We are living in a world of unprecedented innovation. Geospatial data, and the insights derived from it, are transforming the way we live.
In particular, digital terrain modeling and analysis continue to progress, spurred on by new sources of digital elevation data, and rapidly growing computational capacities. Technologies like machine learning are enabling geospatial insights at speeds that were previously unimaginable.
3-dimensional elevation models of the Earth’s surface can be quickly created, shared, and even combined with other products—such as vegetation indices—to answer different questions. More specifically, queries related to elevation. These could be simple questions, like what are the tree canopy heights? Or complex ones, like how much is the above-ground biomass?
Because of the extensive applications of elevation data, it is a data theme in Europe’s Infrastructure for Spatial Information (INSPIRE) and is listed as a global fundamental geospatial data theme by the United Nations committee of experts on global geospatial information management.
Elevation data is collected using different methods, such as Synthetic Aperture Radar (SAR), satellite stereo imagery, Light Detection and Ranging (LiDAR), and surveying techniques. The data may be output as contours, a point cloud, a triangulated irregular network, or a raster surface.
We can classify elevation models as either:
Digital Surface Models (DSMs): These represent elevations of human-made and natural features. A DSM captures the top-most surface of an area, including all exposed objects such as treetops and buildings. They represent the bare ground where there is nothing else above it.
Digital Terrain Models (DTMs): A bare-earth model devoid of human-made and natural features, such as infrastructure and vegetation. Data captured by satellites, helicopters, or drones are DSMs that can be turned into DTMs by removing non-ground objects.
Some derivative products of elevation models include slope, aspect, curvature, shaded relief, and normalized digital surface models (nDSMs).
Leveraging nDSMs can help you uncover insights for diverse applications. In this article, we highlight a few of these insights. We also go over 4 factors to consider before combining elevation models.
Shall we dive in?
A normalized digital surface model is a derivative elevation product obtained by subtracting a DTM from a DSM.
nDSM = DSM - DTM
Image showing a normalized digital surface model (J. Contreras, S. Sickert and J. Denzler, "Region-Based Edge Convolutions With Geometric Attributes for the Semantic Segmentation of Large-Scale 3-D Point Clouds," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 13, pp. 2598-2609, 2020, doi: 10.1109/JSTARS.2020.2998037.) (CC BY 4.0)
The nDSM represents the relative height of features above the surrounding ground surface. In contrast, DSMs and DTMs represent absolute height referenced to a datum, typically the mean sea level, a reference geoid, or ellipsoid.
nDSM height values range from zero at ground level to the vertical height of the tallest feature captured in the DSM.
Normalized digital surface models facilitate height comparison between features and change monitoring between different data acquisitions. They also contribute to:
- Building height estimation
- Urban structure mapping
- Tree height estimation
- Above-ground biomass estimation
Note that in forestry, the term canopy height model (CHM) is usually used to refer to the height of vegetation above the ground. You can think of a CHM as a nDSM that only represents vegetation.
Before going any further, let us define a few key terms you will see mentioned elsewhere in this article.
LiDAR data: Refers to data collected using a LiDAR sensor mounted on a terrestrial, aerial, or space-borne platform. LiDAR sensors acquire x, y, and z coordinates of human-made and natural features by emitting light pulses towards the ground and recording the reflected radiation. The output is a three-dimensional point cloud of the area of interest.
Relative height: The height of a point compared to a local reference surface. For example, the reference surface in a normalized digital surface model is the surrounding ground surface (represented by a DEM).
3D data acquisition: Collection of 3-dimensional (x, y, and z) coordinates of natural or human-made features on the Earth’s surface. 3D data acquisition methods for Earth observation include stereo imagery from both satellite and aerial sources, SAR imagery primarily from satellite sources, and LiDAR data primarily from aerial.
Point cloud data: A collection of many densely spaced 3-dimensional points representing an object or underlying sampled surface. Point cloud data comprises x and y coordinates, z (height) values, and other attributes, such as intensity. Point cloud data is output from LiDAR or generated from imagery.
