California Wildfire Lessons: Integrating Drones for Disaster Management

This article about how drones surveyed wildfire affected areas in California is part of my effort to bring research and knowledge to a broader audience.

The article highlights the work of Greg Crutsinger of Scholar Farms and how he coordinated drone data for an interagency task force of drone pilots and data volunteers for the image collection, processing, and visualization on behalf of Butte, Redding, and Sonoma Counties and their residents. Butte/Alameda/Stanislaus/Contra Costa Sheriff’s Offices, Menlo/La Honda Fire, Stockton/San Francisco/Union City Police, DJI, DroneDeploy, Hangar, Survae, ESRI among others contributed to the effort.

The article was originally accepted as a proceeding for the IEEE International Conference on Image Processing. We were unable to attend the conference and published it through sUAS News in June 2019 instead.

It’s dry. It’s academic. But I still hope it’s helpful.

More than anything else, I hope its sparks thought on how we coordinate police, fire, emergency agencies and communities as we approach this year’s disaster season — especially at a time where there is distance between citizens and authorities.


Drone imagery was deployed by emergency management organizations in California after the 2017 and 2018 wildfires. Although such deployments can lead to high-resolution imagery and novel visualizations for the public, it is still unclear how drone imagery can be integrated into the decision-making workflow for different stakeholders in a natural disaster including emergency managers and first responders, community members, and insurance assessors. We first use the November 2018 Camp Fire in Butte County to explore the deployment of drones in an emergency management scenario. Then, we propose both technical and organizational actions to reduce operational barriers of drone deployment in emergency management scenarios.

Index Terms — drone imagery, emergency response, emergency recovery, decision making

Figure 2. An example of panorama imagery from Paradise, CA. Accessible via

1. Introduction

2017 and 2018 resulted in the worst wildfires in the history of California. The November 2018 Camp Fire in Butte County alone displaced approximately 52k people and caused up to $10 billion in insured losses [1].

Emergency response agencies sought to collection aerial imagery in the days following the Camp Fire, as well as the June 2018 Carr Fire (in Redding County) and October 2017 Tubbs fire (in Sonoma County). Uses of disaster imagery range from helping emergency managers assess physical damage [2] and post-disaster needs [3], [4], allowing insurance adjusters to more rapidly assess the validity of claims [5], to providing visual information to evacuated residents about the state of their community prior to repopulation.

Despite this need, manned aircraft or satellite imagery may be inaccessible due to cloud cover, or lingering smoke in the case of a wildfire. Drones flying at low altitudes have the potential to survey disaster areas before other options are available.

The novelty of drone technology has created ongoing challenges in data collection and processing, such as choices in flight height, resolution, and visualization [6]. More important, the work effort may not reach the full potential of use without prior thought of how the images can be integrated into decision-making processes and emergency incident command systems.

We use the November 2018 Butte County Camp Fire as a case study; the most destructive wildfire in California history that burned over 18k structures, 14k homes, and resulted in 85 fatalities [1]. We describe the methods and challenges to collect, process, and visualize drone imagery from affected areas. We then propose both technical and organizational actions to facilitate the use of drones in future disaster management.

2. Background

On-the-ground, aerial, satellite, and drone imagery can be useful at different stages of disaster management. Imagery use ranges from increased awareness of the situation for enhancing operational decision making of where send resources like supplies and personnel [7], [8] to post-disaster event analysis and structural damage assessment [9], [10].

For example, emergency responders can access publically available traffic, and surveillance camera feeds to view real-time images of disaster areas without having to risk lives [7]. News agencies can portray crowd-sourced images from social media to community members to convey the scale and urgency of a disaster event [11]. Recovery teams can use images to survey structural damage, while insurance adjusters can remotely assess claims with images [12].

