Towards Drone-based Mapping of Volcanic Gases using Gas Tomography

arXiv:2605.27180 · cs.RO · Submitted 2026-05-26 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "Towards Drone-based Mapping of Volcanic Gases using Gas Tomography".

Dev: Volcanoes emit large amounts of CO2, directly influencing human lives, and mapping volcanic gas emissions helps to forecast eruptions and understand their impact on climate and the environment.

Rosa: First, who's behind it and why it matters.

Title and authors: Rosa: Okay, shifting gears slightly, let’s look at the title and authors of this paper we've been discussing, "Towards Drone-based Mapping of Volcanic Gases using Gas Tomography." It really sets the stage for what they are trying to achieve.

Dev: The title itself clearly signals that this is about using drones to map gases from volcanoes, and the inclusion of "Gas Tomography" tells us they’re employing a specific technique to get detailed spatial information about those emissions rather than just getting a single concentration reading.

Taro: I see the authors listed—Marius Schaab, Niklas Karbach, Antonia Rabe, Thomas Wiedemann, Patrick Hinsen, Dmitriy Shutin, Thorsten Hoffmann, and Achim J. Lilienthal. I’m interested in seeing if these researchers have backgrounds that mesh well with both robotics and atmospheric science; that kind of interdisciplinary mix is what makes these kinds of complex mapping papers work.

Rosa: They certainly seem to have a strong combination of expertise, covering computation, chemistry, and navigation across the institutions listed. This suggests a solid foundation for tackling the physical challenges involved in volcanic gas mapping.

Dev: From an engineering perspective, seeing researchers from both TUM and JGU alongside DLR points to a team with deep roots in both high-precision sensing and advanced control systems, which is exactly what’s needed when you’re dealing with things like loop rates and latency mentioned earlier.

Taro: I think the implication of having this kind of diverse team is that they aren't just looking at one problem from an engineering angle but are building a system that bridges the gap between the physical world, like volcanic emissions, and the computational methods needed to interpret that data.

Rosa: Precisely, and when you look at their specific research focus on remote gas sensing, it seems they’re aiming to create a tool that significantly reduces risk in monitoring while also providing richer environmental data than traditional methods.

Dev: That's what I mean; the goal isn't just to monitor the volcano better, but to generate actionable maps of emissions, which feeds directly into forecasting and understanding climate impact as mentioned in the abstract.

Taro: So, when you look at their overall approach—combining drone flight paths with tomography—it suggests they are building a system designed not just for data collection but for spatial inference about a dynamic source.

Rosa: Right, and that’s what we need to keep focused on as we go through the paper; how this combination of methods actually translates into practical, deployable monitoring tools outside of some perfect simulation.

Dev: That's the central question for me, because if it only works perfectly in a lab setting, it doesn't help us when you’re looking at real-world conditions where wind and turbulence are unpredictable.

The paper's summary: Rosa: Now that we’ve talked about the setup, let’s talk about what the paper actually summarized regarding the core research of "Towards Drone-based Mapping of Volcanic Gases using Gas Tomography." Essentially, they are showing how their method addresses the fundamental issue where drone sensors fail due to rotor downwash.

Dev: They summarize that drone-mounted in-situ sensors couldn't detect CO2 emissions because the aerodynamic disturbance from the rotors disperses the gas plume before it hits the sensor, but they show that open-path sensing successfully enabled remote gas distribution mapping.

Taro: That’s a key distinction; they’re saying that while direct measurement near a source is tricky with drones, measuring along an open path allows you to capture the overall distribution, which is much more informative for understanding the volcano's output profile.

Rosa: They detail their methodology by introducing a novel model-based gas tomography reconstruction approach that uses a Lagrangian model to compensate for wind-induced advection, which is what lets them correct the data and create stable maps.

Dev: That Lagrangian model is the technical core they use to handle the wind, and it’s designed to compensate for how the gas moves due to both wind and diffusion as it travels between the sensor and the reflector.

Taro: The paper highlights that by including assumptions described in prior work about these models, they make sure their gas distribution mapping becomes stable against different discretizations or resolutions of the map, which speaks to a level of mathematical rigor they’re applying.

