Mapping Synoptic Phenological Patterns and Trajectories from NDVI Time Series.
Zurita-Milla, Raul1; van Gijsel, J.A.E2
1Faculty ITC, University of Twente, NETHERLANDS; 2Climate Observations division, KNMI, NETHERLANDS

Land surface phenology (LSP) deals with the study of seasonal vegetation patterns as seen by Earth Observation satellites. Characterizing LSP dynamics is a prerequisite to understand the effects of global change on vegetated canopies. Till now, LSP studies have fundamentally relied on the derivation of phenological metrics from vegetation indices time series (mostly NDVI). Such an analysis is done on a per-pixel basis and does not consider spatial relationships. This study presents a novel approach to map synoptic phenological patterns and phenological trajectories for each growing season. This is done at the landscape scale, rather than per pixel. The proposed approach uses self-organizing maps (SOMs) and the Sammon's projection. The Kruger National Park (South Africa) and 13 years of 10-day SPOT VEGETATION NDVI composites are used to illustrate the study. First, we trained a SOM by transforming the NDVI time series into a matrix where the rows contain the 10-day composites and the columns the pixels of the study area. This resulted in a topologically ordered set of phenological synoptic states. Then, we used the Sammon's projection to create a simplified two-dimensional representation of the synoptic phenological states. Finally, we mapped phenological trajectories for each vegetation season to show how phenological development changes among years. This novel approach provides a holistic characterization of phenology at the landscape scale and effectively summarizes the information present in the long time series, thus facilitating further interpretation and analysis.