Resolution

Laser scanners have the potential to collect massive amounts of data from a single scan, often far more than is actually required by the application (e.g. centimetre point spacing is probably not required when capturing the 3D geometry of a series of beds in a cliff section over a kilometre extent). It is therefore important to set the resolution of the scanning system to a sensible level, defined by the scale of the geological features to be measured. Most terrestrial laser scanners sample a roughly gridded point distribution, whereby a scan line (swath) is collected on one horizontal or vertical angle setting, before the beam is deflected to the next line, building up the grid. The resolution of the grid spacing is user-defined, either in angular units representing the separation between the point measurements, or as a distance unit representing the actual grid spacing of the point cloud. The second is more complicated to determine, as the separation of the grid points (the spatial resolution) varies according to distance, whereas angular resolution remains constant; in this case, a grid spacing must be chosen corresponding to a specific range (such as an outcrop face). Laser beam divergence, a constant representing the laser footprint size over distance, also has an important effect on resolution. Many system manufacturers quote the maximum resolution of a scanner according to the smallest angular increment of the beam positioning device (mirrors or rotating head), and the divergence as a separate figure. However, the laser footprint size can play an important role and is often many times larger than the minimum angular increment, giving the appearance of higher resolution (see Lichti & Jamtsho 2006, for more details). The effect of this is that if a resolution is chosen that is too high, the fine details may become blurred as the laser footprint overlaps between point readings (Table 1).

For many geology applications, especially for outcrop-scale studies, it is also important to consider if there is need for the highest resolution dataset (and the associated computing and processing consequences), when a much lower point spacing might suffice, and indeed allow the surveying of a larger area. Designing a survey that is fit for purpose is a key part of the workflow. Despite this, where time constraints and data storage capacity allow, higher than necessary resolution data can be captured for archival purposes, on the assumption that such data will be usable in the future with increased computing power, and in case of unanticipated applications. Resampling these raw scans to a workable level is a valid option.

Site selection and coverage

Laser scanning can be compared with photographic imaging in a number of ways, the first being the requirement for a clear line of sight from the instrument to the target surface. A scan position should be chosen to allow the maximum coverage of the study area, within the maximum range and angular field of view of the instrument. It should be noted at this point that the maximum range of a scanner is affected strongly by the reflecting properties of the target material, and that manufacturer specifications are often given with respect to a certain reflectivity level, with an appropriate range correction factor for stronger or weaker reflectivity materials (Wehr & Lohr 1999). Therefore, the maximum quoted range of a scanner may be given for targets with 80% reflectivity, whereas targets with lower reflectivity may have a maximum range of only 50% or less of this (Riegl 2007). In contrast, high reflectivity targets, such as metal communication aerials or retro-reflective markings, may be obtained from well over the quoted maximum range. Bare rock surfaces generally provide a good return of up to 75% (Wehr & Lohr 1999). Atmospheric conditions also have an effect, with high levels of humidity serving to disperse the laser light, reducing the maximum range further.

A second similarity to 2D imaging is that the best results are gained when the sensor is positioned so that it is close to orthogonal, or normal, to the target, comparable with drawing traditional plane-parallel outcrop sections. Where possible, foreground details such as vegetation or rock debris should not interfere with the line of sight of the scanner; otherwise ‘range shadows’ (holes in the data caused by obstruction from an intermediate object) will be apparent in the same manner as poor subject framing in a photograph. For most study areas, a single scan will not be enough; depending on the complexity of the outcrop, more surface details are obscured as the measurements become more oblique to normal. Because of this, it is usual to capture scans from a number of positions so that a complete dataset is obtained. Often a large amount of overlap may be generated to ensure that complex terrain is covered, with range shadows kept to a minimum (Fig. 3).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

 Overlap between scans is required to fill in range shadows that would otherwise leave holes in the dataset. The example shows Lower Cretaceous carbonates exposed at Apricena quarry, Gargano, Italy, where a number of scans were needed to capture the various faces completely. (a) Schematic illustration of overlap between two scans. (b) Image of outcrop. Two scans were needed, (c) and (d), to cover the faces shown in (a). (e) Merged point clouds; (f) coloured point cloud. Area is c. 250 m × 100 m × 50 m.

