Image Classification Taxonomy

From Classifications, Legends and
Nomenclatures to Parametric
Object Oriented Databases: a
paradigm shift in Geographic
Information
Guillermo VILLA
IGN Spain
CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008

Warning: This presentation is intentionally
provocative and “politically incorrect”
We apologize for the annoyances it may cause
CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 2 2

paradigm= unquestioned* ‘true’ in a
scientific or technical domain
(*) in a certain historical time

Outline
1. Introduction
2. Example: Database of people
3. Example: Land Cover classifications
4. New solution: Parametric Object Oriented
databases. Application to the Land Cover
case
5. Classifications, legends and nomenclatures
“hidden” in other G.I. themes
6. Conclusions

Everybody agrees on the importance of
geographical information (G.I.) but…
Why is geographical information so difficult to
achieve correctly ?

Because …
Our world is infinitely complex, and it is made by multiple

interaction of myriads of phenomenon that go from de atomic
scale to geographical scale
It is constantly changing (plant fenology, agricultural cycles,

climatology, human action…)
G.I. Information uses concepts and terms:

With “fuzzy” definitions and significances
From science and technique as well as everyday use
Changing from one place to another of the world (same term for
different thing and different terms for same thing)

Because …
G.I. needs to “mix” concepts and terms from many different

thematic fields:
Geology, Geomorphology, Edaphology

Hidrography, Hydrology
Botanics, Agriculture
Forestry, Ecology, Chemistry
Urban Planning
Economic Activity
Infrastructures
Architecture
Climatology
Etc….

Historical evolution of geographic
information storing technologies
TIME LINE
Digital
Digital
Paper
Paper
Maps
Maps GIS
GIS
Maps
Maps (CAD)
(CAD)
Databases
Databases
• Historically, the information has been stored in paper maps
• So it was useless to retrieve information

impossible to be stored in a paper map

Amount of information in a paper map
Reality
Needed
Needed
Information
Information
Paper
Paper
Map
Map
Info.
Info.
• The problem is that information you can store in a paper map is

a very small part of information you could retrieve or need
about reality
• Paper maps are a good way to show information, but a very
bad way to store it
Databases designed to contain only
information already present in paper maps
Reality
Map-centric data model

old G.I.
Needed
Needed databases
Information
Information
Paper
Paper
Map
Map
Info.
Info.
• The first information that was introduced in digital databases

was the info already present in paper maps
The first digital databases were designed to store
only the information that was already in the maps

Databases able to store all the
information needed about reality
Needed
database
old G.I.
database
Needed
Needed Paper
Paper
Map
Information
Information Map
Info.
Info.
• Databases that would be able to store all the information

needed are completely different (and more complex)
compared with old G.I. databases
In this presentation we will try to show:
That most of present GI databases are obsolete, because they

were designed only to store information in paper maps
That many of new data models being built today are also obsolete,
because they are “contaminated” (“poluted”) by “map centric” “out
of date” way of thinking. (E.g: the use of classifications, legends
or nomenclatures)
That it is imperative to change all this data models and databases

if we want GI to achieve the goals that we expect in the 21st
century

Outline
1. Introduction
case
“hidden” in other G.I. themes
6. Conclusions

Suppose we need to build a database of “people”

Characteristics considered
Characteristic Possible values Number of

values
gender - man
- woman 2
height - tall
- medium height 3
- small
weight - fat
- medium weigh 3
- thin

Characteristics considered
Characteristic Possible values Number of

values
gender - man
- woman 2
height - tall
- medium height 3
- small
weight - fat
- medium weigh 3
- thin
Classification Number of classes = 2 x 3 x 3 = 18

“Hierarchical classification” of people
2 x 3 x 3 = 18 classes
1. Men
1.1. Tall men
1.1.1. Tall and fat men
1.1.2. Tall and medium-weight men
1.1.3. Tall and thin men
1.2. Medium height men
1.2.1. Medium height and fat men
1.2.2. Medium height and medium-weight men
1.2.3. Medium height and thin men
1.3. Small men
1.3.1. Small and fat men
1.3.2. Small and medium-weight men
1.3.3. Small and thin men
2. Women
2.1.Tall women
2.1.1. Tall and fat women
2.1.2. Tall and medium-weight women
2.1.3. Tall and thin women
2.2. Medium height women
2.2.1. Medium height and fat women
2.2.2. Medium height and medium-weight women
2.2.3. Medium height and thin women
2.3. Small women
2.3.1. Small and fat women
2.3.2. Small and medium-weight women
2.3.3. Small and thin women

But there are many other possible characteristics to be
considered:
- nationality
- age
- study level
- work
- residence
- eyes color
- hair color
- diseases
- marital status
- number of sons
- hobbies
- religion
- etc, etc, etc…..

