the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Oct 2024
Autonomous and efficient large-scale snow avalanche monitoring with an Unmanned Aerial System (UAS)
Abstract. Large-scale monitoring is a crucial task for managing remote mountain environments, especially for hazardous events such as snow avalanches, debris flows or rockslides. One key information for safety-related applications is large-scale information on released avalanches. As avalanches occur in remote and potentially dangerous locations this data is difficult to obtain. Uncrewed fixed-wing aerial vehicles, due to their low cost, long range and high travel speeds are promising platforms to gather aerial imagery to map avalanche activity. However, autonomous flight in mountainous terrain remains a challenge due to the complex topography, regulations, and harsh weather conditions. In this work, we present a proof of concept system that is capable of safely navigating and mapping avalanches using a fixed-wing aerial system (UAS) and discuss the challenges arising for operating such a system. We show in our field experiments that we can effectively and safely navigate in steep mountain environments while maximizing the map quality and efficiency while meeting regulatory requirements. We expect our work to enable more autonomous operations of fixed-wing vehicles in alpine environments to maximize the quality of the data gathered. By enabling the acquisition of frequent and high quality information on avalanche activity, such drone systems would have a large impact of safety critical applications such as avalanche warning, mitigation measure planning or hazard mapping.
Competing interests: One of the co-authors is a member of the editorial board of Natural Hazards and Earth System Sciences.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.- Preprint
(25891 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2024-2728', Thomas Van Der Weide, 04 Feb 2025
Thank you for your work and taking the time to share this research. The paper is structured well and has high quality figures which show the potential for a significant improvement in UAV mapping techniques.
3.1 route planning optimization section
It does make sense to try and maximize the number of ROIs visited per flight but since this can be calculated before take off I'd also like to see it optimized around the number of flights needed to cover all the ROIs. For instance since it’ll take two flights to cover all four ROIs could the route optimization cover E and I with a greater % coverage (mapping cost = 6) in one flight and D and G in another? This would also help ensure there's plenty of flight time since the optimization route from I to D passes right next to A. I'd also like to see some discussion about using a somewhat flexible but realistic flight time/range estimate to inform the route planning optimization instead of these being hard coded values.
3.2 Platform
Does it cruise at the same speed when collecting data as when traversing between study areas?
What is the DEM resolution and what terrain_spacing value was used?
You mention a 50km range but is this with 20% of the battery remaining when landing or how was this estimated?
4.3 Active Mapping
This is done in real-time using the onboard computer which is very cool and would definitely be a useful tool to ensure an area is fully covered before returning to home. In Figure 17 we see that the active mapping collected 67 images over the ROI instead of the 167 images mentioned before but both of these values are still significantly more than the 47 collected in the coverage mapping survey. Were these two surveys the same flight time and range so the resulting difference in the image count is solely because of how well the active mapping performed? Would a cross-grid coverage map have been a better comparison? Also, it should be mentioned somewhere what side lap % was used for the traditional coverage mapping survey.
Because you're working in Avalanche terrain I'm understanding that there's no mention of ground control points for a qualitative comparison between these two mapping methods and their resulting point clouds. One advantage of the coverage mapping method is that (ignoring wind) the UAV is mostly level when imaging which means the onboard GPS has a better sky view and position estimate. I'd be very interested to see some information about how imaging the hillside at an angle influenced your GPS accuracy which translates into your point cloud reconstruction uncertainty. Did you adjust for this difference in positional uncertainty between the methods in Agisoft? Since you used PPK were the images uploaded to Agisoft using their actual positional uncertainties? The Agisoft processing reports should also be included.
5. Discussion
Could this same mapping approach with high roll angles to image the hill side be calculated and used as a mapping tool in the mission planning software prior to taking off?
For more autonomous operations Active Mapping and the on-board computer could still be used to calculate the estimated coverage and direct the UAV to add flight lines to fill in the gaps after the survey if needed and if battery levels are sufficient?
6. Conclusion
"The field tests have also shown that the active mapping planner is capable of mapping an ROI with better reconstruction results, and potentially more efficiently." I do not agree that there's sufficient evidence provided to say that this method gives better reconstruction results. To do that you'd need ground control and validation points. This method has shown that with minimal user input active mapping can legally and safely fly a survey with good coverage.Citation: https://doi.org/10.5194/egusphere-2024-2728-RC1 -
AC1: 'Reply on RC1', Jaeyoung Lim, 20 Apr 2025
Thank you for your work and taking the time to share this research. The paper is structured well and has high quality figures which show the potential for a significant improvement in UAV mapping techniques.
