Model-data fusion of daily ecosystem fluxes (GPP) and forest inventories (DBH) with the process-based dynamic vegetation model BiomeEP

The heterogenous structure and diversity of forest stands shape resource availabilities to individual trees and lead to diversity in stress responses. Heterogeneity in size and/or traits thus strongly determines tree and ecosystem carbon balances. While short-term net ecosystem carbon and water exchange might be well approximated by the average behavior of the top canopy trees, we expect both longer-term structural shifts as well as stress responses to extreme conditions to be more strongly dependent on the demography of below-canopy trees. Dynamic vegetation models (DVM) resolve and track this heterogeneity. They simulate growth, mortality, competition, and carbon allocation strategies. Thereby they propagate changes in environmental conditions into changes in structure of forest ecosystems.

Here, we use daily ecosystem flux measurements of gross primary production (GPP) and multi-annual forest inventories (distribution of diameter at breast height, DBH) as observational constraints to calibrate BiomeEP (Weng et al., 2015). BiomeEP is a process-based DVM grounded in optimality principles for computational efficiency and parameter sparsity. It includes acclimation of photosynthetic capacity via P-model (Stocker et al., 2020), a perfect plasticity assumption for canopy layering (Strigul et al., 2008) and makes use of empirical allometric relationships.

Our model implementation shows single-core runtimes (including 2000 years of spin-up) on the order of seconds for individual sites and thus enables site-specific inference of physiological traits. As next step the PROFOUND data set (Reyer et al. 2020) containing GPP and DBH data from seven European sites will be used for model data fusion. We aim to reproduce observed GPP and DBH distributions (targeting both absolute numbers and relative size distribution) and estimate parameter identifiability as well as across-sites generalizability. Preliminary results demonstrated the sensitivity of targets on the species-specific rate parameters for growth and mortality. Across-site generalizable parameters could pave the way to inform large scale predictions under future environmental changes.