The criterion for near-orthogonality is not limited to the ground position of the instrument relative to the outcrop, but also applies to height. If the instrument is much lower than the outcrop, as is commonly the case when looking up at a cliff section, then line of sight to all areas of the face becomes problematic. An elevated scan position, typically from the opposite side of a canyon or valley, may solve the problem, but is not always possible. This problem is often compounded in bedded, heterolithic outcrops where ledges and ramps result from differential weathering. In such cases, it may be difficult to secure suitable scan positions where the line of sight angle is not oblique; this problem will result in extreme range shadowing and lower data coverage and quality. In this situation, data acquired using aerial photogrammetry or aerial laser scanning may be more efficient, or used to supplement the ground-based scans (Labourdette & Jones 2007).

Finally, the majority of geological problems are 3D. One of the key advantages of lidar data over traditional outcrop photomontages and field sketches is that all the data are spatially referenced in three dimensions. Therefore the technique works best in outcrops that have a high degree of ‘three-dimensionality’; that is, multiple cliff faces at a variety of orientations. Selection of a field area with strong 3D exposure is therefore an important consideration when planning to utilize terrestrial lidar. Enge et al. (2007) described a method for semi-quantitatively assessing the ‘three-dimensionality’ of an outcrop.

Supplementary image acquisition

Raw point clouds, although accurately describing the topography of the outcrop under study, are often difficult to interpret, even if the intensity return is used to colour the data. It is therefore becoming common for high-resolution (>6 megapixel) digital cameras to be integrated into the workflow; to add qualitative information, additional true-colour images are often captured and referenced to the lidar scans. These continuous data add value to the points, and make interpretation easier because image resolution is typically higher than the point distribution (Fig. 4). The imagery may also be used as a supplementary source of measurement in its own right, taking advantage of the metric properties of a photogrammetrically calibrated camera. To use a camera and lens for making accurate measurements, parameters such as the focal length and the distortion of the lens must be modelled (Wolf & Dewitt 2000). It is common practice to fix the focus setting for the duration of a project, usually to infinity, so that the calibration remains stable. In practice, the nominal focal length of a lens is only approximate, and a good calibration can be precise to a fraction of a millimetre. Although this sounds like an insignificant figure, small errors in focal length can result in significant scaling errors over long distances. The effect is that the images will appear to be misaligned with respect to the lidar data, and thus geological interpretation and measurement will be degraded. Similarly, although lens distortion is insignificant for photographic uses, increased distortion away from the lens centre is apparent that will result in misaligned data. Like scan resolution, the choice of lens should be based on the size of the geological features being studied.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

 Detailed view of coloured point cloud (left), triangle mesh (centre) and textured outcrop (right), highlighting the difficulties associated with interpreting detailed geological features using point data alone. Data handling is also more problematic: for such a small area, there are still over 100 000 points in the point cloud, but only a few thousand triangles in the equivalent virtual outcrop.

One popular method for camera calibration is known as self-calibration, where a field of targets with known coordinates is photographed from a number of convergent angles and a mathematical adjustment is carried out to return the calibration parameters (Fraser 1997). The presence of a laser scanner can make the calibration easier to carry out if the camera is mounted in a known position relative to the scanner centre, so that targets are measured simultaneously by both devices.