Thank you for the positive response and valuable feedback. We have tried to clarify the questions and respond to the valuable feedback received by the reviewr as best as we can.
It does make sense to try and maximize the number of ROIs visited per flight but since this can be calculated before take off I'd also like to see it optimized around the number of flights needed to cover all the ROIs. For instance since it’ll take two flights to cover all four ROIs could the route optimization cover E and I with a greater % coverage (mapping cost = 6) in one flight and D and G in another? This would also help ensure there's plenty of flight time since the optimization route from I to D passes right next to A.We thank the authors for the valuable feedback regarding the route planning. The route optimization section was written to a) demonstrate what we envision operating a long endurance autonomous fixed-wing vehicle, and b) show that using fixed-wing vehicles are necessary for the application unlike multirotor vehicles. We have simplified the problem to have uniform mapping cost and simplified the vehicle to consider only range, rather than energy to demonstrate these points.
While global route optimization would be an interesting direction, it is a challenging problem, possibly worthy of more detailed future research. The main value of being able to visit multiple ROIs is the fact that the mapping can happen within a narrow weather window. Moreover, global route optimization spanning over multiple flights can no longer be considered as a simple single-tour orienteering problem. While this aspect was not fully explored in the manuscript, it would be an interesting direction to expand the formulation of the problem to extend the route optimization using novel formulations such as multi-path orienteering. An in-depth exploration of this would be beyond the scope of this work.
I'd also like to see some discussion about using a somewhat flexible but realistic flight time/range estimate to inform the route planning optimization instead of these being hard-coded values.
Thank you for the valuable feedback. As the reviewer has pointed out, a more realistic metric would be to use the amount of energy stored in the vehicle. However, the range of a fixed-wing/VTOL vehicle can depend on various factors on the vehicle, such as aerodynamic efficiency, propulsion efficiency, as well as environmental factors such as wind or the geometry of the reference path.
Therefore, we have chosen to use 3D distance as a budget constraint as it is generally more intuitive, and because it is not vehicle-specific. The underlying method could certainly be applied with other values of the distance budget.
For the improved manuscript, we plan to include a discussion regarding the realistic range and flight time and the implications for route optimization.
Does it cruise at the same speed when collecting data as when traversing between study areas?
The platform is assumed to be flown at a specific air-relative speed (typically the maximum in-air range speed) throughout the whole mission, as fixed-wing vehicles are normally optimized for a certain speed. This would mean that the vehicle will fly at different speeds relative to the ground, depending on the wind conditions. For the revised manuscript, we plan to mention this more explicitly for better clarity.
What is the DEM resolution and what terrain_spacing value was used?
The DEM resolution was selected as 10m, as this limits the amount of data storage required to be small enough for the onboard memory, but also high enough to accurately represent the terrain for navigation. The DEM is loaded as one file, therefore, there is no terrain_spacing value defined.
You mention a 50km range but is this with 20% of the battery remaining when landing or how was this estimated?
The range of the vehicle was defined with a sufficient margin (40%) for a typical mission, and not specifically based on the state of charge of the battery. This was due to the high uncertainty given the large elevation changes and uncertain wind conditions the vehicle would experience during the mission.
This is done in real-time using the onboard computer which is very cool and would definitely be a useful tool to ensure an area is fully covered before returning to home. In Figure 17 we see that the active mapping collected 67 images over the ROI instead of the 167 images mentioned before but both of these values are still significantly more than the 47 collected in the coverage mapping survey. Were these two surveys the same flight time and range so the resulting difference in the image count is solely because of how well the active mapping performed? Would a cross-grid coverage map have been a better comparison? Also, it should be mentioned somewhere what side lap % was used for the traditional coverage mapping survey.
For the coverage plan mapping survey, the sweep spacing was set as 67.4m, which sets the overlap percentage at 70%. This is compatible with the industry standard of 60-80% overlap between sweeps. The trigger rate was set at 1Hz, in order to keep it comparable with the active mapping approach. As mentioned in the manuscript, a path planner was used to find a feasible path between the sweep lines.The active mapping approach triggers images consistently at 1Hz. However, given that the coverage map is not triggering on the path in between the sweep lines, the image count of the active mapping ends up being significantly higher compared to a coverage survey.
