How the Drone Ecosystem Came Together In Support of Camp Fire

Like many people on November 8th, 2018, I was glued to the news about the explosive Camp Fire burning through the town of Paradise, California. At that point, I didn’t know just how devastating the fire would be, let alone that it would become the deadliest and most destructive wildfire in California history. No one did. Eighty-five lives have been confirmed lost with several hundred still unaccounted for at the time this article was written. Roughly 18,000 structures were lost, including about 14,000 homes and 500 commercial businesses.

Watching footage of so much going up in flames, I started a mental checklist to prepare for what would be my third trip in 13 months to assist in the aftermath of fatal wildfires as a drone data analyst.

But let me back up a bit.

I’m not a disaster-drone guy. I am a former ecology professor who left academia for all the promises the drone industry had to offer in 2015. As a scientist, I was confident drones would change how we monitor the earth, and I wanted to be front and center when that happened. Fast-forward through the tumult of the past few years of the drone industry. I founded a small drone business called Scholar Farms which provides training and consulting for plant mapping for agriculture, environmental consulting, and land management. I have an online vegetation mapping masterclass called Phytomappers Pro. It’s awesome. You should take it.

Of all the qualified drone people in the industry, you might wonder, why am I am one of the folks who gets called for these wildfires? The short answer is that one small action of being helpful and responsive led to a domino effect of deeper involvement.

I live in the San Francisco Bay Area, which is in close proximity to where these huge fires have occurred. I provided a small amount of advice for public safety teams that were mapping the tragic Ghost Ship fire in Oakland. This led to work on the Tubbs fire in Santa Rosa, then the Carr Fire in Redding, and finally the Camp Fire in Paradise.

I think my academic background helps with creative problem solving and accumulating evidence to tell my stories. Under the stress of these events, I tend to stay relatively calm and focused (which is less bragging and more a physiological response). I can also give and take a joke in dire situations, using humour as a powerful tool, to balance the tragedy. Not being a total asshole can go a long way, especially when you have smart friends in the drone industry who will answer your call from a disaster zone to give you feedback.

Could my role in the fires be played by any other number of people? Absolutely, and it should be. I’ll write more about this in the near future.

There are too many details to cover in the story of how we mapped Paradise, so I will cover the highlights. In short, it was a world-class effort by the following agencies (apologies if I left anyone out).

Alameda County Sheriff’s Office was granted the exemption to fly in the TFR by the FAA.
All flights were in direct and constant coordination with the Cal Fire manned air team.
Menlo Park Fire sent a large contingent for both mapping and aiding in the search and recovery process.

There were folks from La Honda Fire, Contra Costa County Sheriff’s Office, Stanislaus County Sheriff’s Office, Stockton Police, San Francisco Police, and Union City Police.
Romeo Durscher, Director of Public Safety Integration for DJI, supported all teams with valuable technical and logistical insight.

Andrew Maximow of Firmatek and Michael Dittman volunteered as technical support specialists and data runners to San Francisco for image processing.
The whole operation was on behalf of Butte County Sheriff’s Office, under the mutual aid system.

And a shout out to my wife for solo-parenting our 4-year-old daughter during my week away – while working full time herself.

Contra Costa County Deputy Sheriff Casey Tholborn was among the teams working to document the aftermath of the fire, and compiled dozens of high-resolution photographs on the ground.

Over several days, 16 teams of pilots and visual spotters mapped as much of the structural damage in Paradise as possible, using mainly DJI Phantom 4 Pros. This was done through pre-established flight zones created based on the Cal Fire Structure Damage map. The goal was to map as much damage in the town as we could in the limited time we had to fly. An Alameda County Sheriff acted as the flight director, assigning zones and tracking all team locations at any given time. If a team finished their zone, they were assigned a new one. If any manned aircraft came into the area, the drones were landed immediately for safety reasons.

As teams finished their respective assignments, their SD cards were put into Ziploc bags that had white labels for writing the zone ID, and a red sticker was placed on each bag to indicate the data had not yet been copied to a hard drive. Once the data were backed up, the red sticker would be covered with a green sticker. All the photo files for each zone were put in separate folders, labelled with the appropriate zone, and copied to a hard drive. Each folder was then double-checked against all zones and bags. The bags, stickers, and sharpies cost $9 from the craft aisle at Target.

At the end of the day, all of the data were backed up again on a duplicate hard drive. One hard drive was given to a runner who drove it to the DroneDeploy office in San Francisco. The other hard drive stayed at my side for the entire event. If possible, I tried to keep three backups at any given time.

Jono Millin, the co-founder of DroneDeploy, graciously answered the plea to process all the data and support the relief effort of Butte County. This made all the difference for such a big area. It would have taken a week or more to plow through the 500+ flights, 70K photos, and 466 GB of data on the laptops we had. Amazingly, the DroneDeploy team turned the data around in 24 hours after receiving each hard drive. From my understanding, this involved pulling all-nighters to load all the zones for processing.

As orthomosaics finished, maps went out immediately to the coordinating agencies. SAR teams, as early as the following morning, were looking at the data collected from the previous day.

The next task was to merge all the sections of maps together. Exporting about four dozen orthomosaics (each 1-9 GB) into tiles within ArcGIS was going to take some time. At this point, it was the day before Thanksgiving. My stress levels shot up to incredibly high when it became apparent that the maps weren’t going to happen, and we would need to roll out the data over time. I also hadn’t slept more than a few hours each day for a week. I made the call to Butte County Sheriff’s Office, saying that I could get them 360 panos and the geo-referenced videos now, but the maps would take longer.

It was about 20 minutes after that call that DroneDeploy pulled a huge rabbit out of their hat by being able to merge all of the maps together. They overlaid additional drone data layers, including 360 panoramas and links to the geo-referenced videos, in a public-facing website. Butte County was then able to embed this map on their website, and they released it to the public the next day during their Thanksgiving press brief.

