Automating Volumetric and Thermal Analysis of Volcanoes: My Undergraduate Research Project
Automating Workflows using GIS to Rapidly Analyze Volumetric and Thermal Trends of Volcanoes from Spaceborne Imaging
Hello! My name is Benjamin Schuler and I am an undergraduate Geology major at the University of Pittsburgh in my senior year. I have been working on a research project for approximately two years that involved ArcGIS Pro and volcanic hazard monitoring.
Since I was very young, I always liked finding rocks, however one day I found a large chunk of pegmatite when I was 11 years old in Connecticut and also visited Mt. Vesuvius in the same year. These events would cause me to be further interested in the processes that produced igneous rocks, both surface and subsurface.
When in college, I had been looking for a volcanology related research project for me to participate in as an undergraduate. The only group I could find at the University of Pittsburgh was using something called “remote sensing” to look at volcanoes. After reading up on their research, I realized I would need to take a GIS class to understand how they analyze data from spaceborne instruments and decided to get the GIS certificate. I would end up working on a new project, with Professor Daniel Williams as my advisor, that sought to use the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), on TERRA satellite, to observe the volumetric change of volcanic features.
Developing the workflow
My project built upon work done by Daniel Williams, Ian Flynn, and Mike Ramsey that utilized ASTER previously to observe volumetric aspects of volcanic eruptions. Previously, they had utilized Digital Elevation Models (DEMs), generated by ASTER using photogrammetry, to measure the volume of lava flows or domes that had grown. As the DEMs were generated by ASTER from visual wavelength imaging, the photogrammetry could have errant elevation values caused by shadows, lighting quality, dark terrain, etc. Therefore, to create a smoother surface, they would average together the DEMs that had clear views of the volcanic feature of interest using ArcGIS Pro. An average surface from before the eruption and another from after the eruption would be then subtracted from each other, to generate a difference representing the total volumetric change. The Surface Volume 3d function in ArcGIS would then be applied to the difference to calculate the volume.
However, my project differed in attempting to, despite the known elevation errors that exist in ASTER’s DEMs, observe the volumetric changes over the course of an eruption using individual DEMs. We also wanted to observe relative thermal intensities at the volcano over the course of the eruption using ASTER’s thermal wavelength imaging.
The first volcano I was told to look at and develop a workflow for our analysis was Mt. St. Helens during its 2004-2008 lava spine growth period. The first step would borrow from the previous work done, by generating an Averaged Pre-Eruption DEM (APED) using ArcGIS Pro from clear scenes prior to Mt. St. Helens’ eruption, as seen in a figure I created below:
Step 1
Each scene from during the eruption, a “syn-eruption” scene, would then be subtracted from by the APED to produce difference DEMs:
Step 2
Then the region representing the extent of the lava dome would be extracted from each difference DEM and the Surface Volume function in ArcGIS Pro applied to calculate approximate volume:
Step 3
Automating the workflow
As I was manually conducting the analysis of Mt. St. Helens, I learned in my GIS class about ArcGIS Pro’s Model Builder tool. This caused me to explore the potential of automating the workflow for both our thermal and volumetric observations of volcanoes. As I had no experience in automation, and minimal experience with ArcGIS Pro, I decided I would try to automate the apparently simpler workflow of thermal analysis. Thermal analysis would involve the highlighting of anomalously hot surfaces at the volcanic feature by subtracting an established background temperature from all pixels in the thermal infrared scene captured by ASTER. Though I had eventual success in developing this model, as seen below, it was still significantly slower than hoped. My use of strictly premade modules in Model Builder also did not allow as much flexibility as I would need for volumetric analysis.
This would cause me to take a different approach in the development of an automated volumetric workflow, when I learned the concept of custom scripts in Model Builder. However, this required me learning python from scratch, constantly asking questions of Ai, such as ChatGPT, on the structure of Python code, how to set up a function, basic syntax and then diving into writing my own code to see if I could apply what I learned. This allowed me to learn Python without driving an instructor crazy with constant questions.
After two semesters, I managed to reach the current form of my volumetric model. The workflow would end up going through approximately 2 complete reworks, with individual scripts I wrote being changed overall or tweaked to allow for easier use. There was even the time I had to completely rewrite a couple scripts when a data corruption wiped them.
The final form as it sits today, before I start messing with it again, is seen below:
However, I would also integrate steps 4 and 5 to allow for the quantifying of volumetric errors. This was done by establishing a region of consistently unchanged terrain during the eruption, but proximal to the volcanic feature of interest. Any deviation in elevation from the Averaged Pre-eruption DEM in these regions would be calculated as seen below:
Step 4
The difference in elevations measured from the Averaged Pre-Eruption DEM would then be averaged across the extracted error elevation region to produce an average elevation error. This elevation error would be either added to or subtracted from the APED, producing a Lower Bound APED and Upper Bound APED respectively.
Step 5
In step 5 on the model, the Lower or Upper Bound Elevation Error APED is then used as in steps 1-3 of the model to produce the Lower and Upper bounds of volumetric error. This process is automatically carried out for each Syn-eruption DEM.
This model overall was developed to be as easy to use as possible, with the raw DEMs downloaded from NASA’s Earth Data site only needing to be subdivided into Pre-eruption and Syn-Eruption folders. You will also need 2 shapefiles, one delineating the volcanic feature of interest and the other the elevation error region. Lastly, you need a folder for the output. These folders and shapefiles then all just need to be plugged into the model and for the user to only press “Run.” All results are organized into their own subfolders and renamed for ease of analysis automatically also.
However, the path to this current version of my model was filled with many learning experiences in the order in which you should take certain steps. For example, I originally would clip the extent of the volcanic feature of interest as early as Step 1 from the APED. However, this faced the issue of DEMs not perfectly registering with each other from pixel to pixel. As a result, some Syn-eruption DEMs had pixels that were not fully within the boundaries of the clipped extent and were considered regions of negative loss. After attempting to just filter these out for a month or two, I realized I had implemented the extent clip too early in the process. After only applying extract by mask once in the whole process, on the Difference DEM, this problem was solved and the final result no longer had anomalous pixels demonstrating excessive elevation loss. Some problems, however, were as simple as forgetting to add the “.tif” at the end of automatically renamed files, which would cause nothing to be saved correctly that could be used in the following steps. Another problem was forgetting to change the projected coordinate system setting in all the custom script steps. This could, and did, lead to excessively large volumes being measured when the default “WGS 1984 Web Mercator Sphere” was accidentally used in one or more steps. These issues, and their solutions, could take days to weeks to figure out, but each time you became more experienced in how to effectively develop a custom script tool and workflow.
Future for the work
I would end up presenting my research at American Geophysical Union (AGU) in December 2024 and my further work on it for my senior thesis this spring semester.
For future developments, I plan to improve the quantification of elevation error for each syn-eruption DEM, as it is currently capturing errant elevation errors when clouds intrude on the error region. I also want to find a way to automate a side profile of the topographical change at the volcano to supplement the volumetric analysis. Another plan is to further streamline the overall model and make it even easier to use with less input required. One example of streamlining would be the automatic retrieval of the coordinate system from the DEMs, to avoid mistakes in forgetting to manually set these. I additionally want to revisit my thermal anomaly model to convert it to scripts to speed up the time for it to conduct analysis and allow for more customization. Finally, my advisors and I wish to produce a research paper out of the volcanoes we analyzed using my models and ASTER.
For any questions, feel free to contact me at bschuler108@gmail.com or BES137@pitt.edu.