Figure 15 shows the comparison of the coverage and predicted uncertainty as the flight time progresses. As there is no explicit termination criteria for the active mapping experiment, we can compare the evolution of uncertainty and coverage as if the survey and active mapping has started at the same time.
We believe that it is more important to optimize for flight time, as this prioritizes information gathered during the survey, rather than optimizing the computation cost that is required for the photogrammetry reconstruction after the survey.
A cross-grid coverage map was not selected as a baseline method to compare for two reasons. a) It is not possible to fly in the normal direction of the sweep patterns, as the terrain is steeper than the maximum flight path angle of the vehicle. b) Drawbacks of cross-grid coverage planning are identical to coverage planning, in the sense that cross-grid coverage would also be planned based on a 2D projection of the environment, and it is not able to react to external disturbances. The main advantage of using the active mapping is the adaptability to disturbances on-the-fly and the ability to efficiently map steep terrain without complicated expert planning, which can be tricky in steep environments.
We greatly appreciate the reviewer's comment on this aspect of active mapping. We will mention why cross-grid coverage planning was not used as a baseline, and make it more clear on the images acquired of the results.
Because you're working in Avalanche terrain I'm understanding that there's no mention of ground control points for a qualitative comparison between these two mapping methods and their resulting point clouds. One advantage of the coverage mapping method is that (ignoring wind) the UAV is mostly level when imaging which means the onboard GPS has a better sky view and position estimate. I'd be very interested to see some information about how imaging the hillside at an angle influenced your GPS accuracy which translates into your point cloud reconstruction uncertainty. Did you adjust for this difference in positional uncertainty between the methods in Agisoft? Since you used PPK were the images uploaded to Agisoft using their actual positional uncertainties? The Agisoft processing reports should also be included.
Thank you for the valuable feedback. The GPS position was acquired using RTK. While this was due to the practical limitations of the platform, due to the lack of ground control points the evaluation only requires relative accuracy. We used a helical GNSS antenna which is quite resistant to orientation changes, and no significant GNSS outages were noted during the flights, in either banked or level flight.For further development of this platform for commercial purposes, adding a PPK capability for the registration would be possible, but this would not influence the work that is done in this paper, as the PPK processed data would be unavailable for the onboard planner.
We would be happy to include the Agisoft processing reports and provide an improved description of the GNSS recording and processing if it is still relevant.
Could this same mapping approach with high roll angles to image the hill side be calculated and used as a mapping tool in the mission planning software prior to taking off?We appreciate the valuable suggestion from the reviewer. It would definitely be possible to plan high roll angle maneuvers to be included in the planning prior to the mission. However, because high roll angle maneuvers are states with high acceleration, it may be challenging to follow the plan, especially if the wind conditions vary, or some disturbance causes the vehicle to deviate from its initial plan. One of the significant advantages of the active planning approach is to respond during flight to off-nominal image capture conditions caused by in-flight deviations. While it may be interesting to explore this idea, we expect that the active mapping method would be more robust compared to a preplanned approach, as it can adapt to disturbances and deviations during the survey mission.
For more autonomous operations Active Mapping and the on-board computer could still be used to calculate the estimated coverage and direct the UAV to add flight lines to fill in the gaps after the survey if needed and if battery levels are sufficient?Thank you for the valuable suggestion. Adding sweep patterns would be a feasible approach, where the actions of the onboard planners would become addition of sweep patterns rather than planning maneuvers for the proposed approach.
However, given that the vehicle is operating under tight altitude constraints, adding a sweep line that can be executed only at the end of the survey may require the vehicle to fly large detours. Moreover, coverage planning generates sweep lines based on a 2D projection of the ROI, it may be quite hard to find feasible sweep lines that can be added without significant detours due to the tight AGL constraints posed by regulation.
