A review of vegetation phenological metrics extraction using time-series, multispectral satellite data

Vegetation dynamics and phenology play an important role in inter-annual vegetation changes in terrestrial ecosystems and are key indicators of climate-vegetation interactions, land use/land cover changes and variation in year-to-year vegetation productivity. Satellite remote sensing data have been widely used for vegetation phenology monitoring over large geographic domains using various types of observations and methods over the past several decades. The goal of this paper is to present a detailed review of existing methods for phenology detection and emerging new techniques based on the analysis of time-series, multispectral remote sensing imagery. This paper summarizes the objective and applications of detecting general vegetation phenology stages (e.g., green onset, time or peak greenness and growing season length) often termed ‘land surface phenology’, as well as more advanced methods that estimate species-specific phenological stages (e.g., silking stage of maize). Common data processing methods, such as data smoothing, applied to prepare the time-series remote sensing observations to be applied to phenological detection methods are presented. Specific land surface phenology detection methods as well as species-specific phenology detection methods based on multispectral satellite data are then discussed. The impact of different error sources in the data on remote-sensing based phenology detection are also discussed in detail, as well as ways to reduce these uncertainties and errors. Joint analysis of multi-scale observations ranging from satellite to more recent ground-based sensors is helpful for us to understand satellite-based phenology detection mechanism and extent phenology detection to regional scale in the future. Finally, emerging opportunities to further advance remote sensing of phenology is presented that includes observations from Cubesats, near-surface observations such as PhenoCams and image data fusion techniques to improve the spatial resolution of time-series image data sets needed for phenological characterization.