respiration (RE), evapotranspiration (ET), and transpiration (T), defined as the coefficient of variation, was 61.0%, 20.6%, 12.7%, 14.2%, and 19.9%, respectively. Modeled carbon and water fluxes were highly sensitive to two parameters, maximum net CO2 assimilation rate (Amax) and specific leaf weight (SLWC). They contributed more than two thirds of the uncertainty in predicted NEE, GPP, ET, and T and almost one third of the uncertainty in predicted RE, which should be focused on in further efforts to reduce uncertainty. The results indicated a direction for future model development and data collection. Although there were still limitations in the framework illustrated here, it did provide a paradigm for systematic quantification of ecosystem model prediction uncertainty.