Terrain¶

CloudCompare
- Obtaining CloudCompare and its documentation
- CloudCompare configuration
Airborn LiDAR (ALS)
Other sources
Surface orientation
Hole filling
- Whole site
- Locally
Last requirements

The geometry of the terrain is a main element controlling the propagation of trajectories. That is why the preparation of terrain data is very important and should be treated with care. This step also requires a good part of the time to devote to rockfall simulation projects. In general, it comes down to preparing the elevation data in the form of points distributed in a relatively homogeneous way, free of holes and artefacts, and corresponding only to the ground without infrastructure and vegetation. These operations are detailed in the different sections of this manual concerning the terrain.

Also, the terrain footprint must be large enough to contain all the trajectories (from the source to the stop position). If this is not possible, the terrain should be cut to form a rectangular tile oriented along the X and Y dimensions. Otherwise, beware of results near irregular boundaries of the terrain. Indeed, some trajectories could have been omitted, having been considered as having passed through the terrain rather than having reached the limits of the terrain.

When rocks pass through the terrain falling into holes that have not been filled (or continuing past irregular terrain boundaries), stnParabel restarts their simulation without saving the previous attempt. After two unsuccessful attempts, the position of the source is shifted by a quarter of the maximum diameter (d1) to introduce variability by moving slightly away from the wall. This is repeated for every two additional unsuccessful attempts (for a limit of ¾ of d1, or 8 attempts per source).

With this mechanism, it is possible to make preliminary simulations on raw data with artefacts and/or vegetation and some holes to have a quick overview of the situation for a site. In this case, the terrain data can be quickly prepared according to the method described in the section “Last requirements”, ideally with a subsampling of the points if there are very many. The orientation of the surfaces must also be approximated with the data, which is generally the case by default with data from photogrammetry. The simulated blocks will bounce on all points encountered as if they were part of the ground (and therefore with an rebound behavior against the ground).

Also, again for an overview of the situation for a site, preliminary simulations can be generated without having sufficient terrain point density relative to the size of the blocks (d1 smaller than 4x the point spacing), as long as the points of the terrain are distributed relatively evenly. In this case, the terrain is interpolated for each impact using a neighborhood set at a radius of 2x the minimum average point spacing. The roughness of the elevation data relative to the size of the impactant is then not taken into account as well, which can lead to longer runouts and a little less deviation. The artificial roughness can still be added correctly. Also, interpolation errors can sometimes occur when terrain is interpolated in this way at sharp edges, which can affect impacts.

These errors do not affect normal simulations, with point spacing less than ¼ of the maximum diameter (d1) of the smallest particle to be simulated. Indeed, stnParabel switches to impacts directly against the points of the terrain if the RMS of the interpolation with respect to the points of the neighborhood is too large (bad fit). Therefore, it is recommended to use an appropriate terrain point density for preparing the terrain for simulations, as described in the following terrain sections. CloudCompare software is mainly used for demonstrations, you are free to use other solutions if they allow you to achieve the same results.

CloudCompare ¶

CloudCompare is a very useful open source software for project preparation and for visualizing results. This section first shows how to get it and access its documentation, and then how to configure it for the project.

Obtaining CloudCompare and its documentation ¶

To download it and access its documentation, you can do a simple search on the web and access their page. There is a description of the software, links to download the latest versions, tutorials and complete documentation for all the tools available. The examples in this version of the manual are from version 2.11 beta, which was the most recent at the time of editing the manual.

CloudCompare is particularly suited to the processing of point clouds, but is also compatible with a variety of formats, such as shapefiles and rasters (ASCII, LAS, PLY, OBJ, DXF, GeoTIFF, etc.). The description of all the available tools and plugins can be found in the Tools and algorithms section. From there, we find the exhaustive list of many tools ... And plugins, such as CANUPO, for the automatic classification of point clouds.

If CloudCompare becomes an integral part of your workflow, maybe consider contributing to the open source project!

CloudCompare configuration ¶

It is possible to reset the interface from the display settings before starting. The settings to be changed subsequently are as follows:

Make sure that the DB Tree panel on the left of the interface is large enough to display all the properties of each layer;
Activate the EDL (Eye-Dome Lighting) shader to preview the shading for a better 3D perception of the scene.
Unless you cut the point cloud with the Interactive Segmentation Tool, or you want to produce figures at constant scale (distant objects are the same size as close objects, which allows the use of a scale bar), the 3D perception of the scene is facilitated by a natural display with perspective. The rendering of shadows with the EDL filter is also more contrasted when the perspective is activated, via the option Object-centered perspective.
Activate the automatic capture of a rotation point around the point near the center of the screen ( auto-pick rotation center ).
In the display settings, the text size can be adjusted according to the figures to be produced. One to two significant digits is usually sufficient for the project, so the precision of the displayed number precision can be adjusted accordingly.
Still in the display settings, the option to simplify point clouds when moved can be disabled. Movements may be a little slower depending on the performance of the computer used, but the point cloud display is instantaneous after the movement.
The option to display points in circles instead of square dots can be explored. The rendering of final figures is generally improved, but navigation can be slowed down due to the greater resources used.
The automatic calculation of octrees can be set to always .

Airborn LiDAR (ALS)¶

Preparing the terrain from ALS airborne LiDAR data is probably the easiest thing to do for the project if such data is available.

This data can be easily imported into CloudCompare by simply dragging and dropping the LAS/LAZ file(s) into the active view of the software. The information assigned to the points which are deemed unnecessary for the project can be unchecked at this stage to save RAM memory. The classification, if present, will however be very useful later. If there is no classification, the CANUPO and CSF plugins can be good starting points for cleaning up the data. Outside of CloudCompare, the LAStools commercial tool suite from rapidlasso GmbH, and particularly the lasground tool, may be an avenue to explore.

Then you have to make sure that the local coordinates are close to zero. Otherwise, a transformation must be applied to pass from the original coordinates to local coordinates which are close to zero.

The displayed colors of the imported point cloud should match the classification, unless you have retained other information assigned to the imported points. To extract the points corresponding to the ground, the classification Scalar field must be activated if it is not already active.

You must then use the Filter by value function to extract the points from the ground, and set the minimum and maximum limits to the value corresponding to the soil/terrain/ground category (eg 2).

The classification scalar field can then be cleared for the exported ground points, since they all have the same classification value.

Other sources ¶

Gridded data (raster)¶

It is recommended to prepare your terrain for the project from data in point cloud format. However, if only gridded data (raster) is available, it is possible to convert it to a point cloud for use as terrain.

To do this, it is important to use the terrain data (DTM) and not the surface data (DSM) of the digital elevation model (DEM).

When importing into CloudCompare, make sure that the transformation generates local coordinates near zero.

Cells with no_data_values, if present, are not eliminated by default. To extract the elevation values to keep, use the Filter by value function and extract the elevation values within a range that excludes no_data_values.

Also, the imported points are displayed in 2D, because the information of each cell of the raster is not assigned to the Z coordinate by default. You must use the Set SF to coordinate(s) function to associate the elevation values with the Z coordinates.

The Scalar field containing the information of the cells of the imported raster can now be deleted, since this information has been transferred to the Z coordinates.

Terrestrial LiDAR (TLS)¶

Terrain preparation from data acquired with a Terrestrial LiDAR (TLS) differs slightly from that from Airborn LiDAR (ALS) or digital raster terrain model. In the absence of vegetation, artefacts and large areas of occlusion (absence of data forming large holes), data processing is relatively straightforward. It consists in subsampling the data set with the Subsample tool to the desired point spacing (see the “Hole filling” section). It may also be beneficial to combine TLS data with data from other sources. For example, ALS data is often complementary, as it covers relatively flat terrain, while TLS data better covers cliffs and overhangs.

However, processing TLS data can quickly become complex if there are gaps to be filled and/or the presence of artifacts. Indeed, the artefacts linked to the presence of insects, birds, dust or precipitation must then be cleaned, and the vegetation filtered/classified. The Interactive Segmentation Tool, SOR filter, Noise filter, Density, Curvature and Roughness, and CANUPO, CSF and Virtual broom (for noise on a flat surface) plugins can be good starting points for cleaning up data.

The presence of overhangs in the datasets will also complicate the subsequent steps by requiring the determination of the orientation of the surfaces to then make a 3D mesh in order to fill holes by interpolation. It may be advantageous to segment the terrain to isolate areas with overhangs. The holes of the section without overhang can then be filled directly with the method of the section "Hole filling / Whole site". And the other section, with overhangs this time, can see its holes filled Locally according to the other method “Hole filling / Locally”, with an estimate of the orientation of the surfaces beforehand. In this case, care should be taken not to introduce artefacts at the boundaries of the segments.

Photogrammetry (SfM)¶

Generating the terrain from elevation data from photogrammetry is much like doing it from TLS data. The data however usually comes with the colors from the photographs used to make the 3D model, as well as with an estimate of the surface orientation. If free of vegetation, artifacts and holes, the data can be used almost directly as terrain for simulations. Downsampling the points can speed up simulations if too many are present and texture will need to be removed from the points as described in the “Last requirements” section.