2.1 Drone imagery for disaster management

Aerial drones (a.k.a. Unmanned Aerial Vehicles or UAVs) have emerged for emergency management use in the past decade [13]. Drones have the advantage of operating in areas hazards to humans, such as investigating radiation levels at the Fukushima Daiichi power plant following the 2011 Tohoku Tsunami [6]. Drones are also capable of operating in low altitudes and poor visibility when crewed aircraft cannot fly, or when satellite imagery may not be able to penetrate the cloud cover or smoke. In recent years, drones have become highly portable in size, lower in cost driven by the consumer market, and easier to pilot with smart phone applications. Despite these advantages, drone imagery faces the same operational issues of other media.

2.2 Imagery limitations

There are limitations in both the technology and integration of imagery in disasters. First, image feeds are not necessarily technically reliable. Camera equipment may be damaged due to inclement weather while hazardous conditions (like smoke remaining in a wildfire area) may hinder the ability to capture satellite or aerial imagery remotely. Second, images from non-official sources may not be verified for the validity of its content [14]. Third, imagery that spans a city (or sometimes state) of an affected disaster area currently requires significant computational power to process the raw images into a digestible visualization [15]. Such computational power may not be available in areas with losses of power and other critical infrastructure.

Finally, the use of imagery for different management stakeholders may be developed ad-hoc during the disaster, as in the case of the drone use in the California Fires. As a result, the value of the effort to collect, process, and visualize each image disaster stakeholders is not clearly defined. To address these issues, we present the drone imagery process, potential uses, and challenges from the November 2018 Camp Fire in Butte County.

3. Methods

In the days following the Camp Fire, the emergency response agencies wanted to map as much structural damage as possible in the devastated town of Paradise, CA to assist in search and recovery efforts. The emergency management teams, which consisted of seven police and fire agencies, were granted an exemption to fly drones in the Federal Aviation Administration (FAA) temporary flight restriction (TFR) area surrounding the town of Paradise in Butte County. Drones could operate up to 300 feet above ground level (AGL) while maintaining a visual line of sight with the operator.

Up to sixteen teams, each consisting of a pilot and visual spotter, flew DJI Phantom 4 Pros and Mavic Pros along pre-established missions along the flight paths based on the Cal Fire Structure Damage Assessment map, consisting of an area of over 17,000 acres or 26.2 square miles. Each mission, which included a subset of flights, covered an area of 300 acres. For each flight, 12–20MP cameras on the drones captured overlapping photos for processing into map layers or 360-degree panoramic images Georeferenced video was also collected in 1080p. As part of the overall operation, a lead person coordinated data transfer while another spearheaded flight and safety of the teams.

Figure 1. A view of the Butte County GIS portal with 136 360 degree panoramas and 52 georeferenced video locations. The base map is an orthomosaic of drone images. Accessible via

Hard drives with the images were driven approximately 170 miles from the town of Paradise to San Francisco for processing and visualization by Drone Deploy, partially due to the lack of high-speed internet in the town of Paradise. The resulting visualization included 2D maps, 360 panorama views and georeferenced videos. Butte County embedded the final data layers and hosted them on the county website for public viewing.

4. Results

70,732 photos totaling 477 gigabytes were collected over two days, which were then stitched and presented on the Butte County Website as seen in Figure 1. Each 360 degree panorama is a composite of 23 photos from the 20MP camera resulting in a 45MB JPEG file. Over 500 flights were required to create the base map that covered most of the town of Paradise, 136 selectable locations of 360-degree panoramas, and 52 selectable locations of the georeferenced video.

5. Discussion

The capture of orthomosaic maps, 360-degree panoramas, and georeferenced videos provide several options to explore the images from the days after the fire. The level of utility the suite of images (or one particular image type) provides for community members, first responders, or other stakeholders is still unclear. To better utilize the valuable time of limited personnel and resources during a disaster, we highlight technical, organizational, and policy-based considerations to both improve the speed and quality of image capture as well as relevance of the visualizations to disaster stakeholders.

