Gully Features Extraction Using Remote Sensing Techniques

Abstract

Gullies are large and deep erosion depressions or channels normally occurring in drainage ways. They are spectrally heterogeneous, making them difficult to map using pixel based classification technique. The advancement of remote sensing in terms of Geographic Object Based Image Analysis (GEOBIA) provides new possibilities to extract gullies with relative ease. This study investigates and tests a GEOBIA technique called Imagine Objective for gully features extraction in the Capricorn District Municipality of Limpopo Province. IO extracts gullies by firstly group the pixel information and then subjects them to raster and vector conversion and refinement algorithms. For the purpose of assessing the accuracy of the IO created gullies, reference data were created by manual digitising gullies from SPOT 5 satellite imagery on the background. The error matrix was computed and the results indicated a user's accuracy of 98.67% and 54% for non-gully and gully class respectively; a producer's accuracy of 68.20% and 97.59% for non-gully and gully class respectively; a overall classification accuracy of 76.33% and a kappa statistic of 0.95, 0.36 and 0.52 for non-gully class, gully class and total kappa statistic respectively. Although the accuracy levels are considered moderately acceptable, it is recommended that much higher spatial resolution imagery such as Quickbird be used in future and other functionalities of IO be tested.