Point Data Abstraction Library (PDAL): An open-source library written in C/C++ for working with LiDAR data. PDAL enables users to break down LiDAR operations into steps for code reuse. A specific strength of PDAL is that it can handle any point cloud data format.
With definitions out of the way, let us dive into the insights provided by normalized digital surface models.
An nDSM layer draped over other elevation products (Source: Vermont Center for Geographic Information)
nDSMs reveal patterns and relationships between the bare earth and the objects on it.
Here, we show you the power of normalized digital surface models by highlighting a few benefits and use cases.
An orthophoto is a satellite or aerial image that has been corrected for distortions caused by scale and relief displacement. Like a map, an orthophoto is true to scale enabling direct angle, distance or area measurements. Unlike a map, however, an orthophoto represents actual features and not cartographic representations of those features. It is both a map and an image.
Orthophotos are one of the most widely used geographic products. They can be integrated as base layers for various applications including production and updating of cartographic maps, land use land cover mapping, three-dimensional urban modeling, and precision agriculture.
The process of orthophoto creation is called orthorectification. It is an essential pre-processing step for both satellite and aerial imagery. Orthorectification entails standardization of scale across the image and removal of relief displacement.
Relief displacement is caused by differences in elevation. If the elevation of the terrain and (or) human-made features is known, then geometric distortion can be rectified. In this regard, DSMs/ DTMs are one of the key inputs for transforming vertical aerial photos into orthophotos.
Building detection and mapping is an essential task for applications like urban and rural planning, disaster risk assessment and preparedness, change monitoring, and navigation.
3-dimensional models support efforts geared towards monitoring urban growth for informed decision-making and improved living conditions.
Earth observation satellites collect massive amounts of imagery across the globe. They enable wide-area mapping and monitoring of changes in the environment. However, it is challenging to automatically map buildings from high-resolution satellite imagery because:
Surrounding objects which have similar spectral reflectance and shape as buildings may be wrongly classified as buildings
The shape and structure of buildings vary between countries and regions and between settlements
Historically, spatial resolutions have been insufficient for crowded urban areas
Integrating building heights from nDSMs in the building extraction process has been shown to increase the accuracy of building extraction, reducing the number of objects erroneously classified as buildings. Setting a height limit (threshold) to the normalized digital surface model helps exclude surface features, such as cars, while including the lowest shacks.
Apart from heights, combining a normalized digital surface model with a normalized difference vegetation index (NDVI) enables the identification of non-vegetated surfaces, contributing to greater accuracy in the building extraction process.
Additionally, using texture or color information from imagery with the intensity attribute of a LiDAR-derived normalized digital surface model could help distinguish between building and vegetation areas.
Identifying points or areas that can be seen from one or more observation points is called line of sight or viewshed analysis. Whereas line of sight analysis determines visibility along a line, viewshed analysis determines visibility in all directions of the surrounding area. Both offer insights into urban planning, radio communication, defense, and even flight considerations.
Line of sight or viewshed analysis using a DTM enables the identification of terrain obstructions. A DSM provides additional insights for built-up or forested areas. Ultimately, choosing which elevation model to use depends on your application.
It is important to note that the line of sight is not necessarily visible to humans. For electromagnetic waves, like radio waves, line of sight refers to a transmitter, antenna, and terrain combination that allows transmission and reception of electromagnetic signals.
Let us look at some specific line of sight analysis use cases.
To determine suitable locations for new developments, there is a need to evaluate their impact on the scenery. This entails identifying the visibility of the new development to and from its surroundings—a process known as visual impact assessment.
Visual impact assessment aims to ensure a balance between infrastructure development and views of natural landscapes, historic landmarks, green spaces, water bodies, mountains, and other elements valued by a community.
Normalized digital surface models show the heights of vegetation and buildings. They are crucial inputs when determining the extent of visibility and thus the visual impact of different infrastructure design alternatives.
In recent years, drones have become a common method for 3D data acquisition. In most countries, the drone must always be visible to the pilot.
Thus, viewshed analysis performed using a combination of DSMs and DTMs enables the determination of the optimum combination of pilot location and flight altitude to maximize coverage while maintaining line of sight.
