An algorithm to classify and monitor seasonal variations in vegetation phenologies and their inter-annual change

Abstract

The algorithm presented in this paper classifies vegetation from annual Normalized Difference Vegetation Index (NVDI) time series according to the shape of the temporal cycle. Shape is described using the Fourier components’ magnitude and phase. The degree of an NDVI cycle’s similarity to a predefined reference cycle is measured by the similarity in their amplitude ratios and in their phase differences. Tolerable deviations from the ideal ratios and differences can be set by the user depending on individual accuracy requirements. Tolerable vegetation coverage variation within a shape class is another user defined variable. The algorithm is invariant to cycle modifications including temporal shifts, vertical displacements, and intensity variations, modifications that may be caused by differences in climate, soil (-type, -water, -fertility), or topography, but are unrelated to the vegetation type. The output is a highly consistent clustering of NDVI cycles according to their shapes, which can be linked to distinct vegetation types or land use practices. Intra-class coverage variations in the form of continuous fields measured relative to the reference cycle provide additional information about vegetation covers. Based on the same principles, inter-annual vegetation changes can be monitored with the possibility to distinguish between coverage fluctuations and phenological variations/changes. The algorithms are independent from scene statistics and can be used to create spatially and temporally comparable classifications. Their potential is demonstrated using a 250 m MODIS NDVI time series (version 4) from the Middle East.