Using OWL to classify GIS data?

[Markus Stocker]

Dear all,

Recently I have been thinking a bit on using semantic web
technologies, especially OWL, on GIS data. Specifically, along the
lines of e.g. [1], I tried to apply OWL reasoning and classification
to GIS data. I'm trying a simple experiment with a taxonomy of lakes
[2] which I translated to OWL such that given an individual lake with
its properties richIn/poorIn {Phosphorus, Nitrogen, Humus},
basicRatio, altitude, latitude, volumeDevelopmentIndex a reasoner
classifies the lake w.r.t. the three dimensions (a) physico-chemical
nature of water, (b) climatic zone and (c) type of lake basin.

I believe, given a GIS model with lakes and values for the properties
listed above, a reasoner coupled with the GIS and the ontology should
be able to classify the lakes accordingly, though I have not tested
this yet.

I would like to discuss this idea with someone, so if you have some
thoughts let me know.

Cheers & enjoy the summer,
markus

[1] J.L. Mennis. Derivation and implementation of a semantic GIS data
model informed by principles of cognition. Computers, Environment and
Urban Systems, 27 455--479, 2003

[2] A.R. Zafar. Taxonomy of lakes. Hydrobiologia, 13(3) 287--299, 1959

[Ferdinando Villa]

Hey Marcus,

thanks for sharing. How would that differ from "standard" DL
classification? I haven't looked at the refs but I don't see anything
specifically spatial in the reasoning you outline. I'd be interested in
discussing specific ways how spatial reasoning could help inform
functional taxonomies.

My general feeling about logical modelling in ecology is that the key
problem is the ontology, as the notions you hard-code in properties and
classes (e.g. richIn) depend upon purposes, applications, contexts,
choices of scale etc. This said, I'm greatly interested in approaches
that build the ontologies themselves based on data-driven machine
learning of the relationships and thresholds that get encoded into
ontologies - i.e., conceptualizations that start from a human-provided
'seed' but evolve by learning from data.

I'm deep in the preparation of my move to Bilbao in September. I'm
hopeful that we'll get more chances to collaborate after I'm back to
mothership Europe, although my position and likely professional capacity
has been significantly downsized by the fiscal crisis.

Ciao,
ferdinando

On 7/19/2010 9:13 AM, Markus Stocker wrote:
> Dear all,
>
> Recently I have been thinking a bit on using semantic web
> technologies, especially OWL, on GIS data. Specifically, along the
> lines of e.g. [1], I tried to apply OWL reasoning and classification
> to GIS data. I'm trying a simple experiment with a taxonomy of lakes
> [2] which I translated to OWL such that given an individual lake with
> its properties richIn/poorIn {Phosphorus, Nitrogen, Humus},
> basicRatio, altitude, latitude, volumeDevelopmentIndex a reasoner
> classifies the lake w.r.t. the three dimensions (a) physico-chemical
> nature of water, (b) climatic zone and (c) type of lake basin.
>
> I believe, given a GIS model with lakes and values for the properties
> listed above, a reasoner coupled with the GIS and the ontology should
> be able to classify the lakes accordingly, though I have not tested
> this yet.
>
> I would like to discuss this idea with someone, so if you have some
> thoughts let me know.
>

> Cheers& enjoy the summer,

> markus
>
> [1] J.L. Mennis. Derivation and implementation of a semantic GIS data
> model informed by principles of cognition. Computers, Environment and
> Urban Systems, 27 455--479, 2003
>
> [2] A.R. Zafar. Taxonomy of lakes. Hydrobiologia, 13(3) 287--299, 1959
>

--
Ferdinando Villa, Ph.D.
Research Professor, Ecoinformatics
Gund Institute for Ecological Economics, University of Vermont

From Sep 2009:

Research Professor
Basque Centre for Climate Change
Bilbao, Spain

[Markus Stocker]

Hi Ferdinando,

Thanks for your comments. Some thoughts inline.

On Mon, Jul 19, 2010 at 9:42 PM, Ferdinando Villa <fvi...@uvm.edu> wrote:
> Hey Marcus,
>
> thanks for sharing. How would that differ from "standard" DL classification?