Further, with terrain-following becoming a widespread method for flying drones, loading an accurate DSM beforehand will ensure that the drone flies safely, avoiding collisions with buildings, trees, or other obstacles.
In radio communication, visibility analysis comes in when planning cell phone tower locations—because they need an unobstructed view of other cell phone towers for transmission—to optimize network coverage. Here, nDSMs provide insights into the type and height of obstructions. Specifically, the combination of a highly accurate DTM and DSM is crucial for planning 5G networks whose waves are sensitive to interference from natural and human-made objects.
Note that, to model the propagation of radio waves, you may need additional data besides elevation models.
For example, Intermap’s RF Viewshed machine learning algorithm requires additional details like transmitter power, antenna gain, oxygen loss, and rain to establish whether a transmitter is visible or hidden from a receiver and compute the expected signal strength at the receiver.
Image showing a radio frequency viewshed from a transmitter. The colors represent signal strength with red being high and blue being low. Intermap’s RF viewshed is applicable in telecommunications, visibility impact assessment, or drone flight planning. (Source: UP42)
Factors to Consider in Line of Sight or Viewshed Analysis
- Limits of human sight: for applications where the observer is human (e.g., flying a drone), there is the need to consider a limited observation radius around the observer based on how far the human eye can see.
- The altitude of the observer (human or object) above the terrain: visibility depends on the height of the observer above the ground. Areas or objects visible to a tall observer may be invisible to a shorter one.
- The resolution and accuracy of the 3D elevation model used will affect the accuracy of the viewshed analysis: the higher the resolution and accuracy of the elevation model, the higher the accuracy of your viewshed analysis.
In forestry applications, LiDAR-derived normalized digital elevation models—also referred to as canopy height models—are commonly used because LiDAR provides something that aerial and satellite imagery cannot: Dense 3D point cloud data and vegetation-penetrating capabilities.
Canopy height models help derive individual tree heights, map forest coverage, classify tree stand structures, assess above-ground biomass, and detect canopy changes.
In agriculture, nDSMs enable crop height variability monitoring, which can inform the application of seed, fertilizer, water, pesticides, and herbicides. Further, the generation of visual maps showing crop heights at different development stages could enable the detection of crops whose growth is stunted—an insight that would be difficult to notice using ground-based assessments.
For example, the image below shows the results of a study that used nDSMs to monitor the growth rate of wheat in an agricultural plot. The red clusters in the middle and right-hand show areas where crop height was stunted.
Image showing a LiDAR-derived normalized digital surface model of wheat captured at three different times in the year. (Source: Holman, Fenner H., Andrew B. Riche, Adam Michalski, March Castle, Martin J. Wooster, and Malcolm J. Hawkesford. ‘High Throughput Field Phenotyping of Wheat Plant Height and Growth Rate in Field Plot Trials Using UAV Based Remote Sensing’. Remote Sensing 8, no. 12 (December 2016): 1031. https://doi.org/10.3390/rs8121031.) (CC BY 4.0)
Canopy height models are also used for wildfire risk assessment, specifically, estimation of fire fuel loads. Further, since fire burns faster up-slope, combining slope maps and nDSMs can help model the direction of fire propagation, informing disaster response strategies.
Land cover maps are a prerequisite for effective land management and change monitoring. Applications of land cover maps are broad: water resource management, urban planning, environmental management, and wildlife conservation, to name a few.
Satellite imagery at different spatial, temporal, and spectral resolutions is an essential input for generating wide-area coverage land cover maps. It is, however, a challenge to produce such maps at a finer scale, especially where land cover features exhibit small spectral variations between them.
