Your DTM Looks Smooth. That's Exactly the Problem: Why Breaklines Matter in Engineering Surveys

Your DTM Looks Smooth. That's Exactly the Problem: Why Breaklines Matter in Engineering Surveys

The terrain model looks perfect. Its surface has no jagged edges or depressions, but rather smooth, flowing lines that have been generated by a computer algorithm.

The civil engineer takes a cross-section of the road and wants to know what happened to the ditch.

It is on-site. It is in the orthomosaic. However, it is not in the terrain model (DTM) because the purpose of building this surface was to make it appear smooth, rather than to represent areas where the elevation changes abruptly. These areas are represented by breaklines, and these are the things that separate an aesthetically pleasing DTM from one an engineer can use as a design base.

We have previously discussed what makes a drone-collected data set suitable for use in engineering applications, as well as why a DSM is not necessarily a DTM. Now, we will go a step further: even a properly filtered DTM can remove the features that are most important.

What a Breakline Is

A breakline is a line along which the slope of the land changes abruptly, and which the terrain model must respect as opposed to averaging over the area. Some of the most common types of breaklines include: road edges; crown of roads; bottom of ditches; top of ditches; embankment crest; embankment toe; retaining wall top; retaining wall toe; curb; stream bank; stream center line; and toe of stockpile.

Esri's documentation distinguishes hard breaklines from soft ones. Hard breaklines are areas where there has been some type of disturbance to the continuity of the terrain. Examples include road cuts and streams. Soft breaklines add a line to the surface (such as a boundary) but do not indicate a change in the slope of the terrain. Hard breaklines are what we use to define ditches, road edges, and ridge lines.

Why Interpolation Smooths Over Sharp Terrain Changes

Terrain models are created by making connections between measured points to form triangles (TIN) or by transforming the triangles into a grid. Both of these methods create a surface between two points that is assumed, not measured.

Jerry Mahun's open access mapping text illustrates the issue with a ditch. When points are located on both sides of a ditch, and a triangle side connects those two points, the model treats the ground as uniformly sloped between them. This means the ditch is eliminated from the model. The exact same situation occurs at every point where a triangle side crosses the ditch. Softree's recommendations for improving contouring using breaklines address the same issue; you should not allow triangles to cross over a feature such as a creek bank or the edge of a road because doing so results in a reduction of the slope across the break.

The grid also adds an additional level of smoothing. If there is a narrow ditch that has a width less than the cell size, then this cannot be shown as a separate item. Also, if you have a smooth surface on both sides of a gap, it will still be interpolated; this is the same problem as we have discussed about why photogrammetry does not work well over water: interpolating to fill a hole is not the same as measuring what is inside.

What Goes Wrong Downstream

Contours

Contours are developed directly from the surface and therefore contain any errors contained within that surface. In a tutorial regarding road corridors, Softree demonstrates what occurs when breaklines are turned off; namely that contours flow into the road since the triangles are free to extend across the ditches and road edges without the breakline.

Drainage paths

Water will follow whatever surface you provide to the model. Without preserving the ditch, water can take many routes; flow paths wander, pond, or even jump. Roadway embankments are the best example of this. When we have a topographic DEM, we see that the roadway that passes over the culvert is part of the landscape; therefore, the model sees the embankment as a dam. Breaklines can correct this issue. A Milwaukee-area lidar project describes taking 3D culvert lines and draping them over the point cloud. The resulting breaklines are then used to cut the DEM, so that water passes through the culverts rather than being blocked by the road bank. The Lidar Magazine article regarding validation of breaklines states that proper breaklines and hydro-enforcement are often crucial to accurately model flooding scenarios.

Cross-sections

A cross-section is a slice through the surface. If the surface bridges the ditch, the section shows a straight line where the ditch should be, and the crown and shoulders of the road get rounded off. The section stops matching what the engineer sees on site, and any design built on it inherits the mismatch.

Earthwork

The amount of volume is determined by the shape of the surface. For example, if there is a rounded crest or a fill in a ditch, then the volume will change. To give you an example, a V-shaped ditch that is 1 m wide and 0.5 m deep will have a cross-sectional area of approximately 0.25 sq m. If the surface is able to "bridge" over the top of the ditch, then the model doesn't account for approximately 25 cubic meters of material per 100 meters of ditch. This may seem like very little when considering just one ditch; however, a corridor survey can contain numerous features similar to this for kilometers.

Dense Point Clouds Do Not Guarantee Reliable Surfaces

A million points don't mean quality. A point cloud is just some sample points; a breakline tells you where the features run. These are two completely different things.

As stated in Lidar Magazine, lidar does not inherently produce breaklines, unlike traditional photogrammetric mapping, where measured breaklines were included in the DTM at the time of compilation. Similarly, automated dense image matching will provide you with a point cloud, but will not give you a direct representation of the location of the ditch bottom.

More points don't necessarily equate to a more accurate surface. Agüera-Vega et al., in their study on micro-topographic scale terrain models created from UAVs, showed that reducing the density of the point cloud to 40 percent of its initial value only increased the overall error by approximately 5 percent. It wasn't until the number of points in the point cloud fell below roughly 15 percent of the initial number that the overall error began to rapidly increase. While point density is an important factor in limiting results, interpolation method and grid size are also significant factors.

Noise can complicate things. On a thread on Pix4D's community forum, a member was experiencing problems with a dense point cloud on a flat road producing erratic cross sections. In response, another member said they create breaklines manually and then export them into CAD for use, therefore eliminating any possibility of using those "noisy" points. This solution is practical - select the correct points and explicitly define the line.

As I mentioned previously in "5cm GSD does not mean 5cm accuracy," a description of the amount of data (i.e., resolution, density) being collected is not necessarily a reflection of how well that data accurately reflects the actual condition of the area.

When Breaklines Should Be Captured and Included in the DTM

Drone surveys do not always require breaklines. For example, a single photo that shows how much work has been done, or a fast look at what's going on, does not need breaklines. Breaklines are important when the model you create is going to help with things like design, water control (drainage), land leveling (grading), and/or quantity calculations:

  • roadway and corridor surveys, which include mountain roadways with cuts into the sides of mountains and retaining walls to hold back soil
  • drainage and flood studies
  • grading plans and earthwork volume, including toe of stockpiles
  • site construction with curbs, edge of pavements and retaining walls
  • as-built checks against a design surface

The method you use to collect data will depend on what type of feature you are trying to capture. For example, a drone survey would allow you to digitize 3D lines from the point cloud and image data. However, if the feature is not something that a drone can resolve (e.g., ditches under trees/vegetation/water, vertical faces, tight tolerances), then an RTK GNSS receiver or Total Station is a much better option. This is the hybrid approach to collecting data in practice, and it is especially important when working on mountain roads. Mountain roads have cut slopes and retaining walls, which are precisely the types of features that a straight-down camera does not see well.

The decision on which features are considered breaklines and who will capture those features should occur prior to going into the field. That is part of the conversation we described in the pre-survey checklist: confirm what the deliverable is actually for. Additionally, when you provide your DTM (Digital Terrain Model), you should also include your breakline data. The USGS (United States Geological Survey) requires that all breakline data used within its lidar projects be delivered with the elevation models. This allows the recreation of those models from the point data and line data, without substantial editing. I believe this would be a wise practice to adopt.

How to Evaluate Your Terrain Model Prior to Submitting It to a Civil Engineer

  1. Overlay the contour lines over the orthomosaic. Follow each of the following: roads, ditches, embankments. The contour lines must follow the flow of the terrain and bend with these features; do not allow them to cross over any of them.
  2. Check the triangles (or hillshade) created on the features. The edges of any of these triangles cannot cross over a ditch, road edge, or crest.
  3. Cut a test section across an existing feature such as a ditch, curb, or road crown. Compare the depth of the ditch, the height of the curb, and the crown of the road with measurements taken from the field.
  4. Check the consistency of the breakline elevations. Breaklines need to flow down to allow for drainage instead of producing a wavy pattern. If breaklines intersect at varying heights, you will get an inconsistent-triangle error; according to Softree's documentation, you do not want to ignore this issue.
  5. Run a quick check on the flow path. Is there any evidence of water following the ditch lines? Are there areas where you can see water ponding behind each of the road embankments? When you observe ponding behind a road embankment, it usually indicates that a culvert is not represented in the model.
  6. The features should have independent checkpoints added. Checkpoints can be placed at the ditch bottom, edge, crest, as well as open flat areas. This is where those checkpoints that were excluded from processing will come in handy.
  7. Write down what you have captured and what you have not captured. For each breakline that has been captured, describe how it was captured (e.g., photogrammetric, RTK, total station), and for each area where you could not capture anything due to vegetation or water, describe why you could not capture anything. This is the exact same documentation process used during river survey work, i.e., describe what was measured, what was interpolated, and what was not measured.

The Actual Takeaway

The interpolation has been successful if the DTM is smooth; however, this does not indicate whether or not the actual shape of the ground was maintained during the process. Since all of the following- contours, drainage paths, cross-sections, and earthwork quantities are generated from the same surface, any error made in generating the surface will travel into each of these other items as well.

Before a terrain model is sent to the engineer, the question to be asked should not be "Does this model look clean?" The question to be asked should be "What lines were enforced, how were these lines captured, and how were these lines checked?" Our topographic survey and drone imaging and processing teams can help you create a survey that answers these questions before the very first image or point is captured.

Back to blog

Leave a comment