5.1 Image collection

Several steps can be taken to reduce image collection time. For example, flying drones at higher altitudes can provide a larger field of view, resulting in fewer flights. Higher altitudes are not always possible since other aircraft may occupy the airspace. Operating drones in the evening when there is less air traffic may address this issue, but would require night flying waivers. However, doing so may hinder the use of RGB cameras and be more suited for LiDAR or thermal imaging.

Further, the locations of drone flights were developed post-disaster in the case of the Camp Fire. Planning areas of strategic deployment pre-disaster can minimize the time needed for data capture. The creation of baseline images during non-disaster times can also provide high quality reference datasets for before-and-after change comparisons for damage assessment. Identifying whether a map, 360-degree panorama, or georeferenced video would be most useful in the disaster could also prioritize which data layers are requested and the order of capture.

Pre-planned flights may be more suited for disasters with pre-identified vulnerability areas, such areas prone to flood during a storm or vulnerable to liquefaction during an earthquake. The uncertainty in disasters such as fires and tornados may not allow for the pre-disaster planning of mapping locations. Nonetheless, integrated training and certification of drone use for first-responders or other disaster agencies can ad-hoc teaching during a disaster.

5.2 Image processing

A technical challenge in the case of the Camp Fire was the ability to process the quantity of data on-site as well as a lack of high-speed internet to transfer the images off-site. As a result, hard-drives of the images were driven from the fire location in Paradise, CA to nearby San Francisco. Reducing the image quality to a lower resolution would reduce file sizes, but may hinder some response applications where high detail is required. A development trend in the commercial drone industry is toward more rapid processing and real-time processing solutions, which may solve computational issues in the future. Moreover, edge computing platforms may be one solution to enhance computational infrastructure in remote or disaster-affected areas.

Organizationally, the current operational focus is on drone usage with fewer protocols for data capture, processing, and dissemination. The ability to scale up drone usage first requires not only operators but also data analytical teams proficient in image processing, geographic information systems (GIS), and web-based visualizations. Second, there is a need to coordinate data through city, county, regional, and federal systems to maintain a common operating picture among agencies. Information networks can occur in isolation or ‘silos,’ reducing the speed and effectiveness of data sharing during a disaster event. For improved coordination, future work can develop a centralized hub for inter-agency image processing and coordination that includes fixed camera, drone, manned aircraft, and satellite imagery.

5.3 Image use

In the case of the Camp Fire, the drone images were shown to residents that could not yet return to their home to provide knowledge and situation awareness about the state of their community. Yet, the use of the imagery is not yet fully integrated into the processes of disaster stakeholders.

Potential uses include the ability for off-site agencies to base their decisions. For example, insurance companies may write claims based on the official high-resolution images. Environmental agencies can use these images to coordinate clean-up efforts. Also, the drone imagery serves as a baseline for the recovery efforts to track progress and contractual agreements over time. Subsequent subject matter expert interviews with first responders, community members, and other stakeholders can identify other applications. In any case, the use framework needs to integrate with updates from the fast-changing drone industry to stay relevant with new technology.

Next, it remains difficult for agencies to request drone imagery as a resource or through mutual aid systems during an emergency because drone services have not been integrated into standard operating procedures, contractual frameworks, and reimbursement systems. Without standardizing how resource requests, budgeting, and payments for drone services, it will remain challenging to scale drone solutions during major disaster events.

Finally, there are privacy and regulation concerns associated with large-scale image collection. The release of drone images can show an individual’s possessions of value or secret. At this point, the value of broadly making the information public versus keeping the information private is not clear. Nor is it clear which agency officially owns the rights to the data and its usage and where media requests or requests from private companies should be made. Such issues will need to be discussed among emergency responders, technical personnel, regulators, and the public as drone usage because more prevalent.

6. Conclusion

We study the drone imagery collection efforts in the November 2018 Camp Fires in Butte County, CA and highlight the process, challenges, and future work needed to integrate drones as part of an emergency toolkit. Drones have the potential to provide fast, high-resolution, low-cost images following a disaster, but the value of such efforts will remain unclear without research into potential use cases and capturing the benefits.