"The field tests have also shown that the active mapping planner is capable of mapping an ROI with better reconstruction results, and potentially more efficiently." I do not agree that there's sufficient evidence provided to say that this method gives better reconstruction results. To do that you'd need ground control and validation points. This method has shown that with minimal user input active mapping can legally and safely fly a survey with good coverage.Thank you for the valuable feedback. Our descriptions of ‘better reconstruction’ may have been too general, as we did not specifically calculate the true reconstruction quality using features such as ground control points. Our results do show that the active planning method results in more complete coverage of the target region. We generally agree with the feedback and will revise the manuscript to more accurately and specifically reflect our results.
Citation: https://doi.org/10.5194/egusphere-2024-2728-AC1
-
AC1: 'Reply on RC1', Jaeyoung Lim, 20 Apr 2025
-
RC2: 'Comment on egusphere-2024-2728', Madeline Lee, 09 Mar 2025
The authors have presented a good discussion on the importance of balancing efficiency and safety of UAV-flight operations in rough terrain with achieving the objective of reliable remotely-sensed imagery for avalanche monitoring and mapping. The presented methods may prove to be highly beneficial to reduce scientific data loss when a sensor trigger fails in flight or optimize flight paths to reduce uncertainty / increase image resolution based on real-time feedback of conditions. However, further field testing and quantitative image analysis is needed to fully demonstrate the advantage of the optimization algorithms. Below are minor considerations for the authors.
I. In-text Recommendations
L35: Replace 'manned' with 'crewed' considering UAS is uncrewed aircraft.
L88: Discussion on photogrammetry is presented, however there is no mention in the paper on the implementation of LiDAR and the resolution of photogrammetry to LiDAR, which would be relevant as a potential dataset for validation/comparison. For example: Solbakken, E. et al. (2024) Repeated UAS Lidar Scanning of Snow Surface to Support Site Specific Avalanche Warning.
L168: "We describe the system by describing the system".
L198: The mapping cost is highly variable, where it will differ per ROI. Further, there is no mention on flight direction, which is highly impacted by head or tail winds, especially within valleys.
L263: Acknowledgement of "unaccounted trees and obstructions not included in DEM", however pointing out relevant 'obstructions' is recommended. As this is a paper focused on avalanche terrain, there is no mention of the impact of snow height (except 'snow cover' in conclusion) and avalanche debris. The DEM will represent a 'snow free' interpretation of the surface, however the UAV will be operated above snow covered terrain. This additional height above surface should be acknowledged earlier in the paper and impact on flight altitude above ground level. Additionally, in zones of high avalanche activity, the surface is dynamic and large rock debris may exist causing the DEM to be out of date.
L345: "The extent was outlined by hand". On what dataset (aerial, satellite) was the extent determined?
L438: Consider a comparison with at least a grid difference where areas of disagreement or lower uncertainty can be more clearly identified. Perhaps the areas of disagreement are correlated to slope or flight angle, etc.
L444: "fraction of the time" - What is the time difference needed to accomplish the 67 active mapping images versus the 47 coverage mapping images?
L513: Consider additional quantitative means to validate the quality of the reconstruction. In addition to simple grid differencing, other remotely sensed/geophysical data sets may be acquired by drone. LiDAR, as suggested above, may have a very high resolution result that can act as a baseline to which active and coverage mapping results is compared. Another is ground penetrating radar, GPR. Although GPR is suited for rotary-wing and not ideal for extended coverage by a FW, it can be used for a small scale validation of photogrammetry results. It can also validate the DEMs since GPR will provide a snapshot of the current ground conditions below the snow cover.
II. Suggestions for Figures
Figure 1. Captions under each image pane in addition to the full caption is redundant, and can be minimized to a, b, c
Figure 2. Although mentioned in text, the addition of a scalebar to b) to provide context on wingspan / aircraft size.
Figure 3. ROIs are labeled
Figure 16. Label which sides of the figure are active and coverage mapping with a, b or left, right. Recommend same colour scale for all grids, as viridis is perceptually uniform sequential but rainbow/spectral is not, so anomalies are presented differently.
Citation: https://doi.org/10.5194/egusphere-2024-2728-RC2 -
AC2: 'Reply on RC2', Jaeyoung Lim, 20 Apr 2025
The authors have presented a good discussion on the importance of balancing efficiency and safety of UAV-flight operations in rough terrain with achieving the objective of reliable remotely-sensed imagery for avalanche monitoring and mapping. The presented methods may prove to be highly beneficial to reduce scientific data loss when a sensor trigger fails in flight or optimize flight paths to reduce uncertainty / increase image resolution based on real-time feedback of conditions. However, further field testing and quantitative image analysis is needed to fully demonstrate the advantage of the optimization algorithms. Below are minor considerations for the authors.
