Deriving gross & net primary productivity (GPP & NPP) and carbon turnover time of forests from remote sensing remains challenging. This study presents a novel approach to estimate forest productivity by combining radar remote sensing measurements, machine learning and an individual-based forest model. In this study, we analyse the role of different spatial resolutions on predictions in the context of the Radar BIOMASS mission (by ESA). In our analysis, we use the forest gap model FORMIND in combination with a boosted regression tree (BRT) to explore how spatial biomass distributions can be used to predict GPP, NPP and carbon turnover time (τ) at different resolutions. We simulate different spatial biomass resolutions (4 ha, 1 ha and 0.04 ha) in combination with different vertical resolutions (20, 10 and 2 m). Additionally, we analysed the robustness of this approach and applied it to disturbed and mature forests. Disturbed forests have a strong influence on the predictions which leads to high correlations (R2 > 0.8) at the spatial scale of 4 ha and 1 ha. Increased vertical resolution leads generally to better predictions for productivity (GPP & NPP). Increasing spatial resolution leads to better predictions for mature forests and lower correlations for disturbed forests. Our results emphasize the value of the forthcoming BIOMASS satellite mission and highlight the potential of deriving estimates for forest productivity from information on forest structure. If applied to more and larger areas, the approach might ultimately contribute to a better understanding of forest ecosystems.