Re: About the Spatial Meronymy and Spatial Operators requirements (help needed!)

Hi Frans




Your proposed requirement wording sounds good to me - i.e. including 'mereological' should cover the kind of relationships I have in mind.




To give some background:





My use case relates to processing or visualising statistics related to geographical areas - typically based on some administrative areas, or areas used for statistical reporting.  




Using an example from England (and ignoring some irrelevant detail), the country can be divided into local authority districts - the area that each council is responsible for.  Every point in the country is in exactly one local authority district.  The country can also be broken down into 'output areas' - small regions used for reporting census data.  These exactly cover the country as well, ie no overlaps or gaps.  Output area boundaries are designed to align with districts, so (ignoring the odd exception!) an output area will be entirely within one district.




So we have a hierarchy of areas: England --> districts --> output areas




Most countries have similar arrangements. 




If we have some statistical dataset, such as population, where the measure is countable/aggregatable, then I'd like to know that I can correctly add up the population of all output areas in a district to get the population of a district, or all districts to get the population of England, or all output areas to get the population of England.




Knowing that a subarea is contained within England is not enough.  With that list of areas, if I ask for everything that is contained in England and add them up, I'll count double because I'll get both districts and output areas, so we need some notion of 'levels' in the hierarchy.




And for some other kinds of regions, eg parishes, the coverage is not complete - some places in England have a defined parish, other places don't.  So it's not possible to add up the population of all parishes in England to get the population of England.  We need to know if a collection of regions makes up a mutually exclusive comprehensively exhaustive breakdown of the parent region.




And the hierarchical relationships can change over time.  There is one set of output areas defined for the 2001 census and a different set of output areas defined for the 2011 census.




Looking ahead to a possible solution, the use of XKOS (http://rdf-vocabulary.ddialliance.org/xkos.html) to express these relationships looks promising.  There is an implementation of this for Belgian government statistics at http://id.fedstats.be/conceptscheme/nis#id



Best regards


Bill






Bill Roberts, CEO, Swirrl IT Limited
Mobile: +44 (0) 7717 160378 | Skype: billroberts1966 | @billroberts 
http://swirrl.com

On Thu, May 28, 2015 at 3:55 PM, Frans Knibbe <frans.knibbe@geodan.nl>
wrote:

> All,
> I am in need of assistance for formulating requirements in the UCR document.
> This call for help is triggered by a remark from Bill Roberts in the UCR
> spreadsheet
> <https://docs.google.com/spreadsheets/d/1PSnpJYQDgsdgZgPJEfUU0EhVfgFFYGc1WL4xUX9Dunk/edit?usp=sharing>.
> The remark added to the Spatial Meronymy requirement
> <http://w3c.github.io/sdw/UseCases/SDWUseCasesAndRequirements.html#SpatialMeronymy>
> reads:
> *"This standard should include not only whether A contains B, but to
> express that A can be broken down into B,C,D which exactly cover A and do
> not overlap.*
> *Also, that there can be several different collections of sub-areas that
> make up a parent area. [..]**"*
> My initial thought was that that this further specification could be
> covered by the Spatial Operators requirement
> <http://w3c.github.io/sdw/UseCases/SDWUseCasesAndRequirements.html#SpatialOperators>.
> And then I started wondering why there are two different requirements at
> all. Then I started to try to find some information on the web about
> spatial meronymy and possible relationships with topological relationships.
> Then I started to become overwhelmed. Well, at least I think I found out it
> is probably better to speak of  'spatial mereology'  than 'spatial
> meronymy'.
> One aspect I wondered about is computability. I think that the topological
> relationships that are in use in by the OGC as described by the DE-9IM
> <http://en.wikipedia.org/wiki/DE-9IM> model are computable, i.e. one needs
> quantitative geometries to determine a topological relationship. Can anyone
> confirm or deny that? For example, let's say that there are two spatial
> objects that have no clear boundaries, like the Sahara desert and the
> Tanezrouft <http://en.wikipedia.org/wiki/Tanezrouft>. From a mereology
> perspective, we could say 'The Tanezrouft is part of the Sahara'. Could we
> also make a similar statement from the DE-9IM perspective, e.g. 'The Sahara
> contains the Tanezrouft', if there is no way to compute whether the
> statement is true or false?
> Another thing to consider is the difference between spatial functions and
> spatial properties. A spatial property can describe a relationship (e.g.
> 'object A  overlaps  object B'). A spatial function can determine a
> relationship (e.g. 'return all objects that overlap  object A'). There is a
> need for both and a standard like GeoSPARQL has separated the two, e.g. geo:
> sfContains is a property and geof:sfContains is a function (note the
> different name space prefix). With the requirements phrased as they are now
> the need for standardised spatial properties does not seem covered.
> I am now leaning towards suggesting changing the Spatial Meronymy
> requirement
> <http://w3c.github.io/sdw/UseCases/SDWUseCasesAndRequirements.html#SpatialMeronymy>
> to
> a more general Spatial Relationships requirement:
> 1) There should be a standardised way for expressing spatial relationships
> between spatial entities. These relationships can be topological,
> mereological, directional or distance related.
> The Spatial Operators requirement
> <http://w3c.github.io/sdw/UseCases/SDWUseCasesAndRequirements.html#SpatialOperators>
> perhaps
> needs no change, but we could consider specifying that we understand
> functions or operators to work on numerical data (so that includes raster
> data next to vector data)
> 2) There should be standards for functions or operators working with
> numerical spatial data.
> Rightly phrased requirements are what is needed most at the moment, I hope
> we can agree on them.
> Regards,
> Frans
> -- 
> Frans Knibbe
> Geodan
> President Kennedylaan 1
> 1079 MB Amsterdam (NL)
> T +31 (0)20 - 5711 347
> E frans.knibbe@geodan.nl
> www.geodan.nl
> disclaimer <http://www.geodan.nl/disclaimer>