Earth System Models (ESMs), such as the Accelerated Climate Modeling for Energy (ACME), are needed to understand feedbacks between the terrestrial carbon cycle and climate. Among many factors affecting the responses of the terrestrial carbon cycle to climate, the turnover time of soil organic carbon (SOC) depends on climate, soil properties, and vegetation type. However, estimating SOC turnover times at multiple spatial scales is challenging and subject to large uncertainty. As carbon turnover time can be reliably estimated from radiocarbon data, we strive to improve global predication by comparing measured site-scale 10 and 50 cm depth Δ¹⁴C values with large-scale ACME Land Model (ALM) predictions. Our goal is to identify the main factors causing discrepancies between simulated and observed values.

We first upscale site-scale measurements to large-scale ALM grids and then compare the upscaled and modeled Δ¹⁴C values. At the large scale, we fit the mean observed values as a function of ALM outputs and large-scale surface features. Results show that plant functional type (PFT), soil order, and ground surface variations, such as average slope and standard deviation of elevation, are primary factors for explaining model-observed discrepancies at both depths. This suggests accurately representing PFT and soil order (e.g., by resolution refinement) may reduce discrepancies at the model grid scale. We also perform analysis on the residuals after removal of the mean effects to characterize the site-scale variability and find the Δ¹⁴C residuals at both depths strongly depend on bulk density, mean annual temperature (MAT), and vegetation type. Since soil bulk density and its vertical profile are localized information, characterizing Δ¹⁴C values at the site scale is more difficult and statistical models may play an important role. We finally use ALM to better understand how the global soil Δ¹⁴C values are sensitive to those dominant factors.