FFT analysis on NDVI annual cycle and climatic regionality in Northeast Brazil.

By considering that the climate of Northeast Brazil (NEB) has distinct wet and dry seasons, the mixed radix fast Fourier transform (mrFFT) algorithm, developed at the National Aerospace Centre of the Netherlands, was applied to a monthly Normalized Difference Vegetation Index (NDVI) time series from July 1981 to June 1993, to generate phase, amplitude and mean NDVI data using a 1-year frequency in order to improve the analysis of its spatial variation. The NDVI mean values varied from >0.7, which occurred in northwest and southeast regions, to <0.3 in the northeast and 0.4 in the southwest regions of the NEB. The 90° phase month at its maximum amplitude occurred in August and was observed in both southeastern and northwestern coasts, located at 10.5°S–37.5°W and 4°S–46°W, respectively. It changed rapidly from August, June to May, moved inland and changed gradually from May through to April and from March to February, then moved towards the centre Dry Polygon area. Then it changed gradually from February to January and ended up in December, and moved further southwards. The annual cycle amplitude varied from <0.075 in northwest and southeast regions to >0.25 in the northeast region. By using spatial variations of phase, amplitude and mean NDVI values, 15 climate types were delineated for the NEB. The spatial distribution of climate types in the NEB delineated by the NDVI FFT analysis agreed mostly with the climatic types presented by Hargreaves (Precipitation dependability and potentials for agricultural production in Northeast Brazil. Empresa Brasileira de Pesquisa Agropecuária-EMBRAPA, Brazil, 1974), except regions with higher spatial variability and limited surface meteorological data. Among the three components: phase, amplitude and mean NDVI, the phase image, informing the initiation and duration of rainy season, was the most important component for climate-type delineation. Nevertheless, while the extreme values of amplitude, inferring a high wet–dry rainfall regime contrast, and mean NDVI, inferring annual rainfall total, occurred in the same phase area, amplitude and mean NDVI images were used for further delineation. It is concluded that the application of the mrFFT algorithm to NDVI time series analysis is an effective method for identifying vegetation phenology and hence to delineate the transit spatial change of climate types. Our further study is to construct and validate the alternative El Niño–Southern Oscillation (ENSO) drought onset model for each of the 15 climate types delineated for the NEB. Copyright © 2001 Royal Meteorological Society [ABSTRACT FROM AUTHOR]