The advantage of having correctly calibrated images is that they can be used directly in conjunction with the scan data, as long as they are brought into the same coordinate system. Images may be referenced to the scans in two ways: by using a direct mounting of the camera relative to the laser scanner; or by taking images independently and registering them to the scan data, using tie-points that are identifiable between the two datasets (Fig. 5). A direct mounting has the advantage of a high-accuracy registration, and automatic calculation of the camera position and direction for each image (Fig. 6) In contrast, registration using common points may be less accurate unless tie-points are laid out that are identifiable in each dataset (e.g. retro-reflective targets that give high-intensity laser returns and can be measured in the images, with or without use of the camera flash). However, because the scanner location and lighting conditions may not always be good for photography, independent image acquisition may sometimes be required or even preferred. This is useful as ideal coverage of irregular outcrops may require images to be taken between the scan locations. This is because the optimal angle of incidence for photography is lower than for the scanner, and small occlusions (such as foreground vegetation) may be easily identified and removed in the scan data, but may be distracting in the photography. The ability to acquire and incorporate independent images is also useful if those collected with the scan data turn out to sub-optimal once processing is under way. Then it is possible to revisit an outcrop and take replacement or additional photography as required.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

 Poor lighting conditions can make it desirable to capture images separately from 3D scan data, which are independent of visible light. (a) Camera mounted on scanner. (b) Separate image registered using tie-points. (c) New images captured must have tie-points established so that the camera position and orientation can be found. (d) Correctly registered images can be used in the workflow, to colour points and texture the outcrop model.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

 Typical configuration of image footprints acquired while the camera is mounted in a known position on top of the scanner. Because the coverage of a single image is generally less than the field of view of the scanner, a number of images are required to match the 3D data. (a) Terrain model before image textures are applied; (b) after texturing. The sphere represents the location of the laser scanner.

Relative positioning

Referencing different datasets to the same coordinate system is one of the most important aspects of any study involving spatial information. Scans collected in the field will initially be unrelated in space, and generally assigned an arbitrary coordinate for the instrument position. A number of methods are available to perform the registration, depending on the accuracy requirements of the field study. Traditional photogrammetric surveys have used a network of common control points established between photographic locations to provide adjustment to a single coordinate system. A minimum of three common points is required for each position photo-station. The same theory can be extended to laser scanning, where targets placed and recorded in the field provide control between overlapping instrument positions. A minimum of three common points is required for each position. To help with automation, most systems provide a fine-scanning routine where geometrically shaped targets (such as spheres) are measured at high resolution and matched to a ‘template’ shape to give the best coordinates for the target. It should be noted that the targets do not necessarily have to be placed on the study site, but may be anywhere around the instrument, as long as they are visible from multiple locations. To give the relative position of each scan location, one scan is held fixed and the others adjusted in relation to that using the tie-points. According to the number and distribution of the control targets, this method can provide high accuracy and is recommended for studies where accuracy is paramount.

In practice, however, for many geological applications the accuracy criteria can be relaxed somewhat, removing the need for such tie-points. Instead, sophisticated shape-fitting algorithms have been developed that take advantage of the overlap between adjacent point clouds (e.g. Besl & McKay 1992). Because a large amount of redundancy often exists in multiple laser scans, processing software aims to find the best fit between conjugate areas. This has been implemented either by directly minimizing the distances between the point clouds, or by first fitting surfaces to the point data, and matching these. The advantages of this approach are that no targets are required to be laid out and coordinated in the field (therefore less field time is required; also, this approach is especially useful when the outcrops are inaccessible); that the procedure is largely automated, as the user usually only needs to select three or more common points between two scans as initialization for the adjustment; and that high data redundancies can provide a very precise solution. However, care must be taken in the field to ensure that suitable levels of overlap are collected so that matching can be performed. Around 10% overlap is a minimum level that is needed (Bellian et al. 2005), although this value should be adjusted if there is a great difference in the common areas caused by range shadows at the scan edges that may make identification of corresponding points impossible and the matching fail. Quality control must be performed by the user to ensure that the matching result is valid.

