Update: results of satellite imagery analysis available here.
You gotta love Twitter. Just two hours after I tweeted the above—in reference to this project—a colleague of mine from the UN who just got back from the Horn of Africa called me up: “Saw your tweet, what’s going on?” The last thing I wanted to was talk about the über frustrating day I’d just had. So he said, “Hey, listen, I’ve got an idea.” He reminded me of this blog post I had written a year ago on “Crowdsourcing the Analysis of Satellite for Disaster Response” and said, “Why not try this for Somalia? We could definitely use that kind of information.” I quickly forgot about my frustrating day.
Here’s the plan. He talks to UNOSAT and Google about acquiring high-resolution satellite imagery for those geographic areas for which they need more information on. A colleague of mine in San Diego just launched his own company to develop mechanical turk & micro tasking solutions for disaster response. He takes this satellite imagery and cuts it into say 50×50 kilometers square images for micro-tasking purposes.
We then develop a web-based interface where volunteers from the Standby Volunteer Task Force (SBTF) sign in and get one high resolution 50×50 km image displayed to them at a time. For each image, they answer the question: “Are there any human shelters discernible in this picture? [Yes/No].” If yes, what would you approximate the population of that shelter to be? [1-20; 21-50; 50-100; 100+].” Additional questions could be added. Note that we’d provide them with guidelines on how to identify human shelters and estimate population figures.
Each 50×50 image would get rated by at least 3 volunteers for data triangulation and quality assurance purposes. That is, if 3 volunteers each tag an image as depicting a shelter (or more than one shelter) and each of the 3 volunteers approximate the same population range, then that image would get automatically pushed to an Ushahidi map, automatically turned into a geo-tagged incident report and automatically categorized by the population estimate. One could then filter by population range on the Ushahidi map and click on those reports to see the actual image.
If satellite imagery licensing is an issue, then said images need not be pushed to the Ushahidi map. Only the report including the location of where a shelter has been spotted would be mapped along with the associated population estimate. The satellite imagery would never be released in full, only small bits and pieces of that imagery would be shared with a trusted network of SBTF volunteers. In other words, the 50×50 images could not be reconstituted and patched together because volunteers would not get contiguous 50×50 images. Moreover, volunteers would sign a code of conduct whereby they pledge not to share any of the imagery with anyone else. Because we track which volunteers see which 50×50 images, we could easily trace any leaked 50×50 image back to the volunteer responsible.
Note that for security reasons, we could make the Ushahidi map password protected and have a public version of the map with very limited spatial resolution so that the location of individual shelters would not be discernible.
I’d love to get feedback on this idea from iRevolution readers, so if you have thoughts (including constructive criticisms), please do share in the comments section below.
Thanks for the great post Patrick. I would like to clarify two things.
First, I am unsure about the usefulness of identifying populations within 50×50 km size imagery. Smaller size images would offer a more precise location as to where people are located, and would reduce the chance of error if someone has to scan such a broad area. What would be the benefits of using the large suggested image?
Second, depending on the size of the area needed to be analyzed, should work be restricted to the SBTF? Promoting the task through social media can tap into a significantly larger pool of labor. This project sounds very similar to Galaxy Zoo (http://www.galaxyzoo.org/story) the site where 150,000 users classified galaxies after viewing a short tutorial. If the images are small enough and do not include location information, then is there still an issue of licensing?
Thanks for your comments. I’m a big fan of Galaxy Zoo and know the project well. The 50×50 km size was just arbitrary for illustration purposes. I’m certainly open to opening up the project beyond the SBTF, but the SBTF is already a known & trusted entity in the UN system and the UN colleague who got in touch with me did so with the SBTF in mind. I think we’d want to pilot it within the SBTF, see how that goes and then invite anyone interested in helping out. What do you think?
That makes perfect sense. I’ll be excited to hear if something comes from this.
