Structural analysis of three global land models on carbon cycle simulations using a traceability framework

Abstract. Modeled carbon (C) storage capacity is largely determined by the C residence time and net primary productivity (NPP). Extensive research has been done on NPP dynamics but the residence time and their relationships with C storage are much less studied. In this study, we implemented a traceability analysis to understand the modeled C storage and residence time in three land surface models: CSIRO's Atmosphere Biosphere Land Exchange (CABLE) with 9 C pools, Community Land Model (version 3.5) combined with Carnegie-Ames-Stanford Approach (CLM3.5-CASA) with 12 C pools and Community Land Model (version 4) (CLM4) with 26 C pools. The globally averaged C storage and residence time was computed at both individual pool and total ecosystem levels. The spatial distribution of total ecosystem C storage and residence time differ greatly among the three models. The CABLE model showed a closer agreement with measured C storage and residence time in plant and soil pools than CLM3.5-CASA and CLM4. However, CLM3.5-CASA and CLM4 were close to each other in modeled C storage but not with measured data. CABLE stores more C in root whereas CLM3.5-CASA and CLM4 store more C in woody pools, partly due to differential NPP allocation in respective pools. The C residence time in individual C pools is greatly different among models, largely because of different transfer coefficient values among pools. CABLE had higher bulk residence time for soil C pools than the other two models. Overall, the traceability analysis used in this study can help fully characterizes the behavior of complex land models.