Figure 1.

Conceptual view on the scaling vegetation dynamics (SVD) framework. Vegetation transitions on a single cell are estimated by a Deep Neural Network (b) contingent on environmental factors (a), the current vegetation state (S), the residence time (R) and the spatial context. The model determines transitions by sampling from the DNN‐derived probability distributions for the future state (S*) and the time until state change (ΔR). Human and natural disturbances (c) add an abrupt pathway for vegetation transitions and will be described in future work. Density distributions of ecosystem attributes of interest are linked to combinations of S × R (d). These state‐ and residence time‐specific attribute distributions can subsequently be used to predict changes in the spatial distribution of these attributes based on the simulated vegetation transitions SVD (e)