Please use this identifier to cite or link to this item:
|Title:||Smoothing Time Series of Satellite Derived Vegetation Indices for Global Monitoring of Agricultural Productivity and Food Security|
|Authors:||DELINCE JACQUES; KLISCH ANJA|
|Other Contributors:||ATZBERGER CLEMENT|
|Citation:||Proceedings of the International Workshop on Flexible Modelling: Smoothing and Robustness, FMSR, 12-14 November 2008, Leuven, Belgium p. 4|
|Type:||Articles in periodicals and books|
|Abstract:||A global observation capacity is required for agricultural production forecasts and food security alert systems. The European Commission¿s Joint Research Center (JRC) uses low resolution satellite imagery to map vegetation status. The derived maps are used for near real-time production forecasts as well as for the anticipation of food security problems. The daily imagery used by JRC covers the entire globe at 1 km spatial resolution. An uninterrupted time series is available since 1998. To highlight the response of the vegetation, red and near infrared spectral responses are combined into a widely used vegetation index; the normalized difference vegetation index (NDVI). Growth anomalies are detected at the pixel scale by comparing the actual NDVI with the long term average NDVI. Sensing the Earth surface is not trivial as the electromagnetic radiation, which carries the information about the surface status, is scattered and absorbed by the Earth atmosphere. In addition, clouds may (partly) obstruct the field of view of the sensor. Altogether these perturbations lead to NDVI signals which are far lower of what would have been observed under perfect measurement conditions. To eliminate the strongest perturbations, the daily imagery is generally analyzed as 10-days maximum value composite (MVC) imagery (Holben et al., 1986). In this simple pre-processing step, for a given pixel location, only the highest NDVI value is retained for each 10-day (dekadal) period, thus minimizing the mentioned perturbations. Nevertheless, even dekadal NDVI-MVC images still contain perturbations. Sharp edge lines may appear in regions where insufficient registrations were available for the compositing process. Missing values occur for example at higher latitudes during polar night. Clouds and/or atmospheric conditions with high aerosol load may persist longer than 10 days, leading to sub-optimal MVC outputs which are easily recognized as irregular dips. The oral presentation aims at presenting and comparing three different smoothing strategies: ¿ Best index slope extraction (BISE) algorithm (Viovy et al., 1992) ¿ Weighted least square regression (Swets et al., 1999) ¿ Savitzky-Golay polynomial filtering (Savitzky & Golay, 1964; Chen et al., 2004) The algorithms are currently used at JRC for minimizing the undesired atmospheric/cloud effects, with the ultimate goal to enhance the signal stemming from the land surface. All approaches work within gliding windows of variable size and have been adapted to deal with missing values. Advantages and disadvantages of the different methods will be presented in the context of agricultural production estimates and for deriving phenological indicators useful in global change studies.|
|JRC Institute:||Institute for the Protection and Security of the Citizen|
Files in This Item:
There are no files associated with this item.
Items in repository are protected by copyright, with all rights reserved, unless otherwise indicated.