Comparison of land surface phenology in the Northern Hemisphere based on AVHRR GIMMS3g and MODIS datasets

This paper analyzes inconsistencies in remotely sensed land surface phenology (LSP) reported by AVHRR GIMMS3g and MODIS datasets across land cover types of the Northern Hemisphere. We extracted the start of the growing season (SOS) and the end of the growing season (EOS) from the AVHRR GIMMS3g (1982–2015) and MODIS (2000–2015) datasets, and weekly GPP mean data from the Fluxnet2015 dataset to compare spatial patterns and trends of the phenological indicators in the Northern Hemisphere. We used the same method to fit the time series curves and extract phenological parameters from the two datasets to avoid uncertainties caused by differences in fitting and extraction approach. The results showed that (1) The multi-year means extracted from the GIMMS3g (1982–2015 and 2000–2015) and MODIS (2000–2015) datasets in the Northern Hemisphere display greater differences in the spatial distribution of SOS than EOS; (2) Under the 95% confidence level, GIMMS3g showed a significantly delayed (0.1716 days/year) SOS and advanced (0.9172 days/year) EOS in most parts of the Northern Hemisphere between 2000 and 2015. The SOS and EOS extracted from the MODIS dataset exhibited significantly advanced (0.5861 days/year) and delayed (0.6305 days/year) trends, respectively; (3) From 2000 to 2015, the same significant trends were observed in the SOS and EOS from two datasets, accounting for 1.33% and 1.17% of the total pixels in the Northern Hemisphere, respectively. (4) From 2000 to 2015, a significantly advanced trends in SOS extracted from MODIS (M_SOS) was more frequent than for GIMMS3g (G_SOS) in different land cover types at high latitude. EOS extracted from GIMMS3g (G_EOS) had a significantly advanced trend, and EOS extracted from MODIS (M_EOS) had a significantly delayed trend in different land cover types. The phenological parameters obtained from GIMMS3g are closer to ground phenology than those from MODIS. The results suggest that the phenological parameters derived from different datasets have different effects on the LSP trend