Are locally trained allometric functions of forest aboveground biomass universal across spatial scales and forest disturbance scenarios?

• A tropical forest near Manaus in Brazil was simulated using the process-based model FORMIND. • Old-growth forests, wildfires and high-impact logging as scenarios. • Tree height to biomass allometric models were tested across scales. • Model parameters significantly change across scales. • Localized forest structure adaptation addresses the issue of applying off-scale allometries. Large scale above-ground-biomass (AGB) estimation remains highly uncertain. Multi-sensor, multi-scale and multi-temporal analyses are crucial for capturing the dynamics and the heterogeneity of forests. The European Space Agency’s BIOMASS mission will play a key role in future biomass monitoring. Considering the differences in the spatial scales of input datasets, it is essential to investigate these scale effects. This study examines whether locally trained allometric relationships between forest height and AGB are scale-dependent and how forest disturbances impact these estimates. Using the forest gap model FORMIND, initialized with inventory data from tropical lowland forests close to Manaus (Brazil), we simulated forest height and AGB raster products at resolutions ranging from 20 m to 200 m based on various forest height metrics. Through regression analysis, allometric parameter sets for each resolution step were derived. We then tested the impact of applying these parameters under various conditions, including off-scale and off-scenario usage. Our results show that applying allometric parameters at mismatched spatial scales introduces significant additional errors. This error becomes more prominent as scale differences increase. Additionally, the type and severity of forest degradation scenario strongly influences the estimation quality. However, dynamically adapting allometric parameter sets to local conditions mitigates these errors. Applying the locally trained parameters to varying disturbance scenarios results in substantial errors, underscoring the importance of incorporating local forest structure in AGB models. While using off-scale allometric parameters is possible, it introduces additional challenges. Our study highlights the need for local forest structure products to improve large-scale AGB estimation.