Data assimilation strategy for E3SM land model carbon-nitrogen-phosphorus cycles

Thursday, December 13, 2018 - 08:00
Add to Calendar

Carbon (C), nitrogen (N), and phosphorus (P) interdependently cycle through terrestrial ecosystems, which results in complex dynamics of the dominant controllers of system processes (e.g., vegetation activities and soil biogeochemistry). Critical C-N-P interactions impact the ecosystem through multiple avenues, for example, at leaf level carbon acquisition (photosynthesis), whole plant growing strategy (resource allocation), and root resource foraging strategy (plant-soil nutrient competition). Models of those critical C-N-P processes have been developed and observations are collected across multiple scales (in situ, regional, and global) to explain these dynamics and their effects on global exchanges with the atmosphere.

Nevertheless, one of the greatest challenges is how to effectively assimilate available datasets to improve model representations of C-N-P cycles and their interactions. Here we explored two different data assimilation strategies for the E3SM land model (ELM). One is the traditional approach that assimilates site level observations based on Plant Functional Type (PFT) and extrapolates the optimal model to global scale based on the global PFT distribution. Secondly, we directly assimilate upscaled global datasets and improve model parameterization at a few selected representative gridcells (0.5x0.5 resolution). We found that although the two approaches required similar computational resources, the latter was much more effective since it directly targeted gridcell-scale processes. The traditional parameterization strategy failed mainly due to (1) site level observations typically measure only a few important aspects of the system (e.g., carbon fluxes, but not carbon stocks) and (2) disconnections between in-situ-scale and gridcell-scale dynamics that could not be fully explained by PFT properties. Our results highlight the benefit of assimilating relevant datasets that have a similar scale of interest to the earth system land model to improve model predictability at the global scale.

Link for More Information: