干旱区生态环境知识资源中心

Advances in remote sensing technology, notably in airborne Light Detection And Ranging (LiDAR), have facilitated the acquisition of high-resolution topographic and vegetation datasets over increasingly large areas. Whilst such datasets may provide quantitative information on surface morphology and vegetation structure in riparian zones, existing approaches for processing raw LiDAR data perform poorly in riparian channel environments. A new algorithm for separating vegetation from topography in raw LiDAR data, and the performance of the Elliptical Inverse Distance Weighting (EIDW) procedure for interpolating the remaining ground points, are evaluated using data derived from a semi-arid ephemeral river. The filtering procedure, which first applies a threshold (either slope or elevation) to classify vegetation highpoints, and second a regional growing algorithm from these high-points, avoids the classification of high channel banks as vegetation, preserving existing channel morphology for subsequent interpolation (2.90-9.21% calibration error; 4.53-7.44% error in evaluation for slope threshold). EIDW, which accounts for surface anisotropy by converting the remaining elevation points to streamwise co-ordinates, can outperform isoptropic interpolation (IDW) on channel banks, however, performs less well in isotropic conditions, and when local anisotropy is different to that of the main channel. A key finding of this research is that filtering parameter uncertainty affects the performance of the interpolation procedure; resultant errors may propagate into the Digital Elevation Model (DEM) and subsequently derived products, such as Canopy Height Models (CHMs). Consequently, it is important that this uncertainty is assessed. Understanding and developing methods to deal with such errors is important to inform users of the true quality of laser scanning products, such that they can be used effectively in hydrological applications. (C) 2012 Elsevier B.V. All rights reserved.