For centuries, maps have been flat representations of the world, helpful yet limited. Today, technology allows us to go far beyond flat lines and shaded reliefs. With just a series of photographs captured by drones, satellites, or aircraft, we can build digital elevation models that describe the shape and form of the Earth in astonishing detail. From construction sites to river valleys, from sprawling cities to remote glaciers, digital surface models (DSMs) and digital terrain models (DTMs) provide the foundation for analysis, planning, and visualization. The process of turning ordinary images into these extraordinary 3D products is one of the most transformative developments in geospatial science, and in 2025, it is both more accessible and more powerful than ever.
Understanding DEMs, DSMs, and DTMs
Before diving into the process, it’s important to understand the terminology. A digital elevation model (DEM) is a general term that represents elevation data of the Earth’s surface. Within this category are DSMs and DTMs, each serving different purposes.
A DSM captures everything visible from above, including buildings, trees, and other structures. It is a model of the Earth’s surface as it appears to the eye. This makes DSMs incredibly useful in urban planning, forestry, and telecommunications, where above-ground features matter as much as the ground itself.
A DTM, on the other hand, represents the bare earth, stripped of vegetation and man-made structures. DTMs are essential in applications like flood modeling, hydrology, and geological studies, where the natural terrain is the focus. The ability to generate both DSMs and DTMs from the same imagery means photogrammetry and remote sensing can support diverse industries. The key lies in the process of moving from raw photographs to a structured elevation model.
Capturing the Right Imagery
The journey from photos to DEM begins with the data you collect. High-quality imagery is the bedrock of accurate models. Drones are often the tool of choice, offering flexibility, affordability, and high-resolution data capture. Aircraft and satellites expand this to larger areas, though often with trade-offs in resolution. For photogrammetry, overlap is critical. Forward overlap of around 75 to 80 percent and side overlap of 60 to 70 percent ensures the software can identify enough tie points across images. These tie points allow algorithms to triangulate positions and generate depth information. Consistent lighting conditions also matter. Overcast skies are often best for avoiding harsh shadows that confuse elevation readings.
Flight planning has become easier in 2025, with automated software capable of calculating the optimal path, altitude, and overlap for the desired resolution. Whether mapping a construction site or a mountain slope, careful capture ensures the downstream processing stages can produce reliable DSMs and DTMs.
Structure-from-Motion and Point Clouds
Once the imagery is collected, the next step is processing it through Structure-from-Motion (SfM). This algorithm identifies tie points across overlapping images and calculates both the position of the cameras and the spatial coordinates of the points. The result is a sparse point cloud—a scattered constellation of points representing the scene.
This point cloud is then densified, often using multi-view stereo techniques. Instead of a few thousand points, the software now calculates millions or billions, creating a dense cloud that captures every contour of the terrain and surface features. Each point has x, y, and z coordinates, essentially reconstructing the surface in three dimensions.
From here, the point cloud can be classified. Algorithms distinguish between ground and non-ground points, separating trees, buildings, and other features from the bare earth. This classification is what enables the creation of both DSMs and DTMs from the same dataset. In 2025, artificial intelligence plays a growing role in this stage, automatically filtering noise and improving classification accuracy.
Generating DSMs and DTMs
With a classified point cloud, the creation of elevation models begins. For a DSM, the highest point at each x and y coordinate is chosen. This means rooftops, treetops, and other elevated objects are preserved, giving a view of the surface as it would appear from above. For a DTM, the algorithm selects only the ground points, interpolating where necessary to fill gaps where vegetation or buildings blocked the surface. The result is a smooth model of the terrain itself, showing valleys, ridges, and slopes without the interference of above-ground structures.
Both DSMs and DTMs are typically represented as raster grids, where each cell has a value corresponding to elevation. The resolution of these grids depends on the quality of the input imagery and the density of the point cloud. High-resolution drone imagery can generate models with centimeter-level detail, while satellite data often produces coarser outputs suited for regional studies. Modern software integrates this process seamlessly, offering users side-by-side comparisons of DSMs and DTMs, along with the ability to generate derivative products such as slope maps, contour lines, or hydrological models.
