Tree mortality submodels drive simulated long-term forest dynamics: assessing 15 models from the stand to global scale

Harald Bugmann, Rupert Seidl, Florian Hartig, Friedrich Bohn, Josef Brůna, Maxime Cailleret, Louis François, Jens Heinke, Alexandra-Jane Henrot, Thomas Hickler, Lisa Hülsmann, Andreas Huth, Ingrid Jacquemin, Chris Kollas, Petra Lasch-Born, Manfred J. Lexer, Ján Merganič, Katarína Merganičová, Tobias Mette, Brian R. Miranda, Daniel Nadal-Sala, Werner Rammer, Anja Rammig, Björn Reineking, Edna Roedig, Santi Sabaté, Jörg Steinkamp, Felicitas Suckow, Giorgio Vacchiano, Jan Wild, Chonggang Xu & Christopher P. O. Reyer

Abstract

Models are pivotal for assessing future forest dynamics under the impacts of changing climate and management practices, incorporating representations of tree growth, mortality, and regeneration. Quantitative studies on the importance of mortality submodels are scarce. We evaluated 15 dynamic vegetation models (DVMs) regarding their sensitivity to different formulations of tree mortality under different degrees of climate change. The set of models comprised eight DVMs at the stand scale, three at the landscape scale, and four typically applied at the continental to global scale. Some incorporate empirically derived mortality models, and others are based on experimental data, whereas still others are based on theoretical reasoning. Each DVM was run with at least two alternative mortality submodels. Model behavior was evaluated against empirical time series data, and then, the models were subjected to different scenarios of climate change. Most DVMs matched empirical data quite well, irrespective of the mortality submodel that was used. However, mortality submodels that performed in a very similar manner against past data often led to sharply different trajectories of forest dynamics under future climate change. Most DVMs featured high sensitivity to the mortality submodel, with deviations of basal area and stem numbers on the order of 10–40% per century under current climate and 20–170% under climate change. The sensitivity of a given DVM to scenarios of climate change, however, was typically lower by a factor of two to three. We conclude that (1) mortality is one of the most uncertain processes when it comes to assessing forest response to climate change, and (2) more data and a better process understanding of tree mortality are needed to improve the robustness of simulated future forest dynamics. Our study highlights that comparing several alternative mortality formulations in DVMs provides valuable insights into the effects of process uncertainties on simulated future forest dynamics.

What the paper shows and why it matters

(AI-generated)

Dynamic vegetation models, which simulate how forests develop over decades to centuries, are used to project climate-change effects. Growth alone would change forests fairly smoothly, but mortality could bring fast, strong change. Models represent tree death with theoretical, empirical (fitted to inventory data) or highly mechanistic formulations; which is right cannot be decided in advance, and past data typically cover only a few decades. Sensitivity to mortality had been tested only in a fairly small number of case studies. The paper asks how large the uncertainty from the mortality formulation is, across model types and relative to climate change itself, when past data usually cannot decide between formulations.

The paper argues that the formulation of tree mortality in dynamic vegetation models is a major source of uncertainty in projected forest dynamics, and that agreement with past data does not resolve it. It ran 15 models (eight at stand, three at landscape, four at continental to global scale), each with at least two mortality formulations. Where models could be tested against long inventory series (six of the 15), most matched the records under either formulation, yet equally good fits often diverged sharply under climate change. In the stand-scale models, the sensitivity of basal area (tree cross-section per area) and stem numbers to the mortality formulation was at least of the same order of magnitude as their sensitivity to the climate scenario. In many stand-scale models, sensitivity rose with the severity of climate change and was highest in the models built on empirical formulations, some of which the paper considers likely to be applied beyond their calibration range. The three landscape-scale models were less sensitive to the mortality formulation and, unlike the stand-scale models, more to climate; the paper does not generalise from so few, whose alternative formulations were structurally similar.

**Figure 2.** Simulation of basal area (m²/ha) by the eight stand-scale models at the respective sites (cf. Table 3) for 200 yr into the future (with the exception of 4C, for which the simulation was ended in the year 2100 in all cases). From Bugmann et al. (2019), *Ecosphere*, licensed under CC BY 4.0. Click the image to enlarge.

For about a century ahead, the formulations mostly gave similar trajectories, which the paper reads as good news for management planning. For longer horizons, such as the role of forests in the carbon cycle, using a single mortality formulation understates uncertainty. The paper suggests several conceptually different formulations or formal sensitivity analysis, calls for calibration against multiple data sources, and for more mortality data.