Aim To estimate the effects of full-glacial atmospheric CO₂ concentrations and climate upon leaf area index (LAI), using both global vegetation models and palaeoecological data. Prior simulations indicate lowered LAIs at the Last Glacial Maximum (LGM), but this is the first attempt to corroborate predictions against observations.

Location Eastern North America and eastern Beringia.

Methods Using a dense surface pollen data set and remotely sensed LAIs from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument, we evaluate the ability of analogue-based techniques to reconstruct modern LAIs from pollen data. We then apply analogue techniques to LGM pollen records, calculate the ratio of LGM to modern LAIs (RLAI) and compare reconstructed RLAIs to RLAIs simulated by BIOME4. Sensitivity experiments with BIOME4 distinguish the effects of CO₂ and climate on glacial LAIs.

Results Modern LAIs are skilfully predicted (r(2) = 0.83). Data and BIOME4 indicate that LAIs at the LGM were up to 12% lower than modern values in eastern North America and 60-94% lower in Beringia. In eastern North America, LGM climates partially counteracted CO₂-driven decreases in LAI, while in Beringia both contributed to lowered LAIs.

Main conclusions In both regions climate is the primary driver of LGM LAIs. The decline in eastern North America LAIs is smaller than previously reported, so regional vegetation feedbacks to LGM climate may have been less significant than previously supposed. CO₂ exerts both physiological and community effects upon LAI, by regulating resource availability for leaf production and by influencing the competitive balance among species and hence the composition and structure of plant communities. Pollen-based reconstructions using analogue methods do not incorporate the physiological effect and so are upper estimates of full-glacial LAIs. BIOME4 sensitivity experiments indicate that the community and physiological effects together caused 10% to 20% decrease in LAIs at the LGM, so simulated RLAIs that are 80-100% of reconstructed RLAIs are regarded as consistent with data.