What would be the number of classes needed to store all
this information ?
250 * 100 * 4 * 100 * 250 * 5 * 4 * 20 * 4 * 10 * 20 * 10 =

250 * 100 * 4 * 100 * 250 * 5 * 4 * 20 * 4 * 10 * 20 * 10 =
= 8,000,000,000,000,000
=8 * 1015 classes !!

• Would these classes be useful in practice?
• Would it be possible to implement them in an
information system?
⇒ not at all !!

This is called by computer engineers
the ‘class explosion’ problem

So… this is clearly not the way to go…

Outline
1. Introduction
case
“hidden” in other GI themes
6. Conclusions

3.1. The actual ‘paradigm’*:
Land Cover Classifications
(*) paradigm= unquestioned ‘true’ in a certain technical domain
December 2007 INSPIRE D2.6 workshop Ispra 26 26

‘Hierarchical classification’ of people
2 x 3 x 3 = 18 classes
1. Men
1.1. Tall men
1.1.1. Tall and fat men
1.1.2. Tall and medium-weight men
1.1.3. Tall and thin men
1.2. Medium height men
1.2.1. Medium height and fat men
1.2.2. Medium height and medium-weight men
1.2.3. Medium height and thin men
1.3. Small men
1.3.1. Small and fat men
1.3.2. Small and medium-weight men
1.3.3. Small and thin men
2. Women
2.1.Tall women
2.1.1. Tall and fat women
2.1.2. Tall and medium-weight women
2.1.3. Tall and thin women
2.2. Medium height women
2.2.1. Medium height and fat women
2.2.2. Medium height and medium-weight women
2.2.3. Medium height and thin women
2.3. Small women
2.3.1. Small and fat women
2.3.2. Small and medium-weight women
2.3.3. Small and thin women

Corine - Moland Classifications
LEVEL 1 LEVEL 2 LEVEL 3 LEVEL 4

1. ARTIFICIAL 1.1. Urban fabric: 1.1.1. Continuous urban fabric: 1.1.1.1 Residential continuous dense urban fabric.
AREAS Most of the land is covered by structures and transport network. Residential structures cover more than 80% of the total surface. More than 50% of
Buildings, roads and artificially surface areas cover more than the buildings have three or more stories.
80% of the total surface. Non-linear areas of vegetation and bare 1.1.1.2 Residential continuous medium dense urban fabric.
soil are exceptional Residential structures cover more than 80% of the total surface. Less than 50% of
the buildings have three or more stories.
1.1.1.3 Informal settlements
1.1.2 Discontinuous urban fabric 1.1.2.1 Residential discontinuous dense urban fabric.
Most of the land is covered by structures. Buildings, roads and Buildings, roads and artificially surface areas cover between 50% and 80% of the
artificially surface areas are associated with vegetated areas and total surface area of the unit.
bare soil, which occupy discontinuous but significant surfaces. 1.1.2.2 Residential discontinuous sparse urban fabric.
Between 10% and 80% of the land is covered by residential Buildings, roads and artificially surface areas cover between 10% and 50% of the
structures. total surface area of the unit. The vegetated areas are predominant by but is not
land dedicated to forestry or agriculture.
1.1.2.3 Residential urban blocks
1.1.2.4 Informal discontinuous residential structures
This is the same methodology used in Land Cover

Classification databases. E.g: Corine LC, Moland,
Anderson,…

3.2. Problems experienced during the design, production and use
of Land Cover classifications (CORINE, Moland, etc…)

1) Class “explosion” (in detailed nomenclatures)
Proliferation of “difficult to use” classes, due to the multiple

“crossings” of several classification criteria
This has been a great problem in the effort to develop and use of
Spanish 5 level 84 classes developed for CLC2000

2) Class definition incoherencies
Complex definitions, in which several “classification rules” apply

simultaneously, cause incoherencies:
- “overlapping definitions” (some polygons can be assigned to more than
one class)
- “un-definitions” (some polygons cannot be assigned to any of the
classes)
- E.g.: Moland 1.1.1.1. Residential continuous dense urban fabric:

- Most of the land is covered by structures and transport network.
- Buildings, roads and artificially surface areas cover more than 80% of the total
surface
- Non-linear areas of vegetation and bare soil are exceptional
- Residential structures cover more than 80% of the total surface
- More than 50% of the buildings have three or more stories.”