Applications Driving Innovation
The creation of DSMs and DTMs is not just a technical exercise; it underpins countless real-world applications. In urban planning, DSMs provide insights into how buildings interact with sunlight, wind, and viewsheds. Telecom companies rely on them to optimize line-of-sight for antennas and cell towers. Forestry managers use DSMs to measure canopy height and assess biomass.
DTMs are equally powerful. Hydrologists model how water will flow across landscapes, predicting flood zones and informing infrastructure planning. Geologists study terrain features to understand fault lines, landslides, and erosion patterns. Archaeologists use DTMs to uncover hidden features beneath dense vegetation, revealing ancient structures invisible to the naked eye. In 2025, climate science is another major driver. By generating DTMs from imagery over time, scientists track changes in glaciers, coastlines, and river systems with unprecedented accuracy. The ability to create time-series models helps monitor environmental change and guide policy decisions. The versatility of DSMs and DTMs ensures their place as foundational tools in modern decision-making, bridging the gap between raw imagery and actionable insights.
Challenges in Accuracy and Processing
Despite its transformative potential, generating DSMs and DTMs from imagery comes with challenges. Accuracy depends heavily on image quality, overlap, and ground control. Without sufficient ground control points or RTK/PPK-enabled drones, vertical errors can creep into the models, sometimes by several centimeters or more.
Vegetation remains a persistent obstacle. Distinguishing ground from non-ground points in dense forests can be difficult, leading to errors in DTMs. Similarly, reflective or homogenous surfaces like water bodies can confuse algorithms, leaving gaps or distortions. Processing power is another factor. Dense point clouds and high-resolution models require significant computational resources. While cloud processing has made this more manageable, large projects can still take hours or days to complete. Data storage and management are ongoing concerns, as even a single project can generate gigabytes or terabytes of data. In 2025, advances in artificial intelligence and edge computing are helping mitigate these challenges. AI-driven classification improves ground separation, while distributed processing platforms make handling massive datasets more efficient. Still, users must balance ambition with practicality, tailoring their workflows to the resources available.
The Future of Elevation Modeling
Looking ahead, the process of generating DSMs and DTMs will only grow more refined. The integration of imagery with LiDAR is one of the most promising trends. While imagery provides rich visual detail, LiDAR delivers unparalleled penetration through vegetation. Combined, the two methods create models that are both accurate and visually realistic.
Real-time DEM generation is another frontier. Advances in onboard processing for drones and satellites are moving toward models being created in the field, allowing immediate feedback and decision-making. This could be invaluable in disaster response, where knowing terrain conditions within minutes can save lives. AI will continue to reshape the landscape of DEM creation. From automated flight planning to adaptive classification, artificial intelligence is making the process smarter, faster, and more reliable. Coupled with the expanding accessibility of drone technology, this democratization ensures that elevation modeling is no longer limited to specialists. Students, small businesses, and local governments can all harness the power of DSMs and DTMs.
By 2030, the boundary between DSMs and DTMs may blur further, with dynamic models that can toggle between surface and terrain in real time, adjusting to the needs of different users. The future is not just about static maps but about interactive, adaptive systems that bring elevation data to life.
Elevating the World Through Imagery
The transformation from photos to DSMs and DTMs is a story of innovation, patience, and precision. It begins with carefully captured images, stitched together through Structure-from-Motion, expanded into dense point clouds, and refined into elevation models that capture both the seen and the unseen. Along the way, challenges of accuracy, vegetation, and processing power are met with creativity and technological advances. In 2025, the ability to generate DSMs and DTMs with just imagery is more than a technical feat—it is a bridge to understanding and shaping the world around us. From monitoring climate change to designing cities, these models are tools of insight and action. The true power of photogrammetry lies not just in capturing what is visible but in revealing what lies beneath, allowing us to elevate our understanding of the Earth itself.
