Accounting for carbon allocation shifts after drought improves NBE predictions

Drought affects carbon fluxes by influencing photosynthesis, respiration, and disturbance, as well as by impacting the size of the different vegetation and soil pools. Changes to how much photosynthetic carbon is allocated to foliage, woody components, roots, and more persist even after meteorological conditions have returned to normal. These shifts in carbon allocations in turn affect future photosynthesis through changes in leaf area and water uptake ability, for instance. However, the magnitude of post-drought recovery effects on terrestrial carbon fluxes are often overlooked and poorly modeled. This is especially the case in tropical ecosystems as there are large uncertainties in the tropical ecosystem sensitivity to climate forcing such as drought. We hypothesize that a key driver of carbon flux predictive error in land surface model simulations is their representation of changes in carbon allocation during and after water stress. Most models use static carbon allocation schemes which, given their assumption of uniformity through time, do not account for the impact of drought. Those that do have dynamic allocation are often poorly parameterized due to the difficulty of in situ carbon allocation measurements. We investigate first whether it is possible to constrain dynamic carbon allocation based on observations of fluxes and model data fusion. We then investigate whether the constrained dynamic carbon allocation model has improved predictions of modeled net biosphere exchange (NBE) during drought and drought recovery. To do so, we implement a dynamic carbon allocation scheme within the CArbon DAta MOdel fraMework (CARDAMOM) data assimilation system, which robustly optimizes the parameters and carbon cycle states of an intermediate-complexity ecosystem model based on a suite of observational data on carbon fluxes and pools. We test the dynamic allocation scheme at a wet tropical (French Guiana) and a dry tropical (Cumberlands Plain) flux tower site. We find at the Cumberland Plains flux tower that the dynamic allocation model outperforms the static allocation model in predicting NBE. Furthermore, we retrieve significant carbon allocation shifts during drought periods at this site with increasing allocation to autotrophic respiration, wood biomass, and labile biomass and decreasing allocation to foliage biomass and fine root biomass relative to non-drought conditions. Our results demonstrate the importance of accounting for stress-induced carbon allocation shifts in land surface models as well as the ability to infer carbon allocation shifts from flux measurements.