Abstract. Forecasting atmospheric CO2 concentrations on synoptic timescales (∼ days) can benefit the planning of field campaigns by
better predicting the location of important gradients. One aspect of this,
accurately predicting the day-to-day variation in biospheric fluxes, poses a
major challenge. This study aims to investigate the feasibility of using a
diagnostic light-use-efficiency model, the Vegetation Photosynthesis
Respiration Model (VPRM), to forecast biospheric CO2 fluxes on the timescale of a few days. As input, the VPRM model requires downward shortwave
radiation, 2 m temperature, and enhanced vegetation index (EVI) and land
surface water index (LSWI), both of which are calculated from MODIS
reflectance measurements. Flux forecasts were performed by extrapolating the
model input into the future, i.e., using downward shortwave radiation and
temperature from a numerical weather prediction (NWP) model, as well as
extrapolating the MODIS indices to calculate future biospheric CO2
fluxes with VPRM. A hindcast for biospheric CO2 fluxes in Europe in
2014 has been done and compared to eddy covariance flux measurements to
assess the uncertainty from different aspects of the forecasting system. In
total the range-normalized mean absolute error (normalized) of the 5 d
flux forecast at daily timescales is 7.1 %, while the error for the model
itself is 15.9 %. The largest forecast error source comes from the
meteorological data, in which error from shortwave radiation contributes
slightly more than the error from air temperature. The error contribution
from all error sources is similar at each flux observation site and is not
significantly dependent on vegetation type.