[Talk-GB] UPRN to postcode lookup

[Mark Goodge]

On 02/07/2021 12:34, Robert Whittaker (OSM lists) wrote:

> It certainly looks useful for determining postcodes of individual
> properties, and possibly spotting errors in existing addr:postcode
> tags, with greater precision than from the centroids in Code-Point
> Open. One warning though (as someone else has already noted) I think
> they assign postcodes to non-addressable objects, including historic
> UPRNs that no longer exist on the ground. We'd need to be careful that
> the UPRNs we're using to infer postcodes are current and correspond to
> the property we think they do.

Having had a bit of a look through the data, I think we're on safe 
ground so long as we stick to normal (ie, not "large user") postcodes at 
street level. There are cases where different UPRNs at the same 
coordinates have different postcodes, but, as far as I can tell, these 
all appear to be large user postcodes assigned to different entities at 
a single physical location. More particularly, they almost all appear to 
be PO Box addresses, where the coordinates are those of the sorting 
office but the postcode is unique to the user.

I haven't, yet, found any cases where UPRNs sharing the same coordinates 
have different normal postcodes. So the presence of historic UPRNs and 
non-addressable objects doesn't seem to be an issue here. 
Non-addressable objects are given a nominal postcode that's the same as 
the nearest addressable object.

Also, the ONS data has a fair number of UPRNs that have no postcode 
linked to them. I randomly sampled a bunch of these at an authority 
where their own planning search facility will take a UPRN as an input, 
and all of them either do not appear at all, or, judging by the map and 
aerial photos, are UPRNs of properties that have been replaced by more 
recent development (eg, fields, rural buildings such as barns, and 
former brownfield commercial premises). So this may go some way to 
identifying the elusive historic UPRNs!

Mark