EMERGENCY IMAGERY
April/May 2015 – No. 76
The Australasian magazine of surveying, mapping & geo-information
EMERGENCY IMAGERY many eyes make light work
Official publication of
inside Predicting floods An unexpected use for GPS
Fire UAV Drones for disasters
Locate anywhere Three words to replace lat/long
feature alongside longer term projects such as identifying population centres in Africa to optimise food and medical deliveries, locating invasive species across Hawaii, or helping in the search for the missing Malaysia Airlines’ flight MH370. “What’s interesting is that, maybe not in your backyard, but in somebody’s backyard, there are one or two disasters every week,” said Shay.
Many eyes make light work SIMON CHESTER
Helping every day
A
s Severe Tropical Cyclone Pam made its way across Vanuatu on March 13 2015, its 250km/h winds – the equal strongest of any cyclone seen in the Southern Hemisphere – destroyed much of the islands’ vegetation and infrastructure, and ended up tragically taking the lives of 15 people, whilst injuring many others. After a disaster like this strikes, there’s a need for the emergency responders to quickly gain access to a status report of the area in question, so as to best allocate resources based on urgency. This often requires sending people out to assess the damage. An alternative to this method is to examine a satellite or aerial image captured shortly after the event, but, as it’s a largely manual process, this takes a significant amount of time. However, there is one ingenious solution to this problem pioneered by a company called Tomnod: engage ‘the crowd’ to do the searching for you. “Traditional methods of post-crisis data collection usually require sending crews out in cars to document damage,” said Caitlyn Milton, crowd coordinator at Tomnod, which has been part of DigitalGlobe since 2013. “This process can take several weeks and sometimes returns incomplete or inaccurate data. “Crowdsourcing satellite imagery on Tomnod not only expedites this process to a matter of days – or hours – but it also keeps rescue crews out of danger while producing more complete and accurate results.” Tomnod works because it can engage its user-base to manually scroll through countless kilometres of satellite imagery and tag specified objects through its specially designed, web-based interface. But why would people want to do this? “The news sometimes leaves us feeling completely helpless,” said Caitlyn. “Tomnod offers a place for like-minded people to contribute their effort towards something tangible, with real outcomes.”
16 position April/May 2015
BEFORE
BEFORE
AFTER
AFTER
The idea for Tomnod had an unlikely beginning: it was initially created as a method to help an expedition search for the tomb of Genghis Kahn in Mongolia. In fact, Tomnod means ‘big eye’ in Mongolian. “I was in Mongolia in the summers of 2009, 2010, 2011, and again in 2013, searching for the tomb of Genghis Kahn as part of a National Geographic sponsored expedition,” said Shay HarNoy, CEO and founder of Tomnod. “We dreamt up Tomnod when we came back from Mongolia in 2010. “After searching for the tomb, we got a lot of buzz. We had a tonne of crowdsourced tags from our website (http://
exploration.nationalgeographic.com), but back then we didn’t have any idea about reliability or metrics or anything like that. We just looked at a heat map and then zoomed in and looked at the imagery ourselves. “When we came back from Mongolia, though, some folks were like: ‘Hey, this is really cool! Can you apply this for other applications?’ and a group in Long Beach asked us about it for some of the disaster response work that they’d previously done,” said Shay. Since that time, Tomnod has acquired nearly two million worldwide users who have helped during crises by spotting damage caused by a natural disaster,
During almost any of these weekly disasters, Tomnod can roll out a ‘campaign,’ where imagery of the affected area captured by DigitalGlobe satellites is rolled out to Tomnod users, who are tasked with searching for signs of damaged infrastructure, or something similar. “Tomnod has launched six campaigns in Australia, covering nearly all of the major crowdsource-able events that have occurred in the country since we debuted this capability in 2013,” said Caitlyn. “Most recently, we crowdsourced damage from the Sampson Flat bushfires, which burned across Adelaide, and Cyclone Marcia, which struck Queensland. We are currently crowdsourcing before-andafter satellite imagery of Vanuatu in order to map damage from Cyclone Pam (see images in this article). “The type of data we collect on Tomnod depends on the event and the data that is most needed by resources acting on the ground. During the Sampson bushfires, it was important for ground crews to know where every burned building was located, and whether there were felled trees blocking major roads that could hinder relief efforts. During Cyclone Marcia, Tomnod users helped map buildings and roads that sustained flooding or other damage.” Once the crowd has sifted through the imagery and marked the points of interest, algorithmic checks then take place to make sure that the data is sound – this is an area that Shay takes a great interest in. “I have taken ownership of geospatial big data at DigitalGlobe,” said Shay, whose title at DigitalGlobe is now ‘senior director of geospatial big data.’ “That’s the term that we’re using to talk about all the data that’s been collected by our satellites – but is lost inside of the pixels – and how we can convert that into information. “Part of that is machine learning and automatic feature extraction, and part of it is crowdsourcing. The real magic happens when you combine the two, and we’ve started doing that with Tomnod. “We’re constantly testing, modifying, and QA-ing feature extraction algorithms and machine learning algorithms, then having the crowd take it that last step of the way. It’s very easy to get to, say, www.spatialsource.com.au 17
feature 85% accuracy with algorithms, and then incrementally it gets really, really hard. I don’t have the patience to wait for that, so we use the crowd to take it that last 14-15% or however close we can get it to the perfect answer. It’s a closed loop of machines feeding the crowd, and the crowd training the machines.” After the crowdsourced data has been run through the algorithms, it is then packaged into useful datasets and made available to first responders, often free of charge. “We make Tomnod data accessible to both governmental and non-governmental organisations,” said Caitlyn. “Between the recent Sampson bushfires, Cyclone Marcia, and Cyclone Pam, data was sent directly to several relief organisations from Australia, New Zealand, and France to help mobilise aid distribution and calculate total loss.” Of course, Tomnod doesn’t only provide data to first responders: it also creates datasets that are sold to enterprise users, too. “Sometimes we might provide the data for free for the people responding to the event, but for enterprise users, we have another curation layer of data that’s packaged up for their needs,” said Shay. “After we source the initial data from the crowd, we have some expert analysts go through it, and we place quality guarantees on it so we’re able to re-package it for sale.”
BEFORE
Timely data = positive difference But it is providing timely data to first responders where Tomnod can make a positive difference during crises, as Gary Maguire, manager of business & location intelligence at the SA Department for Communities & Social Inclusion, can attest. “We needed to get information as quickly as possible, as you can’t get onto the ground while the fire is happening, nor shortly after,” said Gary. “The information DigitalGlobe gave us was within a day of the imagery becoming available on Tomnod. With it, we were able to get the first look of what was there.” This proved a stark contrast to the time frame of traditional field collection methods. “The full ‘rapid assessment’ wasn’t done until 22 days after the Tomnod data arrived, due to safety issues,” said Gary. “There were 3,466 taggers active on the campaign, and they came back with the locations of 112 damaged buildings and 16 hotspots across the ground. This ‘first pass’ helped us understand how many people were affected in the area.” However, it’s not just Tomnod’s timeliness that helps make a difference in these situations; it’s the cost savings, too.
18 position April/May 2015
AFTER “If each tagger spends a few hours on the Tomnod site, that’s the equivalent to several thousand hours of work by emergency responders. Gauge that on an average GIS salary of maybe $50 per hour, and that’s easily $340,000 of work effort. From that side alone, Tomnod is hugely valuable for first responders and planners,” said Gary. “People are quite engaged to help other people in time of need. Whether they are in the US, or Spain, or France, people are happy to help out. It’s a social good to help, and that’s what the crowdsourcing at Tomnod does – it leverages people with an interest in a particular topic or subject
matter to give up half an hour or an hour to create real value,” said Gary. Indeed, humanity’s strength is in its community, and Tomnod uses that sense of community to help others out when they need it most. “Nearly 2,000,000 people worldwide have volunteered on Tomnod to help strangers during times of need,” said Caitlyn. “The community is extremely vibrant and full of dedicated people who not only volunteer on Tomnod, but also learn how to become better image analysts, share insight, and ultimately help make the world a better place. It’s what we call the ‘power of the crowd’ and it’s helping us to see a better world.” ■
The Australasian magazine of surveying, mapping & geo-information
EMERGENCY IMAGERY many eyes make light work
Official publication of
inside Predicting floods An unexpected use for GPS
Fire UAV Drones for disasters
Locate anywhere Three words to replace lat/long
feature alongside longer term projects such as identifying population centres in Africa to optimise food and medical deliveries, locating invasive species across Hawaii, or helping in the search for the missing Malaysia Airlines’ flight MH370. “What’s interesting is that, maybe not in your backyard, but in somebody’s backyard, there are one or two disasters every week,” said Shay.