This is a very interesting post and idea. The folks at the Satellite Sentinel project you blogged about back in Dec 2010 may have some insights to share. They have created a team and information work flow—to process, interpret images, and transform them into products for wider distribution among the humanitarian community. Those with experience in population estimation (particularly with estimates using aerial and perhaps ground) during humanitarian crisis may be a good group to reach out to as well. In order to bridge from shelter to actual population estimates, average household (HH) size is necessary, and currently is determined by some form of ground truthing. (particularly if the environment includes empty shelters)
Perhaps also thinking about how they UN/NGOs and similar agencies present their findings in the crisis community may be a great way to think about a plugin (or something like it) for the Ushahidi platform. Perhaps have the map accompanied by a sit rep population report more like how it’s currently presented to and by UN agencies and NGOs. Field operations may be interested in one of the tiles if it fits their specific interest, but others may be more interested in the aggregation of certain tiles for planning and the margin of error for specific regions.
Check out the: CIEDERS Demographic Methods In Emergency Assessment A Guide For Practitioners, UNHCR’s handbook-3rd Edition, Chapter Operations including Annex 1.
Thanks very much, Jen, just got off a very good call with the Satellite Sentinel, there’s an excellent opportunity to collaborate. More soon.
Hi, reading this thread with great interest. I am with the SBTF, Media Monitor Coordinator.
The SBTF is not only a known and trusted entity, we stick together and work other projects together, which in turn strenghthens our buiness as well as our personal relationships.
To just add a “pool” of people takes away the working relationships that already exist. I for one, would not be comfortable working along side someone that I “think” may be trying to jeopardize the deployment at hand.
As I also continued on with the UN deployment of the Libya Crisis map, its a very delicate situation at times, and trust is a must ingredient within the actual mapping of an event.
SBTF is a close knit group.. the best I have had the pleasure of working with, and I would hate to see any of the trust and respect of that group
When we get new folks, they go thru training, we chat on skype, we get to know each other personally as well as professionally. I find it best to work with my friends and collegues other than a group of strangers. You cant ‘learn” in a short tutorial what we at SBTF already have and our new volunteers bring to the table.
Always comes to mind, the bartender, straight out of bartending school who can make a mean drink but has no idea how to communicate to his peers or his clients.
Thanks for reading this, and thanks for the great post and comment.
Thank you for expanding on this Leesa. I am familiar with the SBTF and hold it in very high regard. It has done some impressive work in Libya (Congratulations on your recent award!); however, the SBTF is restricted by where it can commit its limited resources. I perceive this particular project with less risk than the Libya Crisis Map, and considering the vast area that may be required to scan, tools like Amazon’s Mechanical Turk may be of value.
I completely agree that the SBTF should not be opened up in a way that would diminish the trust that it has worked so hard to build, but it should be mobilized only when it can most benefit a project.
Is it possible to get the difference between two imageries (ex.yesterday and today)?
It would be nice if data should be updated somehow semi-automatically.
Peter, I especially like the redundant review process that enhances validation. Providing bite size projects also makes it easier for volunteers to jump in, even for a little while and enables a sense of contribution, necessary for volunteer capacity building. Keep up the great work!
Interesting idea. Note that both the European SAFER project and UNOSAT have already published various analysis results, which may be useful to get some idea on task complexity, avoid duplication, etc.
I reckon you intend to use very high resolution data (~ 1 m pixel detail). These images are typically (strips of) some 10 km wide. Getting 50 by 50 sqkm image blocks is pretty costly (not just dollarwise) and impossible to feed to the average SBTF user at that resolution. So, blocks should be more like 1 by 1 km @ 1m resolution. Makes more sense also in terms of precision of location. The resolution could be somewhat less detailed, but not below 2-3 m in order to recognise the (temporary) settlements. But still, loads of data, given the huge space to look at. Google has published some data which shows that huge areas are spectacularly empty, btw.
