Projecting Forest Dynamics Across Europe: Potentials and Pitfalls of Empirical Mortality Algorithms

Abstract

Mortality is a key process of forest ecosystem dynamics and functioning strongly altering biomass stocks and carbon residence times. Dynamic vegetation models (DVMs) used to predict forest dynamics are typically based on simple, largely data-free ('theoretical') mortality algorithms (MAs). To improve DVM projections, the use of empirically-based MAs has been suggested, but little is known about their impact on DVM behavior. A systematic comparison of eight MAs (seven inventory-based, one 'theoretical') for the pan-European tree species Pinus sylvestris L. was carried out within the DVM ForClim for present and future climate scenarios at three contrasting sites across Europe. Model accuracy was furthermore evaluated with empirical data from young and old-growth forests. We found strongly diverging mortality patterns among the MAs for present climate. Based on their behavior, we identified two distinct empirical MA groups that were related to their structure (i.e., variables considered), but not to their geographic origin (i.e., the environmental conditions they were calibrated to). Under climate change, MAs based on a competition index produced ecologically inconsistent results, while MAs based on growth showed more plausible and less extreme behaviors. Furthermore, MAs based on growth reached a higher accuracy for projecting young and old-growth forest dynamics. Our results demonstrate that using empirical MAs in DVMs has a high potential to better predict forest dynamics, but also a risk of yielding implausible results if their structure is inadequate. For DVM applications across large spatiotemporal scales, we thus suggest using MAs based on growth, particularly under future no-analogue climates.

Bayesian calibration of a growth‐dependent tree mortality model to simulate the dynamics of European temperate forests

“…, Thrippleton et al. ). Direct parameter estimates (DPE) for this model were derived using single‐tree inventory data from the calibration plots ( n = 4,663).…”

Stand‐scale climate change impacts on forests over large areas: transient responses and projection uncertainties

“…In particular, assumptions on sapling densities have shown to be of key importance for model output but are associated with high uncertainty (Huber et al 2018, 2020). Moreover, there is considerable uncertainty with a large influence on model projections regarding the formulations of mortality and of the allocation of volume increment to height vs. diameter growth (see also Rasche et al 2012, Hülsmann et al 2018, Bugmann et al 2019, Thrippleton et al 2020 b ). Hence, we applied eight model versions (Appendix : Table S2) that represent a factorial combination of different formulations introduced by Huber et al (2020) with respect to the establishment probability, the allocation to height vs. diameter growth, and the background mortality.…”

“…S1), we focus on residual stand BA (Data and Data ). Due to the large number of simulations (128 per stratum, see Table 1), results are presented for one particular model version (i.e., model version 22, Appendix : Table S2), which has been evaluated and shown to provide accurate results when applied to a wide range of situations (Isler 2019, Huber et al 2020, Thrippleton et al 2020 b ). In addition, we assessed the robustness of the projected BA changes across the eight model versions by estimating the inter‐model variability of the responses for the different forest strata under different climatic and site conditions (i.e., standard deviation of the share of strata falling into a given impact category; for details, see caption of Appendix : Fig.…”

Capturing ecological processes in dynamic forest models: why there is no silver bullet to cope with complexity

“…Instead of an alternative theoretical mortality formulation, other types of formulations could have been used, as, for example, empirical formulations. Such formulations have however been explored in other recent studies (Hülsmann et al 2018, Bugmann et al 2019, Thrippleton et al 2020). In this study, when comparing the alternative to the standard formulation, we expect the alternative formulation to yield more realistic forest structure for long‐term simulations of unmanaged stands, where mortality of large trees is particularly relevant (Worrall and Harrington 1988, Nagel and Diaci 2006, Trotsiuk et al 2012, Holzwarth et al 2013, Hobi et al 2015).…”