A Cross-Site Evaluation of Alternative FPAR Products for Use in Diagnostic Carbon Flux Models

Input requirements for diagnostic carbon flux models include estimates of FPAR (the fraction of incident photosynthetically active radiation that is absorbed by the canopy). These estimates are available from a variety of satellite-borne sensors, but in their raw form may contain significant artifacts associated primarily with clouds. In this study we evaluated the sensitivity of simulated gross primary production (GPP), ecosystem respiration (Re), and net ecosystem production (NEP) to alternative formulations of FPAR from the MODIS sensor. These alternatives included the original product (FPARorig), a version based on filling and interpolation (FPARntsg), and a version with a smoothing based on the TIMESAT algorithm (FPARts). Estimates of daily GPP, Re, and NEP from multiple eddy covariance flux towers over multiple years were assembled as reference data. In almost all cases, model output based on FPARntsg and FPARts reduced bias on an annual basis and RMSE for daily values relative to model runs using FPARorig. At a wet conifer site, FPARts maintained the highest and most consistent FPAR, a pattern consistent with the dense coniferous forest canopy there. Limitations of FPARts were apparent at a grassland site with an abrupt fall off in greenness associated with onset of the dry season. Model effectiveness in capturing the interannual variation in NEP was also enhanced in many cases with the adjusted FPAR products.