Impacts of two types of errors on the predictability of terrestrial carbon cycle

Abstract

AbstractInitial errors and model parameter errors are two of the main factors that produce uncertainties in numerical simulations and predictions. It is crucial to determine in advance which of these types of errors should be reduced to improve numerical simulations and increase their prediction skill. In this study, a fundamental issue related to studies of the predictability about terrestrial carbon cycle is discussed. The relative importance of initial errors and model parameter errors in causing the prediction uncertainty of net primary production (NPP), which is part of the terrestrial carbon cycle, is assessed. The errors in NPP predictions related to initial errors and parameter errors are evaluated within the Lund–Potsdam–Jena (LPJ) model using the conditional nonlinear optimal perturbation approach for these two types of errors (CNOP‐I and CNOP‐P, respectively). This method explores the upper bounds of the prediction errors caused by initial errors and model parameter errors at a particular time in the future. The contributions of initial errors and model parameter errors to NPP prediction uncertainty are found to depend on climate backgrounds and the prediction timescale. For the moist and semimoist regions of China, initial errors play a more important role than model parameter errors in the prediction uncertainty of NPP for short prediction times. As the prediction time scale increases, model parameter errors lead to large errors in NPP predictions. However, in the arid and semiarid regions of China, the uncertainty in NPP predictions caused by model parameter errors is always larger than that caused by initial errors. The above results can be explained by analyzing variations in photosynthesis and autotrophic respiration. The numerical results also suggest that the skill of predictions of the terrestrial carbon cycle may be improved by reducing the different types of errors for different climatic regions and prediction timescales.