Assignment of hh building loads lead to many synth. buildings (census data <-> OSM)

[nesnoj]

While assigning loads to buildings in map_houseprofiles_to_buildings() too many synthetic buildings are created in some cases for multiple reasons:
(A): Residential building geom overlaps with cell but their centroids do not (clipping uses centroids)
(B): No residential building geom overlaps with cell but other non-residential buildings and/or building centroids (following the current definition)

Same issue can be observed in CTS, see openego#953

Examples

1. (A)
   

2. (A+B) only tag "retail" available
   

3. (A+B) only tag "dormitory" available
   

Legend:

Approach

• Include further residential tags e.g. "dormitory", "terraced"
• Determine cells without residential building assigned via centroid but surface coverage (intersect), assign pop linearly using surface area
• Use buildings with non-residential tags (potential mixed use) as fallback e.g. "retail", "office", "commercial"

[nesnoj]

New approach:

• Extend OSM tags: regular building types and retirement and assisted homes
• Use commercial and retail buildings as fallback in populated cells without res. buildings
• Use building parts in hh load profile allocation instead of centroids: Buildings are clipped with census cells and resulting parts are now used for allocation instead of centroids only resulting in less synthetic buildings to be created. This reduced the total count of synth. buildings from 14k to 6.4k (SH run).
• Adjust the number of synth. buildings created in a cell: Use median of profile/building rate of adjacent cells instead of the global rate median to better represent the local conditions. This did not change the total count of synth. buildings (6.4k -> 5.6k) significantly but alters the regional distribution.