Validating Long-term Consistency of MODIS Vegetation Index Time series Using Ground-based Radiation Sensor Data

Spectral vegetation index (VI) time series obtained from satellite sensors such as Moderate Resolution Imaging Spectrometer (MODIS) are essential for vegetation phenology analysis, landcover change detection, and terrestrial biogeochemical cycle modeling on a global scale. The MODIS data record has exceeded ten years and is expected to contribute further to change detection analysis and climate studies. It is therefore of great importance to evaluate how well MODIS VI time series capture inter-annual variability in vegetation seasonal dynamics across different biomes. A variety of methods have been developed to examine the sensitivity of MODIS VI products using in situ radiation sensor data from ground observation networks such as FLUXNET. Validation using ground based VIs has been attempted, but might not be appropriate because of the potential differences including view angle, band pass differences, and spatial representativeness. The use of phenological metrics calculated from VIs would be a good estimate to compare inter-annual variability between MODIS and ground VIs.

A new approach to validate MODIS time-series VIs were examined by comparing over five years of vegetation start of season (SOS) estimated from up to ten years (2003-2012) of MODIS and ground based VIs across five different biomes. Daily ground based NDVI and a two band EVI (EVI2) were calculated from incoming and outgoing photosynthetically active radiation and global solar radiations at ten AmeriFlux sites located in North America. Obtained daily ground based VIs and eight-day composited MODIS VIs (NDVI, EVI, and EVI2) were fitted to asymmetric double logistic function in each year, and SOSs were calculated using two methods; inflection point of first derivative (SOS_inf) and half amplitude (SOS_ratio) of fitted curves.

Although site specific differences were seen in the direction and range, SOS from MODIS and ground VIs showed an analogous pattern over time across five biomes. MODIS SOS_ratio plotted against ground SOS_ratio closely aligned with 1:1 line (slope > 0.80, R²>0.85), whereas SOS_inf showed apparently narrower angle caused by the large variability in arid grassland sites (slope ≈ 0.5, R²<0.60). Comparing SOSs from both datasets after the Gram-Schmitt orthogonalization, the standard deviation (s.d.) of SOS differences were smaller than s.d. of interannual variability except for evergreen needle leaf forest and arid grassland sites. In particular, MODIS tracked ground based SOSs with a week differences on average at deciduous broad leaf forest sites (3-7days) and clopland sites (6-9days) regardless of the method. SOS_ratio at wooded savanna site and temperate grassland site also showed good agreement (5-7days). These results indicate that MODIS VIs are capable to monitor inter-annual variability of surface vegetation at four biome types with half amplitude method. The narrow slopes suggested that the early in the year, the early MODIS detects SOS, and vice versa. The further effort on time series screening and the increase of number of data point could improve the issue seen in evergreen forest and arid grassland sites.