Figure 1.

Schema of integrating big data into SOC distribution maps by CLM5. In data assimilation (A), we used NPP, soil temperature, and soil water potential as model inputs to drive the CLM5 model. The model output (i.e., analytical solution of vertical SOC content calculated through Equation 2 under the steady state assumption) will be compared with the observed SOC content in the data assimilation framework, the results of which will further guide the adjustment of parameter values of CLM5 to improve model representation of SOC distribution. The figure was adapted from https://www.hzg.de/institutes_platforms/cosyna/models/data_assimilation/index.php.en. Two data assimilation methods were used and were further developed to generate SOC distribution maps across the conterminous United States (B). Optimized parameter values at individual sites in the site-by-site method were either randomly sampled to generate the continentally homogeneous parameter distribution (i.e., all the sites use the same posterior parameter distribution to generate SOC distribution) or used in training, validating, and testing a neural network to predict parameter values at each grid of the map (continentally heterogenous parameter). The one-batch method can directly generate the continentally homogenous parameter distribution by taking all observational SOC profiles as one batch in data assimilation. The continentally homogeneous distributions by the random-sampling method and one-batch method and gridded maps of parameters by the neural networking method were then applied to CLM5 to generate the SOC distribution of the conterminous United States.