The leaf phenology product presented here shows the amplitude of annual cycles observed in MODIS (Moderate Resolution Imaging Spectroradiometer) normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) 16-day time-series of 2000 to 2013 for Meso- and South America. The values given represent a conservative measure of the amplitude after the annual cycle was identified and tested for significance by means of the Fourier Transform. The amplitude was derived for four sets of vegtation indices (VI) time-series based on the MODIS VI products (500m MOD13A1; 1000m MOD13A2). The amplitude value can be interpreted as the degree in which the life cycles of individual leaves of plants observed within a pixel are synchronised. In other words, given the local variation in environment and climate and the diversity of species leaf life cycle strategies, an image pixel will represent vegetation communities behaving between two extremes:
* well synchronized, where the leaf bud burst and senescence of the individual plants within the pixel occurs near simultaneously, yielding a high amplitude value. Often this matches with an area of low species diversity (e.g. arable land) or with areas where the growth of all plants is controlled by the same driver (e.g. precipitation).
* poorly synchronized, where the leaf bud burst and senescence of individual plants within a pixel occurs at different times of the year, yielding a low amplitude value. Often this matches with an area of high species diversity and/or where several drivers could be controlling growth.
Publication date: 2017-07-07
The key underlying method is the Fourier Transform (FT) a well established and efficient methodology to analyse time series data. The FT can be used for the detection of regular cycles, even when hidden in a noisy background and when the time series in question contains data gaps. The FT estimates a power spectrum that indicates the average variance in a time series at different time scales or frequencies. Standard linear de-trending of the data before application of the FT allows an analysis of a 'stationary' time series as the mean and variance should ideally be more-or-less constant.
We used two approaches to evaluate the amplitude values. The first involved comparing the full extent of the amplitude data grids with the independent International Geosphere-Biosphere Programme (IGBP) land cover map. This was done with histograms showing the distribution of the amplitude values found within each of the IGBP cover classes. The second approach compared published in situ site observations that describe canopy leaf life cycles through leaf flushing and falling, LAI and litter fall or other observations, with the corresponding amplitude values of these sites.