Many eyes make light work SIMON CHESTER
Helping every day
A
s Severe Tropical Cyclone Pam made its way across Vanuatu on March 13 2015, its 250km/h winds – the equal strongest of any cyclone seen in the Southern Hemisphere – destroyed much of the islands’ vegetation and infrastructure, and ended up tragically taking the lives of 15 people, whilst injuring many others. After a disaster like this strikes, there’s a need for the emergency responders to quickly gain access to a status report of the area in question, so as to best allocate resources based on urgency. This often requires sending people out to assess the damage. An alternative to this method is to examine a satellite or aerial image captured shortly after the event, but, as it’s a largely manual process, this takes a significant amount of time. However, there is one ingenious solution to this problem pioneered by a company called Tomnod: engage ‘the crowd’ to do the searching for you. “Traditional methods of post-crisis data collection usually require sending crews out in cars to document damage,” said Caitlyn Milton, crowd coordinator at Tomnod, which has been part of DigitalGlobe since 2013. “This process can take several weeks and sometimes returns incomplete or inaccurate data. “Crowdsourcing satellite imagery on Tomnod not only expedites this process to a matter of days – or hours – but it also keeps rescue crews out of danger while producing more complete and accurate results.” Tomnod works because it can engage its user-base to manually scroll through countless kilometres of satellite imagery and tag specified objects through its specially designed, web-based interface. But why would people want to do this? “The news sometimes leaves us feeling completely helpless,” said Caitlyn. “Tomnod offers a place for like-minded people to contribute their effort towards something tangible, with real outcomes.”
16 position April/May 2015
BEFORE
BEFORE
AFTER
AFTER
The idea for Tomnod had an unlikely beginning: it was initially created as a method to help an expedition search for the tomb of Genghis Kahn in Mongolia. In fact, Tomnod means ‘big eye’ in Mongolian. “I was in Mongolia in the summers of 2009, 2010, 2011, and again in 2013, searching for the tomb of Genghis Kahn as part of a National Geographic sponsored expedition,” said Shay HarNoy, CEO and founder of Tomnod. “We dreamt up Tomnod when we came back from Mongolia in 2010. “After searching for the tomb, we got a lot of buzz. We had a tonne of crowdsourced tags from our website (http://
exploration.nationalgeographic.com), but back then we didn’t have any idea about reliability or metrics or anything like that. We just looked at a heat map and then zoomed in and looked at the imagery ourselves. “When we came back from Mongolia, though, some folks were like: ‘Hey, this is really cool! Can you apply this for other applications?’ and a group in Long Beach asked us about it for some of the disaster response work that they’d previously done,” said Shay. Since that time, Tomnod has acquired nearly two million worldwide users who have helped during crises by spotting damage caused by a natural disaster,
During almost any of these weekly disasters, Tomnod can roll out a ‘campaign,’ where imagery of the affected area captured by DigitalGlobe satellites is rolled out to Tomnod users, who are tasked with searching for signs of damaged infrastructure, or something similar. “Tomnod has launched six campaigns in Australia, covering nearly all of the major crowdsource-able events that have occurred in the country since we debuted this capability in 2013,” said Caitlyn. “Most recently, we crowdsourced damage from the Sampson Flat bushfires, which burned across Adelaide, and Cyclone Marcia, which struck Queensland. We are currently crowdsourcing before-andafter satellite imagery of Vanuatu in order to map damage from Cyclone Pam (see images in this article). “The type of data we collect on Tomnod depends on the event and the data that is most needed by resources acting on the ground. During the Sampson bushfires, it was important for ground crews to know where every burned building was located, and whether there were felled trees blocking major roads that could hinder relief efforts. During Cyclone Marcia, Tomnod users helped map buildings and roads that sustained flooding or other damage.” Once the crowd has sifted through the imagery and marked the points of interest, algorithmic checks then take place to make sure that the data is sound – this is an area that Shay takes a great interest in. “I have taken ownership of geospatial big data at DigitalGlobe,” said Shay, whose title at DigitalGlobe is now ‘senior director of geospatial big data.’ “That’s the term that we’re using to talk about all the data that’s been collected by our satellites – but is lost inside of the pixels – and how we can convert that into information. “Part of that is machine learning and automatic feature extraction, and part of it is crowdsourcing. The real magic happens when you combine the two, and we’ve started doing that with Tomnod. “We’re constantly testing, modifying, and QA-ing feature extraction algorithms and machine learning algorithms, then having the crowd take it that last step of the way. It’s very easy to get to, say, www.spatialsource.com.au 17
feature 85% accuracy with algorithms, and then incrementally it gets really, really hard. I don’t have the patience to wait for that, so we use the crowd to take it that last 14-15% or however close we can get it to the perfect answer. It’s a closed loop of machines feeding the crowd, and the crowd training the machines.” After the crowdsourced data has been run through the algorithms, it is then packaged into useful datasets and made available to first responders, often free of charge. “We make Tomnod data accessible to both governmental and non-governmental organisations,” said Caitlyn. “Between the recent Sampson bushfires, Cyclone Marcia, and Cyclone Pam, data was sent directly to several relief organisations from Australia, New Zealand, and France to help mobilise aid distribution and calculate total loss.” Of course, Tomnod doesn’t only provide data to first responders: it also creates datasets that are sold to enterprise users, too. “Sometimes we might provide the data for free for the people responding to the event, but for enterprise users, we have another curation layer of data that’s packaged up for their needs,” said Shay. “After we source the initial data from the crowd, we have some expert analysts go through it, and we place quality guarantees on it so we’re able to re-package it for sale.”
BEFORE
Timely data = positive difference But it is providing timely data to first responders where Tomnod can make a positive difference during crises, as Gary Maguire, manager of business & location intelligence at the SA Department for Communities & Social Inclusion, can attest. “We needed to get information as quickly as possible, as you can’t get onto the ground while the fire is happening, nor shortly after,” said Gary. “The information DigitalGlobe gave us was within a day of the imagery becoming available on Tomnod. With it, we were able to get the first look of what was there.” This proved a stark contrast to the time frame of traditional field collection methods. “The full ‘rapid assessment’ wasn’t done until 22 days after the Tomnod data arrived, due to safety issues,” said Gary. “There were 3,466 taggers active on the campaign, and they came back with the locations of 112 damaged buildings and 16 hotspots across the ground. This ‘first pass’ helped us understand how many people were affected in the area.” However, it’s not just Tomnod’s timeliness that helps make a difference in these situations; it’s the cost savings, too.
18 position April/May 2015
AFTER “If each tagger spends a few hours on the Tomnod site, that’s the equivalent to several thousand hours of work by emergency responders. Gauge that on an average GIS salary of maybe $50 per hour, and that’s easily $340,000 of work effort. From that side alone, Tomnod is hugely valuable for first responders and planners,” said Gary. “People are quite engaged to help other people in time of need. Whether they are in the US, or Spain, or France, people are happy to help out. It’s a social good to help, and that’s what the crowdsourcing at Tomnod does – it leverages people with an interest in a particular topic or subject
matter to give up half an hour or an hour to create real value,” said Gary. Indeed, humanity’s strength is in its community, and Tomnod uses that sense of community to help others out when they need it most. “Nearly 2,000,000 people worldwide have volunteered on Tomnod to help strangers during times of need,” said Caitlyn. “The community is extremely vibrant and full of dedicated people who not only volunteer on Tomnod, but also learn how to become better image analysts, share insight, and ultimately help make the world a better place. It’s what we call the ‘power of the crowd’ and it’s helping us to see a better world.” ■
