Extended Data Fig. 1: Graphical overview of the analysis approach.

We initialized the simulated forests (16 ×16 km grid cells) using recent species distribution maps³⁸ and age class information extrapolated from NFI data for continental Europe⁵⁴. Forest growth was simulated based on soil- and climate-sensitive NPP estimates for each tree species, derived from a deep neural network trained on simulations of a process-based forest growth model. NPP values were mapped to yield tables to obtain information on merchantable timber volume and mean tree diameter. We simulated an even-aged clear-cut system (pictograms in top row), with the rotation length varying between cells. To calculate optimal rotation length per cell, we converted extracted timber volumes into economic cashflow and computed the net present value for each possible rotation length (from 0 to 260 years in 10-year intervals), assuming a discount rate of 1.5%. The optimal rotation length was defined as the one that maximizes net present value. After final harvest, we assumed the area was regenerated with the same tree species. To quantify the effect of disturbances, we explicitly simulated two types of disturbances: First, climate-sensitive, background disturbances derived from empirically parameterized hazard probabilities⁵⁹ (center row), and second, stochastic extreme disturbance events informed by observations from remote sensing⁶¹ and scaled to future scenarios using Taylor’s power law equations⁶³ (bottom row). In the event of a disturbance, the revenues from timber were reduced for background disturbances, and set to zero for extreme disturbances, representing the combined effects of market price responses, wood devaluation, and increased harvesting costs in the wake of disturbances³². Figure created with BioRender.com.