Ladamer - Results

Vegetation Fractions

An essential component of the land degradation status assessment product is the availability of a homogeneous information layer on vegetation density, which overcomes some limitations of the usually used vegetation indices, like i.e. the NDVI. Thus, the NDVI is not only influenced by the vegetation cover but also by the background (soil, rock), which can show NDVI values up to 0.3 for non vegetated areas. Therefore, the detection of sparse vegetation cover is limited which become especially apparent in low vegetated areas like the semi-arid to arid areas of the Mediterranean. Due to these problems it is preferable to find a measure for vegetation abundances which proves to be a better indicator for vegetation cover.

In LADAMER a spectral unmixing strategy was implemented based on the inverse relationship between the vegetation index NDVI and the land surface temperature and which was developed in the frame of the EU-funded project MEDALUS. This method was performed on the 1 km NOAA AVHRR MEDOKADS dataset and the implementation of the spectral unmixing model was supported by WP5.
The study demonstrated that the vegetation abundances show a higher dynamic than the NDVI (Figure 1). The NDVI utilizes only a range from 0.05 to 0.7, where the limitation of the lower values is caused by the background signal and the upper boundary results from the saturation of the NDVI with dense vegetation cover. In addition, the NDVI values for non vegetated areas like the Sahara are significantly above 0, whereas the vegetation abundances supply better estimates for non vegetated areas due to the reduction of the background's influence.

This enhanced information layer will serve as an input layer for the modelling of the land condition index performed by WP3 as well as for the products from WP2 itself like time series analysis and land cover classification (D2.2).

Figure 1: NDVI (upper left) and derived vegetation abundances (upper right) and differences of NDVI and vegetation abundances after z-transformation (below) for August 1989 for the Iberian Peninsula; based on the MEDOKADS data archive (provided by D. Koslowsky, TU Berlin).

Trend classes:

Figures 2a and 2b show the results of the trend analysis for the Pathfinder and the MEDOKADS time series. It was performed using linear regression on the annual NDVI mean values covering a time period from 1989 to 1999. The significance of trends was examined implementing a t-Test. As the Pathfinder data set is already trend adjusted, the MEDOKADS dataset still shows the "greening of the deserts"-effect. Nevertheless both datasets reveal the same areas of negative and positive trends on the Iberian Peninsula, i.e. Valencia.

Figure 2a: Direction and significance (T-Test) for the linear regression performed on the annual means of the NOAA AVHRR Pathfinder time series for the Mediterranean area covering the time period 1989-1999.
Figure 2b: Direction and significance (T-Test) for the linear regression performed on the annual means of the NOAA AVHRR MEDOKADS time series for the Iberian Peninsula covering the time period 1989-1999.

Land use classification:

During the second year of LADAMER a sophisticated approach was implemented to enable an automatic land cover classification. This approach comprises different techniques like i.e. neural networks and classification trees as well as the inclusion of additional information layers like for example climatic data for a climatic stratification for the area under investigation. Thus, it has become possible to generate yearly land cover maps from the MEDOKADS NOAA AVHRR archive. Figure 3 shows the classification results for the years 1989 and 2002. The classes are referenced by CORINE 1990, so that the classes from the automated classification are directly comparable to the CORINE classes. Due to the coarser geometric resolution of the NOAA data the number of classes is reduced in comparison to the CORINE dataset, which bases on visual interpretation of Landsat TM images with a geometric resolution of 30m.

Figure 3: Results from the automated land use classification for 1989 (above) and 2002 (below) based on the 1 km MEDOKADS NOAA AVHRR data archive (provided by D. Koslowsky, FU Berlin) for a reduced number of CORINE land use classes.