Reedbed monitoring using classification trees and SPOT-5 seasonal time series

The Camargue, the Rhône river delta in south of France, has lost 40,000 ha of natural areas, including 33,000 ha of wetlands over the last 60 years, following the extension of agriculture, salt exploitation and industry. Reed development and density in Camargue marshes is influenced by physical factors such as salinity, water depth, and water level fluctuations, which have an effect on reflectance spectra. Classification trees applied to time series of SPOT-5 images appear as a powerful and reliable tool for monitoring wetland vegetation experiencing different hydrological regimes. The resulting tree provided a cross-validation accuracy of 98.7% and a mapping accuracy of 98.6% (2005) and 98.1% (2006). Misclassifications were partly explained by digitizing inaccuracies, and were not related to biophysical parameters of reedbeds. The resolution of SPOT-5 scenes provides an adequate scale for acquiring detailed field data within homogeneous stands, allowing to optimize the time spent for data collecting and to properly locate the sampled plots on the ground and on the scenes. Our results demonstrate that it is possible with a good field campaign to avoid repeated sampling for a long-term cost-efficient monitoring of reed marshes. The accuracy and reliability of our models provide a vision where the roles are reversed: the field campaigns become a complementary tool in wetland monitoring using satellite remote sensing.