One of the challenges in obtaining accurate estimates of CO2 fluxes at the Earth’s surface is the still high uncertainty related to biospheric carbon exchange. Estimates for the net uptake of CO2 by the biosphere (NEE) are typically obtained by assimilating measurements of atmospheric CO2 mole fractions into a framework that includes biospheric surface fluxes. Due to the sparse coverage of observation locations and the poorly known error covariance structure, the resulting spatial NEE patterns remain to a large extent determined by the patterns predicted by the original biosphere model. As these can be validated at only a limited set of surface eddy-covariance sites, global NEE fluxes remain only weakly constrained by observations.

We aim to alleviate these limitations by making use of sun-induced fluorescence (SIF) and temperature observations, which are both available at high spatial and temporal coverage. In Rödenbeck et al. (2018) it was shown that the mean seasonal cycle and long-term trend of NEE can generally be represented by a simple statistical function, and that interannual variability in NEE can be introduced through a linear regression of temperature anomalies onto NEE anomalies. Additionally, remotely-sensed SIF recently emerged as a powerful proxy for GPP anomalies on regional to global scale. 

Inspired by these findings, the CarbonTracker data assimilation system (which is based on a sequential ensemble square root filter algorithm) was modified to allow for direct optimization of statistical function parameters that describe long-term and seasonal NEE, and monthly anomaly sensitivities.  A single set of these parameters, valid for the full temporal window, is optimized per ecoregion subject to an atmospheric CO2 constraint in a global inversion. Like for temperature, we will show that SIF anomalies can be used in a similar way to capture NEE anomalies on interannual time scales. We will present how this new set-up exploits the spatiotemporal patterns of SIF and temperature to improve NEE estimates.