The ground beneath our feet: bringing ‘bare earth’ mapping to life with FABDEM
Bristol-based flood modelling company Fathom has combined forces with the University of Bristol FloodLab to launch FABDEM. This is the first global Digital Elevation Model (DEM) with forests and buildings removed at a 30m resolution.
Initially designed to create a more accurate, detailed data set for flood modelling, landslide modelling and location analysis, the GIS applications for FABDEM are proving to be far wider than first thought. From planning linear assets like transport routes, telecoms and energy infrastructure in data-scarce areas, to forestry and agriculture projects and creating artificial environments for flight simulation and gaming, FABDEM’s potential applications spread across multiple industries.
Fundamentally, FABDEM provides engineers and GIS specialists with the ability to map and model any project off-site. Using FABDEM, engineers can conduct initial remote site surveys and guide site selection. Here, providing a reliable source of data in advance of conducting physical surveys eliminates the need for multiple site visits in the initial stages of large-scale projects, delivering significant benefits in both time and cost.
Free weekly newsletter
Fill out your e-mail address to receive our newsletter!
By entering your email address you agree to receive our newsletter and agree with our privacy policy.
You may unsubscribe at any time.
Digital Elevation Models (DEMs): A User’s Guide
In their simplest form, Digital Elevation Models (DEMs) are a representation of elevations of a topographic surface above a chosen reference surface, most commonly on a regular grid. There are freely available global DEMs based on satellite measurements, at horizontal resolutions down to 1-arc-second (approximately 30m).
The problem lies in the fact that global DEMs are prone to large errors, often including elevations that represent layers above the Earth’s terrain (e.g. vegetation or buildings). The reality is that many applications require ‘bare earth’ elevations, representing the elevation of the ground beneath vegetation and buildings.
The freely and globally available Copernicus DEM, GLO-30, was released in 2020. It works at 30m resolution and uses more recent and accurate measurements compared to previous global DEMs, leading to suggestions it should become the gold standard of global DEMs.
At Fathom and the University of Bristol, we set out to use this gold standard model to create an effective DEM that strips vegetation and buildings for a true representation of ground-level terrain. FABDEM (the FAB standing for Forests and Buildings Removed) was developed as a ‘bare earth’ version of Copernicus DEM.
FABDEM: So what?
When our team first developed FABDEM, we intended to use it to advance our global flood modelling capabilities. Through the removal of surface objects, the dataset is ideal for the simulation of natural perils, such as flooding, where results are dramatically influenced by variables like ground elevation.
However, since its release, we have received significant interest from a wide range of industries and use cases, demonstrating the many other applications where FABDEM will prove incredibly useful.
Alongside water risk, enquiries have ranged from animal migration and pandemic modelling, all the way to 3D simulations for gaming and aerospace. Although we were aware of the common uses where FABDEM could be applied, it has been fascinating to observe the many niches where its data is applicable. Ultimately, anything that requires a 3D representation of the earth’s surface can find purpose in FABDEM.
The most interesting use case that we have found to date is the application of our terrain data in humanitarian aid and response. Functions such as supply logistics and relief mapping are all made more efficient through the use of DEMs. If you are aware of any other areas that might be of interest, please do get in touch and let us know.
How FABDEM compares to other digital elevation models
An important element of the paper we published describing FABDEM was its validation against LiDAR and comparison against other global DEMs. Firstly we compared the dataset against Copernicus DEM, to show that the elevation artefacts from forests and buildings are removed or severely reduced in FABDEM.
We also focused on a comparison against MERIT DEM. As the only other global DEM that removes forest heights alongside other errors, MERIT DEM is the closest conceptually to FABDEM. Our comparison against MERIT found lower errors compared to reference data, and spatial inspections show clearer representations of features in the landscape in FABDEM.
It is important to note that current global DEMs are based on observations many times coarser than LiDAR surveys, so FABDEM is not a product designed to rival local LiDAR-based DEMs. However, as high-quality local DEMs cover less than 1% of the globe, there is an urgent need for improvements to global DEMs for the rest of the world.
Machine learning in DEM building
To produce FABDEM we used machine learning methods to estimate the heights of vegetation and buildings in the DEM, relative to ground level, and then subtracted these estimates from the Copernicus DEM elevations.
The two key inputs to the machine learning algorithm were high-quality reference data, representing the ground elevations and predictor datasets which could be used to estimate the vegetation and building heights:
Reference data
In addition to global DEMs, many places have higher resolution local DEMs, for example, produced by LiDAR flown by aircraft. These LiDAR DEMs are extremely accurate and can offer horizontal resolution on the order of centimetres.
For the vegetation and building removal in FABDEM, we used local DEMs from 12 different countries to train our machine learning algorithm.
The use of training data from diverse locations is important here, as training a model on limited types of landscapes will tend to over-fit and produce results which are of less use in other locations.
Predictor data
Predictor datasets are used by a machine-learning algorithm to develop rules or connections between the values in the predictor datasets and the target (heights of forests or buildings).
As predictor datasets differ between forests and buildings, we applied separate machine learning models for each of these cases. For example, our team used Forest Height and Canopy Cover datasets as key predictors for forest removal.
However, when looking at urban removal, building footprints (World Settlement Footprint), population density, and travel times to urban centres were used among other datasets.
Although each of the individual predictor datasets are not perfect in isolation, by combining different useful data, we give the machine learning algorithm the information needed to make reliable estimates.
More about FABDEM
FABDEM combines a wealth of datasets used in combination to enable the reliable prediction of ‘bare earth’ representations of global terrains. If you are looking to accurately predict the influence of topographical components in altering or implicating large scale projects, it is essential to have a model that maps ground level, or ‘bare earth’. Applying FABDEM’s improved terrain information works to level up the accuracy of flood models and many other applications across engineering, GIS mapping and geospatial analysis.
For those interested in the data, you can download the research for free here or you can learn more about Fathom’s options for licensing here.
Hawker, L., Uhe, P., Paulo, L., Sosa, J., Savage, J., Sampson, C., & Neal, J. (2022). A 30 m global map of elevation with forests and buildings removed. Environmental Research Letters, 17(2), 024016. DOI: https://doi.org/10.1088/1748-9326/ac4d4f
About the authors:
Dr Peter Uhe, Senior Developer, Fathom
Dr Uhe has over a decade of experience working in academia and climate-related flood risk. His career has varied from the University of Oxford, CSIRO in Australia and the University of Bristol. Now Dr Uhe works within the technical team at Fathom where he contributes to the organisation’s understanding of the impact of climate change on flood inundation models.
Access his Google Scholar here.
Dr Laurence Hawker, Senior Research Associate, University of Bristol
Dr Hawker works in the University of Bristol’s School of Geographical Sciences where he has specialised in the development of hydrodynamic models in data-sparse regions and risk communication. His work has varied from the development of an intermediate scale flood model of the Mekong Delta to developing event response reports for the FCDO.
Access his Google Scholar here.
