When Charlie Kelly first sent a message indicating he wouldn’t return home that evening, his partner felt distressed. It was September 6, 2023, a Wednesday, and 56-year-old Kelly, an avid hillwalker, had departed from their home in Tillicoultry, close to Stirling, before Emer Kennedy went to work. His intention was to climb Creise, a 1,100-meter peak overlooking Glen Etive, a remote Highland valley renowned from the James Bond film Skyfall.
The weather was unexpectedly mild for the time of year, leading Kelly to believe he might have the opportunity to “bag” a second Munro, referring to the Scottish peaks exceeding 3,000 feet. During his leisure time from his role as a forensic psychologist for the Scottish Prisons Service, he had been steadily pursuing his goal of conquering the peaks. “He had this book where he would log them,” Kennedy recalls. “But we were set to depart for a holiday in two and a half weeks, so this was the final Munro he aimed to climb before winter arrived.”
Hiking wasn’t a particular passion of Kennedy’s. When they had first met four and a half years earlier, their connection grew from a mutual love of Celtic Football Club and their distinctively quirky sense of humor. She had come to admire Kelly’s intellect—his vast knowledge of football, Robert the Bruce, and Doctor Who. In return, he cherished her ability to laugh at his “terrible jokes,” as she describes. He also valued her support for his interests, even those they did not share. “One of the last things he said to me the night before was, ‘You let me be me,’” she recalls.
So, when Kelly mentioned he wouldn’t be coming down from the hill until after dark, Kennedy felt anxious, but she had faith in his ability. “Charlie was a very resourceful person,” she notes. “In his job, he was a trained negotiator for situations involving hostage-takers or escaped prisoners. He usually avoided taking unnecessary risks.” He reassured her that there was no need to seek assistance, as he had packed sufficient food, ample water, and enough warm attire. He intended to wait for daylight to make his descent.
On a Thursday at work, Kennedy found herself checking her phone during every available break. Early in the morning, Kelly had sent her a message, and as his day unfolded, he continued to send upbeat updates whenever he found a signal. As evening approached, around 8 p.m., he sent a message stating that his battery was nearly depleted. However, he reassured her that he could see the lights of the Glencoe Ski Center, where his car was parked, and estimated it would take him about half an hour to reach it. That was the final communication anyone received from Charlie Kelly.
In the days that followed Kelly’s disappearance, Glencoe Mountain Rescue initiated what they would later call a “Herculean” search operation. This extensive effort included the use of sniffer dogs, quad bikes, multiple helicopters, and drones equipped with both infrared and conventional cameras. The search team was composed of professionals from the Coastguard, Police Scotland, and the Royal Air Force, along with numerous highly skilled volunteers from ten different Mountain Rescue teams. Often, the number of people actively searching on the hillside reached as high as 50 at a time. On Saturday, September 9, they discovered his backpack, but after that, there were no further leads.
The turning point in the search came more than six weeks later, when Dan Roach and David Binks, two members of a Mountain Rescue team from the Lake District in northern England, learned about the ongoing search through online updates. In their free time, they had been working on new piloting and image analysis software intended to enhance the efficiency of drone searches for missing persons. After two years of development, which Binks described as involving “some hideous maths,” they finally had a working prototype ready for testing.
“I’d heard about the search and kept thinking it would be an excellent opportunity to test our system,” Binks recounted. “Then, while I was returning home from vacation, Dan called me with the same idea.” They reached out to Glencoe Mountain Rescue to offer their assistance. “I don’t think they expected much to come of it,” Binks reflected. However, given their exhaustive efforts with all available tools, they were open to the new approach.
On October 24, Binks, Roach, and Dan Parsons, a friend and fellow member of Roach’s MR team who contributed to initial testing, made their way to Glencoe. There, they joined forces with local MR volunteers, launched two drones, and successfully located Charlie Kelly’s body within the first hour of their search.
Mountain rescue volunteers frequently spring into action to locate missing hikers, with drones playing an increasingly vital role in search and rescue operations.
Mountain Rescue in the UK is often dubbed the nation’s fourth emergency service. However, unlike the police, fire brigade, or ambulance services, it operates entirely with volunteers. The upland regions of the country are serviced by a network of teams comprised of locals from diverse backgrounds. Each team functions as a distinct registered charity, managing its own fundraising, training, and equipment.
