A time series processing tool to extract climate-driven interannual vegetation dynamics using Ensemble Empirical Mode Decomposition (EEMD)

Interannual changes of vegetation are crucial in understanding ecosystem dynamics under global change. However, there is no automated tool to extract these interannual changes from remote sensing time series. To fill this gap, the Ensemble Empirical Mode Decomposition (EEMD) framework was refined and implemented to decompose time series of Normalized Difference Vegetation Index (NDVI) and reconstruct their interannual components. The performance of EEMD-based interannual NDVI detection was assessed using simulated time series, and its sensitivity to model and data parameters was determined to provide a basis for remote sensing applications. The sensitivity analysis highlighted application limitations for time series with low interannual to annual amplitude ratios and high irregularity in timing of growing seasons, as these factors have the strongest effects on the overall performance. However, within these limitations, the detected interannual components correspond well to simulated input components with respect to timing of episodes and composition of time scales. The applicability on real world NDVI time series was demonstrated by mapping the coupling between precipitation variability, interannual vegetation changes, and the El Niño Southern Oscillation and Indian Ocean Dipole phenomena for ecoregions in East and Central Africa. In most areas where precipitation was found sensitive to oceanic forcing, the EEMD detected vegetation changes matched the predicted response, except in dense forest ecosystems.