Characterization of surface roughness of bare agricultural soils using LiDAR.

摘要

The current definition and parameterization of surface roughness have the advantages of being simple, but it is generally accepted that the way soil roughness needs to be measured and described for the modeling of microwave phenomena is not yet fully understood. The research hypotheses tested in this work, that explain these limitations are: the existence of systematic and random errors that are not properly accounted for in the measurements; the inadequate practice of using 2D profiles to derive 3D characteristics of the surface; the scaling up of in situ roughness measurements under the assumptions of homogeneity and isotropy. To test these hypotheses, Scanning light detection and ranging (LiDAR) technology was employed.;LiDAR enables the digitization of surface height variations in three-dimensions (3D) and thus allows for an improved characterization of surface roughness. To address the above mentioned challenges in the characterization of surface roughness with LiDAR technology, a multi-step approach was followed. The first step was to investigate the factors that affect the precision and accuracy of roughness parameterization from 2D measurements. The next step was to analyze and test the current assumptions of roughness characterization following the traditional 2D formulation. This was followed by developing methodologies to characterize roughness from 3D information; that is truly representative of the entire surface. Finally, the issue of scaling was addressed by developing methodologies to use airborne LiDAR to derive millimeter level roughness maps of large areas and to prove the heterogeneity and anisotropy of roughness characteristics over large areas.;The issue of the accuracy and precision of roughness parameters was studied by performing two accuracy assessments and a direct comparison between a meshboard and the ground-based LiDAR. The first accuracy assessment was based on computer generation of random rough surfaces and the second based on measurements employing roughness references. Results indicate that to obtain accurate and repeatable parameter values it is necessary to properly characterize the instrument random errors. The height variation measurements obtained with any instrument are the result of the addition of two random processes: the surface roughness and the instrument random error. If the instrument error is not properly characterized and considered, the computed roughness parameter values will be corrupted and thus inaccurate. Of the parameters, the RMSh is the least sensitive to instrument noise and it was determined that this parameter can be derived from ground-based LiDAR with an accuracy of better than 1mm. The achievable accuracy in retrieving the ACF and associated CL depends on the relative magnitudes of the surface's roughness and the instrument error. Correlation lengths can be accurately determined to better than a cm if the surface RMSh is larger than 1 cm.;With regards to the assumptions used to characterize roughness using the traditional 2D formulation, it was found that agricultural surfaces exhibit multi-scale roughness characteristics. This contradicts the single-scale assumption. However, it is possible to obtain roughness at a particular scale if the proper detrending techniques are applied. It was also determined that the exponential and Gaussian ACF models are just two limiting cases, and that the majority of surfaces have characteristic ACFs intermediate between these two models. In contrast to what has been commonly reported, no correlation was found between the RMSh and CL. However, it was found that at small scales there is a possible negative correlation between RMSh and the maximum observable CL.;3D characterization of surfaces of agricultural fields reveals that they are generally even more multi-scale in terms of their roughness than is evident from the 2D formulation. Roughness parameters obtained from the 2D formulation underestimate the characteristics of the surface; by 25% in terms of the RMSh and 30% in terms of the CL. This is because profiles generally do not record the extremes of the surface in a single transect and do not necessary follow the trend of the entire surface. The assumptions of homogeneity and isotropy were proved to not be valid even for small areas. 3D digital elevation models (DEMs) derived from ground-based LiDAR allow for the characterization of roughness with advance tools in the spatial-temporal domain. When the characterization of millimeter level surface roughness of large areas is required, data from high resolution airborne LiDAR can be used. RMSh derived from airborne data was within 1 mm of the RMSh derived from ground-based LiDAR data. (Abstract shortened by UMI.).