Digital Elevation Models (DEM) are recognized as necessary spatial datasets in many applications dealing with geographic data. They can be useful for visualization and simulation applications, and any landscape modeling projects. DEMs are commonly produced using data collected with remote sensing techniques, but they may also be built from land surveying. 

DEMs are spatial representations of reality, which means that they approximate the reality of the terrain. For this reason, they can contain errors (artefacts) that are important to consider and correct. Otherwise, these errors inevitably propagate in the different applications that use the data. Consequently, this could highly affect high stake projects. Even smaller, simpler projects could end up with mistaken data. 

AI and Machine Learning do not provide all the answers 

Stereo auto-correlation techniques are used to fix overlapping images when assembling a DEM. However, not even supercomputers like the Blue Waters supercomputer found at the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign are able to fix the generation of artefacts. These artefacts can come in the form of spikes, pits, false landforms. Most definitely, they can decrease the quality of the dataset.               

For the moment, the sole effective and reliable way of correcting artefacts is by doing it manually. It is however expensive and resource-demanding. It also requires a level of geographic and topographic comprehension and expertise that most people do not have…  making the task even harder. 

PlanetObserver’s methodology (And why it is more reliable)  

The good news is that here at PlanetObserver, our team of experts has you covered and have the eye and technical skills to correct and make an exceptionally reliable medium resolution global elevation dataset.  

This is possible due to the team’s expertise in geographic analysis, terrain topography, and data management. This allows PlanetObserver’s team to identify the artefacts in the source data quickly using different techniques and replacing them with correct data. 

Consequently, PlanetObserver has put together PlanetDEM, a corrected and global Digital Elevation Model with 30-meter resolution. The product comes in the form of a Digital Surface Model (DSM) that captures the surface including natural and human-made structures such as vegetation and buildings. This DEM is mainly based on ALOS World 3D-30m dataset (JAXA) and corrected with NASADEM data (NASA).  

PlanetObserver has applied different techniques to correct the source data.

First step: Detection and removal of spikes and holes

Spikes and holes are mostly caused by the low correlation between the AW3D30 images. This step is performed in a semi-automated process that allows to automatically detect spikes and holes and replace them with NASADEM auxiliary data.  

Second step: Detection and removal of artefacts

Artefacts are usually created during the processing of AW3D30 source product. Indeed, if the surface is covered with low-contrast materials, such as water, snow and ice, the correlation between the images becomes weak.  This induces a shift in the images, and consequently, artefacts appear at this location.  

This process requires human expertise. This expertise has been developed by PlanetObserver during the production of PlanetDEM 30 Plus elevation data, PlanetObserver’s previous global DEM based on SRTM 1-arc-second dataset. Based on the experience of PlanetObserver’s experts and the existing auxiliary layer information in the ALOS World 3D-30m product, anomalies and outliers can be identified and corrected. Data corrections have been done mainly by using the Delta Surface Fill (DSF) replacement method.  

As a result, PlanetObserver managed to correct most of the anomalies that commonly appear in many DEMs of the same resolution, as illustrated below.  

Visual comparison between PlanetDEM (PlanetObserver), NASADEM (NASA), ALOS World 3D-30m (JAXA) and GLO 30 (Copernicus) – Tahiti Island (French Polynesia) 

The result: a reliable global elevation dataset 

PlanetObserver’s expertise and knowledge allows today to target historically problematic areas of global DEMs and make the necessary corrections to come up with a global and reliable elevation dataset. 

Thus, 35.5% of the tiles of AW3D30 base source product have been corrected by PlanetObserver. See areas in yellow and red in the image below. 

Correction of AW3D30 source product (areas in yellow and red corrected)

In practice, due to its reliability, PlanetDEM global elevation dataset can be used in various industries covering defense, mission preparation systems, flight simulation, GIS applications (energy, land planning, infrastructure, etc.) and orthorectification of satellite images. 

About the author 

Guy Boussougou Boussougou has a PhD in environmental geography and remote sensing obtained at the University of Reunion in 2017. After a year as a research engineer (PhD position) at the French National Institute for Agricultural Research (INRA, 2018-2019), he is since 2020, in charge of research projects and development of innovative products at PlanetObserver, Clermont-Ferrand (France). He’s been working on the processing and production of a global DEM (PlanetDEM) from open source DEMs in collaboration with the Earth Observation Laboratory (Lab’OT) of the French National Center for Space Studies (CNES) via the SME support program, Connect by CNES.  

Learn more about PlanetDEM