Digital elevation models transform business processes

Fathom was the SME winner of The Stack’s Tech for Good 2021 awards

Mapping the Earth’s topography, or the shape of the Earth’s surface, was primarily of interest to the military, civil engineers, and academics. With the advent of smartphone apps, however, these datasets have entered the public consciousness, in the form of countless services that allow us to seamlessly navigate and interact with complex geographic and geospatial data. Or just follow and share our Sunday morning bike ride.

In the corporate world, geospatial data and analytics are now firmly on the agenda of corporate boards, writing Dr Andrew SmithChief Operating Officer, Fathom. From CEOs and CIOs to risk managers and data scientists, terrain datasets known as digital elevation models (DEMs) are used in business processes far beyond consumer smartphone apps. .

Enterprises use DEMs in risk mitigation, compliance, engineering, infrastructure, and automotive use cases. In utilities and NGOs, DEM datasets are used to map potentially dangerous areas, plan future emergency response, and coordinate rescue and aid missions following natural disasters.

Digital Elevation Models or DEMs Explained

DEMs take the form of a digital surface model (DSM) or a digital terrain model (DTM), the former containing surface artifacts such as buildings and trees and the latter retaining only information about the underlying ground or the bare earth itself. These datasets can be collected in a variety of ways, including laser altimeter (LiDAR) data collected through aircraft surveys and data collected from space using near-infrared, radar, and visible sensors. On a global scale, it is to this last category of data collection that we are limited. Consequently, this has presented significant challenges in obtaining an accurate image of the Earth’s surface over the past decades.

Global terrain data falls under the DSM category, retaining surface features as radar beams bounce off the tops of trees, buildings, and other artifacts. For some applications, such as risk mitigation, the presence of these artifacts prevents us from using them accurately. Think of the difference it would make if you were in a tall building during a flood – the difference of just a few meters can have a huge impact on the devastating effects of a flood. Simply put, without an accurate map of the entire surface of the Earth, mapping and avoiding the most dangerous places becomes very difficult. Yet recent research suggests that if we avoid using and building on hazardous areas, we can avoid the majority of future risks. As we try to manage the impact of climate change on changing weather patterns, identifying risk areas becomes increasingly important.

The data challenge scarcity

Mapping of this type is precisely what we do at Fathom, building flood models that cover the entire planet. Our data is used in multiple sectors, including (re)insurance, engineering, financial markets, enterprise risk management and disaster response. Until recently, our ability to build models on a global scale was severely limited by the availability of accurate field data. In data-poor areas, we relied on a legacy dataset derived from a radar product collected by NASA’s space shuttle, called the Shuttle Radar Topography Mission (SRTM). This dataset is over 20 years old and is riddled with instrumentation errors, as well as superfluous surface features, including newly constructed city buildings. These must be removed in order to build an accurate flood pattern.

That all changed last year when a new global terrain dataset called Copernicus GLO-30 (COPDEM30) was released by the European Space Agency. COPDEM30 is a ~30m resolution DSM collected by the TanDEM-X mission, a mission whose sensors operate at a native resolution of ~12m. The precision and accuracy of the missions’ sensors have enabled COPDEM30 to become the new benchmark in global terrain mapping, containing far more information than previous datasets. However, the data still represents a DSM; to unlock true data functionality, all surface features had to be removed. This is where global hydrology, and in particular the flood risk modeling community, came into the picture.

Figure 1: Example of different terrain data sets in Rockhampton, Australia. On the left is MERIT DEM, a terrain dataset derived from SRTM. In the middle is the recently released FABDEM and to the right is a locally derived aircraft-surveyed LiDAR terrain dataset that can be taken as the reference.

Machine learning in cartography

By striving to build global flood hazard models over the past 20 years, the flood modeling community has become de facto experts in processing DSM data into DTM data. A year after the release of COPDEM30, a team of flood modelers released the Copernicus Deleted Forest and Buildings DEM (FABDEM), a global DEM derived from a reprocessing of COPDEM30. The team behind FABDEM trained machine learning (ML) algorithms on a suite of predictor variables to estimate artifact location and to characterize artifact size. These ML-trained algorithms were deployed to remove all estimated surface artifacts in the COPDEM30 dataset. The result was a global DEM, representing a significant advance in our ability to model the Earth’s bare surface.

An obvious area of ​​application for this new dataset is in flood modeling, especially in data-poor areas. Flood models generally simulate the propagation of water across the Earth’s surface. They therefore need an accurate terrain dataset on which to simulate water flow. The advent of FABDEM ensures that even in areas where no local information is available, useful flood risk information can be derived. This means we can deliver meaningful and actionable risk information anywhere in the world. For example, we can deploy FABDEM to avoid construction and development in regions most prone to flooding, including a recent project where Fathom’s global data was used to identify safe locations to build schools in rural areas. from Peru.

Figure 2: Simulation of flood risk in the three different terrain datasets shown in Figure 1.

But the relevance of this data goes far beyond flood modeling and far beyond all forms of hazard modeling. This data can be deployed wherever an accurate model of the Earth’s surface is required; from infrastructure planning and industrial visualization to navigation apps on the smartphones of billions of people. The next time you use an app to map your commute to the office or your morning bike ride, the underlying data might have been produced by a small group of scientists in the world of flood modeling.


FABDEM lets you create powerful maps and 3D simulations of the earth’s surface and accurately model natural hazards. Derived from Copernicus GLO 30 and built using machine learning methods, FABDEM is the world’s first digital elevation model to remove forests and buildings at 30 meter resolution to give an accurate picture of the terrain at level. of the ground. Developed in collaboration with the University of Bristol, FABDEM’s high-resolution digital terrain data ensures its value in the insurance and engineering sectors, as well as for GIS mapping and geospatial analysis. For more information visit:

Comments are closed.