This decentralized system has benefits, according to Mike Park, who is the elected CEO of MR England & Wales, an organization focused on enhancing collaboration among teams. “However, the drawback is that it can foster individualistic attitudes where everyone thinks they know best,” he notes. Park, who previously led Cockermouth MR in the Lake District—where Roach and Parsons are members—has been a volunteer for nearly 42 years. He points out that MR members, predominantly men, often become overly protective of their areas of expertise, resulting in a slower adoption of new ideas than he prefers. “There’s this longstanding culture where it feels like you must serve for 20 years before others will take you seriously,” he remarks.
When consumer camera drones burst onto the scene in the early 2010s, their potential for search and rescue missions appeared straightforward. They enabled teams to survey extensive areas at a fraction of the price of using helicopters. However, Roach, an early drone enthusiast with experience in photography, held reservations. A passionate climber sporting long, curly hair and a Pedro Pascal mustache, he became a part of Cockermouth Mountain Rescue (MR) a decade ago and remains one of the team’s younger members at 34. “Everyone believed, ‘This is fantastic because I can see everything.’ But then reality set in; you need to be quite close to what you want to see, and you need to view it on a very small controller,” he explained.
In the context of searching for a needle in a haystack, merely having an aerial perspective on the haystack might not be particularly useful. Roach contended that teams should leverage drones’ capability to follow pre-programmed flight paths to automate the search as much as possible. Despite his insights, he encountered pushback from the veteran members of the MR community. “I recall one MR conference at Leeds University, where Dan arrived with his long hair and skateboard,” Mike Park recalls, “and someone jokingly asked if he’d gotten lost—they assumed he was a student who had returned too early.”
Undeterred by the resistance to his ideas, Roach continued to persevere with support from his fellow volunteer and friend, Dan Parsons. Like Roach, Parsons, who is 33, had experience with drones, having utilized them for surveying in the forestry sector. Together, they began exploring off-the-shelf piloting and photography software and discovered a program that could identify specific color pixels, like the red of a hiker’s jacket. “Ultimately, we created this highly technical, albeit clunky system,” Roach reminisces. “It functioned, but it wasn’t refined. You had to juggle three different software components to conduct a search.” The concept truly began to evolve when Roach met David Binks, a retired software developer who volunteers with Duddon & Furness MR, located roughly 30 miles south of Cockermouth. Roach shared his experiences of attempting to automate drone searches during a meeting with Lake District pilots, and Binks approached him afterward. “He said, ‘I hadn’t considered it that way. Let’s discuss it further,’” Roach recalled.
This collaboration formed an unexpected but effective dynamic. Roach, known as “the ideas guy,” radiated Tigger-like energy, while Binks presented a more subdued and analytical presence. At 57, Binks had over two decades on Roach but, crucially, possessed not only software development experience but also a history of creating software effectively utilized by MR teams. Since 2007, he had developed a system called MR Maps that enabled teams to monitor their members’ locations in almost real-time through regular updates from their radios. In a time before GPS-enabled smartphones were commonplace, this innovation transformed operations. Following initial discussions in late 2021 and early 2022, Binks concluded that the most effective way to realize Roach’s concept was to create it as a supplementary feature for MR Maps.
Thanks to innovative image-analysis software, it has become easier to locate individuals even amidst dense vegetation and rocky terrains.
“I understood how the mathematics would function due to my background in simulation software for offshore oil rigs and wind farms,” Binks shares. “So, I recognized it was achievable. However, there were numerous nights where I would start at 9 in the morning, and before I knew it, it would be 3 in the morning, and I would have been engrossed in the work all day.” Perhaps reflecting its homemade roots, and the fact that it was developed in C++, Roach describes the finished software as having “a Windows 95 aesthetic.” Nonetheless, this basic interface masks a remarkable level of complexity.
Binks elaborates on their primary challenge, which consisted of three parts. “The first part is maneuvering the drone in such a way that the ground is photographed effectively.” The camera must be oriented directly downward, and every inch of the ground should appear in multiple frames to ensure that objects are not hidden by walls or rocks. “We maintain a 55 percent overlap, meaning that an object should appear in an average of nine frames,” Binks notes. The pre-programmed flight paths are set automatically at whatever altitude the operator selects, employing the 2-meter by 2-meter lidar data that supports the contours of the UK’s Ordnance Survey (OS) maps, which are the official maps of the UK government (provided to MR at no cost). The software also assists the pilot in positioning themselves to maintain a line of sight with the drone during flight—a legal requirement in the UK.