We thank the reviewer for the positive response. We agree that further field testing would fully demonstrate the advantage of the algorithm. While additional flight tests this season would not be possible due to the snow season being concluded, we plan to work on future research directions to advance the platform.
L35: Replace 'manned' with 'crewed' considering UAS is an uncrewed aircraft.
Thank you for the recommendation. We will fix the terminology for UAS from ‘manned’ to ‘crewed’ in the revised manuscript.
L88: Discussion on photogrammetry is presented, however there is no mention in the paper on the implementation of LiDAR and the resolution of photogrammetry to LiDAR, which would be relevant as a potential dataset for validation/comparison. For example: Solbakken, E. et al. (2024) Repeated UAS Lidar Scanning of Snow Surface to Support Site Specific Avalanche Warning.
Thank you for the valuable feedback. We agree that lidar surveys and comparison to photogrammetry reconstruction results would have resulted in a better dataset for validation and comparison. However, we believe that the suitability of photogrammetry for snow depths has been investigated and shown by previous works [1, 2].
Moreover, the platform is unfortunately not capable of flying a relevant Lidar. Previous work (such as https://nsidc.org/data/aso) using airborne lidar sensors utilises large, high-power and high-cost LiDAR devices on manned-scale aircraft. These sensors are too large for the platform used in this work. Smaller, lower-cost sensors are becoming available but are still significantly more expensive than the camera used in this work. Generating a complete mapping solution with low-cost Lidar is currently a topic of interest for us, but unfortunately is out of the scope of this paper.
We would like to stress that the platform demonstrates the first regulation-compliant autonomous flight in steep terrain, under the new EU regulations [3].
[1] Eberhard, L. A., Sirguey, P., Miller, A., Marty, M., Schindler, K., Stoffel, A., & Bühler, Y. (2020). Intercomparison of photogrammetric platforms for spatially continuous snow depth mapping. The Cryosphere Discussions, 2020, 1-40.
[2] Bühler, Y., Adams, M. S., Bösch, R., & Stoffel, A. (2016). Mapping snow depth in alpine terrain with unmanned aerial systems (UASs): potential and limitations. The Cryosphere, 10(3), 1075-1088.
[3] European Union (2019), Commission implementing regulation (eu) 2019/947 of 24 may 2019 on the rules and procedures for the operation of unmanned aircraft. Official Journal of the European Union, 2019, vol. 62, pp. 45–71.
L168: "We describe the system by describing the system".
Thank you for pointing out the error in the manuscript. We will address this mistake in the updated manuscript.
L198: The mapping cost is highly variable, where it will differ per ROI. Further, there is no mention on flight direction, which is highly impacted by head or tail winds, especially within valleys.
The comment on mapping cost being highly variable is accurate. We have simplified the example to only consider 3D distance, considering the kinematics of the vehicle. In realistic scenarios, many more factors can be accounted for, as pointed out by the reviewer, such as the area of the ROI, wind direction. However, with state-of-the-art methods, this can be challenging. First, wind fields are highly non-uniform due to the rugged terrain around avalanche risk terrain, but current weather predictions only provide kilometer-scale wind information. Therefore, the wind field is highly uncertain when starting the survey. Second, there is no reliable method that can produce coverage paths that respect the vehicle kinematics with tight AGL constraints as required for the regulation. Moreover, it is not predictable due to the active mapping path being unpredictable after reacting to unknown disturbances. Therefore, the analysis was simplified to demonstrate using long-range aerial vehicles is advantageous for the vision we are presenting.
L263: Acknowledgement of "unaccounted trees and obstructions not included in DEM", however pointing out relevant 'obstructions' is recommended. As this is a paper focused on avalanche terrain, there is no mention of the impact of snow height (except 'snow cover' in conclusion) and avalanche debris. The DEM will represent a 'snow free' interpretation of the surface, however the UAV will be operated above snow covered terrain. This additional height above surface should be acknowledged earlier in the paper and impact on flight altitude above ground level. Additionally, in zones of high avalanche activity, the surface is dynamic and large rock debris may exist causing the DEM to be out of date.
