Evaluation of time-series and phenological indicators for land cover classification based on MODIS data
Remote Sensing for Agriculture, Ecosystems, and Hydrology XIII
In the context of defining a procedure for near real time land use/land cover (LULC) mapping with seasonal updated products, this research examines the use of time-series and phenological indicators from MODIS NDVI. 16-day NDVI composites from MODIS (MOD13Q1) covering the period from 2001 to the present were acquired for three test sites located in different parts of Europe. The newly proposed Whittaker smoother was used for filtering purposes. Metrics of vegetation dynamics (such as minimum,
... (such as minimum, maximum and amplitude, etc.) were extracted from the filtered time-series. Subsequently, the capability of three data sets (raw, filtered data and phenological indicators) was evaluated to separate between different LULC classes by calculating the overall classification accuracy for the years 2002 and 2009. Ground truth data for model calibration and testing set was derived combining existing land cover products (GLC2000 and GlobCover 2009). Based on these results, the benefits of using phenological indicators and cleaned data for land cover classification are discussed. cleaning procedure of raw NDVI data, (ii) an innovative data smoothing algorithm using the Whittaker filter, (iii) and the extraction of phenological indicators. Subsequently, we used the filtered data to investigate the classification performance with various subsets of multi-temporal observations. For the classification a Least Square Support Vector Machine (LS-SVM) algorithm, developed by Suykens et al.  , was implemented. LS-SVM represents a variant of the original SVM formulation  to reduce complexity and computation times. Finally, benefits of using phenological indicators and cleaned data for land cover classification are discussed.