Solar Induced Fluorescence (SIF) has significant potential to constrain the carbon cycle in land surface models. This requires either that the target variables in the model are first retrieved from the SIF data (for example, by estimating Gross Primary Productivity, GPP), or that the SIF is directly predicted from the model itself using a so-called observation operator. In the retrieval problem it is difficult to guarantee that assumptions made in the retrieval scheme are consistent with the assumptions in any given land surface model. Observation operators, on the other hand, offer the potential to enforce that consistency, but this comes with additional complexity. Ideally, observation operators should themselves be consistent with the assumptions inside the land surface model. If that is not the case, mismatches between the modelled and observed SIF can arise purely due to the observation operator, potentially resulting in biases. With a perfectly consistent system, we can be confident that any discrepancies are due to the underlying land model itself, and hence the discrepancies with the observed SIF inform us about land surface model.

This presentation describes an observation operator that is physically consistent with the two-stream radiative transfer scheme of Sellers (1985) commonly used in land surface models to represent the interaction of sunlight with vegetation canopies. We describe the derivation of the new observation operator and how it can be used to predict SIF. The scheme is numerically efficient, and can be easily extended to work with vertically inhomogeneous canopies. We show results from the JULES model (the land surface scheme of the UK’s flagship climate model UKESM) for both GPP and SIF at eddy covariance sites.