It doesn't. As I see it, it's nothing more than applying DL
classification to the special case of GIS. I suppose it could,
however, play together with, e.g., PelletSpatial which would provide
(RCC) spatial reasoning. Combined, you might query for "lakes within
US states that are externally connected to Vermont and have certain
physico-chemical properties for water." The query has a spatial
reasoning part ("within" and "externally connected") and a conceptual
reasoning part ("certain physico-chemical properties").

> I haven't looked at the refs but I don't see anything specifically spatial
> in the reasoning you outline. I'd be interested in discussing specific ways
> how spatial reasoning could help inform functional taxonomies.
>
> My general feeling about logical modelling in ecology is that the key
> problem is the ontology, as the notions you hard-code in properties and
> classes (e.g. richIn) depend upon purposes, applications, contexts, choices
> of scale etc. This said, I'm greatly interested in approaches that build the
> ontologies themselves based on data-driven machine learning of the
> relationships and thresholds that get encoded into ontologies - i.e.,
> conceptualizations that start from a human-provided 'seed' but evolve by
> learning from data.

Yes, this is interesting and something I might look into next, not
least perhaps because our group here is more on the data-driven ML
rather than on the ontological reasoning side. A first problem I need
to overcome here, though, is getting the data. The lake taxonomy
example could be a start for a use case: given a GIS model with lakes
and their physico-chemical properties of water etc. learn
relationships (e.g. richIn) and thresholds (e.g. basicRatio) and
distill an ontology, or part of it. Right now, though, I'm in the
situation of having a described taxonomy that can be translated to OWL
and not having a moderately large dataset from which an ontology could
be distilled. That said, lots of other datasets might be interesting
here.

> I'm deep in the preparation of my move to Bilbao in September. I'm hopeful
> that we'll get more chances to collaborate after I'm back to mothership
> Europe, although my position and likely professional capacity has been
> significantly downsized by the fiscal crisis.

Good luck and I'm looking forward to hear from you.

Ciao,
markus

[Tara Athan]

Hey Ferdinando- I have finally read enough of the literature to have a
clue what you are talking about here regarding machine-learning for
ontologies. I am considering exploring how (and if it is possible) to
convert between GML application schemas and OWL ontologies. Beyond the
mechanics of parsing, there are issues because the people who develop
profiles, such as OGC O&M, have not aligned their concepts with an upper
ontology such as DOLCE, as shown in [1]. But it might be possible to
resolve the inconsistencies using context. And having a tool to
translate back and forth might help those developing new app-schemas, or
revising old ones, avoid such mis-alignments in the future, as well as
to extract ontologies from the structure of published datasets. If this
works, it could provide a source of "local" ontologies that could be
compared for matching concepts to link to other ontologies and expand
the knowledge base. Datasets served through application schemas are
still the exception, but seem to be getting more attention. I would be
interested to know your thoughts on this.

Tara

[1] PROBST, F. 2006. An ontological analysis of observations and
measurements. In Proceedings of the Fourth International Conference of
Geographic Information Science (GIScience). Muenster, Germany, pp.304�20.

[Tara Athan]

Unless someone complains, I propose we use the aries group for now - perhaps a dedicated group will be built in the future and we can revisit the decision then. This mailing list is not much used, so I don't see any harm in it, and the topic is certainly relevant to Artifical Intelligence for Ecosystem Services. I am perhaps even more isolated than you - I am a Distance Learning Student - nobody here to talk to about anthing technical at all, so I welcome the opportunity.

Thanks for the twitter and wikipedia links.� I wasn't aware of the iEMSs conference - too bad, it will probably be a while until it is in North America again. Talks related to ontologies seem to appear under "Integrated Modeling Technologies" and I notice IEMSs as an Integrated Modeling google group
http://groups.google.com/group/commiem?hl=en
but it is not even viewable unless you get approved for membership by some moderator. Doesn't seem very open or inclusive to me.

I have a question on the eutrophic lakes. What are you choosing as your definition of eutrophic - bloom and anoxia, or abundance of nutrients. I can imagine situations where the abundance of nutrients is present, but due to other conditions (e.g. presence of toxic substances), a bloom and resultant anoxia does not happen.

