ABSTRACTAimLand surface models (LSMs) currently represent each plant functional type (PFT) as an average phenotype, characterised by a set of fixed parameters. This rigid and constant representation is a limit in understanding the dynamics of highly diverse ecosystems, such as permanent grasslands, and their response to global change.LocationFrance.Time Period2001–2019.Major TaxaGrassland plant species.MethodsWe incorporated spatially explicit trait variability at the France scale in the ORCHIDEE land surface model to assess how the net primary productivity (NPP) will spatially vary over the years. More precisely, we focused on three key functional traits that govern the NPP of grassland ecosystems: specific leaf area (SLA) and leaf nitrogen content (LNC), as measured traits, and leaf lifespan (LLS) as an estimated trait. Community‐weighted means (CWM) were implemented in various combinations with prescribed and spatially varying traits. We compared the outcomes of each NPP simulation to remotely sensed proxies of productivity by using the MODIS satellite‐driven NPP products.ResultsThe sensitivity of NPP to traits depends on climate conditions, such as temperature and water limitation. Considering trait variability decreases the NPP in the most productive regions (plains) and increases the NPP in the less productive regions (mountains) compared to the case with constant trait values. This leads to a more homogenous NPP across France. Compared to the observed MODIS NPP and FLUXCOM GPP, the simulation using varying traits improves the spatial NPP and GPP variations in several regions and most climate conditions.Main ConclusionsBased on the existing trait data, we revealed that incorporating the CWM of traits in an LSM such as ORCHIDEE can be effectively performed. Improving the modelling and predictions by considering the relationships between biodiversity, functional biogeography, and ecosystem functioning is essential in current and future ecological research.