Optimal model complexity for terrestrial carbon cycle prediction

Abstract. The terrestrial carbon cycle plays a critical role in modulating the interactions of climate with Earth system, but different models often make vastly predictions its behavior. Efforts to reduce model uncertainty have commonly focused on structure, namely by introducing additional processes and increasing structural complexity. However, extent which increased complexity can directly improve predictive skill is unclear. While adding may realism, resulting are encumbered greater number poorly determined or over-generalized parameters. To guide efficient development, here we map theoretical relationship between skill. do so, developed 16 structurally distinct spanning an axis incorporated them into model–data fusion system. We calibrated each at six globally distributed eddy covariance sites long observation time series under 42 data scenarios that resulted degrees parameter uncertainty. For combination site, scenario, model, then predicted net ecosystem exchange (NEE) leaf area index (LAI) for validation against independent local site data. Though maximum evaluated lower than most traditional biosphere models, range explored provides universal insight inter-relationship uncertainty, parametric forecast Specifically, only improves if parameters adequately informed (e.g., when NEE observations used calibration). Otherwise, degrade intermediate-complexity optimal. This finding remains consistent regardless whether LAI predicted. Our COMPLexity EXperiment (COMPLEX) highlights importance robust observation-based parameterization land surface modeling suggests characterizing fluxes will be key improving decadal high-dimensional models.