[HOT] Interim Report: Typhoon Haiyan (Yolanda) Damage Assessment

Dan S danstowell+osm at gmail.com
Sun Feb 16 21:47:51 UTC 2014


Hi Robert,

Thanks for pointing me to the raw data. I've run a crosstabulation
(matching up objects by their osm ID) and it seems to look like this:

                no    damaged    destroyed
no            85               82           63
partial    170             119          111
major     173             160         196
total         88               31           70

This is rather disappointing - there is some correlation between the
two types of annotation but a lot of noise. For example the osm 'no'
category seems to be more likely to be _any_ of the other categories
in your observations rather than 'no'!

I quantified the predictability of one from the other (using mutual
information) and it confirms this, comes out rather low. My analysis
code is at https://gist.github.com/danstowell/9040956 - just a quick
evening script, check it before relying on it...

Best
Dan


2014-02-14 14:53 GMT+00:00 Banick, Robert <Robert.Banick at redcross.org>:
> Hi Dan,
>
> There's a "download raw data" button on the side of the website at
> americanredcross.github.io/OSM-Assessment. Feel free to download and play
> with the data further -- and do contribute back anything new you find.
> We're running more intensive stats analysis on the data now and will be
> working that into the final report.
>
> Cheers,
> Robert
>
> Robert Banick | Field GIS Coordinator | International Services | Ì
> American Red Cross <http://www.redcross.org/>
> 2025 E Street NW, Washington, DC 20006
> Tel 202-303-5017 | Cell 202-805-3679 | Skype robert.banick
>
>
>
>
>
> On 2/14/14 4:18 AM, "Dan S" <danstowell+osm at gmail.com> wrote:
>
>>Hi Robert,
>>
>>Thanks for this. It's great to have some concrete analysis of what we
>>did, and the extent to which it's accurate.
>>
>>Is there a full cross-tabulation available of the numbers for
>>building-osm-status vs building-true-status? In the report there's a
>>table of over/underrepresentation, and some other stats, but it
>>doesn't give me a complete summary of whether our errors were in the
>>form of "bias" (e.g. consistently labelling things worse or better
>>than they are) or "variance" (e.g. the labelling tends to be a bit
>>random). I'd be really grateful if you could provide the full
>>cross-table.
>>
>>Dan
>>
>>
>>2014-02-12 14:22 GMT+00:00 Banick, Robert <Robert.Banick at redcross.org>:
>>> Dear HOT Communuity,
>>>
>>>
>>>
>>> The American Red Cross and the REACH Initiative are pleased to present
>>>an
>>> interim assessment report on the validity of the building damages
>>>assessed
>>> through OpenStreetMap in the weeks following Typhoon Haiyan. You can
>>>find a
>>> print copy attached and a more interactive website version at the above
>>> link.
>>>
>>>
>>> The results were unfortunately negative and underline real limitations
>>>in
>>> OpenStreetMap's ability to capture these results in the present.
>>> Neverthless, this report identifies strong promise in the OSM model of
>>> crowdsourcing and highlights the investments needed to make that
>>>potential
>>> possible.  It's our sincere hope that funders, NGO partners and most
>>> especially the OpenStreetMap community will rally around these
>>>investments
>>> so that OSM can play an even stronger and more operationally useful
>>>role in
>>> future disaster responses.
>>>
>>>
>>> We are indebted to the US Agency for International Development's Office
>>>of
>>> Foreign Disaster Assistance (OFDA) for funding this assessment and look
>>> forward to future partnerships to improve the utility of open data and
>>> OpenStreetMap in particular for disaster response.
>>>
>>>
>>>
>>> With all the best,
>>>
>>> Robert Banick, Dale Kunce and Clay Westrope
>>>
>>> American Red Cross & REACH Initiative
>>>
>>>
>>> Robert Banick | Field GIS Coordinator | International Services | Ì
>>>American
>>> Red Cross
>>>
>>> 2025 E Street NW, Washington, DC 20006\
>>>
>>>
>>> _______________________________________________
>>> HOT mailing list
>>> HOT at openstreetmap.org
>>> https://lists.openstreetmap.org/listinfo/hot
>>>
>



More information about the HOT mailing list