In the early stage of numerical model development, owing to limited computation capabilities simplified formulation were used to simulate physical processes. Hence, in the 80s and 90s, climate models are structured to simulate atmospheric and oceanic processes at the scale of the Earth system. It is only in the 2000s, that, for instance, ecosystems components have been included in the IPSL- Earth system climate model which has resulted in a continuous increase of model complexity. Today, land-cover classes are commonly employed to consider ecosystems diversity by vegetation types such as forest, grassland or crops and by climatic zones, i.e., tropical, temperate and boreal areas. Still, some uncertainties remain in global ecosystem models due to ecosystem variety among a single type of land-cover class.Using three studies, I will present results on the significant of ecosystem heterogeneity for the estimation of greenhouse gases and of the soil organic carbon stocks. The first study shows that spatial variability in soil properties is significantly driving the relationship between soil organic carbon decay and soil moisture content implicating notable uncertainty in global soil organic carbon stocks estimates by the ORCHIDEE ecosystem model. In the second study, the diversity of northern peatlands vegetation will be shown to control methane emissions. Indeed, three transport processes namely diffusion, ebullition and aerenchyma transport enable methane produced in soil to escape to the atmosphere. While diffusion and ebullition depend essentially on physical processes, aerenchyma transport efficiency lean on plant species. The third study focus on the contribution of lakes on climate and the evaluation of lakes cover classes level on greenhouse gases emissions.