7. References

[1] A. B. Leverkus, P. G. Murillo, V. J. Doña, and J. G. Pausas, “Wildfires: Opportunity for restoration?,” Science (80-. )., vol. 363, no. 6423, p. 134 LP-135, Jan. 2019.

[2] M. Vona, G. Cascini, M. Mastroberti, B. Murgante, and G. Nolè, “Characterization of URM buildings and evaluation of damages in a historical center for the seismic risk mitigation and emergency management,” Int. J. Disaster Risk Reduct., vol. 24, pp. 251–263, Sep. 2017.

[3] C. F. Barnes, H. Fritz, and J. Yoo, “Hurricane Disaster Assessments With Image-Driven Data Mining in High-Resolution Satellite Imagery,” IEEE Trans. Geosci. Remote Sens., vol. 45, no. 6, pp. 1631–1640, Jun. 2007.

[4] S. Platt, D. Brown, and M. Hughes, “Measuring resilience and recovery,” Int. J. Disaster Risk Reduct., vol. 19, pp. 447–460, Oct. 2016.

[5] A. G. Rumson and S. H. Hallett, “Innovations in the use of data facilitating insurance as a resilience mechanism for coastal flood risk,” Sci. Total Environ., vol. 661, pp. 598–612, Apr. 2019.

[6] S. M. Adams, M. L. Levitan, and C. J. Friedland, “High Resolution Imagery Collection Utilizing Unmanned Aerial Vehicles (UAVs) for Post-Disaster Studies,” in Advances in Hurricane Engineering, 2012, pp. 777–793.

[7] C. Surakitbanharn et al., “Cross-referencing social media and public surveillance camera data for disaster response,” in 2018 IEEE International Symposium on Technologies for Homeland Security (HST), 2018, pp. 1–9.

[8] A. Vidan and G. Hogan, “Integrated sensing and command and control system for disaster response,” in 2010 IEEE International Conference on Technologies for Homeland Security (HST), 2010, pp. 185–189.

[9] C. M. Yeum, S. J. Dyke, and J. Ramirez, “Visual data classification in post-event building reconnaissance,” Eng. Struct., vol. 155, pp. 16–24, Jan. 2018.

[10] C. M. Yeum and S. J. Dyke, “Vision-Based Automated Crack Detection for Bridge Inspection,” Comput. Civ. Infrastruct. Eng., vol. 30, no. 10, pp. 759–770, Oct. 2015.

[11] C. Reuter and M. A. Kaufhold, “Fifteen years of social media in emergencies: A retrospective review and future directions for crisis Informatics,” J. Contingencies Cris. Manag., vol. 26, no. 1, pp. 41–57, 2018.

[12] J. de Leeuw et al., “The Potential and Uptake of Remote Sensing in Insurance: A Review,” Remote Sens., vol. 6, no. 11, pp. 10888–10912, Nov. 2014.

[13] S. Dashti, L. Palen, M. P. Heris, K. M. Anderson, T. J. Anderson, and S. Anderson, “Supporting disaster reconnaissance with social media data: A design-oriented case study of the 2013 Colorado floods.,” in ISCRAM, 2014.

[14] W. Sherchan, S. Pervin, C. J. Butler, J. C. Lai, L. Ghahremanlou, and B. Han, “Harnessing Twitter and Instagram for disaster management,” IBM J. Res. Dev., vol. 61, no. 6, p. 8:1–8:12, Nov. 2017.

[15] F. Ofli et al., “Combining Human Computing and Machine Learning to Make Sense of Big (Aerial) Data for Disaster Response,” Big Data, vol. 4, no. 1, pp. 47–59, Mar. 2016.

Stanford d.schooler and more. Researching the intersection of tech, policy, and society around the globe since being old enough to vote.