Sensitivity of ecosystem goods and services projections of a forest landscape model to initialization data

Projections of indicators of forest ecosystem goods and services (EGS) based on process-based landscape models are critical for adapting forest management to climate change. However, the scarcity of fine-grained, spatially explicit forest data means that initializing these models is both a challenge and a source of uncertainty. To test how different initialization approaches influence the simulation of forest dynamics and EGS indicators we initialized the forest landscape model LandClim with fine resolution empirical data, coarse empirical data, and simulation-derived data, and evaluated the results at three spatial scales (stand, management area and landscape). Simulations were performed for a spruce (Picea abies) dominated landscape in the Black Forest, Germany, under current climate and a climate change scenario. We found that long-term (>150 years) projections are robust to initialization uncertainty. In contrast, shorter-term projections are sensitive to initialization uncertainty, with sensitivity increasing when EGS are assessed at smaller spatial scales, and when the EGS indicators depend on the spatial distribution of individual species. EGS dynamics are strongly influenced by interactions between the density, species composition, and age structure of initialized forests and simulated forest management. If EGS dynamics are strongly influenced by climate change, such as when climate change induces mortality in drought-sensitive species, some of the initialization uncertainty can be masked. We advocate for initializing landscape models with fine-grained data in applications that focus on spatial management problems in heterogeneous landscapes, and stress that the scale of analysis must be in accordance with the accuracy that is warranted by the initialization data.