abstract
- Phenology is an important indicator of annual plant growth and is also widely incorporated in ecosystem models to simulate interannual variability of ecosystem productivity under climate change. A comprehensive understanding of the potentials of current algorithms to detect the start and end for growing season (SOS and EOS) from remote sensing is still lacking. This is particularly true when considering the diverse interactions between phenology and climate change among plant functional types as well as potential influences from different sensors. Using data from 60 flux tower sites (376 site-years in total) from the global FLUXNET database, we applied four algorithms to extract plant phenology from time series of normalized difference vegetation index (NDVI) from both MODIS and SPOT-VGT sensors. Results showed that NDVI-simulated phenology had overall low correlation (R2<0.30) with flux-derived SOS/EOS observations, but this predictive strength substantially varied by fitting algorithm, sensor and plant functional type. Different fitting algorithms can produce significantly different phenological estimates, but this difference can also be influenced by sensor type. SPOT-VGT simulated better EOS but no difference in the accuracy of SOS was found with different sensors. It may be due to increased frequency of data sampling (10 days for SPOT-VGT vs. 16 days for MODIS) during spring season when rapid plant growth does not help SPOT-VGT more sensitive to growth. In contrast, more frequent data acquisition favors better modeling of plant growth in autumn when a gradual decrease in photosynthesis occurs. Our study results highlight that none of these algorithm can provide consistent good accuracy in modeling SOS and EOS with respect to both plant functional types and sensors. More importantly, a rigorous validation of phenology modeling against ground data is necessary before applying these algorithms at regional or global scales and consequently previous conclusions on regional SOS/EOS trends should be viewed with caution if independent validation is lacking.