Large-scale heterogeneity of Amazonian phenology revealed from 26-year long AVHRR/NDVI time-series

Abstract

Depiction of phenological cycles in tropical forests is critical for an understanding of seasonal patterns in carbon and water fluxes as well as the responses of vegetation to climate variations. However, the detection of clear spatially explicit phenological patterns across Amazonia has proven difficult using data from the Moderate Resolution Imaging Spectroradiometer (MODIS). In this work, we propose an alternative approach based on a 26-year time-series of the normalized difference vegetation index (NDVI) from the Advanced Very High Resolution Radiometer (AVHRR) to identify regions with homogeneous phenological cycles in Amazonia. Specifically, we aim to use a pattern recognition technique, based on temporal signal processing concepts, to map Amazonian phenoregions and to compare the identified patterns with field-derived information. Our automated method recognized 26 phenoregions with unique intra-annual seasonality. This result highlights the fact that known vegetation types in Amazonia are not only structurally different but also phenologically distinct. Flushing of new leaves observed in the field is, in most cases, associated to a continuous increase in NDVI. The peak in leaf production is normally observed from the beginning to the middle of the wet season in 66% of the field sites analyzed. The phenoregion map presented in this work gives a new perspective on the dynamics of Amazonian canopies. It is clear that the phenology across Amazonia is more variable than previously detected using remote sensing data. An understanding of the implications of this spatial heterogeneity on the seasonality of Amazonian forest processes is a crucial step towards accurately quantifying the role of tropical forests within global biogeochemical cycles.