Laser-scanning and 1D wavelet transform for artificial drainage detection in Mediterranean rural landscapes

Artificial drainage networks such as ditches networks are landscape elements that control many hydrological transfers. To introduce these elements in hydrological models, we need to develop methods that allow their representation in space. We aimed here to assess the utilities of LiDAR data in the detecting and characterizing of ditches in Mediterranean cultivated rural landscapes. To perform the detection, we combined jointly laser-scanning and parcel boundaries data and developed a methodology based on : 1- estimation from last pulse laser points of terrain altimetric profiles that are perpendicular to parcel boundaries; 2- 1D wavelet transforms of the profiles ; 3- a supervised classification of profiles with CART on the wavelet coefficients that are relevant for the size of the shapes we want to detect. The methodology we developed was applied in France on a typical basin of the Mediterranean vineyard landscapes. Compared to visual interpretation of the profiles, we obtained satisfactory detection rate. According to these results, high resolution laser-scanning data appears reliable for ditches detection but, due to vegetation and sampling resolution, misclassification on ditches typology is still important.