Error assessment of extracting morphological parameters of bank gullies by manual visual interpretation based on QuickBird imagery

Abstract: High-resolution satellite images such as those from Ikonos and QuickBird are increasingly available and are considered a valuable tool for extracting the consequences of gully erosion, and QuickBird imagery has been used to map gully features. To evaluate the accuracy of extracting morphological parameters of bank gullies including area, perimeter, and length from QuickBird imagery by manual visual interpretation, and to analyze the causes of errors, 20 bank gullies were selected in Hegou catchment and Qiaogou catchment located in the Loess Plateau of northern Shaanxi province, respectively. 3D laser scanner was used to measure the topography of bank gullies within two weeks after the QuickBird images were taken. The interval among measurement points of 3D laser scanner was set as 0.15 m. 3D laser scanner had been increasingly used to measure the topography as it can provide a high accuracy, non-contacted, penetrating and rapid method that enables the rapid production of accurate, high-resolution digital elevation model (DEM). Based on Delaunay triangulated irregular networks (TIN), then DEM with the pixel size of 0.15 m × 0.15 m was created using the 3D Analyst extension of ArcGIS 9.3. On this basis, the values of morphological parameters extracted from DEM were taken as actual values and errors of morphological parameters of bank gullies extracted from QuickBird imagery were determined. The average absolute error, maximum absolute error, average relative error, maximum relative error, and maximum relative error among different interpretation persons were used to assess the accuracy of manual visual interpretation of QuickBird imagery. The results showed that maximum relative error of area and perimeter ranged from 4.1% to 22.2% and 1.4% to 22.2% in Hegou catchment, respectively, while maximum relative error ranged from 2.4% to 22.6% for area and 4.5% to 29.7% for perimeter in Qiaogou catchment, respectively. Although maximum relative error appeared to be large for some gullies, average relative error of area and perimeter were both about 5% in two study areas. In addition, by overlaying the two valley regions extracted from QuickBird imagery and DEM, the area that the distance offsets was larger than 0.6 m (QuickBird image one pixel value) remained below 4%. Average relative error of gully length in Hegou catchment and Qiaogou catchment were about 2% and 5%, respectively, with corresponding average of the average absolute error of 0.5 m and 0.75 m. Correlation analysis showed that average relative error, maximum relative error and maximum relative error among different interpretation persons were negatively correlated with the area and length of bank gully (P<0.05), which meant that the larger the area and length of bank gully were, the smaller the errors were. This suggested that choosing large gullies would reduce the errors when the morphological parameters were extracted manually from QuickBird imagery. The accuracy of manual visual interpretation was also influenced by the vegetation types near the gully boundary. Catchments which land was covered with native grassland had higher precision than those were covered with shrub and grass. Overall, QuickBird imagery can provide reliable data source for extracting morphological parameters of bank gullies. The results can provide the references for assessing gully erosion using QuickBird imagery at larger spatial scales rather than direct field measurements.