Coffee Talk - FathomDEM, a new global 30 m DTM

Published: Mar 11, 2025 by ISG Board

FathomDEM, a new global 30 m DTM

Dr. Chris Lucas
Fathom company

January 20th, 2026
14:00 (UTC)

Recording available in our YouTube channel

Bio: Chris Lucas is the Principal Machine Learning Engineer at Fathom, a global water intelligence company. Holding a PhD in High Energy Physics, his 11 years in industry span from software engineering to ML research, contributing to projects such as scaling medical diagnosis models and developing new computer vision techniques for vehicle analysis. At Fathom, he leverages his combined ML, software, and scientific expertise to advance multiple levels of the company’s modeling stack, from terrain modeling (DEMs) to generative AI for climate downscaling and hydrological modeling with graph neural networks.

Abstract: Accurate digital elevation models (DEMs) are foundational inputs for a vast array of geomorphometry applications, including natural hazard modeling, glaciology, and infrastructure planning. However, existing global DEMs, such as Copernicus DEM (COPDEM), often contain surface features like trees and buildings, limiting their effectiveness as Digital Terrain Models (DTMs). This talk introduces FathomDEM, a new global 30 m DTM created using a novel machine-learning methodology. We utilized a hybrid vision transformer model within a U- Net architecture to perform pixel-wise regression, analyzing and correcting the height biases in COPDEM. This approach differs significantly from previous methods (like the pixel-by-pixel correction used for FABDEM) by leveraging 2D spatial information (context) as an inductive basis, essentially employing ‘computer vision’ to achieve more spatially coherent and robust corrections. FathomDEM was trained on extensive, diverse LiDAR reference data and has been rigorously validated, demonstrating: Improved Accuracy, surpassing the accuracy of existing best-ranked global DEMs; Excellent Performance in Specific Landscapes, showing reduced error even when compared to specialized coastal DEMs (e.g., DeltaDTM); High Utility in Downstream Tasks, when utilised in flood inundation modeling, FathomDEM achieves increased accuracy, approaching the performance levels of models derived from high-resolution LiDAR data. Join this session for an informal discussion on the methodology behind FathomDEM, its novel use of ML for artifact removal, and its potential to improve applied geomorphometry tasks globally.

First round of the DEMIX exercise published final report

Global digital elevation models (DEMs) have become routine operational inputs across mapping, environmental monitoring, modelling and Earth-observation workflows. Their broad coverage and global availability have made them data that many practitioners simply take for granted. But knowledge of the terrain and topography of Earth’s surface is fundamental for monitoring and understanding terrestrial ecosystems and the planet’s habitability. For that reason, the CEOS Working Group on Calibration and Validation (WGCV) maintains a subgroup dedicated to digital topography and the quality of DEMs, the ‘Terrain Mapping Sub-Group (TMSG).

DEMs are representations of elevation in the form of a georectified grid, at global scale commonly derived from space-based interferometric synthetic aperture radar (InSAR) or stereoscopic optical observations, while regionally and locally airborne Laser induced detection and ranging (LIDAR) has become the primary source. DEMs can be classified as digital surface models (DSMs) when depicting the lower surface of the atmosphere or digital terrain models (DTMs) when depicting the top of the lithosphere (the Earth’s crust).

A wide variety of DEM products exist – each having different characteristics which suit different applications. It can be challenging to understand which of the available DEM products are fit for purpose for certain applications or regions. That is the practical gap addressed by the Digital Elevation Model Intercomparison Exercise (DEMIX) undertaken by TMSG. By exposing where widely used global DEMs agree, where they differ and how those differences affect rankings, the exercise delivered evidence to users for selecting a DEM rather than defaulting to one by habit.

DEM products were compared through a ‘wine contest,’ which identified requirements of openness, reproducibility, adaptability, and statistical rigour. The characteristics, evaluations, and ranking for each DEM were collected in a GIS database with over 50,000 entries. A quantitative assessment was carried out for each global DEM based on pixel-by-pixel differences of geomorphometric parameters against finer spatial resolution (1-5 m resolution) reference DEMs.

The study incorporates an unprecedented amount of high quality reference data covering a broad range of landforms and surface types, comparing their characteristics without resampling or interpolation. Users are invited to consult the results of DEMIX in order to make informed choices when needing to use a DEM, or in some cases a combination of the most relevant DEMs for their applications.

The final rankings of the wine contest, presented below, compared the six DEMs based on land cover (forest, urban, or barren), slope (cliff, steep, gentle, or flat), differences in elevation, slope and roughness, and statistical metrics.

Led by the European Commission’s Joint Research Centre (JRC), the CEOS Working Group on Calibration and Validation (WGCV) Terrain Mapping Subgroup (TMSG) established DEMIX in partnership with the International Society for Geomorphometry. Several members of the International Society of the Geomorphometry contributed significantly and we hosted various secssions at our latest conference. The exercise was performed over a three-year period starting in 2020, with community-wide calls for participation and an assembly of experts producing a number of peer-reviewed publications, culminating with a JRC reference report published in July 2026.

