Effects of bias in an LAI data assimilation system on carbon uptake and hydrological variables and over Europe

Abstract

Vegetation is a major control on land-atmosphere fluxes of carbon and water. An improved representation of vegetation in land surface and dynamic vegetation models can therefore improve both short-term weather predictions as well as long-term climate projections.  State update data assimilation (DA) of remotely sensed leaf area index (LAI) is one way to obtain vegetation state estimates consistent with physical constraints from a land surface model and observational data. Most LAI DA studies so far used bias-blind DA systems, i.e. they did not explicitly take bias between observations and model into account. However, if the observations are biased against the land surface model, this might hamper  the performance of the DA system, because it can induce instabilities in the model. We therefore examined the effect of bias on an LAI DA system, and compared a bias-blind LAI DA system with bias-aware approaches.   For this purpose, we assimilated the Copernicus Global Land Service (CGLS) LAI into the Noah-MP land surface model over Europe in the 2002-2019 period.  We find that in areas with large LAI bias, the bias-blind LAI DA by design leads to a reduced bias between observed and modelled LAI and GPP, but it also has strong impacts on soil moisture, leading to a worse agreement with independent, satellite-derived ESA CCI soil moisture. Furthermore, the bias-blind LAI DA produces a pronounced sawtooth pattern due to model drift between update steps. This drift also propagates to short-term estimates of GPP and ET. Furthermore, internal DA diagnostics indicate suboptimal DA system performance.  The bias-aware approaches avoid the negative effects of the bias-blind assimilation, and still improve anomaly estimates of LAI. Therefore, bias-aware LAI DA might be a useful method to consider in LAI DA, especially when anomalies of LAI or GPP are of interest.  Our results furthermore show that LAI of CGLS and Noah-MP show strong disagreement especially in dry climates. Model calibration or DA methods that include parameter updating could be an alternative to bias-aware DA to reduce these discrepancies. Our results can guide such efforts, and highlight the need for multiple constraints.