Extraction of Phenological Parameters from Temporally Smoothed Vegetation Indices

Within the MARS Crop Yield Forecasting System (MCYFS; Royer and Genovese, 2004) of the European Commission vegetation indicators like NDVI, SAVI and fAPAR are operationally derived for daily, decadal and monthly time steps. Besides low resolution sensors as SPOT-VGT and NOAA-AVHRR, medium resolution data from TERRA/AQUA-MODIS or ENVISAT-MERIS are used at pan-European level. In case of available time series, esp. NOAA-AVHRR (since 1981) and SPOT-VGT (since 1998) difference values of the indicators (e.g. relative or absolute differences) and frequency analysis of the indicators (e.g. position in historical range or distribution) are calculated. The exploitation of the data is performed at full resolution, at grid level of the MCYFS or regional unmixed means (C-indicators) are used. Therefore a database has been set-up in order to provide the indicators based on a weighted average for each CORINE land cover class within an area of interest. The current work aims to develop a strategy for an optimal use of the different sensors and thus derived indicators at different aggregation levels for the ingestion into the MCYFS. As a first step smoothing algorithms have to be applied to the time series to diminish noise effects. Therefore, existing methods as simple sliding windows, piecewise linear regression or fitting of polynomial functions are employed and compared. Thereafter, the time series analysis is performed with the aim to establish relationships between indicators profile features and the crop phenology.