A new ForestPaths study, written by project partners Lund University and Natural Resources Institute Finland (Luke),  alongside 21 other universities and institutions spanning across 11 countries, shows the importance of Vegetation Demographic Models (VDMs) for simulating forest recovery, structure and competition dynamics, for projecting carbon cycling, with added implications for potential exploration of large-scale forest management strategies in earth system models. Although these models are increasingly incorporated into these Earth System Models, they differ in their demographic assumptions and until now, there has been no quantitative comparison of their performance.  

In the study, nine VDMs, namely BiomeE, BiomeEP, CABLE-POP, EDv3, ELM-FATES, JULES-RED, LPJ-GUESS, ORCHIDEE, and SEIB-DGVM, were benchmarked against observational data from boreal, temperate and tropical forest sites. Model performance was evaluated based on predictions of tree growth, carbon turnover, biomass stocks and tree size distributions. All modelled forests were subjected to a total forest disturbance and their recovery monitored for at least 420 yr. The analysis showed that post disturbance carbon recovery trajectories varied considerably among models, but largely remained within the range of the observations. Initial forest regrowth rates differed substantially, spanning 0.03–0.60 kgCm−2 yr−1 in boreal forests, 0.18–0.70 kgCm−2 yr−1 in temperate forests and 0.35–1.10 kgCm−2 yr−1 in tropical forests. These differences are probably influenced in part by each model’s initial forest state after the disturbance. While the models reproduced mature forest carbon content well, this often was caused by compensating effects between overestimated growth and underestimated mortality rates.  

All in all, the benchmarking shows that VDMs generally replicate carbon pools and forest structure in mature forests reasonably well and capture emergent ecological processes such as self-thinning, as well as plausible interactions between forest structure and carbon fluxes along the forest recovery phases. At the same time, variability remains high in the duration of stand development phases and there is divergence of growth rates and carbon turnover times from observations. However, because VDMs resolve the forest structure and these growth and mortality dynamics much more realistically than simpler vegetation models, these models offer an opportunity to use similar types of observations as were used for benchmarking here as more direct “points of contact” between models and data, necessary for model constraining, which means that ultimately they can be much better constrained. More data is needed to constrain especially the early post- disturbance forest structure and recovery dynamics, where models vary a lot but which are phases that many forests of the world will increasingly find themselves in this first multi-model benchmarking effort of its kind identifies constraining growth and mortality rates as well as earl forest recovery as a clear opportunity to strengthen predictions of forest responses to environmental change. 

Read the full study here.