Tara

Markus Stocker wrote:

Hi Tara,

On Thu, Nov 18, 2010 at 2:45 AM, Tara Athan <tara...@gmail.com> wrote:
  

Hi Markus- I am considering a slightly different perspective than you
outline below- which is good, we don't want to be doing exactly the same
thing :)

I am aiming to harvest metadata, express as ontologies and then look for
matches with other datasets and domain ontologies. I am not, at this time,
thinking about how to make use of the data itself, although that is
certainly an important line of study - just more than I want to tackle to
start. So I would say my approach is "metadata-driven" rather than
data-driven.

My use-case, at least what I am thinking about at the moment, is metadata
for sensor data based on the OGC O&M [1].

[1] http://www.opengeospatial.org/standards/om

But I think it is also applicable to your use case - let me try
The Harvester discovers a new resource called "californiaLakes" serving WFS
1.1 in GML3 format, but the resource is not semantically annotated. The
Harvester obtains the application schema for the resource through the
DescribeFeatureType request, and discovers that they are importing an
application schema about water chemistry from another namespace, and this
application schema has already been mapped to a domain ontology, according
to the FeatureTypeSemanticsCatalog. Voila, you have some semantic
information about the "californiaLakes" resource. The Converter creates a
californiaLakes ontology based on the californiaLakes schema, and this
ontology refers to some water chemistry classes from the domain ontology.
Next the Assimilator searches a suite of known ontologies looking for
classes that are related to the water chemistry classes in the same way that
classes in the californiaLakes ontology are, and ranks them on some
similarity scale. If the similarity is above some threshold, other actions
are triggered (details fuzzy here), which might include displaying similar
models to a user who will then examine free text metadata to verify the
match and further develop ontologies and/or application schemas manually,
or involve further ML, such as diving into the data to examine the
similarity at a deeper level ... which is where you come in.
    

Yes, I think this is quite accurate.

  

Question: do we want to continue discussing this topic on the aries mailing
list in order to perhaps attract other interested parties? Or would you
prefer to keep it private?
    

Sure, we can move the discussion to the list. I'm also open for other
forms to communicate or to bring people together. With a broader
perspective, I recently started @envinf [1] and there is a Wikipedia
page for environmental informatics [2] and I invite anyhow to join
(well, for Wikipedia it's obvious, of course :)). As far as I know,
students within iEMSs are planning to start a group, though I haven't
heard anything more on this lately (don't know if you attended the
conference, but I haven't and I think it was a mistake ...).

By for now,
markus

[1] http://twitter.com/envinf
[2] http://en.wikipedia.org/wiki/Environmental_informatics

  

Tara


Markus Stocker wrote:

Hi Tara,

Great to see that Ferdinando is inspiring people for ML and
ontologies. I just decided a few weeks back that I'll focus on this
topics also, fyi.

I think, if I have to give a summary in simple words on how I see the
bridge between ML and ontologies, at least right now, is that
"knowledge gathered from data ought to feedback into an ontology." So,
knowledge from a given--what Ferdinando called "seed"--ontology,
conceptualization of a domain (and, possibly, a specific modelling
ontology) is used in a ML (data-driven) environment *and* knowledge
acquired from the ML environment ought to be fed back into the
ontology. This might be an iterative process.

For example, imagine a small ontology on lake classification,
eutrophic, oligotrophic, heterotrophic, etc. Now, the (seed) ontology
says eutrophic lakes are "rich in" nitrogen. Good, but what does "rich
in" exactly mean? This information may be learned from a local dataset
on nitrogen concentrations of lakes. The key here is, while trivial,
this information should feedback into the overall ontology.

As we are apparently both looking into this area, I would love to
discuss more tightly as here I am essentially with no one to discuss,
and that's not good :>

Let me know what you think.