Absolute positioning

Registration of the scans relative to each other allows the data to be combined, viewed and interpreted in a single coordinate system. For many applications this may be enough for carrying out all measurement, interpretation and modelling tasks. However, a major advantage of using this approach is the ability to relate disjointed study areas relative to each other, in a national or global coordinate system. In addition, absolute positioning allows further registered data to be integrated with the laser scans. An example of this is where a particular stratigraphic unit crops out sporadically, and cannot be tracked continuously with overlapping scan data. This poses a problem for relative registration, as the various areas of outcrop must be related in space. GPS positioning is therefore used to provide absolute registration, by positioning reference targets in the scan data, or by coordinating the instrument positions.

A number of techniques are available for GPS positioning, from handheld devices that measure a single point position, to relative positioning that measures the coordinates of an unknown point relative to a second receiver operating from a known control point (see, e.g. Hofmann-Wellenhof et al. 1997). Relative, or differential, positioning requires more sophisticated equipment than handheld devices, with survey grade systems being essential to achieve high-accuracy solutions. Both single- and dual-frequency receivers may be used, with dual-frequency receivers giving a higher accuracy measurement over longer distances between the base station and unknown point, with less occupation time. However, the high cost of a dual-frequency system may not be justifiable for simply positioning a lidar instrument to better than 0.1 m level.

Registration by positioning the instrument itself is the more common method, because it only requires a GPS antenna to be set at a known offset to the scanner measurement centre (see Fig. 2), and observed for around 30 min for a single-frequency system and a good satellite configuration (Hofmann-Wellenhof et al. 1997), relative to a local base station. No targets are required to be set out and measured by the scanner. A minimum of three points is required to transform the scan data to the global system; therefore three overlapping scans are needed, each with a GPS scanner coordinate. A single GPS coordinate will give the correct scanner position, but the orientation will be unknown. A poor satellite configuration or obstructions to the sky view in canyons may mean that longer observation times or further observation stations may be necessary. Accurate instrument positions are then determined by post-processing the recorded data series relative to the base station. Finally, the GPS positions are used to transform the scan data from relative local coordinates into the global system. Where fewer than three scans are overlapping, targets should be located in one or more scans and measured using kinematic GPS (real-time or post-processed; see, e.g. McCaffrey et al. (2005) for review). This allows the orientation of at least one scan to be fixed and the others registered relatively.

Points and DEMs

To facilitate interpretation, the 2D digital images acquired with the integrated camera can be used to give a true colour value to the lidar points; this is especially useful as a low-resolution preview of the entire area. Using the image calibration, position and orientation, perspective projection is used to project each 3D laser point into the corresponding registered image. The colour of the image pixel is queried and assigned to the projected 3D point.

Registered point clouds may be combined and manipulated in any way desired by the user to gain the geological interpretation or geometric measurements required by the application. Points of interest may be selected from the vast overall point cloud, such as only the points corresponding to a single layer, by drawing a polygon round the desired points to form a new ‘mini’ point cloud. Quantitative measurements may be carried out, from simple distances between two points, fitting best-fit planes through the point clouds, and calculating strike and dip measurements from selected points. Available software dictates the features that are available.

Generally, two forms of the lidar data are of most use in outcrop studies: the point cloud data and DEMs (also called surfaces or meshes) formed from the original points. Use of the point cloud provides the highest accuracy, as all the collected data are available to interrogate. However, the sheer quantity of 3D points is often a barrier to working with the raw data, as the features of interest may be massively oversampled with respect to the topography of the outcrop. An example is an outcrop face where the geometry of bed boundaries is to be digitized. Although the vertical topography (cliff face) may be relatively smooth, the laser scanner does not account for this and records data equally on rugged and smooth surfaces, resulting in a large quantity of points. To record the bed boundaries, it is the line geometry that is most important, rather than the surface roughness, which may be more a product of weathering or manmade quarrying. To save on the computer resources required to load large point clouds (a merged dataset may contain more points than can be loaded), decimation of the raw point data and creation of a DEM is valid. Texturing, or draping, the digital imagery from the camera onto the created DEM results in a photorealistic model that is of extreme value for interpretation, visualization and education. Use of a textured virtual outcrop model also addresses another problem with using points alone: that of accurately identifying fine-scale features. Although the point cloud may have very high resolution, at a certain interpretation level, on zooming in, the point spacing becomes too large to accurately identify fine-scale features (Fig. 4). A textured model uses the higher resolution and continuity of the digital imagery to ‘fill’ the gaps between the points, making interpretation easier.