Link to Butte County map (click on Drone Images at the top)

Credit for Map visualization: Casey Miller at Mapbox

I still can’t believe we turned around so much data in 48 hours and made it public the following day. It was a Hail Mary full court shot, right at the buzzer. I won’t lie, I shed a couple of tears when I saw it had all come together after so much work. I called Butte County back to tell them we had actually pulled it off.

Also deserving credit is Hangar in Austin, TX, that assisted in processing 150+ 360 panoramas. Mapbox assisted in a rapid visualization of data to use while working in the field. Survae assisted in helping to visualize all major roads in the geo-referenced videos. City of Redding GIS also provided support. ESRI’s Disaster Response Program supported the Butte County GIS team in the story-boarding for public visualization. ESRI will continue to support the migration of these map layers into ArcGIS to facilitate interagency use in the coming days and weeks. John Cherbini used his blazing-fast internet to help me download 240 GB of map sections in record time, and I drove a hard drive of all data to Butte County GIS for the final data handoff.

At this point, you might notice the growing number of teams involved in this effort. You might think to yourself, hmmm…that seems to be a significant number of people to get this done. You’re damned right it was!

Knowing the drone industry as I do, let me head off a few FAQs:

Q. Why use DJI Phantom 4 Pros and not fixed wings for such a large area?
A. Because we were limited to 300 ft. AGL and visual line of site under the FAA waiver and by Cal Fire air command. The smoky conditions didn’t help with visibility. Moreover, all these teams owned Phantom 4s, and the 20 MP camera is good. If we could have popped up a few eBees to cover bigger areas, we would have.

Q. Why not bring in manned aircraft to map?
A. Manned aircraft wouldn’t have helped. Visibility was terrible all the way to San Francisco, which was almost 2.5 hours away. You can imagine how bad it was close to the fire. Cal Fire was still containing the fire, and air attack was active at the time. The agencies involved wanted information to provide to the tens of thousands of evacuated people in order to facilitate repopulation of what was left of some neighborhoods and recovery planning. Time was definitely of the essence.

Q. Why not bring in private contract drone teams or volunteer pilots to help?
A. This is only the third active wildfire that UAV teams have been allowed access to in California. Flying drones in a wildfire is still very sensitive (for good reasons). I suspect flying is going to be restricted to public safety teams for a while to come. At this point, it’s actually the lack of geospatial analysts that is causing the biggest bottleneck.

Q. Why don’t you show more photos of the agencies involved?
A. Many of these agencies have social media policies that need to run through their public information offices. I don’t post pictures of badges or faces without the agencies’ formal approval, and that takes time.

Back to the story – There were many logistical challenges to overcome along the way to mapping Paradise. Perhaps the worst was the thick haze of smoke that hung over the town until the very last day of mapping. It does a number on the lungs (they called the cough “the Camp Fire Crud”). I did my best to keep a mask on when I wasn’t communicating with teams, but it still felt like I had put down a lot of unfiltered American Spirits by the end of the day.

As expected, there were all the standard drone challenges of flight planning for up to 16 drone teams at a time. Managing SD cards, charging drone and tablet batteries, apps crashing, and safety coordination were just a few of them. Thankfully, the telecommunications providers brought in mobile towers, and we had cell reception and a basic internet connection. This helped tremendously for downloading apps, setting up software accounts, caching base maps on site, and those unplanned firmware updates that will get you every time.

As you can imagine right after a fire, Paradise neighborhoods were still fairly unsafe, with trees and poles leaning precariously on power lines. There were many partially burned structures waiting to collapse. While calm winds over those days failed to blow the smoke and haze away, it also meant trees weren’t falling around us while we mapped.

Credit to all the teams for an excellent performance. By midday the first day, we fell into a rhythm, knocking out acre after acre, zone after zone, safely and efficiently for two full days. If I had to guess, I would have guessed we mapped 6,000 acres. In fact, it ended up being 17,000 total acres covered, or 26.2 sq, miles (so much for what I said about being a scientist!).

Obviously, there were bumps along the way, many lessons learned, and still more work to refine the process. But the drone teams are only getting better with the (unfortunate) practice we are getting. In the coming weeks, we’ll be working hard to debrief and refine the process. I learn more every time and try to run micro-experiments during the events to test new methods as we go.

Most of the smoothing that remains is on the data-side of things. That’s not just in the image processing. It’s also in the visualization, adding and updating different layers, sharing among agencies, and coordinating the steps to making it public. This time the fire was a 2-hour drive from San Francisco with technical support waiting and high-speed internet. These type of resources are likely the exception and, in that light, I’d say we accomplished a successful proof of concept more so than a decided overall win for the industry. The final win is when there is a repeatable process for earthquakes in Haiti, mudslides in Mexico, or hurricanes in the Southeastern U.S.

I won’t pretend I have this whole disaster-drone response thing all figured out. I know full well there is a wealth of experience in the emergency management and drone communities to draw on, from the drone deployment side to the geospatial side. This article is my simple way of saying that I’m willing to be a part of that bigger conversation and provide my own two cents.

There were so many smaller moments during the week I was in Butte County. I don’t have space here to tell them, some aren’t for the public, and some are other’s narratives they can tell themselves.

Finally, permeating this whole situation is the tragedy and loss for the communities of Paradise, Magalia, and Concow. In the moment, I tried to keep my head down and stay hyper-focused on the job of pushing the data over the goal line. Now that I am back home with a little more sleep, the sheer scale of the loss is coming back into clear focus. That sadness stays with the soul much longer.

I’m honoured to have been able to help, in what small way I could, in this and other fires. I’ll be there next time if needed.