Mapping tillage practices over a peri-urban region using artificial neural networks applied to combined spot and asar/envisat images

摘要

This study aimed at assessing the potential of combining synchronous SPOT4 and ENVISAT/ASAR images for mapping tillage practices of bare agricultural fields over a 220 km²-peri-urban area located in the western suburbs of Paris (France). The approach relied on topsoil roughness measurements combined with information about tillage operations: 28 reference zones demarcated according to soil map information, the visual interpretation of the SPOT4 infrared coloured image and their standard deviation of surface height were related to the backscattering coefficient of the ASAR image (R² 0.70). They were then used for training/validating neural networks on co-registered 20 m-SPOT/ASAR 6 bands with 15 bootstrapping iterations. The overall mean validation accuracy was 94.9%, while the producer's and user's mean validation accuracies were 91.6-81.5% and 61.8-75.4% for smooth and rough surfaces respectively. The SPOT/ASAR synergy thus enabled to map soil tillage operations with reasonable accuracy.