DEM creation from point clouds

Creating a virtual outcrop model involves finding a best-fit surface through the raw points, to produce a triangulated mesh. This is similar to methods used for aerial terrain data, where a point set is triangulated to form a grid or triangular irregular network (TIN) model, but different from the ‘surface gridding’ approach utilized by most (but not all) seismic data and subsurface reservoir modelling packages commonly used today. The advantages of using TINs have been discussed in detail by McCullagh (1998). The major difference, and indeed difficulty, with the data acquired from a terrestrial laser scanner is that the point clouds are truly 3D, having points on vertical and overhanging surfaces. This is in stark contrast to conventional aerial data, where it is usual to have only one point for each (x, y) DEM point. It is easy to form a DEM from such a point distribution, as a 2D Delaunay triangulation finds the best criteria for triangle creation automatically (e.g. McCullagh 1998). However, running such an algorithm on 3D lidar points will result in an erroneous surface model, where points of all elevations are connected on 2D adjacency, irrespective of range or obstructions. Three-dimensional surface reconstruction is therefore a complicated procedure, with much research in the fields of computing, and it is yet to be solved to allow fully automated algorithms. Nevertheless, software is available that can make DEMs from input scans. The process is facilitated by using the geometry of the collected data as an aid in determining the mesh; triangulation of a single scan is a simple matter, because the point cloud is a rough grid of points with angles and distances known relative to the instrument position. Therefore a 2D triangulation is possible, with a maximum edge length filter to prevent connecting points that are far apart in range. Integrating several scans is more problematic, as it is difficult to automatically triangulate complex overlapping point clouds, and editing is usually required.

Some preparation of the point cloud data is usually necessary prior to mesh creation. Because the original dataset is very dense, and may contain vegetation and other non-geology points, a pre-processing stage is carried out. Points may be manually selected and removed from each of the point clouds, a potentially time-consuming procedure if many unwanted points exist. Some automation is possible. Filtering of the point cloud may be carried out according to the intensity of the returns, where enough resolution of the digitized signal is provided by the instrument to distinguish different materials. Generally, the more noise caused by vegetation or registration errors there is in the point cloud, the worse the success rate of automatic meshing procedures will be. To combat this, preparation of the point cloud for meshing may be continued by smoothing and decimating the overall point set; because of overlap between different scans required to obtain full coverage of the outcrop, much redundancy may be present, resulting in an uneven point density. The point cloud may be decimated on surface curvature, or by using an octree, where a 3D grid is formed and each cell is populated using an average value of all points found within the cell volume (e.g. Girardeau-Montaut et al. 2005). The result is a more regular point cloud, with little loss of accuracy. In the experiences gained so far (Table 2), it is possible to significantly reduce the original point cloud, whilst maintaining an accurate mesh. Indeed, aggressive decimation must be carried out to make models that can be loaded and visualized at interactive frame rates. This is a decision-making process that must be evaluated for each model, and for the size of the area and geological features that are being studied. This is a limitation with current hardware that will improve with time, but in general it can be said that if the utmost accuracy is required, the point clouds must be broken down into smaller areas to be able to be meshed and textured.

Some mesh editing is usually required after automated processing has been used, to fill holes and remove errors where the algorithm was unsuccessful. Although only a subset of the original scan data may have been used to make the mesh, this may still represent a significant number of points, leading to a large number of triangles (around double the number of input points). Loading textured outcrop models is intensive on PC resources, and the RAM and graphics card specification becomes critical. Therefore, a final mesh editing stage is intelligent decimation to reduce the number of triangles in areas of low surface roughness by representing them with larger triangles, whilst keeping smaller triangles in detailed areas (Buckley & Mitchell 2004). A paradox of outcrop modelling is that often the smoothest surfaces are of most interest (i.e. a vertical outcrop face), so care should be taken during decimation that these areas are not over-simplified.