Rosa: So, in simple terms, the summary is this: drone sensors fail locally due to wind turbulence, but by using open-path sensing and then applying a model that accounts for wind movement—the Lagrangian model—they can reconstruct an accurate map of where the gases are distributed across an area.

Dev: It’s about moving from a single point measurement to a spatial distribution map, which is fundamentally different in terms of what information you get back from the sensor deployment.

Taro: It's about gaining context; instead of just knowing *if* gas is there at one spot, you get an idea of the pattern of emissions across that area.

Rosa: That’s the main takeaway from their summary—they’ve developed a method to overcome the physical hurdle of aerodynamic interference and translate remote measurements into meaningful spatial distribution data.

Dev: And it sets up a very clear picture for us regarding how they are trying to make this work in practice, which brings us nicely into how they actually improve the system.

The paper's improvements: Rosa: Moving on to what the authors suggest as improvements for "Towards Drone-based Mapping of Volcanic Gases using Gas Tomography," they point towards refining their current approach to make it even more practical and accurate. They focus heavily on optimizing the compensation mechanisms.

Dev: They explicitly suggest implementing a machine learning optimization routine specifically to tune the parameters within the Lagrangian compensation method, particularly optimizing that time delay parameter t, which is crucial for minimizing discrepancies between measurements and reconstructions.

Taro: Tuning that time delay sounds like they’re trying to find the sweet spot where the model best reflects reality, especially since they acknowledge that their current compensation might not be perfect under all conditions.

Rosa: They also recommend developing a more advanced wind model, which is necessary because the current one is what allows them to compensate for advection, but a better model would certainly improve accuracy when dealing with complex atmospheric dynamics.

Dev: From an engineering standpoint, improving the wind model means you are reducing the reliance on just a simple compensation equation and instead having a more sophisticated understanding of how those atmospheric forces actually behave in real-time.

Taro: I wonder if they should also consider incorporating data from other sources, maybe combining this with those short-snapshot in-situ sensor readings to create that quantitative metric we discussed earlier for validation.

Rosa: Combining the outputs to compare the map against these short snapshots would give them a direct way to evaluate how well the tomography is actually capturing the true source versus just noise or dispersion effects.

Dev: If they can establish that quantitative metric, it moves this from being just an internal validation exercise to a strong tool for assessing the overall reliability of their remote sensing system.

Taro: That kind of cross-validation is essential for any autonomous system deployed in an unpredictable environment; you need to know when the model is trustworthy versus when you need to fall back on direct, albeit limited, measurements.

Conclusion: Rosa: So, wrapping up this discussion on "Towards Drone-based Mapping of Volcanic Gases using Gas Tomography," the paper concludes by summarizing the main implications of their work for volcanic monitoring and future application. They emphasize that this approach successfully overcomes propeller downwash limitations when used with open-path TDLAS sensing and Lagrangian compensation.

Dev: The main implication is that we have a viable pathway to generate spatially distributed gas maps from aerial platforms, which could significantly enhance our ability to forecast eruptions and better understand the local environmental impact.

Taro: It’s about giving us a more comprehensive view of the emissions pattern, not just a point reading, which is valuable for long-term geological studies of these active sites.

Rosa: And they suggest that future work should involve an optimization method to minimize error by varying t and developing a more advanced wind model to refine the accuracy further.

Dev: That points toward continuous improvement in the system's predictive capability, focusing on refining those parameters for better real-world performance.

Taro: I think having that kind of refinement roadmap is what makes this paper relevant beyond just proving a single concept works; it shows they are thinking about how to harden the system for real operational use.

Rosa: Indeed, "Towards Drone-based Mapping of Volcanic Gases using Gas Tomography" provides a concrete framework for how to use remote sensing to get detailed spatial data from challenging environments like active volcanoes. We'll keep an eye on their next steps as they work on those optimizations.

Dev: Agreed, it’s a solid contribution to the field because it shows that combining advanced sensing and computational modeling can yield useful spatial reconstructions even when direct measurements are hampered by physical factors.

Taro: It's a good piece of work showing the practical application of complex physics to solve an environmental problem.

