@Article{Bonan_GBC_20191001,
 author		= {Gordon B. Bonan and Danica L. Lombardozzi and William R. Wieder and Keith W. Oleson and David M. Lawrence and Forrest M. Hoffman and Nathan Collier},
 title		= {Model Structure and Climate Data Uncertainty in Historical Simulations of the Terrestrial Carbon Cycle (1850--2014)},
 journal	= GBC,
 volume		= 33,
 number		= 10,
 pages		= {1310--1326},
 doi		= {10.1029/2019GB006175},
 day		= 1,
 month		= oct,
 year		= 2019,
 abstract	= {The divergence among Earth system models in the terrestrial carbon cycle has prompted interest in how to reduce uncertainty. Previous studies have identified model structural uncertainty arising from process parameterizations and parameter values. The current study highlights the importance of climate forcing in generating carbon cycle uncertainty. We use simulations in which three models (CLM4, CLM4.5, CLM5) with substantially different carbon cycles are forced with two climate reconstructions (CRUNCEPv7, GSWP3v1) to examine the contributions of model structure and climate to uncertainty in the carbon cycle over the period 1850--2014. Climate uncertainty for global annual net biome production exceeds one-third of total uncertainty (defined as the sum of climate and model structure uncertainty) in the first half of the twentieth century, but declines after the 1950s. Global annual gross primary productivity, net primary productivity, heterotrophic respiration, and vegetation and soil carbon stocks have substantial climate uncertainty (relative to total uncertainty) throughout the simulation period. Climate forcing contributes more than one-half of total uncertainty for these carbon cycle fluxes and stocks throughout boreal North America and Eurasia, some mid-latitude regions, and in eastern Amazonia and western equatorial Africa during the decade 2000--2009. Comparison with observationally-based datasets of the carbon cycle using model benchmarking methods provides insight into strengths and deficiencies among models and climate forcings, but we caution against overreliance on benchmarking to discriminate among models. The conceptualization of uncertainty arising from this study implies embracing multiple feasible model simulations rather than focusing on which model or simulation is best.}
}