How to thin the ground layer using Flai web app

In LiDAR (Light Detection and Ranging), the laser beam interacts with surfaces in three main ways: absorption, transmission, and reflection. Most LiDAR systems rely on reflected light, as they measure the time it takes for the beam to bounce back to the sensor.

If a surface absorbs the laser light (e.g., dark materials or water), the reflection is weaker or may not occur at all. Transmission happens when the beam passes through materials, such as glass or water. In the case of reflection, we distinguish between specular reflection (on smooth surfaces like metal or water) and diffuse reflection (on rough surfaces like leaves, soil, or concrete). Diffuse reflection is crucial for LiDAR, as it enables accurate detection of terrain topography and vegetation.

Try Flai

Point clouds generated using LiDAR technology allow for highly accurate surface mapping. A thicker ground layer in point clouds typically results from several factors, including:

• Surface reflectivity: Different ground surfaces and objects have varying reflective properties, affecting the number of reflections and, consequently, the density of the point cloud.

  
  

• Measurement quality and density: A higher number of measurements or greater point density provides a more detailed surface representation. Using high resolution results in a denser point cloud.

  
  

• Topography: Uneven or complex surfaces will have more points, as each small irregularity is detected as a separate point. Flat surfaces will have fewer points.

  
  

• Angle and direction of the laser beam: When the laser beam strikes a surface at different angles, it can cause scattered and multiple reflections, increasing the number of captured points.

  
  

• Filtering and data processing: During post-processing, different filtering techniques can be applied, which influence the final density of the point cloud.

Figure 1: Different point dispersion depending on the surface. Where the ground is smoother, such as on asphalt surfaces, the dispersion is minimal and no additional filtering is necessary. In the case of uneven terrain, where grass is often present, the dispersion is greater and filtering is needed to obtain a very thin layer of ground.

Once you've successfully classified your point cloud data—especially identifying ground points—you might notice something: the ground layer can appear thicker than expected. This is common and largely due to how LiDAR interacts with real-world surfaces. Variations in reflectivity, measurement density, terrain complexity, and even the angle of the laser all contribute to a denser ground layer.

To improve the clarity and usability of your data, especially for applications like digital terrain models (DTMs), you might want to thin that ground layer. Flai has therefore developed two different approach that works for thinning ground. When we talk about filters, it is very important to remember that their result also depends to a large extent on the input data itself and, in the case of ground, also on the slope and roughness of the surface.

If you want to make the ground yourself a bit thinner you can do this directly in the Flai web app using different processes.

Different approaches to thin the ground thickness

1. Morphological filter

   
   

Morphological filtering for thinning ground in point clouds involves applying mathematical morphology operations—such as erosion and dilation—to reduce ground point density while preserving the essential terrain structure. This process helps remove redundant or closely clustered points, making the dataset more manageable without losing key elevation features. It's especially useful in preprocessing steps for digital terrain model (DTM) generation or ground segmentation, where maintaining the shape of the surface is more important than retaining every individual point.

How to set up your flow correctly

The setup is simple, and it gives you full control over the thinning process. Here's how you can build your own flow in the Flai web app:

1. Start with a point cloud reader – Load your dataset.

2. Add a filter node – Use the Refine ground using simple morphological filter node.

3. Finish with a point cloud writer – This saves your thinned result.

That’s your basic three-step flow.

💡Make sure your dataset is already classified and that ground points are correctly labeled. If you haven’t done this yet, check out our blog post: Introducing Flai Web App: Automatic point cloud classification platform powered by AI.

Figure 2: How to prepare the flow for the use of a ground thinning filter - using simple morphological filter

Choosing the Right Parameters

Within the morphological filter node, you’ll find several parameters that can be adjusted depending on your data’s characteristics—like slope, surface roughness, and point density. Based on extensive testing, we’ve found a sweet spot of settings that work well in most cases (see Figure 3 and Figure 4). These control how aggressively the filter removes points while preserving elevation features.

Figure 3: Example of parameters used for datasets with units in metres

Figure 4: Example of parameters used for datasets with units in feet

Results

Here’s what the morphological filter looks like in action. In the image below, the yellow points are the ground points that were kept, while the bright green points were removed during the thinning process.

This helps make the ground layer thinner and cleaner by getting rid of extra points that aren’t needed. The result is a more accurate and easier-to-use surface for things like digital terrain models.

Figure 3: Using the Morphological Filter. Points that have been excluded from the Ground class during the filtering process are marked in bright gold.

2. Ground thinning using Flainet

   
   

If you're looking for a more automated approach to ground thinning, the FlaiNet ground thinning model is a great option. It takes your existing ground predictions and refines them—removing outliers and duplicate layers that can occur due to alignment mismatches. In such cases, FlaiNet keeps only the lowest valid points, ensuring a clean, single-layer ground surface.

Setting up your flow is straightforward:

1. Point cloud reader – Load your dataset.

2. FlaiNet ground thinning node – Select the Ground thinning using FlaiNet node.

3. Point cloud writer – Save the thinned output.

   
   

Figure 5: How to prepare the flow for the use of a ground thinning - using FlaiNet

That's it—no need to fiddle with complex parameters. All you need to do is specify the classification numbers: the class you want to thin (e.g., ground = 2), the new label for thinned ground (often the same = 2), and a label for non-ground points (3 in the example shown). FlaiNet takes care of the rest.

Figure 6: Example of configuration panel for FlaiNet ground thinning

💡 Tip: This method works best on relatively smooth or gently sloping terrain. On highly stepped or rugged surfaces, results may vary, as sharp elevation changes can make it harder to distinguish outliers from valid ground points.

Results

The image below shows a before-and-after comparison:

• On the left, the original ground points include more points in class Ground

• On the right, after applying FlaiNet, you can see a much cleaner, thinner ground surface with only the lowest valid points kept.

This automated approach is great for quickly cleaning up your dataset—especially when dealing with alignment issues or overlapping scans.

Figure 4: The left side shows the ground points before the Morphological Filter was used. On the right is the ground after the filter has been used.

3. Combination of both approach

   
   

In some cases, especially when working with challenging terrain or noisy datasets, it may happen that neither the morphological filter nor the FlaiNet approach alone produces fully satisfactory results. That’s perfectly normal—no single method fits all scenarios, particularly when dealing with varied topography, dense vegetation, or alignment issues between scans. In such situations, combining both approaches can be a practical and effective solution.

For example, you might first apply the FlaiNet ground thinning model to clean up obvious outliers and remove duplicate ground layers that often result from minor misalignments in overlapping flight lines. FlaiNet excels at identifying the lowest, most reliable ground points and separating them from noisy layers above. Once this initial cleanup is done, you can follow up with a morphological filter to fine-tune the surface, smooth out remaining irregularities, and further reduce residual ground thickness.

Figure 8: How to prepare the flow for the use of a ground thinning - using FlaiNet + Morphological filter

Try Flai

Ground thinning is a crucial step in preparing high-quality LiDAR data, especially when generating digital terrain models or conducting precise topographic analysis. While the appearance of a thick ground layer in point clouds is a common byproduct of how LiDAR interacts with real-world surfaces, it can be managed effectively using the right tools and methods.

Flai offers two flexible approaches to tackle this: the simple morphological filter, which gives you manual control over thinning parameters, and the FlaiNet model, which provides a more automated, intelligent refinement of ground points. Depending on your dataset and project needs, you can choose one or combine both methods for optimal results.

No matter which path you take, the key is to understand your data—its density, terrain characteristics, and classification quality—and choose the thinning strategy that brings out the most accurate and usable representation of the ground. With the tools available in the Flai web app, you're well equipped to do just that.