Arid Land cover change trend analysis with series of satellite images for desertification monitoring in Northern Africa

摘要

A time series of 27 Landsat images (MSS and TM) over an ecological study area in Tunisia has been georeferenced and resampled. Conversion of DNs into ground reflectance has been done in two steps: full atmospheric correction of the most recent images and regression method with pseudo-invariant features detected through statistical analysis. Three different approaches of change detection have been tested : change in classes through time, vegetation and colour indices linear trend analysis and Fourier decomposition. Strong degradation and then recovery has been evidenced during the 30 years time span, demonstrating the necessity of long term monitoring for desertification assessment.