Binks further explains, “The second component of the software is determining the grid reference of any specific pixel within an image,” which can be achieved using rotational vectors to find where they intersect with the ground. The third part, according to Binks, is “actually analyzing the images to identify what is noteworthy.”
Roach and Parsons developed a basic version that required users to input specific colors for the search, but Binks had a crucial insight that transformed the approach. “Instead, I can instruct it to count the pixels of any color and highlight color clusters that are unusual.” For example, a hiker’s red jacket would remain prominent against a green backdrop, Binks notes, “whereas in a landscape predominantly made up of rocks, green would be an unusual color.”
The system is crafted to function completely offline, making it suitable for use in remote areas. While images must be downloaded from the drone following each flight, the analysis can occur in the field, allowing unusual clusters of colored pixels to be flagged for human examination as the drone moves on to another area. “There are many false positives, but a human can quickly sift through them,” remarks Binks. “Usually it’s just oddly colored moss, stones, or even a sheep,” adds Roach. “Yet, every now and then, we do come across a person.”
Operators have the capability to define specific areas for drone focus and can establish automated flight paths for searching.
The rugged, boulder-laden terrain behind the Glencoe Ski Center served as an ideal testing ground for MR Maps’ innovative features. Between the ski lifts and the narrow road that trails down to a dead-end in Glen Etive, the landscape is largely characterized by rocks, dry tussocks, and the occasional deer. “The terrain here is quite intricate,” states Brian Brathurst, one of the deputy team leaders at Glencoe MR, his accent and understated manner hinting at his Zimbabwean roots. “There are extensive boulder fields, intricate gully systems, and you could easily walk within five or ten meters of someone without noticing them.”
As the search effort approached the end of its second week, the likelihood of discovering Kelly alive became increasingly slim, resulting in a natural reduction in the intensity of the operation. “Unfortunately, people have their own lives to live,” Brathurst remarks. “However, in our area, we tend not to give up easily on these matters. We dedicated every day we could to it. For a span of six weeks, there was always someone out there—individuals finishing their work in the afternoons to explore unsearched regions. It felt like the search never really stopped.”
When Binks, Roach, and Parsons joined the initiative, they were thoroughly prepared. “Without the groundwork laid by Glencoe, we would not have been able to achieve what we did,” Binks reflects. Working in tandem with the two Dans controlling one drone while Binks operated another, they allocated the most promising search locations among themselves and set out from various starting points. Despite the extensive efforts invested in the software, a sense of nervousness lingered. “We had practiced finding our jackets in fields and simulated the process,” Parsons recalls, “but we’d never actually discovered anything that we hadn’t previously placed there.”
David began his drone operation in the first designated search area, dubbed “Area A.” However, roughly 10 minutes into the flight, the drone began to drift out of view. Frustrated that complications arose so quickly, he recalled the drone and mentally noted adjustments for the software. “I figured, all right, I’ll shift to Area B now. But while that’s happening, I’ll review what I captured in Area A.” As he sifted through the flagged images, he suddenly spotted something unexpected. “Two of the photos contained Mr. Kelly,” Binks recalls, feeling a wave of disbelief.
Dan Parsons, David Binks, and Dan Roach, the minds behind the MR Maps-enabled drones.
When the recovery team from Glencoe arrived at the location, they discovered that Kelly had fallen several meters, resulting in fatal injuries to his head and chest. According to Brathurst, it remains uncertain exactly when he died. However, combine harvesters were operating higher up the valley on the evening of September 7. Binks speculated that Kelly might have been heading toward the lights, mistaking them for the Ski Center, which could have led to his unfortunate fall. “But we’ll never know exactly,” he acknowledged.
It was evident that Kelly had ended up far from where he believed he was. Nevertheless, the area where he was ultimately located had been thoroughly searched. Unfortunately, he had tumbled into a gully that obstructed the view of the rescuers on the ground. Additionally, the muted hues of his clothing made him nearly invisible from the air. “He was dressed in a light brown top and dark blue trousers,” Binks noted. “To the human eye, that’s really difficult to spot. However, to the computer, it contrasts sharply against the surroundings—and thus stands out prominently.”
Brathurst, who was one of the volunteers searching that area on foot, shared, “We literally walked within probably 20 or 30 meters of him.” Yet, they had failed to see anything.
Even for seasoned hikers, the wilderness can present perilous challenges.
