Hi Jean-Marc,
Thanks for bringing up the topic in a constructive manner! I do agree
it's valuable to question and examine some of our basic assumptions
sometimes. Please do keep in mind that to some extent this is still a
developing field. Having access to this type & level of data we are
creating is novel in a lot of contexts, and creating as comprehensive
and reliable datasets as we can is also a method of the making it
possible for people to start developing and implementing the use cases
for this data. So to address the "why are we mapping buildings"
question, let me sketch two current use cases where the building
footprint data is being used by NGO and gov't partners:
* Malaria elimination (and several other health-related use cases).
We've been working in southern Africa and Mali with various partners
to digitize buildings. This data is amended with data collection and
field mapping exercises, where we record (amongst other attributes)
building and roof materials, number of rooms, and the number of
sleeping spaces (where the latter two are not public/OSM data). The
building outlines (and thus size and shape of buildings) in
conjunction with this data allow for much better extrapolation to
inform what type of interventions to apply to which building, to
inform procurement and distribution of bed nets, insecticide,
logistics of spraying teams, etc (see
https://www.hotosm.org/updates/field-surveying-in-botswana-to-support-the-national-malaria-programme/).
* Rural electrification (including mini-grids and other sustainable
energy options). Information on estimated number of households, size
and density of villages, estimated number of
public/commercial/industrial buildings, estimated relative economic
activity/wealth indicators, in combination with datasets on current
electricity grid and grid expansion planning feeds into analysis on
what which sites would be most attractive/feasible for small,
standalone solar, hydro and wind-powered grids.
In some areas, where just having data is the priority and urgency, we do
start out by just marking `landuse=residential` afaik (see Congo/Ebola
recently). There are however also other datasets available that are
relatively reliable in identifying inhabited areas (such as WorldPop,
GPW, HRSL, etc) that can also serve as the basis. So following up and
continuing with digitizing building outlines where time and (relative)
lack of urgency allows does provide a much improved starting point for
additional and more 'advanced' use of these datasets. Even without any
further data on building use, type, or materials, what improves the use
for these analyses are having access to a) the size of the building, in
order to (for example) estimate which are residential, which are too
small & thus more likely storage/shed/pen etc, and which are large &
more likely to be commercial, industrial or public buildings, and b)
shape of the building (for example, round vs square). And to acknowledge
another point, yes, AI/ML will come into the equation (relatively soon,
even; way earlier than "ten years") and we will need to think about how
to deal with this type of 'generated' data, with the OSMF.
Further down the line, the real value in having a building dataset that
includes geometry is that it allows you much more accurately attach
additional attributes/data to these polygons, in the process creating a
historical record of the existence of this specific building (which is
also useful for land rights/ownership purposes), and enter into a
process of refinement and enrichment of that data.I definitely agree
that trying to get "all" buildings mapped is a herculean task - how many
would there be in total, 3 billion or so? At which point we get into the
really difficult task of trying to keep this dataset updated and
accurate. That doesn't take away from your point that 'low-quality'
mapping, due to a number of reasons and causes, is a large problem. We
are/should be working to improve mapper retention, upskilling, and make
validation more fun/attractive, but this is a topic others can speak to
better than me. Hope this helps a bit in understanding some of the
reasoning!
best,
Paul
On 4-7-2018 10:16, Jean-Marc Liotier wrote:
On Tue, July 3, 2018 7:20 pm, john whelan wrote:
I think my concern is more about the 'then a miracle occurs' in the
project plan to clean up the buildings.
Yes because, among other reasons:
- For most people, verifying is not as gratifying as creating
- Correcting entirely incorrect geometries is many ways more work than
re-creating them from scratch
I am not concerned about the most egregious cases: cars & trucks modelled
as buildings, duplicates & superposed, rubbish heaps and vague shadows as
building=yes, buildings found in old imagery... Those I delete with no
hesitation.
I am not concerned either about minor simplifications or errors such as
the shape being traced on the roof of the building rather than its base -
those I let them be and correcting them capitalizes on a good foundation.
I am concerned about the cases where a building does exist in reality, the
shape is less than ten meters from its position, some of the shape
overlaps the building's position on the imagery and some of the shape
resembles some of the building. In those cases, there is some value in the
record: approximate position and area of the building. But there is also
the liability of having introduced a low-quality object in the database.
I am convinced that the immense majority of those buildings will never be
corrected. In ten years, we can expect massive campaigns of automated
image recognition to produce new building layers - but even then the
extensive conflation will be an horribly tedious job.
Meanwhile, for areas with reliable imagery, I can imagine Maproulette
jobs: something in the spirit of "Does this building at least partially
overlap one in the imagery and does it approximately resemble the one in
the imagery ?". Those jobs could be designed at national or regional
levels - under control of the local communities. They could be a way of
systematic quality control. But maybe I'm horribly deluded about how many
people would volunteer for such a mind-numbing task. Also, looking at
buildings one at a time is very inefficient compared to panning through an
area on JOSM - but then again, JOSM-enabled contributors that might be
motivated for this are not exactly in plentiful supply either.
And that does not even answer the question: what to do with the
"low-quality shape but actually exists" cases ? I am at a loss to answer
that.
_______________________________________________
HOT mailing list
[email protected]
https://lists.openstreetmap.org/listinfo/hot
_______________________________________________
HOT mailing list
[email protected]
https://lists.openstreetmap.org/listinfo/hot