Thank you for pointing out the accuracy of the DEM. Indeed, unmodelled objects in the terrain, such as vegetation, artificial buildings or other objects may result in obstructions as pointed out in the manuscript. We will include a note to this effect in the revised manuscript.
For safety, the minimum distance of 50m serves as a margin to keep the vehicle safe from unknown obstacles such as vegetation. The unmodelled obstacles or outdated DEM may influence the accuracy of the predicted uncertainty. A feasible research direction would be to add onboard perception capabilities on the vehicle to correct the wrong uncertainty predictions based on the updated geometry. However, we would like to stress that using DEMs would provide the best uncertainty prediction given the information that is available. Moreover, specifically on snow covered terrain susceptible to avalanches, the change in the normal direction would not be significant, making the affect on the predicted uncertainty minimal.
L345: "The extent was outlined by hand". On what dataset (aerial, satellite) was the extent determined?
The extent was outlined by hand over the orthoimagery, where the avalanche outline would exist, and taking into account operational constraints of the experiments. For example, the safety pilot was required to be able to discern the vehicle orientation at all times, limiting the operation distance to less than 1 km.
L438: Consider a comparison with at least a grid difference where areas of disagreement or lower uncertainty can be more clearly identified. Perhaps the areas of disagreement are correlated to slope or flight angle, etc.
Thank you for the valuable feedback. We agree that comparing the grid difference would be valuable. However, given that the uncertainty metric does not take into account the appearance of the texture, low predicted uncertainty may not necessarily correlate with the predicted uncertainty. Moreover, the fisher information metric represents the aggregate information, incorporating viewpoints from diverse distances and flight angles, finding correlation to a single quantity may be challenging.
L444: "fraction of the time" - What is the time difference needed to accomplish the 67 active mapping images versus the 47 coverage mapping images?
The flight time comparison between active mapping and coverage mapping is done in Figure 15. The time it took to acquire 67 images for active mapping was at 45.63s and 47 images for coverage mapping at 135.73s. The active mapping was run up to a similar time to coverage planning, such that the incremental mapping capabilities of the active mapping planner could be compared. We will explicitly mark the timestamp in Figure 15 for the reconstructions shown in Figure 17, such that the flight time and number of images gathered are more intuitive.
L513: Consider additional quantitative means to validate the quality of the reconstruction. In addition to simple grid differencing, other remotely sensed/geophysical data sets may be acquired by drone. LiDAR, as suggested above, may have a very high resolution result that can act as a baseline to which active and coverage mapping results is compared. Another is ground penetrating radar, GPR. Although GPR is suited for rotary-wing and not ideal for extended coverage by a FW, it can be used for a small scale validation of photogrammetry results. It can also validate the DEMs since GPR will provide a snapshot of the current ground conditions below the snow cover.
Thank you for the valuable suggestion on evaluating the quality of photogrammetry results. Unfortunately, we are no longer able to access the avalanche terrain, due to the fact that the season has passed. We also believe that using photogrammetry for snow depth estimation has been sufficiently shown (see comment to lidar above), and the primary goal of this paper is to demonstrate a method for efficiently collecting photogrammetry data using a UAV. We plan to continue our research on this topic and further evaluate the quality of the photogrammetry using different modalities of sensors.
Figure 1. Captions under each image pane in addition to the full caption is redundant, and can be minimized to a, b, c
Figure 2. Although mentioned in text, the addition of a scalebar to b) to provide context on wingspan / aircraft size.
Figure 3. ROIs are labeled
We thank the reviewer for pointing out improvements regarding the formatting of the paper. We will improve the pointed out aspects of the paper.
Figure 16. Label which sides of the figure are active and coverage mapping with a, b or left, right. Recommend same colour scale for all grids, as viridis is perceptually uniform sequential but rainbow/spectral is not, so anomalies are presented differently.
We thank the reviewer for pointing out improvements regarding the formatting of the paper. We will improve the pointed out aspects of the paper.
Citation: https://doi.org/10.5194/egusphere-2024-2728-AC2
-
AC2: 'Reply on RC2', Jaeyoung Lim, 20 Apr 2025
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 327 | 206 | 317 | 850 | 24 | 22 |
- HTML: 327
- PDF: 206
- XML: 317
- Total: 850
- BibTeX: 24
- EndNote: 22
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1