Fusion of Time Series of Geostationary Satellite and VIIRS Observations for Detecting Land Surface Phenology

Abstract:

Land Surface Phenology (LSP) derived from satellite data quantifies the seasonal dynamics of vegetation activity including phenological transition timing and greenness magnitude in vegetation communities. Accurate LSP detection requires high temporal frequency of cloud-free observations from satellites. Although polar-orbiting satellites (such as VIIRS) provide daily observations across the globe, the cloud-contaminated observations consecutively longer than two weeks are frequent in much of the world, which results in large uncertainties in LSP detection. The cloud impact can be mitigated using observations from the Advanced Baseline Imager (ABI) onboard Operational Environmental Satellite (GOES) systems (GOES-16 launched in November 2016 and GOES-17 launched in March 2018). The main shortcoming of ABI observations is the coarse spatial resolution. The pixel size of ABI’s red band is 500 m at nadir whereas near infrared and shortwave infrared bands vary from 1 km at nadir to larger than 3 km at large View Zenith Angles. Therefore, this study is to generate synthetic high spatiotemporal resolution time series by fusing ABI and VIIRS data for LSP detection. Specifically, we investigate diurnal variation of surface spectral reflectances from ABI in the central United Sates during 2018 and calculate the diurnal angularly-dependent EVI2 (2 band enhanced vegetation index) using the kernel-driven model of bidirectional reflectance distribution function in order to generate daily EVI2 observations. The highly temporal cloud-free ABI EVI2 time series is considered as the temporal shape of vegetation phenological development. The temporal ABI EVI2 shape is then used to match the shape of 500m VIIRS time series. This spatiotemporal shape matching approach uses all cloud-free observations in fine and coarse resolution time series to build up a pixel-dependent model that is applied to generate cloud-free 500m ABI-VIIRS EVI2 time series. This time series is analyzed in the improvement of LSP detections.