4.4 Elevation and Surface Representations: DEMs, DSMs, DTMs, TINs, and Point Clouds
Key Takeaways
- Digital Elevation Models (DEMs) represent bare-earth topography, Digital Surface Models (DSMs) capture the topmost surface including vegetation and buildings, and Digital Terrain Models (DTMs) supplement bare earth with geomorphic breaklines.
- Subtracting a bare-earth DEM from a first-return DSM generates a Normalized Digital Surface Model (nDSM), isolating the true heights of above-ground structures and vegetation.
- Triangulated Irregular Networks (TINs) construct continuous vector surfaces using the Delaunay triangulation empty circumcircle criterion, maximizing the minimum interior angle of triangles to eliminate slivers.
- Hard breaklines define abrupt slope discontinuities (cliff edges, retaining walls) that TIN triangle edges cannot cross, whereas soft breaklines preserve linear features without altering surface slope.
- LiDAR active sensors emit laser pulses and record multiple discrete returns: first returns map canopies and building roofs (DSM), intermediate returns capture foliage, and last returns penetrate to the ground (DEM).
Elevation and Surface Representations: DEMs, DSMs, DTMs, TINs, and Point Clouds
Quick Summary: Digital representation of three-dimensional topography is a cornerstone of geospatial engineering. Geographic elevation surfaces are captured and modeled through five primary structures: Digital Elevation Models (DEM), Digital Surface Models (DSM), Digital Terrain Models (DTM), Triangulated Irregular Networks (TIN), and LiDAR Point Clouds. Mastering the functional differences between bare-earth and first-return surfaces, the mathematics of Delaunay triangulation in TINs, and LiDAR return mechanics is vital for GISP exam success.
The Elevation Model Hierarchy: DEM vs. DSM vs. DTM
In professional practice, terrain terms are frequently conflated. However, USGS and international mapping standards establish precise, non-interchangeable definitions for each elevation product.
THE ELEVATION MODEL PROFILE COMPARISON
Forest Canopy Building Roof
(Tree) [=====]
/ * * \ | |
/ * * * \ | |
/ * * * * \ | |
---------+---------+------------+-----+---------- <-- DSM (First Return Surface)
| | | |
| Trunk | | |
~~~~~~~~~+~~~~~~~~~+~~~~~~~~~~~~+~~~~~+~~~~~~~~~~ <-- DEM / DTM (Bare-Earth Surface)
^ ^
Canopy Filtered Out Building Filtered Out
* nDSM (Normalized DSM) = DSM - DEM ===> Yields Height of Tree & Building!
1. Digital Elevation Model (DEM)
In strict technical usage, a Digital Elevation Model (DEM) is a regular raster grid representing the bare-earth bare terrain of the Earth. All natural and anthropogenic surface features—including tree canopies, dense vegetation, buildings, industrial infrastructure, bridges, and highway overpasses—have been digitally filtered out and removed. Each cell value records the elevation of the bare ground surface above a specified vertical datum (typically NAVD88).
2. Digital Surface Model (DSM)
A Digital Surface Model (DSM) captures the elevation of the topmost reflective surface encountered by the remote sensor. A DSM captures:
- The upper canopy of forests and tree crowns
- Building roofs, industrial complexes, and residential structures
- Electrical transmission towers and power lines
- Bare terrain in areas devoid of vegetation or structures
3. Digital Terrain Model (DTM)
A Digital Terrain Model (DTM) is an enhanced bare-earth model. While a DEM typically consists of a uniform raster grid of elevations, a DTM incorporates vector breaklines and mass points that capture linear discontinuities and geomorphic morphology:
- Breakline Integration: DTMs embed explicit linear features defining ridge tops, valley lines (thalwegs), stream centerlines, road edges, retaining walls, and shoreline boundaries.
- Morphologic Fidelity: By forcing the surface interpolation to conform to linear breaklines, a DTM prevents the smoothing or flattening of sharp topographic edges, producing a superior representation of drainage networks and steep terrain.
4. Normalized Digital Surface Model (nDSM / Canopy Height Model)
The Normalized Digital Surface Model (nDSM)—often termed a Canopy Height Model (CHM) in forestry—is generated through a simple raster map algebra subtraction:
Because the bare-earth elevation is subtracted from the top surface elevation, the resulting nDSM represents the true absolute height above ground level (AGL) of all surface objects. In an nDSM, ground surfaces have an elevation of zero, buildings appear at their true physical structural heights, and trees reflect their true biological canopy heights. The nDSM is the standard tool for urban 3D building extraction and forest biomass estimation.
Triangulated Irregular Networks (TIN)
A Triangulated Irregular Network (TIN) is a vector-based topological data structure that represents a continuous three-dimensional surface as a contiguous mesh of non-overlapping triangular facets. Unlike uniform raster grids, TINs adapt dynamically to the complexity of the terrain.
