The use of satellite NDVI data for the validation of global vegetation phenology models: application to the Frankfurt Biosphere Model

An algorithm based on a three-spline function fitted to measured NDVI courses (normalized difference vegetation index) was developed to analyze a given NDVI annual course with respect to leaf shooting and leaf abscission times of deciduous vegetation. In contrast to algorithms which are based on modified second derivatives of the NDVI time course to detect shooting or abscission, the proposed algorithm takes into account the whole annual time course and is therefore less sensitive to noise in the NDVI-signal. In the present study this algorithm was used to validate the phenology results for the deciduous vegetation of a global equilibrium run of the prognostic Frankfurt Biosphere Model (FBM, spatial resolution 0.5° × 0.5°) driven by a climatology which represents a mean seasonality of the driving variables. The mean value of the area-weighted frequency distribution of the difference between the shooting date deduced from NDVI and the shooting date calculated by the FBM for the deciduous vegetation types is −4 days, indicating that in the global mean the FBM predicts leaf shooting less than one week too late. A 75% fraction of the area under consideration shows predicted shooting dates lying within a range of ± 30 days compared to the satellite-derived dates. The distribution has its maximum at a difference of 0 days (i.e. the FBM exactly fits the NDVI deduced shooting day for these areas). This result supports the general assumption that at least in global scale models phenology can be successfully deduced from carbon flux balance considerations.