Abstract: Earth System Models (ESMs) are the primary tool for evaluating and predicting terrestrial soil organic carbon (SOC) dynamics. The model's fundamental controls of SOC decomposition were commonly represented as regulated by climate (temperature and moisture) and soil properties (such as nitrogen, soil texture, etc.) on top of a prescribed decay rate of each conceptual SOC pool. Recent studies revealed that SOC turnover (or decay rate) might be dominantly controlled by edaphic factors rather than climate, and this is inadequately represented in ESMs. Therefore, in this study, we used a long-term (~580d) lab incubation of multiple NEON soil samples (4 replicates * 2 layers * 20 sites), including daily CO2 efflux and 13C data, to constrain SOC decomposability parameter values (here we include decay rate - K, and microbial carbon using efficiency - CUE for each conceptual SOC pool). The lab incubation database revealed that the daily CO2 effluxes varied greatly among sites, among layers, and even for samples within the same site but from different plots. Using an improved and well-tested process-based SOC model (CN-SIM), we optimized the sample-specific parameter values for K and CUE to reproduce the daily CO2 efflux pattern from the lab incubation. Then, we trained a Random Forest model (RFM) to identify key soil factors (among 29 N-related, geochemical, and microbial properties) and learn the relationship between the sample-specific K/ CUE and key soil properties. The key soil features responsible for CN-SIM model decay rates (pH, C: N ratio, Ca, Fe, fungi: bacteria ratio) and microbial CUE (pH, inorganic N, silt, a fungal community index, particulate organic matter) were identified. Our trained RFM could explain 91% and 86% (p < 0.05) of variation in decay rate and CUE, respectively, for testing samples. Our findings indicate a potential to quantify SOC decomposability and reduce model uncertainties by using soil geochemical and microbial properties.