Rosa: And that’s a wrap on this discussion for now. We’ll be ready for the next paper when we get it, Dev and Taro, keep an eye out!

School of Computation, Information and Technology, Technical University of Munich (TUM) · Department of Chemistry, Johannes Gutenberg-University (JGU) · Institute of Communications and Navigation, German Aerospace Center (DLR)

cs.RO

Submitted: 2026-05-26

Updated: 2026-06-01

DOI: 10.1109/ISOEN68725.2026.11665396

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 69/100

The gist: Volcanoes emit large amounts of CO2, directly influencing human lives, and mapping volcanic gas emissions helps to forecast eruptions and understand their impact on climate and the environment.

Key concepts

Gas Tomography: This is a specific technique used in the paper to get detailed spatial information about volcanic gas emissions rather than just a single concentration reading. It involves using drone-mounted sensors and modeling the gas distribution to create maps of where the gases are located across an area.
Lagrangian Model: This is a model used in the research to compensate for wind-induced advection, which is how gas moves due to wind. It helps correct the data collected by sensors, allowing researchers to create stable maps of gas distribution even when facing unpredictable atmospheric forces.
Open-Path Sensing: This method allows for remote gas distribution mapping by measuring emissions along an open path rather than directly near a source. The hosts noted that this is key because drone sensors fail locally due to rotor downwash, but open-path sensing captures the overall gas plume distribution.
Drone Downwash: This refers to the aerodynamic disturbance created by a drone's rotors. This disturbance disperses volcanic gas plumes before they reach in-situ sensors, which is why direct measurement near a source is difficult for drones.

Terminology

Gas Tomography — 43× in this paper · explained in 2 episodes
Accuracy — 3× in this paper · explained in 1 episode
Core — 2× in this paper · explained in 2 episodes
Johannes Gutenberg-University — 1× in this paper
Technical University of Munich — 1× in this paper
Tunable Diode Laser Absorption Spectroscopy — 1× in this paper

Summary

Volcanoes emit large amounts of CO2, directly influencing human lives, and mapping volcanic gas emissions helps to forecast eruptions and understand their impact on climate and the environment. While drone-based gas sensing significantly reduces risks in volcanic monitoring, it faces technical limitations when measuring gas because rotor downwash disperses the gas plume before detection. Gas Tomography using remote gas sensing addresses this challenge. At the Salinelle dei Cappuccini mud volcanoes, the authors demonstrate that while drone-mounted insitu sensors failed to detect CO2 emissions due to aerodynamic disturbance, open-path sensing successfully enabled remote gas distribution mapping. They present a novel model-based gas tomography reconstruction approach that incorporates a Lagrangian model to compensate for wind-induced advection. The resulting gas distribution maps align with manually collected insitu measurements, confirming that model-based gas tomography effectively overcomes downwash limitations and enables accurate mapping of volcanic emissions.

The paper focuses on the challenges of drone-based open-path gas distribution mapping using model-based gas tomography, specifically targeting CO2 degassing close to the ground at the mud volcanoes Salinelle dei Cappuccini. The mud volcanoes emit mainly CO2 but also minor amounts of CH4, H2, and He. The amount of CO2 varies between approximately 2 × 10 squared kg/d to 2 × 10 4 kg/d, depending on the state of activity. The authors conducted different experiments to evaluate the differences between drone-based and manual gas sampling, as well as in-situ and open-path sensing. They demonstrate the advantages of drone-based sampling, especially open-path sampling, over manual sampling. Finally, they introduce a simple method to mitigate the impact of wind on our gas tomography.

Remote gas sensing addresses downwash issues because the TDLAS sensor and the reflector remain outside of the plume, not affecting the natural dispersion of the gas plume. Furthermore, remote sensing increases detection chances: while an in-situ sensor only detects the gas concentration at a single point, a remote sensor integrates the gas concentration along its beam. This increased footprint allows for gas tomography, which can be mitigated by collecting multiple measurements from different angles, enabling us to perform gas tomography and reconstruct a gas distribution map in a given area. By including model assumptions as described in prior work, the gas distribution mapping becomes stable against different discretisations or resolutions of the map.