Virtual outcrop formation

The final stage in lidar data preparation is the integration of the triangle meshes with the registered digital camera imagery to form a virtual outcrop model, and the quality control of that model. The images are used as textures, which are mapped (or ‘draped’) onto each triangle in the mesh, based on the projection of each triangle vertex into the digital image. Because a large number of images may be available, criteria for choosing the most suitable image part for each triangle must be used. These criteria are based on the direction the camera is pointing with respect to the triangle orientation, distance from the camera to the triangle, and quality of the image texture itself. Different lighting conditions or poor image exposures will severely reduce the visual impact of a virtual outcrop model, and may make interpretation difficult. A key issue is when adjacent triangles are textured with image data taken from very different angles or distances and under different lighting conditions. This highlights the triangles and distracts from the overall virtual outcrop quality. Therefore, attention to image acquisition while in the field is extremely important, more important than scan acquisition, which is light-independent, and should be given priority to achieve a suitable end product without performing additional and time-consuming manual image enhancement. Once suitable images have been chosen and optionally enhanced, the triangle texture mapping procedure is largely automatic, resulting in a 3D photorealistic model. The workflow to achieve the virtual outcrop model is summarized in Figure 7.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 7.

 Summary of the workflow for using lidar data in outcrop studies, from field acquisition to creation of a product suitable for interpretation and digitization of geology. Timings represent the approximate weighting of the different field and processing tasks for an area of around 0.5 km², whereby it can be seen that data collection may represent only around a quarter of the time needed to obtain a virtual outcrop model. More ‘noise’ (e.g. vegetation) or complex topography may significantly increase the time required for point cloud and mesh editing.

Interpretation and measurement

The processed virtual outcrop model (or point clouds) can be visualized in a 3D environment, where the user is able to change the virtual camera viewpoint to suit their needs. In this way, inaccessible areas in the field become accessible, and interpretation and measurement can be performed, which can then be integrated with traditional field data. Although visualization is extremely valuable, it is the potential for adding quantitative information that makes spatial data collection technologies so useful. The aim for many studies is to model an outcrop analogue within subsurface, geocellular reservoir modelling software, requiring accurate geometry from the lidar data. It is relatively simple to interpret the geology directly onto a virtual outcrop model, defining interest points or linear features as required. These features can be later imported to the reservoir modelling software, to be used in the generation of surfaces that provide the framework for the geocellular models. For example, a surface may be tracked in three dimensions across a wide area (and multiple virtual models), using the GPS absolute positioning to relate the different sections in space. The surface is represented by 3D lines, and the geology reconstructed by interpolating and extrapolating the line points to fill in the geometry between exposed areas. The lidar geometry therefore provides a basis for building geological models that allows more detailed studies to be carried out, on smaller-scale features.

Figure 8 shows an example from the Ferron Sandstone, Utah, USA, where the objective was to investigate the geometries of seaward dipping, delta front clinoforms. The aim was to map the clinoforms, to extract numerical data on their thickness and lateral extent, and to use the mapped data as a framework for building 3D geocellular models. The geocellular models were built in software typically used for modelling subsurface hydrocarbon reservoirs, which allow fluid flow to be simulated under reservoir conditions. Thus the detailed geometries observed in the outcrop can be used as an analogue for subsurface cases where geometric data are sparse or lacking. The sedimentology and stratigraphy of the Ferron Sandstone have been extensively documented by Ryer (1981) and Anderson & Ryer (2004).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 8.

