Japan Flooding - Getting Oriented


If you’re looking for a good place to start on the Japan Flooding campaign, it looks like the river overtopped its levee here http://www.tomnod.com/campaign/japan_flood_2015/map/4boxhy26 and the debris field spreads east (through the river-side village, you’ll find houses moved whole from wherever they used to be), turns immediately south around the next village that seems up on a slight ridge, then east toward an irrigation canal where it spreads north and especially south. To the south there are whole villages along and near the irrigation canal that are surrounded by water. Eventually you’ll get all the way south to the city of Joso, where the worst of the flood damages seemed to occur.

Some news items about the flood event thus far:
Japan Times http://www.japantoday.com/category/national/view/floods-stretch-limits-of-japans-elaborate-water-networks
BBC (one of the photos here is that location of the levee breach) http://www.bbc.com/news/world-asia-34205873
New Scientist (another excellent photo of levee breach) https://www.newscientist.com/article/dn28165-did-global-warming-play-a-role-in-japans-devastating-floods/

If I could make a suggestion for admins, that might be to make a topographic layer available for the image area. Flood studies almost always start with a topo, some idea of where the breach occurred, and the image. Searching for flood damage is far more efficient that way–just keep going across flats and downhill from the levee breach.


PS There is another area of flood damage well north of that levee breach. Go to http://www.tomnod.com/campaign/japan_flood_2015/map/4boxay3b and you’ll see a location where another irrigation canal connects with the river, effectively making a gap in the levee protection. All around the world, people build homes and farms within the river floodplain. It looks like the river rose high enough above the levee gap to flood that location as well.


I agree with Matthew! Topography is the key to a fast response here.
Dear Tomnod team: It´s very time consuming to mark every flooded house and road. In this case it would be much better to make the whole area visible from the beginning and then to enable drawing of polygons around the affected areas. This would take a fraction of the time needed to flick through all tiles. I know, the tile-statistics would be different, but the effect would be, that a helpful map would be available within hours and not like now within days. The picture shows an example how such a fast response map could look like.
In a second phase the more detailed search for affected houses and roads could still be done…


We’ve considered having the crowd draw a polygon around affected areas… In this campaign, tags on damaged buildings and roads helps to understand where there are affected populations -and their densities (within the broader affected area). Tags like blocked roads can help responders navigate, as well.

We’ve considered having Nodders draw polygons to help identify affected areas where less is known about an event’s impact & where more detailed tag data may not be as necessary (environmental impact of oil spills, burn areas that aren’t near populated areas etc). This is something we may test in the future-


@martins_binocolo and @HappyMapper

Thanks for your thoughts on this–I fully agree with Martins that this could be a 2-stage response, something like flood zone outlining for the first 24 hours and then exploration within that zone for the remainder of the campaign, or maybe people could choose which of those they’d rather do for that 1st 24 hours until a flood zone consensus is reached. I know they’re quite different approaches, polygons vs points, as HappyMapper indicates. Points for road blockages are certainly needed immediately for rescue navigation.

One thing with the step of outlining the flood zone: There is very likely some guiding information against which our tomnod polygons could be compared. In a country like the US or Japan, flood defenses are extremely well-planned and -engineered after a flood survey. The 100-year (and maybe even 500-year) flood zones are already delineated on some government map (a Flood Insurance Rate Map in the US). If those are readily available, that’s a good addition to a topo layer (which is available everywhere via NASA and other agencies). A convergence of tomnod flood area polygons could be checked against the pre-defined flood zone to find areas outside of the expected flooding, places where heavy rain has pooled but not necessarily with the devastation in the valley above Joso here.


@Matthew: I fully agree! This would reduce the number of tiles for the second phase quite a lot. A similar approach could be used for fires. For the Clearwater Fire I used a map published in the Internet (see part of it below) showing the latest known extent of the fire to restrict myself to the most important areas.


@martins_binocolo: Nice, and excellent approach! Part of my PhD work is oriented on making just such maps, based on Landsat multispectral calculations and before/after image comparisons. The data latency is several days though, even with an improved USGS data access for Landsat 8, so not much help for rapid-response and real-time analysis. My own PhD work is historical (1984-2013) as I develop the detection methods–I’m not trying to generate a real-time product (yet ;-)).



Yet… If Nodders self-restrict to one area based on maps, I see 2 problems:

  1. we could only see what we expect in that area
  2. we could find what we least expect in the ignored areas

In Tomnod, I generally pick a corner to start. In Clearwater, I happened to pick bottom-left, cut the map in half and proceeded up the left side. This ended up being thick smoke-clouds. But I don’t consider that 2.5k useless work. What isn’t found is just as important as what is found.

While Tomnod does time-critical work, a campaign like Japan isn’t completely immediate… Local SAR will be going door to door for weeks, as they did in New Orleans. I don’t see much added benefit in thinking data gets immediately over to rescuers, when rescuers are already doing many tasks. AS data comes in, then they can access and reassess their efforts. At least that is my take on things.

I’d rather go tile by tile so there’s been eyes on that area.


@cageycat cc @martins_binocolo

We’re not suggesting that nodders focus on one designated area at the expense of others. In the case of a flood or fire event, it’s important to get a sense of the whole map, not just one location. The main image we have for the flood in Japan shows numerous locations of damage and blocked roads, not just the valley between the levee breach and Joso. All of those areas get mapped and examined, the idea is to find them quickly (step 1: area delineation) and then focus (step 2: point marking). Same with fires, given their jumpy behavior in many areas of the world.

No area in a map would get ignored, and within the flooded/burned areas there may be inclusions with no damage at all (there is active research on this regarding forest fires, btw). What we suggested was simply a hierarchical approach, starting at image scale and working down to the finest scales. Others, such as yourself, may still prefer to start at the (sub)tile scale and work toward larger scales. Both are needed for this kind of work–the difference in approach can (should) fit with the spatial scale of the thing we’re mapping.