If the points have artifacts and/or gaps to fill, the processing can quickly become complex, but remains similar to that required for TLS data. Therefore refer to the description for TLS data to prepare the terrain from SfM data. An introduction to generating SfM models with the commercial software Agisoft Metashape Professional is given in the following video tutorial:

Surface orientation ¶

Once the point cloud corresponds to the imported terrain, it may be necessary to calculate the orientation of the surfaces if the point cloud has overhangs. However, this is usually not the case if it comes from airborne LiDAR data or from a gridded digital terrain model ( raster ). Then, it is not necessary to evaluate the surface orientation at this step, since it will be done in the next step by filling in the holes for the whole site.

For a dataset with overhangs, it is necessary to evaluate the orientation of the surfaces if this information is missing from the dataset, or of poor quality (data produced by photogrammetry generally already includes the surface orientation).

This can be done by using the Normals/Compute tool and then making sure the normals are oriented correctly (that they do not show up as black in CloudCompare).

Hole filling ¶

For simulations, the terrain should be free of holes and the points relatively evenly spaced. The point spacing should correspond to approximately 20-25% of the maximum diameter (d1) of the smallest rocks to be simulated.

As this can require a large number of points, it can be useful at this stage to generate several terrains with point spacings adapted to the different scenarios (simulations with large rocks will be much faster if the terrain is suitable for them rather than to use the same terrain as for the small rocks).

Whole site ¶

To adjust the point density and for filling the gaps, a mesh can first be created and then converted back to points at the desired density. If the terrain data does not have an overhang, the 2.5D triangulation approach is the fastest to use, and does not require surface orientation first, as this is determined during mesh creation. (The Poisson Surface Reconstruction plugin can be used instead of the Mesh/Delaunay 2.5D tool if the terrain has overhangs in order to generate a triangulated mesh in 3D.)

This step can place great strain on computer resources, especially RAM, due to the potentially large number of points to be generated from the mesh, before these are downsampled to the desired spacing with the Edit/Subsample tool. It takes about 3 to 4 GB of RAM occupied per 100 million points.

The number of points to be generated by the Mesh/Sample points tool can be estimated beforehand according to the surface of the terrain in order to assess whether the RAM of the computer will be sufficient. The terrain surface can be evaluated with the Mesh/Measure surface tool. It will be necessary to segment the mesh if the memory is not sufficient, to treat each segment individually, to sub-sample them with the Edit/Subsample tool and to merge their generated terrains.

Mesh to points and subsampling settings for CloudCompare¶

Locally ¶

For filling the holes locally, it is necessary to cut (with the Interactive Segmentation Tool ) a portion of the ground slightly larger than the portion to be filled. Then, the segmented portion to be filled can be treated in the same way as for the whole site (see the section “Whole site”).

To add the points thus generated to the rest of the terrain, they must first be segmented to remove the excess due to the previous cut from the slightly larger portion of the terrain and remove the interpolation artifacts near the perimeter of the area. But before selecting the segment In (I) option and excluding the points outside the zone with the Confirm and delete hidden point (Del) option, the selected zone must be saved (just after having made the right click allowing to disconnect from the green polygonal segmentation line).

The original points of the segmented terrain must also be cut out using the same cutout previously saved using the Import polyline from DB for segmentation option. This time, the points must be kept outside the perimeter with segment out (O) then Confirm and delete hidden point (Del).

At this point, three point segments should have been created from the start: 1) the original part of the terrain outside the first cut; 2) the local part of the terrain interpolated in order to fill the hole, without the excess of points previously selected around its periphery; and 3) the portion of the original terrain covering that excess strip. These three parts should complement each other without being superimposed since the same polygonal cutout lines have been applied to them, either by keeping the inside or outside part of the cutout perimeter.

The two parts of the original terrain can be merged and downsampled to the desired point spacing to speed up simulations (see graph in “Whole site” section). The interpolated segment can finally be merged with the other two already merged segments to reform a uniform terrain, the hole of which has been Locally filled by interpolation.

Last requirements ¶

Finally, we must ensure that we only have the position of the points for the terrain and their orientations: [X, Y, Z, Nx, Ny, Nz].

Any colors [R, G, B] and scalar fields must be removed before exporting the terrain points in ASCII format.

The ASCII file must be named "mnt.txt" or "dtm.txt" in order to be correctly taken into account for the simulations.