Context and objectives
A major challenge facing ecologists studying the earth as a dynamic system is the mapping of vegetation quantities over time as over large areas in order to parametrize biogeochemical cycle and climate models. The role of remote sensing data as key source of quantitative information for the regional and the global scale is no longer discussed.
Today, the data interpretation relies on a fitting of a linear or polynomial semi-empirical models based on the relationships between the field-measured variable and the sensor signal. This simplistic calibration between observations at non-compatible spatial resolutions can no longer be supported. According this approach a point measurement is directly related to a pixel signal corresponding to an area as large as one square kilometer. This research proposes to tackle the upscaling issue between observation level using the recent evolution of geostatistics.
The overall objective is to develop methods for estimating LAI variables derived from satellite data based on geostatistics concepts. The latter would provide the basic approach to take advantage of the spatial autocorrelation of these variables.
The goal is a functional characterization of vegetation cover by biophysical variables for various aggregation levels from local to regional scale. At the same time this study would also develop a method of quantitative assessment of seasonality and spatial variability of the vegetation cover. Moreover, the proposed approach gives a way for new upscaling methods between different observation levels in order to enhance the semi-empirical modeling based on relationships between measured and remotely sensed variables.
Expected scientific results
Products of the research are both methodological and thematic.
The first project output is to make available methods of geostatistics adapted to remote sensing data analysis and especially to scaling up between different levels of observations. A new calibration model of semi-empirical relations between remotely sensed and locally observed variables is developed. The prediction performances of geostatistical approach for mapping quantitative variables of vegetation is compared with those of classical calibration techniques that ignore the space-time dependence of observations. Finally, a procedure of field sampling optimization based on spatial structure of variables as on signal pattern is also proposed.
From the thematic point of view, this research allows for the first time to document the temporal variability of LAI of a humid tropical forest at a scale compatible with climatic models.
This research suggest that structural information can be retrieved from images whose pixel size exceed that of the elementary objects constituting the target, at the condition to dispose of images of the same target but with different geometries of observation