DELAUNAY TRIANGULATION & CIRCUMCIRCLE CRITERION
Valid Delaunay Triangle Invalid Non-Delaunay Triangle
(Empty Circumcircle) (Violates Circumcircle Criterion)
Node B Node B
/ \ / \
/ \ / \
/ * \ / * \ <-- Node D falls
/ Center \ / Center \ INSIDE the
Node A ------------ Node C Node A ------------ Node C circumcircle!
( Node D )
Circumcircle contains NO other sample points * Edge must be flipped!
Mathematical Construction: Delaunay Triangulation
TIN surfaces are constructed using Delaunay Triangulation, governed by the Empty Circumcircle Criterion:
- Circumcircle Rule: For any triangle in the network, a circle drawn through its three vertices $(A, B, C)$ must contain no other sample points from the dataset within its interior.
- Angle Maximization: Delaunay triangulation mathematically maximizes the minimum interior angle of all triangles in the mesh. This avoids long, thin, needle-like sliver triangles, ensuring that facets remain as equiangular and compact as possible.
- Edge Swapping: If a fourth sample point falls inside the circumcircle of an existing triangle, the shared diagonal edge between adjacent triangles is "flipped" or swapped to restore Delaunay validity.
Breaklines in TIN Modeling: Hard vs. Soft
TIN models integrate linear breaklines to control surface interpolation across physical features:
| Breakline Class | Structural Function | Behavior in Triangulation | Real-World Examples |
|---|---|---|---|
| Hard Breakline | Defines a sharp discontinuity in terrain slope and aspect (slope break). | Triangle edges are forced to follow the line. Triangles cannot cross the line; surface interpolation breaks across the boundary. | Cliff faces, road curbs, retaining walls, ridge lines, fault lines, building perimeters. |
| Soft Breakline | Preserves linear cartographic features without establishing a sharp slope break. | Triangle edges are forced to conform to the line, but slope interpolation continues smoothly across the edge. | Study area boundaries, administrative map sheet edges, contour line traces. |
| Replace (Void) Surface | Flattens an enclosed polygon area to a constant horizontal elevation. | Eliminates surface slope inside the polygon boundary; triangles inside are assigned a constant $Z$-value. | Lakes, reservoirs, large calm water bodies, building excavation foundations. |
Advantages and Disadvantages of TINs
- Variable Resolution Efficiency: In flat plains, a TIN requires only a few widely spaced nodes to represent the terrain. In rugged, mountainous terrain, the node density increases dramatically. A TIN captures complex terrain with a fraction of the data storage required by an equivalent uniform raster grid.
- Disadvantages: High computational processing overhead to construct and edit; complex topological data structure; inefficient for continuous map algebra operations compared to raster DEMs.
LiDAR Point Clouds: Multi-Return Active Remote Sensing
Light Detection and Ranging (LiDAR) is an active remote sensing technology that measures distances by emitting rapid pulses of laser light toward the Earth's surface and timing how long it takes for reflections to return to the sensor photodiode.
LIDAR MULTI-RETURN PULSE DYNAMICS
[ Aircraft / Sensor Platform ]
|
| Laser Pulse Emitted
v
.-'''''-.
.' * * * '. <--- First Return: Top of Forest Canopy (DSM)
/ * * * * * \
| * * * * * |
\ * * * / <--- Intermediate Return: Mid-Story Foliage
'. Trunk .'
'-.....-'
============= <--- First Return on Hard Surface: Building Roof (DSM)
| Building |
| |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- Last Return: Bare Earth / Ground Surface (DEM)
The Mechanics of Laser Pulse Returns
As an emitted laser pulse travels toward the ground, the beam spreads into a conical footprint. When the pulse encounters a semi-permeable surface (such as a forest canopy), parts of the light energy reflect off upper leaves while the remainder continues downward:
- First Return: The earliest energy reflected back to the sensor. Originates from the highest surface feature encountered—the top of the tree canopy, power lines, or building roofs. First returns are used to generate the DSM.
- Intermediate Returns: Reflections generated as the laser pulse filters through tree branches, secondary understory, and foliage layers.
- Last Return: The final pulse reflection. In vegetated terrain, a portion of the laser pulse penetrates through gaps in the foliage to reflect off the actual ground. Last returns (when classified as ground) are used to interpolate the bare-earth DEM.
- Single Returns: When a pulse strikes an impenetrable, solid flat surface (such as an asphalt parking lot or open bare soil), all energy reflects at once, producing a single return that is simultaneously the first and last return.
