Future Predictions of Ecosystem Changes in California’s Sierra Nevada over the coming century using Remote-Sensing Constrained Terrestrial Biosphere Model Simulations

Reliable predictions of ecosystem dynamics and carbon stocks depend on accurate initialization of ecosystem states in process-based model simulations. Unlike traditional potential vegetation simulations which assume that ecosystems equilibrate with long-term climate, observation-initialized simulations integrate the impacts of previous history of disturbance events and human activities on ecosystem structure and composition. However, observation-constrained initialization is challenging at regional scales due to limited availability of spatially-comprehensive measurement data. In this study, we assimilate remote-sensing estimates of canopy structure from Global Ecosystem Dynamics Investigation (GEDI) and canopy composition from AVIRIS imaging spectrometry into Ecosystem Demography version 2 (ED2), a cohort-based Terrestrial Biosphere Model. We drive model simulations with future climate scenarios and rising atmospheric CO2 concentrations to predict ecosystem responses to environmental changes over an elevational transect region in California’s Sierra Nevada by the end of the century. Our simulations suggest that predictions are significantly impacted by ecosystem initial condition at the multi-decadal (50+ year) scale. The impacts are stronger in dense-canopy forests at mid-to-high elevations than woody savannahs at low elevations. Under a hotter and drier future climate with CO2 enrichment, ecosystems across the elevational transect are predicted to act as a net carbon sink but with marked changes in composition. Aboveground biomass (AGB) is predicted to increase at low elevations due to increasing abundance in both deciduous and coniferous trees. However, at mid-to-high elevations, AGB increases are caused by increasing abundance of coniferous trees but large declines in the abundance of deciduous trees. Our research demonstrates how large-scale remote-sensing data can be assimilated into process-based model simulations to improve future predictions of ecosystem dynamics.