The objective of DEMIX was to propose a procedure to rank the available free and open global DEMs, taking into consideration user needs while promoting the implementation of FAIR (Findable, Accessible, Interoperable and Reusable) principles. The exercise compared DEMs based on criteria for land cover and terrain slope categories, representative testing, and a statistically sound ranking approach. The report outputs tailored recommendations regarding available DEM products that are not limited to one domain, geographic area, or landscape type, with flexibility for different user needs and applications.

A major challenge in the comparison of the global DEMs were the varying formats, data, and metadata contents. The adoption of a common set of standards would enhance the quality and interoperability of global DEMs and streamline the exchange and utilisation of DEM data for both providers and users. DEMIX makes the following recommendations for data providers:

● Adopt a standardised grid layout, with complete grid definitions and encodings following ISO/OGC rules.

● Include vertical datum information as part of raster DEM files.

● Clearly indicate the product version in the file names.

● Participate in future DEMIX rounds, which serve as a neutral platform for independently evaluating products before their release.

● Provide high quality, finer resolution, accurate, and multi-temporal elevation data to be used as a reference alongside DEM products.

Global DEMs are indispensable operational geospatial datasets, but they should not be treated as error-free or interchangeable. DEMIX delivered to the global community a systematic and transparent method to help choose elevation products more intelligently and help define what better elevation measurements should look like in the future.

Benchmarking Elevation Plus Land Surface Parameters Finds FathomDEM and Copernicus DEM Win as Best Global DEMs (2025) and subsequent discussion (2026a, 2026b)
Ranking of 10 Global One-Arc-Second DEMs Reveals Limitations in Terrain Morphology Representation 2025
Novel Approach for Ranking DEMs: Copernicus DEM Improves One Arc Second Open Global Topography (2024)
Digital Elevation Models: Terminology and Definitions (2021)
The Digital Elevation Model Intercomparison eXperiment DEMIX, a community-based approach at global DEM benchmarking (2021)

Building the world’s largest repository of FAIR drone mapping data

Paul Mead GeoNadir

September 2nd, 2026
11:30 (UTC)

Bio: Paul Mead is Co-founder and Head of Business & Strategy at GeoNadir, a platform building the world’s largest repository of FAIR drone mapping data to support environmental decision-making. A non-technical co-founder, Paul focuses on strategy, commercial development, and building high-performing teams — a discipline shaped by 11 years in the New Zealand Army, including leading a bomb disposal team in Lebanon and service in Afghanistan and the Asia-Pacific region. Since 2013 he has built and led several small businesses and start-ups, with drones as a central technology since 2016. Alongside GeoNadir, Paul co-founded She Maps, an internationally recognised drone and geospatial STEM education organisation, and has helped raise over $7M in equity and non-equity funding across his ventures. He holds a Master of Philosophy from the University of Queensland and is a Chartered Manager.

Abstract: GeoNadir set out to build the world’s largest repository of FAIR drone mapping data, and ended up building something more powerful. Drones have democratised data capture, but the same democratised ability to generate decision-grade insights from the data is lacking. I’ll walk through how we’re closing that gap: the data infrastructure behind hosting large volumes of UAV imagery and derived products, and how we’re moving from manual analysis toward automated, repeatable geospatial analytics.

Geomorphometry with GRASS

Dr. Corey T. White Center for Geospatial Analytics at NCSU, USA

April 1st, 2026
14:00 (UTC)

Recording available in our YouTube channel

Bio: Corey White is a geospatial research software engineer with the Center for Geospatial Analytics at North Carolina State University and a maintainer of the open-source GRASS project. He is also the founder of OpenPlains Inc., an open-source web-based platform for interactive geospatial modeling. His work sits at the intersection of geomorphometry, environmental modeling, UAS mapping, and open-source software development, with a focus on making interactive terrain analysis accessible and reproducible. He has applied GRASS to problems ranging from post-disaster topographic change detection following Hurricane Helene to designing a participatory geospatial modeling platform to address ‘wicked’ socio-environmental problems. Corey is an active member of the OSGeo community and an advocate for open geospatial science.

Abstract: GRASS is a modern, actively developed, and extensible geospatial processing engine for geomorphometry. This session draws on post-Hurricane Helene landscape change analysis and ongoing addon development for DEM data fusion, terrain uncertainty modeling, and overland flow simulation to showcase what GRASS does well and where it fits in contemporary geomorphometric practice. Beyond its core analytical capabilities, GRASS’s Python, R, and Jupyter integration make it straightforward to build reproducible, scriptable terrain workflows that move easily from exploration to publication. Join this informal discussion to explore how GRASS serves not only as a research tool but as a community platform where geomorphometric methods can be developed, shared, and reused through its open addon ecosystem.

Coffee Talk - FathomDEM, a new global 30 m DTM

Further Reading: