A database with refined plant functional type attribution at eddy covariance sites for enhanced land surface model simulations

Abstract. Measurements and estimates of carbon, water, and energy fluxes between the biosphere and the atmosphere, provided by eddy covariance (EC) techniques, are crucial to understand ecosystem functioning and for evaluating and improving land surface models (LSMs). Because most LSMs still rely on the vegetation description in the form of plant functional types (PFTs), accurate characterization of PFT fractions at EC sites is essential to reduce uncertainty in simulated fluxes, as well as in model parameters when EC data are used for calibration (data assimilation). However, the PFT information directly usable by LSMs is not provided by networks tower teams, leading to each LSM having its own methodology to prescribe PFTs.

This study presents a PFT attribution methodology and an associated database for 229 sites from the FLUXNET dataset and from regional networks, relying on the combination of multiple data sources, including literature review, information from network websites, satellite imagery, and a land cover map at 300 m. The resulting database of PFT fractions (based on 13 generic PFTs and 15 ORCHIDEE PFTs) across EC sites is associated with a confidence flag, based on the consistency between the various sources. Since in situ data do not always come with gap-filling information or can sometimes contain errors in the gap-filling process, we applied an automatic identification and removal of spurious data leading to a data coverage flag. Then, to assess the impact of this new PFT attribution on LSM simulations, we ran the ORCHIDEE LSM for simulating gross and net carbon fluxes, as well as energy fluxes. Those simulations led to the introduction of a model-data agreement flag (based on variability, correlation, and bias). The methodology, transposable to other LSMs with their own associated PFTs, provides a traceable framework for improving the PFT fractions attribution; the flags introduced aim at identifying the most relevant sites for validation and model calibration. The database is publicly available in CSV and XLSX format at https://doi.org/10.14768/6e99ff4b-3205-4038-8f9a-713e3e33753d (Tartaglione et al., 2026) and can be interactively explored at https://data.ipsl.fr/ecsitepftexplorer/.