Using ecosystem experiments to improve vegetation models

NATURE CLIMATE CHANGE | VOL 5 | JUNE 2015 | www.nature.com/natureclimatechange The response of the terrestrial biosphere to increasing atmospheric CO2 concentration (Ca) is a major uncertainty in models projecting future climate change, because of the critical feedback between terrestrial ecosystem carbon (C) cycling and the atmosphere1–3. Current Earth system models disagree strongly on the size of this feedback2. This disagreement results from differences in the projected increase of plant production owing to CO2 fertilization, including its interactions with terrestrial nitrogen (N)4–6 and phosphorus7 cycles, as well as differences in the turnover times of C in terrestrial ecosystems. These differences among models imply that our predictive understanding of the effects of Ca on terrestrial C storage remains very low. FACE experiments aim to investigate how terrestrial ecosystems respond to elevated atmospheric CO2 concentration (eCa). In general, FACE experiments are ecosystem-scale, long-term, multi-investigator experiments that provide an extraordinarily rich source of data on plant and soil processes mediating ecosystem-level responses to eCa (refs 9,10). However, the full richness of these experiments has rarely been exploited to constrain model uncertainty, with model outputs typically only being compared against the response of net primary production (for example, refs 11–13). The FACE Model–Data Synthesis (FACE-MDS) project14 aimed to benefit from the wide range of complementary data sets available from these experiments to better constrain eCa responses in models. To do so, we followed an ‘assumption-centred’ approach, in which we studied the underlying assumptions that models use to represent key ecosystem processes, and evaluated which assumptions best represented the experimental data. We used 11 process-based models, including four stand-scale ecosystem models (DAYCENT15, ED216, GDAY17 and TECO18), five land surface models (CABLE19, CLM420, EALCO21, ISAM22 and O-CN23) and two dynamic vegetation models (LPJ-GUESS24 and SDGVM25). These models were compared with data from two Using ecosystem experiments to improve vegetation models