[Talk-us-massachusetts] Some observations and preliminary stats on MassGIS address import

[Yury Yatsynovich]

Thanks for creating the wiki page, Angela!

I did some simple spatial joins of MassGIS address points with existing
buildings without addresses (either house number or street name missing)
using python (geopandas and osmnx). Here are some stats from such joins by
counties:

County    OSM buildings w/o addresses    MassGIS points    Matched
points    1-to-1 matches    Share of matched points    Share of 1-to-1
matches
Barnstable    188370    192619    157235    121330    0.82    0.77
Berkshire    89520    79138    62324    39290    0.79    0.63
Bristol    233590    284883    247625    112776    0.87    0.46
Dukes    23647    20330    13639    11236    0.67    0.82
Essex    268122    379036    328679    147073    0.87    0.45
Franklin    50091    40853    31336    18146    0.77    0.58
Hampden    206111    225352    198221    100958    0.88    0.51
Hampshire    78709    73675    60417    34122    0.82    0.56
Middlesex    429556    779421    594647    219696    0.76    0.37
Nantucket    14048    12963    8408    6445    0.65    0.77
Norfolk    361719    337260    302017    147594    0.90    0.49
Plymouth    233374    244200    207421    134116    0.85    0.65
Suffolk    118763    457611    426855    36796    0.93    0.09
Worcester    383670    407001    347611    173147    0.85    0.50
TOTAL    2679290    3534342    2986435    1302725    0.84    0.44

Most of the address points (84%) lie within boundaries of buildings w/o
addresses, almost half of which (44%) are unique address point within the
corresponding buildings.

For those 1-to-1 matches the address info can be added directly to the
buildings. For the many-to-1 matches (several addresses within one
building) the options are either creating separate address points within a
building or combining addresses of all such points and adding it to the
building (e.g. addr:housenumber = 11,13,15 as suggested in
https://wiki.openstreetmap.org/wiki/Addresses#Buildings_with_multiple_house_numbers).
It seems that the second approach has been used for the existing addresses
in Malden and Boston. If points in the many-to-1 matches have different
street names then adding separate address points seems to me the only
solution.

Having visualized the data in QGIS I've noticed couple of issues:
1) some points lie outside, yet, very close to a building so that it is
unambiguous to which building they belong. Maybe creating a small buffer
(5-10m) around such points and merging them with unique buildings that
these buffers intersect can help match them.

2) some points, as it was mentioned in previous messages, are assigned to
parcels. E.g. there are many cases when a group of buildings (a house and,
say, barns)  have an address point next to them. It could be possible to
identify a house among those building manually and add the address only to
it, yet, as there might be several thousands of such cases, it can be very
time consuming. So, for parcels I would suggest simply adding an address
point in the middle of a parcel (as it is placed in MassGIS) without
identifying the exact building to which the address belongs.

-- 
Yury Yatsynovich
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/talk-us-massachusetts/attachments/20180717/bdc6e15c/attachment.html>