Cheers,
markus

On Wed, Nov 17, 2010 at 7:42 PM, Tara Athan <tara...@gmail.com> wrote:


Hey Ferdinando- I have finally read enough of the literature to have a clue
what you are talking about here regarding machine-learning for ontologies. I
am considering exploring how (and if it is possible) to convert between GML
application schemas and OWL ontologies. Beyond the mechanics of parsing,
there are issues because the people who develop profiles, such as OGC O&M,
have not aligned their concepts with an upper ontology such as DOLCE, as
shown in [1]. But it might be possible to resolve the inconsistencies using
context. And having a tool to translate back and forth might help those
developing new app-schemas, or revising old ones, avoid such mis-alignments
in the future, as well as to extract ontologies from the structure of
published datasets. If this works, it could provide a source of "local"
ontologies that could be compared for matching concepts to link to other
ontologies and expand the knowledge base. Datasets served through
application schemas are still the exception, but seem to be getting more
attention. I would be interested to know your thoughts on this.

Tara

[1] PROBST, F. 2006. An ontological analysis of observations and
measurements. In Proceedings of the Fourth International Conference of
Geographic Information Science (GIScience). Muenster, Germany, pp.304�20.


Ferdinando Villa wrote:


Hey Marcus,

thanks for sharing. How would that differ from "standard" DL
classification? I haven't looked at the refs but I don't see anything
specifically spatial in the reasoning you outline. I'd be interested in
discussing specific ways how spatial reasoning could help inform functional
taxonomies.

My general feeling about logical modelling in ecology is that the key
problem is the ontology, as the notions you hard-code in properties and
classes (e.g. richIn) depend upon purposes, applications, contexts, choices
of scale etc. This said, I'm greatly interested in approaches that build the
ontologies themselves based on data-driven machine learning of the
relationships and thresholds that get encoded into ontologies - i.e.,
conceptualizations that start from a human-provided 'seed' but evolve by
learning from data.

I'm deep in the preparation of my move to Bilbao in September. I'm hopeful
that we'll get more chances to collaborate after I'm back to mothership
Europe, although my position and likely professional capacity has been
significantly downsized by the fiscal crisis.

Ciao,
ferdinando

On 7/19/2010 9:13 AM, Markus Stocker wrote:


Dear all,

Recently I have been thinking a bit on using semantic web
technologies, especially OWL, on GIS data. Specifically, along the
lines of e.g. [1], I tried to apply OWL reasoning and classification
to GIS data. I'm trying a simple experiment with a taxonomy of lakes
[2] which I translated to OWL such that given an individual lake with
its properties richIn/poorIn {Phosphorus, Nitrogen, Humus},
basicRatio, altitude, latitude, volumeDevelopmentIndex a reasoner
classifies the lake w.r.t. the three dimensions (a) physico-chemical
nature of water, (b) climatic zone and (c) type of lake basin.

I believe, given a GIS model with lakes and values for the properties
listed above, a reasoner coupled with the GIS and the ontology should
be able to classify the lakes accordingly, though I have not tested
this yet.

I would like to discuss this idea with someone, so if you have some
thoughts let me know.

Cheers& enjoy the summer,
markus

[1] J.L. Mennis. Derivation and implementation of a semantic GIS data
model informed by principles of cognition. Computers, Environment and
Urban Systems, 27 455--479, 2003

[2] A.R. Zafar. Taxonomy of lakes. Hydrobiologia, 13(3) 287--299, 1959






    

[Markus Stocker]

On Fri, Nov 19, 2010 at 9:14 PM, Tara Athan <tara...@gmail.com> wrote:
> I have a question on the eutrophic lakes. What are you choosing as your
> definition of eutrophic - bloom and anoxia, or abundance of nutrients. I can
> imagine situations where the abundance of nutrients is present, but due to
> other conditions (e.g. presence of toxic substances), a bloom and resultant
> anoxia does not happen.

Abundance of nutrients. AFAIK, Finland has an own classification
scheme, though the EU has been trying to harmonize them for member
countries since 2000 or so. I found [1] quite curious (though perhaps
not really in use) as the description is quite detailed and rather
straightforward to translate to OWL.

Cheers,
markus

[1] http://www.springerlink.com/content/x0v57vk3m3mx944n/

> Geographic Information Science (GIScience). Muenster, Germany, pp.304–20.