Thankfully, combining LiDAR-derived normalized digital surface models with multispectral imagery improves classification accuracies by:
Helping to separate different vegetation classes using heights, e.g., differentiating grassland from shrubland
Distinguishing features of similar height but with different reflectance properties using LiDAR point cloud intensity values
nDSMs enable the reconstruction of building roof forms, and the estimation of tilt, orientation, and surface area of the roof. These provide insights into:
The amount of solar radiation received by a building and hence, where to place solar panels
The likelihood of a building overheating because of thermal radiation
Compliance with zoning, development, and city planning regulations
Flooding caused by sea-level rise, tsunamis, or overflowing rivers can have devastating socio-economic effects on a community. With climate change, flood risks are an increasing concern. As such, highly accurate, high-resolution 3-dimensional elevation models are essential for:
Estimating and predicting flood depth and flood extents
Identification of affected infrastructures
Emergency response planning
Development of preventive measures
Flood risk modeling involves using DTMs to estimate flood depth and extents during different flooding scenarios and overlaying DSMs to identify and visualize critical infrastructures at risk.
Also, nDSMs are valuable for assessing the possible impacts of flooding through building submersion. Used in combination with population data, you can also predict the number of people at risk.
Map showing the buildings affected by flooding in red (Source: Using OpenStreetMap Data and UP42 to Map the Impact of Flooding)
The generation of normalized digital surface models calls for the combination of DSMs and DTMs.
When combining different elevation datasets:
Use data from the same provider, if possible
Ascertain that they are referenced to the same horizontal and vertical datum
Ensure they have similar horizontal and vertical units
Below are other factors to consider.
Remote sensing technologies that provide elevation data include LiDAR, radar, and stereo photography.
LiDAR has vegetation-penetrating capabilities and offers high vertical accuracies. It is thus suited for vegetation applications. Radar is more effective in cloudy conditions and, like LiDAR, has vegetation-penetrating capabilities. However, it has lower vertical accuracies. Stereo photography only captures the top-most surface heights.
Because of the characteristics of each data collection method, the resulting elevation model (DSM or DTM) may suffer vertical biases which affect the accuracy of the results obtained.
Horizontal resolution—commonly referred to as spatial resolution—means the smallest size and distance between features observable in the elevation model. It is the land area represented by a single grid cell. The elevation model's vertical accuracy is tied to the horizontal resolution.
A higher horizontal resolution implies the representation of more topographic details—and hence a higher vertical accuracy—as opposed to a lower resolution. This, however, depends on the area. For example, you would need a high-resolution elevation model to correctly model urban areas, while you would adequately capture flat and open terrain at lower resolutions.
Additionally, the resolution is both a function of the data collection method used and the processing and interpolation techniques applied to derive the digital elevation model.
Considering all the above factors, it is important to ensure that the combined DSM and DTM not only have the same horizontal resolution but are also from a common source.
Vertical and horizontal accuracy refers to the positional accuracy of the elevation model to a specified vertical and horizontal datum, respectively.
The horizontal (spatial) resolution affects the vertical accuracy in that, generally, high vertical accuracies call for high horizontal resolutions.
The requirements for vertical accuracy vary by application. For example, applications like sea-level rise modeling and deformation monitoring, where small elevation changes lead to large variations in the results, call for higher vertical accuracies.
There are two types of vertical accuracies:
Absolute vertical accuracy: This refers to the difference between the elevation stated in the digital elevation model vis-à-vis the correct elevation obtained from a vertical datum.
Relative vertical accuracy: Refers to the elevation difference between two points measured on the digital elevation model vis-à-vis the elevation difference between the same two points on the reference surface.
The vertical accuracies depend on the terrain (e.g., flat or rugged) and land cover (e.g., buildings or vegetation). As such, a digital elevation model’s reported accuracy may not hold for different terrain or surface characteristics.
Further, since DTMs are created by removing surface features from DSMs, uncertainties in the complete removal of surface features may affect their accuracy—especially in urban or densely vegetated areas. Thus, DTMs and DSMs with similar horizontal resolutions may have dissimilar vertical accuracies.
The Earth’s surface changes over time because of human-made and natural activities. Inevitably, DTMs DSMs of the same area captured during different periods may have varied elevations where such changes have occurred. When generating nDSMs, combine elevation models with minimum temporal differences.
What makes normalized digital surface models so powerful?
Well, nDSMs can augment your insights, enabling better decisions for better outcomes. Combining them with different datasets can help you uncover new opportunities and novel ways to disrupt and be disrupted.
Sign up today and take advantage of our advanced analytics tools to uncover insights in vast troves of geospatial data.