The authors collected two datasets using the TDLAS sensor to reconstruct 2D gas distribution maps close to the ground. For one dataset, they positioned the sensor at five different locations and aimed at a handheld reflector, allowing for a two-dimensional tomography slice at a plane parallel to the ground at approx. 1.5 m height. For the second dataset, they used a reflector mounted on the drone, which allowed for rapid relocation of the reflector, even to different altitudes. Both datasets required around 1 hour to acquire.

To compensate for changing wind conditions that lead to false reconstructions in gas tomography, they employ a Lagrangian model describing dispersion by wind and diffusion. They compensate by offsetting the sensor and reflector locations: x new t = x t − wt∆t, where ∆t is chosen as 3 s for the collected data at the mud volcanoes, meaning it took around 3 s for the emitted gas to reach our measurement plane. wt is the averaged wind speed over timespan [t − 3 s, t].

The results show that the locations of the gas detections generally match the gas distribution map. The authors note that our gas tomography reconstruction adjusts for wind, while in-situ sensor detections are plotted at their exact point of acquisition, and that "gas tomography captures a time-averaged gas distribution map, whereas the in-situ sensor captures short snapshots at specific locations. Hence, gas tomography and in-situ sensing each measure different properties of the mud volcanoes, such that the significance of a direct comparison is limited. However, since there is no ground-truth gas map of the mud volcano, a comparison between in-situ sensing and gas tomography serves as a reasonable option to evaluate the validity of the gas tomography. They also observe that both datasets have similar maximum values, and CO2 occurs at approximately the same locations, although they acknowledge that Human precision and the reflector size limit the maximum range of the open-path sensor; thus, the resulting reconstruction area of the drone-based dataset is smaller." The conclusion is that propeller downwash limits in situ detection of gas sources beneath a drone, and this limitation is overcome by open-path TDLAS sensing combined with a Lagrangian wind compensation method. They suggest future work should involve an optimization method to minimize error by varying ∆t and a more advanced wind model. Additionally, future measurement campaigns should include longer in-situ sensing periods with more sensors to allow for a significant comparison between gas tomography and an averaged map captured by in-situ sensors.

Improvements for AI systems

Here are specific improvements to AI systems based on the findings of this scientific paper, and what those improved systems could achieve:

Improve drone-based remote sensing for volcanic monitoring by incorporating a model-based gas tomography reconstruction approach that explicitly compensates for aerodynamic disturbances (rotor downwash) using a Lagrangian wind model. This allows the system to generate accurate 2D gas distribution maps from open-path TDLAS measurements, overcoming the limitation where in-situ sensors fail to detect low-concentration or small source emissions beneath the drone.
Develop an AI algorithm that integrates multiple open-path sensor datasets (from different drone flight paths or manual tripod placements) to perform robust, model-based gas tomography reconstruction. This improved system could generate a stable, high-resolution 2D gas distribution map of volcanic emissions close to the ground, which is significantly more accurate than single point measurements from in-situ sensors alone.
Create an AI framework that utilizes the Lagrangian wind compensation method (Equation 1) to automatically correct for changing wind fields during data acquisition. This system could process raw TDLAS measurements and generate a time-averaged gas distribution map, effectively mitigating the false reconstructions caused by atmospheric advection, leading to a more reliable source localization estimate.
Improve the ability of autonomous monitoring systems to distinguish between direct source emissions and atmospheric dispersion effects by comparing the output of model-based gas tomography with short-snapshot in-situ sensor readings. This improved system could provide a quantitative metric for evaluating the validity of remote sensing data versus localized measurements, helping to validate or refine future gas emission models.
Implement a machine learning optimization routine to tune the parameters within the Lagrangian compensation method (specifically optimizing the time delay parameter, ∆t) to minimize discrepancies between the TDLAS measurements and their tomographic reconstructions. This would enhance the precision of mapping volcanic plumes under varying wind conditions, leading to more accurate forecasting of eruption characteristics.
Build an AI-driven data processing pipeline capable of automatically generating source localization probabilities based on the reconstructed gas distribution maps. This system could be used to rapidly identify potential degassing spots and estimate release rates for specific gases (like CO2) in real-time from aerial drone surveillance data, significantly speeding up early warning systems for volcanic activity.