Creating a digital elevation model (DEM) from open data
Any sufficiently advanced technology is indistinguishable from magic. (c) Arthur C. Clarke
Building a digital elevation model (DEM) using the open source PyGMTSAR (Python InSAR) software is a prime example of a technology that looks like magic. Indeed, it only takes one click to create a satellite DEM. This example and many other interactive InSAR examples with 3D visualizations are available on Google Colab, in Docker images, and even on GitHub Actions at https://InSAR.dev
This digital elevation model is created from Sentinel-1 radar images in just one minute on an Apple Air laptop and can be accessed in minutes on internet-connected iOS, Android, and other devices. This speed and accessibility makes the technology suitable for anyone, without the need for a fast internet connection, large storage space, or a powerful computer. Examples are available at https://insar.dev
Introduction
Publicly available radar satellites such as Sentinel-1 scan our planet day and night, storing the reflected signal as a raster image for subsequent processing. These images store complex values of the phase of the reflected signal and can be used to obtain a variety of information about the state and movements of the earth's surface (from which the signal is reflected) and the properties of the atmosphere (through which the signal passes twice on its way – from the satellite to the surface and back). The simplest option for processing is to calculate the amplitude of the signal, obtaining an image in shades of gray (or pseudo-colors), much like an optical survey, but independent of day and night and cloud cover. The true potential of radar satellites is revealed if complex raw data is processed. The wavelength of the Sentinel-1 radar is 56 mm, which means that the received data registers precise heights and displacements of the earth's surface on the order of fractions of a millimeter, which can be extracted. The idea of obtaining surface displacements is very simple – if the surface drops a millimeter down, it moves away from the satellite and the radar beam needs to travel an additional distance of 1 millimeter on the way to the surface and back, a total of 2 millimeters, and this tiny delay can be recorded and processed. In addition to surface displacements, the so-called atmospheric phase is preserved in the images, that is, the delay of the signal as it passes through the atmosphere. In the case of strong atmospheric turbulence or cloudiness, the atmospheric phase delay can be equivalent to hundreds of millimeters of surface displacement, so it is necessary to accurately separate the different sources of change in the registered signal in order to achieve high accuracy of surface monitoring. Of course, for calculations of such accuracy, it is necessary to know the satellite orbit with the appropriate detail and be able to compensate for errors in the orbit values. In addition, solid-state lunar and solar tides have a significant impact, significantly shifting the value of the planet's surface; this correction can be calculated analytically or compensated for, based on the fact that in the spatial spectrum its influence manifests itself only in the long-wave part. There are many other sources of errors, such as the so-called radiometric noise inherent in the radar technology itself, which must be taken into account, analyzed and, if necessary, compensated for in various ways. This is what the science and technology of satellite interferometry deals with.
In the simplest case, for illustrative purposes and training, a pair of “ideal” images is selected on a well-reflecting rocky area with low cloud cover, while in industrial use, various methods of processing a series of images are used. Accuracy in fractions of a millimeter is not required to construct the relief, so for a well-chosen area, data analysis is performed automatically without the need to select any parameters. In the presence of vegetation and cloud cover, additional operations are required to separate these components of the signal and assess the accuracy of the results.
results
For example, a “convenient” area for analysis in Turkey was selected, with minimal influence of clouds and vegetation. It can be seen that the contours of the two selected images do not completely coincide; to compensate for this, preliminary alignment of the images on a rough relief model is required.
Suggested Python Jupyter Notebook 'PyGMTSAR Elevation Map: Erzincan, Turkey' on https://insar.dev is hosted on Google Drive and runs on publicly available free Google Colab resources, allowing you to perform all processing online in a web browser: loading and unpacking the original radar images and rough terrain (NASA SRTM or Copernicus Global DEM), combining multi-temporal images, calculating the phase difference, constructing the topographic phase and calculating the corresponding terrain heights. Note that the processing requires DEM data, which is used only to combine the original radar images on the surface of the 3D model of the planet. The resulting DEM can be much more accurate than the one used in the processing, so this allows you to use any open terrain data to obtain a detailed model.
The constructed interferogram is a phase map specified in the interval [-π, π] and for the selected territory is clearly distinguishable. The correlation map corresponds to the probability measure of the phase values - low correlation corresponds to low accuracy of calculations and high correlation means accurate results.
The obtained results can be compared with the originally used global relief model, in case of large deviations this may mean that the analysis is not correct and it is necessary to check each step of calculations, find and compensate for unaccounted factors. In this case, the result is valid and the observed differences correspond to the difference in the accuracy of detailing.
Conclusion
Currently, there is only one global publicly accessible (data is available to everyone and covers almost the entire surface of the planet) satellite in orbit, Sentinel-1, since the second one failed. However, almost a decade of imagery from two devices is available, allowing us to build a detailed relief and study surface displacements, infrastructure, earthquake consequences and many other processes and phenomena on our planet. At the end of the year, two new Sentinel-1 radar satellites are expected to launch, as well as a joint NISAR satellite from the American NASA and the Indian Space Agency (one satellite has a double set of equipment, which makes it very interesting in terms of obtaining a simultaneous double set of images). I also continue to develop PyGMTSAR (Python InSAR) – a Python library that performs InSAR processing extremely efficiently and affordably. You can find a large PyGMTSAR community at LindedIn And Patreonas well as many interactive examples online on the project website https://insar.dev (GitHub repository mirror https://github.com/AlexeyPechnikov/pygmtsar). So InSAR technology is evolving both technically and in terms of publicly available data, allowing the entire planet to be monitored remotely and with sub-millimeter accuracy. Isn't this technology pure magic?
