Skip to main content
This guide shows how to create a Digital Terrain Model (DTM) from point cloud data using PDAL Python. A DTM represents the bare-earth surface by filtering out vegetation, buildings, and other above-ground features.

​
Overview

The DTM creation workflow involves:
  1. Reading the point cloud file
  2. Removing noise
  3. Cleaning up invalid values
  4. Classifying ground points using SMRF
  5. Writing the result as a raster with the GDAL writer
If your point cloud already has ground points classified, you can skip directly to the reader and writer stages.

​
Complete DTM creation example

import pdal

pc_path = 'https://github.com/PDAL/data/raw/refs/heads/main/autzen/autzen.laz'
out_file = 'autzen_dtm.tif'


# read
reader = pdal.Reader.las(pc_path)

# remove noisy points
lownoise_filter = pdal.Filter.range(
    limits='Classification![7:7]', tag='lownoise'
)
highnoise_filter = pdal.Filter.range(
    limits='Classification![18:]', tag='highnoise'
)

# saving incorrectly labeled returns here, some people want this, some don't
prepare_ground = pdal.Filter.assign(
    value=[
        'Classification=0',
        'ReturnNumber=1 WHERE ReturnNumber < 1',
        'NumberOfReturns=1 WHERE NumberOfReturns < 1',
    ],
    tag='prepare_ground_classifier',
)

# classify ground
smrf_classifier = pdal.Filter.smrf(tag='ground_classifier')

# write with gdal, resolution in feet for autzen
gdal_writer = pdal.Writer.gdal(
    filename=out_file,
    where='Classification == 2',
    data_type='float32',
    resolution=10,
    output_type='idw',
    window_size=3,
    pdal_metadata=True,
)

# collect pdal stages and execute pipline
pipeline = (
    reader
    | lownoise_filter
    | highnoise_filter
    | prepare_ground
    | smrf_classifier
    | gdal_writer
)
pipeline.execute()

​
Workflow breakdown

1

Read the point cloud

Start by reading your point cloud file:
reader = pdal.Reader.las(pc_path)
This works with various formats including LAS, LAZ, and others supported by PDAL.
2

Remove noise

Filter out points classified as low noise (7) and high noise (18+):
lownoise_filter = pdal.Filter.range(
    limits='Classification![7:7]', tag='lownoise'
)
highnoise_filter = pdal.Filter.range(
    limits='Classification![18:]', tag='highnoise'
)
The ! operator excludes points matching the classification range.
3

Prepare for ground classification

Clean up invalid return values:
prepare_ground = pdal.Filter.assign(
    value=[
        'Classification=0',
        'ReturnNumber=1 WHERE ReturnNumber < 1',
        'NumberOfReturns=1 WHERE NumberOfReturns < 1',
    ],
    tag='prepare_ground_classifier',
)
This ensures all points have valid return numbers before ground classification.
4

Classify ground points with SMRF

Use the Simple Morphological Filter (SMRF) to identify ground points:
smrf_classifier = pdal.Filter.smrf(tag='ground_classifier')
SMRF assigns classification value 2 to points identified as ground.
5

Write the raster output

Use the GDAL writer to create a GeoTIFF raster from ground points:
gdal_writer = pdal.Writer.gdal(
    filename=out_file,
    where='Classification == 2',
    data_type='float32',
    resolution=10,
    output_type='idw',
    window_size=3,
    pdal_metadata=True,
)
Key parameters:
  • where='Classification == 2' - Only use ground points
  • resolution=10 - Output resolution in data units (feet for this example)
  • output_type='idw' - Use Inverse Distance Weighting for interpolation
  • window_size=3 - Window size for IDW interpolation
6

Execute the pipeline

Chain all stages together and execute:
pipeline = (
    reader
    | lownoise_filter
    | highnoise_filter
    | prepare_ground
    | smrf_classifier
    | gdal_writer
)
pipeline.execute()

​
Ground classification with SMRF

The Simple Morphological Filter (SMRF) is a progressive morphological filter that identifies ground points based on the assumption that the ground surface is continuous and relatively smooth. SMRF works by:
  1. Creating an initial estimate of the ground surface
  2. Progressively refining this estimate using morphological operations
  3. Classifying points below the threshold as ground (classification 2)
For more details on SMRF parameters and tuning, see the PDAL SMRF filter documentation.

​
GDAL writer for raster output

The GDAL writer converts point cloud data into raster format. Key options include:
  • output_type - Interpolation method: min, max, mean, idw, count, stdev
  • resolution - Output resolution in the same units as your point cloud coordinates
  • data_type - Output data type: float32, float64, int32, etc.
  • window_size - For IDW, the size of the search window
For complete GDAL writer options, see the PDAL GDAL writer documentation.
Make sure the resolution value matches your data’s coordinate system units. The example uses feet, but you may need to adjust for meters or other units.

Build docs developers (and LLMs) love

Get started for freeTalk to us