 Geological interpretation: study of clinoform surfaces in the Ferron Sandstone, Utah, USA, for building a highly detailed reservoir model that captures the subtle geometric relationships between small-scale surfaces. (a) Virtual outcrop model. (b) Interpreted surface boundaries are traced directly onto the virtual outcrop, resulting in 3D lines. (c) Surfaces built from the 3D lines within reservoir modelling software can be visualized and analysed simultaneously with the outcrop model. (d) A 3D grid created from input geology surfaces in reservoir modelling software (3.5× vertical exaggeration). (e) Example of quantitative analysis possible using digitized clinoform data: study of clinoform length and frequency in the Ferron outcrop, an area characterized by steeply dipping clinoforms. Study area is c. 200 m × 200 m × 25 m.

The study area was 2 km² and includes around 3 km of near-continuous cliff section located at different orientations, ensuring both close to depositional strike and dip coverage. The data for the virtual outcrop comprised 18 scans collected over 5 days. The total dataset included c. 63 million raw points, 470 registered images, and an additional 45 images taken separately from the scanner. The point data were meshed and textured to form a virtual outcrop model that was then used for interpretation. Clinoform contacts were digitized on the virtual outcrop, with the user able to rotate and translate the model until the best orientation was found for seeing the geology. The result of this was a set of detailed line features (vertical separations from decimetre to metre scale), which were analysed (Fig. 8e) and imported to geocellular modelling software for building surfaces using the lines as constraints. These surfaces, combined with conventional sedimentary logging carried out in the field, were used to model volumes representing the true clinoform geometry, in greater detail than previously possible. Such a study demonstrates the contribution of lidar techniques, as the vertical outcrop face would have made in situ measurements hazardous and extremely time consuming. Here, the lidar acquisition could be combined with logging and interpretation to build up a truly detailed dataset in only a few field days. The curved nature of the outcrop and the accurate 3D mapping with lidar allowed the clinoform surfaces to be accurately recreated in a way that is not possible in two dimensions. The dataset could be explored in much greater detail back in the office and, if necessary, a further field campaign could be carried out at a later date to verify interpretations made on the virtual outcrop.

Building surfaces from interpreted line features is only one measure that can be captured using the virtual outcrop model. Additional examples are strike and dip measurements, plotting fracture networks and obtaining cross-sections of the outcrop surface. Again, the application defines the required products. Table 2 outlines further projects and datasets captured by the authors, and the geological problems that benefited from the use of lidar.

Software for lidar processing and interpretation

The above discussion has focused on the generalities of lidar processing, without being specific to the software tools that are available. Such software tools are transitory, and description of the various merits of each is out of the scope of this paper. A number of commercially available software tools exist for processing lidar data, although it should be noted that these packages have not been developed specifically for geology applications. Indeed, most have not been developed for terrestrial survey applications and the long-range laser scanners that are of most use in geology, and instead have been driven by industrial applications (such as scanning manufacturing components in the case of PolyWorks by InnovMetric, one leading package). Such close-range applications are normally devoid of much of the data noise found in real-world environments, and therefore some algorithms and procedures are not yet perfected, making the development of a processing workflow not entirely as simple as selecting an off-the-shelf package. Software usually offers import of native scanner formats, registration by targets and surface matching, mesh creation (with variable success) and editing, as well as texturing using imagery. Additionally, some software (PolyWorks by InnovMetric; RiSCAN Pro by Riegl) allows digitization of computer-aided design (CAD) features, which can be used for interpretation and input to modelling.

Because of the large amount of data that are generated, more sophisticated handling and visualization strategies are required, provided by the dedicated software packages. The true 3D nature of point clouds and virtual outcrop models means that this form of data is not suitable for inputting to most geographical information systems (which are inherently 2.5D, where only one height value is permitted for each x, y position) that many geologists use. Therefore a developing part of the workflow remains the standardization of data formats for sharing project results. The lack of commercial solutions directly providing a geological interpretation framework for lidar data has meant that in-house software has been created for more advanced data conversion, processing and interpretation tasks.