If locating Charlie Kelly in the initial search area required a fair amount of luck, it also highlighted the impressive capabilities of MR Maps. The program has demonstrated its effectiveness in numerous other searches in subsequent months, primarily by assisting in the exclusion of certain areas. “Dan likes to claim that we have a 100% success rate,” Parsons states. “Although we’ve only discovered one individual, we have never surveyed a region where someone was present without locating them.”
Currently, the software has predominantly been utilized in the Lake District. However, encouraged by Mike Park, Roach, Parsons, and Binks have begun supporting other teams throughout the country when needed. When the well-known British TV physician Michael Mosley went missing on the island of Symi in Greece, Park even engaged in discussions with his Greek counterparts about the possibility of the trio flying out to assist. “The terrain seemed ideal for our system as it was rocky, and this can navigate around all the rocks,” Binks remarks. Ultimately, Mosley was located before their assistance could be deployed, but it’s clear how their technology could lessen the excruciating wait for families like Mosley’s, along with alleviating the strain on search teams.
In the UK, Mountain Rescue is facing increasing demands. “The number of callouts has been on the rise for several years,” mentions Ian Bunting, operations director for the overarching organization MR England & Wales, “but we’ve observed a significant surge since Covid.” He explains that hiking and outdoor activities gained immense popularity once team sports were prohibited, and with international travel restricted, the national parks saw a notable influx of visitors. “Coupled with that is the growth of social media,” Bunting adds, which has brought attention to specific locations.
Cockermouth, the team led by Dan Roach and Parsons in the Lake District, experienced an unprecedented number of callouts last year. In North Wales, the Llanberis MR team, which operates around Yr Wyddfa, the region’s tallest peak, publicly cautioned that their volunteers were at risk of “burnout” following over 300 incidents in 2023—an increase from just 100 in 2008. As a 30-year volunteer with Edale MR in the Peak District, Bunting acknowledges the strain can be substantial. “There are members of my team who responded to over 100 incidents individually,” he reveals. “With all the training, equipment inspections, and additional responsibilities, I logged around 600 hours last year.”
“The crux of the matter is that volunteers are just that—volunteers,” Bunting admits. “The wonderful thing about volunteering is that you can choose to decline. However, those who volunteer typically do not turn down opportunities.” In this context, any resource that helps minimize the time spent in search activities is a valuable addition to the MR toolkit, he explains. The program developed by Binks and Roach might not be needed for most callouts, where the individual’s location is known. However, its ability to reduce the hours consumed by resource-intensive “formal searches” (which account for approximately 10 to 15 percent of annual callouts, based on data provided to WIRED by MR England & Wales) holds significant promise.
According to Lucintel, a market research firm, the global search and rescue drone market is valued at around $4 billion. Despite the numerous potential applications for their product in remote mountainous areas worldwide, Binks and Roach have no intention of commercializing it. In the UK, Binks has consistently offered MR Maps free of charge to anyone involved in search and rescue operations, including police forces, several of which are now trialing the drone integration. “It’s intended for search purposes, so I’m perfectly fine with that,” Binks states.
“Monetizing the product has never really been my focus,” he shares. His motivation is straightforward: “I simply enjoy the challenges of problem-solving.” Ultimately, he emphasizes that the only significant cost of development was the time invested by himself, Roach, and Parsons—a resource they, like countless other MR volunteers in the UK, are more than willing to contribute to assist those in distress.
David Binks’ MR Maps software initially allowed rescue team members to monitor “pings” from their radios.
When two uniformed police officers knocked on Emer Kennedy’s door in late October 2023, they simply stated, “You know why we’re here,” she recalls. She had been holding onto hope that the news would be different. After six weeks, her primary concern was whether they would find him before winter set in. And if not, would he ever be discovered?
People gathered from near and far to pay their respects at Charlie Kelly’s funeral. Colleagues from across the nation and friends from universities around the globe attended. The Pogues performed as his casket was carried out, and donations were collected for a charity that would benefit the Glencoe Mountain Rescue team.
“Even though they operate voluntarily, they are remarkable professionals,” remarks Kennedy. “If that Mountain Rescue team hadn’t developed that software and been willing to test it, who knows what would have happened?” She trails off, reflecting. Ultimately, she expresses relief, saying, “I’m just thankful we have closure. It means his children can move forward with their lives. And we were able to say goodbye—his friends and family were able to say goodbye.”
This article is featured in the November/December 2024 edition of WIRED UK magazine.