As said by others, getting the data to crowd mapping apps other than Google Mapmaker will be a licensing challenge. Technically, it is not impossible, though, given enough eyeballs. Would be useful, btw what you consider to be a shelter, collecting maybe interactively from the results.
Collecting at high resolution and then presenting results only aggregated may not satisfy the SBTF user experience, but I understand the potential security risk.
Hope this helps,
Very good to know, many thanks, Guido
Automatic shelter detection on very high resolution satellite imagery can be an easy task . Several teams with image processing capabilities (including UNOSAT) have already fast and reliable approaches. An idea would be to start from an automatic approach and then call for the SBTF to validate and cross-check this information on a set of statistically representative tiles.
I really like that approach, thanks for sharing, Christina.
A number of great minds have been working this problem for more than a year. First, it is Humanitarian OpenStreetMap’s mission, and much work has gone into building the job tasking server and other tools to manage micro tasking and auditing. Ask Kate Chapman.
Second, Satellite Sentinel Project has developed techniques to deal with the multiple levels of analysis necessary to probe complex emergencies, including tactics for ground truthing imagery analysis of human security issues. Talk with Nathaniel @ HHI about those.
Third, there are other toolsets from Grassroots mapping for handling collective analysis of images, including some if the stats to compare accuracy versus other volunteers and known benchmarks. MIT ‘s Center for Collective Intelligence also has studies on this type of work when applied to visual diagnosis and other tasks. Talk with Jeffrey Warren.
And as for the frustrating day that generated the idea, there may be some breakthroughs soon which will make those frustrations diminish. Until then, I owe you a white russian or two next time I am in DC.
You own me two white russians, John.
Hey Patrick … once you and John have finished rocking the White Russians, would be really interested to hear your substantial response to John’s comment about HOT … it really does fit our mission, and we have the tools.
Thanks Mikel, I’m always happy to talk and explore ways to collaborate. Lets find out how we can best serve as force multipliers for each other. You have my number, feel free to call any time. Just to be clear, this project is not about tracing satellite imagery and placing the data on a public map. We’re tagging specific features for a UN partner by leveraging micro-tasking and mechanical turk services. We’re then passing this information to them, and them only.
Following this discussion with great interest, keeping in mind how these ideas could be integrated into the Humanitarian Dashboard Project, and more broadly, help improve needs assessments and assessment coordination in difficult to access areas. Exciting read, at least from my small OCHA cubicle!
Hey Marcus, good point re integration with the Humanitarian Dashboard, which I blogged about here a couple years ago:
If this pilot is successful for Somalia, we should definitely talk to see how integrate the analysis with the Dashboard.
My colleague Eduardo Dias is busy implementing something very similar. Earthwatchers will be using crowd sourcing with remote sensing to monitor deforestation on Borneo. Participants get a parcel of forest which they can monitor over time. The difference lies in the time scale. In Earthwatchers the participants monitor their piece of forest over a longer time for changes.
Eduardo presented the application recently: http://www.youtube.com/watch?v=QE2MKq1F3rA
Wow, Edwin, just watched the YouTube video, many thanks for sharing! I love what Eduardo is doing so I just signed up to help out online. Very excited about this, thanks again for sharing!
I would like to be involved. Kepp me posted
Pingback: On Genghis Khan, Borneo and Galaxies: Using Crowdsourcing to Analyze Satellite Imagery | iRevolution
This looks very interesting. Curious as to how the info will be used by the UN arm. Is it for information purposes only (ie: to get a sense of how the IDP camps have spread out) or is it to actually distribute aid to the extended areas? “muggles strike again” – that’s funny Harry.
Pingback: Irene Recovery Map: For Ordinary People Helping Ordinary People – The Ushahidi Blog
Pingback: Crowdsourcing Satellite Imagery Analysis for Somalia: Results of Trial Run | iRevolution
Pingback: Crowdsourcing Satellite Imagery Analysis for Somalia: Results of Trial Run