Carbon exchange in tropical land, as seen from both bottom-up and top-down perspectives

The tropical land (< 23° absolute latitude) is crucial to global carbon dynamics. In tropical forests, net flux is a small residual between large gross fluxes. In grasslands and savannas, carbon flux is critically dependent on seasonality of precipitation. Here, we investigate behavior of bottom-up and top-down models of carbon dynamics, and evaluate the physical processes that control behavior at the surface as a function of what these models suggest.

Bottom-up models, or Land Surface Models (LSMs) predict CO2 flux from meteorological forcing. Top-down (inversion) models convolve CO2 observations, from either surface or satellite data, with transport models to obtain an `optimized' CO2 flux. LSM flux is frequently used during inversions as a `prior', or first-guess estimate of surface flux. As LSMs generate flux in response to environmental forcing, they provide a physically-based representation, or an answer to the question `why', but this answer is unfortunately poorly constrained with observations in tropical land regions. Inversions provide a `what' answer, but do not have the ability to dig into mechanisms and processes. By combining LSM and optimized flux, we have an opportunity to use the `what' answer from inversions as a tool to better understand the `why' of bottom-up models.

Optimized flux from a suite of models that participated in the OCO-2 inversion exercise represent the top-down flux. We use the ensemble mean flux, and the individual inversion models utilize their own estimates of prior flux and modeled transport. Our LSM flux is obtained from the SiB4 and ORCHIDEE models. We confront flux from the individual LSMs with optimized flux, on a pan-tropical, continental (South America, Africa, Maritime Continent), and northern hemisphere/southern hemisphere basis.

We find neither LSM superior to the other when compared to optimized flux from the inversion exercise. On a pan-tropical basis, ORCHIDEE has responsiveness similar to the optimized flux, while SiB4 is more `sluggish', or not showing as much amplitude in annual cycle of flux. However, ORCHIDEE has much larger drawdown of CO2 than either SiB4 or the optimized flux suggest. When broken out on a regional/hemispheric basis, the comparison becomes more heterogeneous. Both models perform worst statistically in South America, best in Africa.