ASPRS Standard LAS Classification Codes
The American Society for Photogrammetry and Remote Sensing (ASPRS) maintains the standard LAS file format for storing 3D point cloud records. Points are categorized into standardized numerical classification codes:
| ASPRS Class Code | Official Classification Name | Physical Surface Feature Represented |
|---|---|---|
| Class 1 | Unassigned / Unclassified | Raw, unclassified points or points not meeting specific class criteria. |
| Class 2 | Ground | Bare-earth terrain surface (used directly to interpolate DEMs and DTMs). |
| Class 3 | Low Vegetation | Ground cover, turf, and low shrubs ($\le 0.5\text{ meters}$). |
| Class 4 | Medium Vegetation | Shrubbery, brush, and small trees ($0.5\text{ to }2.0\text{ meters}$). |
| Class 5 | High Vegetation | Forest canopy, mature tree crowns ($> 2.0\text{ meters}$). |
| Class 6 | Building | Anthropogenic structures, commercial rooftops, residential houses. |
| Class 7 | Low Point / Noise | Sub-surface sensor artifacts, multi-path water reflections below terrain. |
| Class 9 | Water | Canals, rivers, lakes, oceans (points exhibiting low reflectance). |
| Class 17 | Bridge Deck | Bridge structures spanning across water or roadways (removed from bare-earth DEM). |
Point Cloud Storage Formats: LAS vs. LAZ
- ASPRS LAS: An open, binary file format that stores 3D point records with millimeter coordinate precision, alongside return number, number of returns, scan angle, intensity, classification, and GPS time.
- LAZ (LASzip): A completely lossless compressed format developed by Martin Isenburg. LAZ achieves a $7:1\text{ to }10:1$ compression ratio without altering a single coordinate or attribute byte. Because LAZ is completely lossless, it has become the standard archiving and distribution format for public LiDAR repositories.
Comprehensive Comparison Matrix of Surface Models
| Surface Model | Data Model Architecture | Surface Represented | Structural Elements | Storage Efficiency | Breakline Support | Primary GIS Applications |
|---|---|---|---|---|---|---|
| DEM | Raster Grid | Bare-Earth Terrain | Regular rectangular grid of elevation cells. | Predictable; uniform cell storage. | No explicit breaklines (smoothed across cells). | Hydrologic modeling, regional slope/aspect analysis, flood inundation. |
| DSM | Raster Grid | Topmost Reflective Surface | Grid cells capturing tree crowns, roofs, towers. | Predictable; uniform cell storage. | No explicit breaklines. | Viewshed obstruction analysis, telecom line-of-sight, solar potential. |
| DTM | Raster Grid or Hybrid Vector | Bare-Earth Terrain + Morphology | Grid cells integrated with linear vector breaklines. | Moderate; requires companion vector files. | Full support for hard and soft breaklines. | Civil engineering, highway corridor design, cut-and-fill calculations. |
| TIN | Topological Vector | Continuous Piecewise Surface | Irregular nodes, directed edges, planar triangular facets. | High in varied terrain; low in flat terrain. | Full support via forced Delaunay triangle edges. | Local slope modeling, volumetric excavation modeling, CAD integration. |
| Point Cloud | Discrete 3D Vector Points | All reflective surfaces (Multi-layer) | Unordered collections of $(X, Y, Z)$ points with attributes. | High storage requirements (mitigated by LAZ). | Implicit in dense point samples. | Raw source data for DEM/DSM generation, 3D building modeling, forestry. |
Summary of Common Exam Traps
[!CAUTION] Exam Trap 4.10: Using a DSM for Hydrologic Modeling. Running watershed delineation, flow direction, or flood inundation modeling on a Digital Surface Model (DSM) is a catastrophic analytical error. A DSM treats forest canopies as solid elevated walls and buildings as mountains, causing modeled water to unnaturally "flow" around tree canopies and dam behind highway bridge decks. Hydrologic flow routing must always be performed on a bare-earth DEM.
[!CAUTION] Exam Trap 4.11: Believing TINs are Raster Data Structures. Because TINs model continuous surfaces, examinees often classify them as rasters. TINs are strictly topological vector models composed of coordinate nodes, vector edges, and triangular polygon faces.
[!CAUTION] Exam Trap 4.12: Assuming LAZ Compression is Lossy. Candidates frequently assume that because LAZ achieves massive 80–90% compression ratios, it must be lossy like JPEG. LAZ (LASzip) is strictly lossless compression; every 3D coordinate and classification byte is preserved with exact bit-for-bit mathematical fidelity upon decompression.
A municipal stormwater engineer is tasked with delineating drainage basins and calculating overland flow paths to evaluate neighborhood flood risk. An intern suggests using a newly acquired high-resolution Digital Surface Model (DSM) generated from drone photogrammetry. Why should the engineer reject the DSM in favor of a bare-earth Digital Elevation Model (DEM)?
When constructing a Triangulated Irregular Network (TIN) surface from a set of irregularly distributed mass elevation points, which mathematical criterion governs Delaunay triangulation to prevent long, thin sliver triangles?
An environmental forestry agency processes airborne LiDAR data to model canopy height and estimate standing timber biomass. The sensor records multiple returns per laser pulse. Which pulse return sequence and ASPRS classification workflow is used to derive the canopy height model (nDSM)?