Combining remote sensing data with process modelling to monitor boreal conifer forest carbon balances

Abstract

Approaches combining satellite-based remote sensing data with ecosystem modelling offer potential for the accurate assessment of changes in forest carbon balances, for example, in support of emission credits under the Kyoto Protocol. We investigate the feasibility of two alternative methods of using satellite-derived data to constrain the behaviour of a dynamic ecosystem model, in order to improve the model's predictions of the net primary production (NPP) of conifer forests in northern Europe (4–30°E, 55–70°N). The ecosystem model incorporates a detailed description of forest stand structure and biogeochemical processes. The satellite product comprises multi-spectral reflectance data from the VEGETATION sensor. The first method combines satellite-based estimates of FPAR, the fraction of incoming photosynthetically active radiation absorbed by vegetation, with the model's predictions of the efficiency with which trees use the incoming radiation to fix carbon. Results obtained using this method averaged 0.22 kg C m −2 yr −1 for the NPP of conifer and mixed forests across the study area, and compared well with forest-inventory-based estimates for Sweden. The second method uses forest stand descriptions derived by application of an inverse radiation transfer scheme to VEGETATION data to prescribe stand structure in the ecosystem model simulations. Predictions obtained by this method averaged 0.31 kg C m −2 yr −1, somewhat high compared to forest inventory data for central and northern Sweden. Simulations by the ecosystem model when driven only by climate, CO 2 and soils data, but unconstrained by satellite information, yielded an average NPP of 0.41 kg C m −2 yr −1, which is likely to be an overestimate. Summed over the study area, the NPP estimates amounted to 0.16–0.23 Gt C yr −1, around 6–9% of the NPP of all boreal forest globally or 0.3–0.4% of terrestrial NPP globally. The investigated methods of combining process modelling and products derived from remote sensing data offer promise as a step towards the development of operational tools for monitoring forest carbon balances at large scales.