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


Lake Maggiore - Italy, 3-6 April 2000

SPATEM: The analysis of annual sequences of VEGETATION data at the landscape scale.

Agustin Lobo and Nicolau Pineda
Instituto de Ciencias de la Tierra "Jaume Almera" (CSIC)
Lluis Solé Sabarís s/n, 08028 Barcelona, Spain

Paper (pdf file, 1.11 M)

The instrument VEGETATION acquires images with global and daily coverage of the Earth surface at 1 km2 resolution from the same SPOT platform that hosts HRVIR, the sensor acquiring high spatial resolution images of selected areas. VEGETATION and HRVIR imagery have similar spectral characteristics, are acquired with similar angles and can be geo-corrected to the same projections. In SPATEM we have approached the integration of multi-temporal VEGETATION images into products derived from the high spatial resolution images in a forested Mediterranean landscape. Our approach has included both methodological aspects as well as the implications of such integration for the applications of Earth Observation.

We used the high spatial resolution imagery and a method based on image segmentation and discriminant analysis to produce a detailed land-cover map, emphasizing on forest types, and we calculated the cover fractions of the classes within each VEGETATION pixel. After a general analysis of the geometric and optical quality of the multi-temporal sequence of VEGETATION images and an optimization of the annual sequence of vegetation indices, we modeled the annual cycle of the vegetation indices of each class. We also performed an inverse analysis, in which we estimated the cover fractions from the multi-temporal sequences of the VEGETATION pixels and the models, using an Spectral Mixture Analysis. Results were poor at the pixel level, probably as a result of our models not being at extremes in the feature space, but the average estimate for the entire area of study was very accurate. We discuss some possible improvements to increase accuracy at the 1 km2 pixel level.

We also discuss the characteristics of the different annual cycles considering both the ecological attributes of the classes and their projections on a reduced plane of phenologic variability. In order to obtain a more comprehensive view of different annual cycles, we also enlarged our area to cover 17 digital vegetation maps at 1:50 000 scale within the area covered by our VEGETATION imagery (0 to 5 E, 40 to 45 N). In this case, we selected those 1 km2 pixels that were at least 90% contained within one single type of vegetation according to the maps. We found that VEGETATION data can define distinct annual cycles of vegetation indices for very detailed vegetation types. This ability will have important consequences, not only for improving global-scale land-cover mapping, but also because a better understanding of the annual cycles will let us formulate and test improved models of vegetation dynamics, beyond the elementary vegetation units that are currently used.