Super typhoon Haiyan makes landfall
Super typhoon Haiyan
Subscribe here to receive special images


Lake Maggiore - Italy, 3-6 April 2000

Intermediate Scale Approach for Estimating Vegetation Canopy Leaf Area Index using SPOT4/VGT Spectral Bands.

F. Cipriani, E. Cubero-Castan
MEDIAS Toulouse France

Paper (pdf file, 250 k)

The Leaf Area Index, defined as the one-sided green leaf area per unit ground area, has until now been either globally estimated, with rather coarse ground resolution data (typically about five kilometers, with NOAA14/AVHRR), or locally estimated, upon forest sized areas, and unmixed kind of vegetation species, through high resolution data (SPOT/HRV or SPOT/HRVIR or LANDSAT/TM). This type of land surface parameter need though being also estimated on an intermediate scale, typically one kilometer, in order to increase, with dedicated sensors as SPOT4/VGT, the precision of previous large scale estimations of vegetation behaviour, as climate-atmosphere-vegetation interactions models are now reaching the meso-scale. Moreover, the ability of daily coverage of the vegetation cover remains a powerful means of dynamical modelling, which is worth exploiting, in the case of derived parameters as the LAI, when vegetation phenology is involved. Eventually, this intermediate scale is worth considering, in order to reach the combination of local experiments consistency with the large remote sensed both temporal and spatial coverage abilities.

From this point of view, a one kilometer scale processing approach is being developped, using SPOT4/VGT decade synthesis (one kilometer ground resolution), covering one year of data (april 1998 - april 1999), over France. The decade synthesis provide both a sufficient amount of data over the year, for a temporal evolution study, as well as an already corrected data set, from the geometric, radiometric, and atmospherical effects. These data sets are associated with the Corine Land Cover data set raster map (250 meters resolution), which provides a coarse description of the land cover, both static, and inaccurate in terms of vegetation species description, but sufficient for a first approach.

The VGT data sets and the CLC map are first commonly geo-referenced, using USGS map projection routines. Thus each VGT pixel can be associated with a composition of land occupation components. The Red and Near-infra-red signals of the sensor are then unmixed (Faivre, 1996), in order to calculate the contribution of each land cover component, within the VGT pixel. The Normalized Differenced Vegetation Index can therefore be computed, for each component, and the LAI estimated from the NDVI, trough a global scale parametric model (Sellers, 1994). The parameters used are the NDVI temporal extrema values for each land cover class, computed from the VGT data sets, and the maximum value of the LAI, over the year. The NDVI extrema are taking into account a North-South gradient over France, being computed within one degree latitude bands. The maximum LAI values depends on the forest composition gradient, over the studied area (Solmon, LA), the other class values remaining constant. The LAI values computed are eventually integrated, in order to obtain a one kilometer LAI data set, over France. LAI maps are produced for each decade of the year where a VGT synthesis exists, allowing thus temporal profiles to be constructed.

The process actually follows the validation procedure. Straight on ameliorations could be yielded by data pre-analyzing, the improvement of the LAI model used, and a precision growth by taking into account the LAI-NDVI saturation (and limits), and error due to the integrated surfaces. Moreover, a classification procedure has to be engaged to replace the CLC map. This procedure would involve high-resolution SPOT4/HRVIR data crossing with SPOT4/VGT data.

Key Words : Remote sensing, Land surface parameters, Meso-scale, SPOT4/VGT, NDVI, LAI, Spatial heterogeneity, Signal unmixing, Land cover classfication.