3) Important information lost
Information stored in the database is much less than information

acquired by the photointerpreter:
E.g.: The photointerpreter evaluates a certain polygon’s trees percentage

as 75 %, and in consequence he labels it as Corine 3.1.1. “Broad-leaved
forest”.
… But the user only receives the information

that trees are “more than 30 %”

4) Mixed classes problems
Need to create “mixed classes”, due to the obligation to assign a

single class to mixed polygons
E.g: “Mixed forest”, “Complex cultivation patterns”,…
- Mixed classes provide little information to user:

E.g.: “Complex cultivation patterns”,…???
- Mixed classes lead to erroneous conclusions in the use of the

database:
E.g.: If one wants to know how many vineyards there are in a certain
region, he will search for the class 2.2.1: “Vineyards”. But there can also
be plenty of vineyards “hidden” in other classes as: 2.2.3: “Olive groves”
(which includes the association of olives trees and vines); 2.4.2: “Complex
cultivation patterns”, etc…

5) Important spatial variations not registered
Important spatial variations in certain parameters values do not

appear in the database if this variations do not “cross” the “threshold
line”
- E.g.: Urban areas with very

different levels of
building densities (as
10 % and 50 %) have to
be assigned to the same
Moland class (1.1.2.2.
Residential
discontinuous sparse
urban fabric.)

6) Important temporal changes not detected
Important temporal changes do not appear in the changes

database, because:
- These changes do not “cross” the “definition rule” threshold.

E.g.: If the building density of a polygon has increased from 11% to 79 %
in the time between 2 revisions of the database, this polygon is labeled as
Corine´s 1.1.2. “Discontinuous urban fabric”, in both databases, and so no
change is registered.
and/or:
- These changes are “hidden” in
polygons assigned to dominant
classes or to mixed classes.

7) Parameters and Indicators calculations not possible
Many parameters and “Indicators” could be calculated from the

values of the parameters appearing in class definitions (sometimes
“crossing” them with exogenous information such as population,
etc…).
Eg:
• building density (m3/m2)in an area
• m2 of building per person in an area
• average height of buildings in a town
• % of impervious surface in an area
• % of trees in a forest
• m2 of green areas per person in an area
• land take by transport infrastructures in a city
• etc…
Need for calculation of parameters and Indicators
New data sets:

Data specifications Databases - Parameters
- Agri-environmental Indicators
Data
Data
model
model
Database e.g: Land Cover Dataset
e.g: Urban sprawl
1 Database
1 Dataset
1 A
1 A
Interpretation
Data Dataset
Data
model Database Multilayer Dataset
B Environ-
model
2 Database
2 B Environ-
2 2 Multilayer
Geoprocessing mental
Multilayer mental
Geoprocessing Processes
Geoprocessing Processes
World
World models
models
Data Dataset e.g: Climatic

Data Dataset
N
Model Database N
Model
n Database
n change
n n
Actions
--Laws Predictions
Predictions
Laws of
ofFuture
Political
--Policies Future
Policies decisions
e.g: Kioto Protocol e.g: Future climate

Calculations not possible …
Land Cover Classifications do not allow calculating these indicators,

because de real values of the different parameters are not stored
in the database (only “intervals”):
E.g: You can not divide “Artificially surfaced areas are more than 80 %” by a
surface

8) Mosaicking different databases not possible
It is impossible to compare, “translate” or mosaic two databases

built with different Nomenclatures:
E.g.: When we have a 3.1.1. “Broad leaved forest” polygon in Corine (defined
as having more than 30 % of canopy closure)
→ there is no way to know if it should be labeled as “Forest” in a

database with a different Nomenclature where “forests” are defined as
“more than 50 % of canopy closure”.

9) Automatic generalizations not possible
If one wants to derive a smaller scale database (through a

“generalization” process), one must aggregate polygons into bigger
ones.
But there is no way to automatically derive the class of the
resultant polygon:
E.g: A polygon classified as Corine´s 1.1.1 (“Continuous urban fabric”)

aggregated with a polygon classified as 1.1.2 (“Discontinuous urban
fabric”) could result in an aggregated polygon that should be classified as
1.1.1 or 1.1.2.
There is no way to know, without repeating the interpretation !

3.3. UN FAO´s LCCS (Land Cover Classification System)
approach

LCCS approach
Land Cover Classification System (LLCS) has been developed

by UN-FAO in an effort to solve the problems associated with
existing Classification databases.
LCCS establishes a standardized comprehensive system that

allows one to build “a priori” classification “nomenclatures”
and “legends” in a more rigorous and coherent way

LCCS: our opinion
LCCS correctly exposes some of the problems of Land

Cover classifications mentioned before
In particular, it addresses and solves quite well

Problem Number 2 (Class definitions incoherencies)

LCCS: our opinion (2)
.. But there are still many remaining problems with LCCS

approach…

Problems remaining
• Plenty of information lost

• Mixed classes problems
• Spatial variations not registered
• Temporal variations not detected
• Parameters and Indicators calculations not possible
• Database mosaicking not possible
• Automatic generalization not possible

LCCS: our opinion (3)
LCCS makes the same mistake than traditional HC LU/LC

nomenclatures:
It tries to classify each polygon in one and only one class, using
the values of the “classifiers” to put sequentially the polygon to
one side or the other of the “classification rules”
(It is important to note that the values that this classifiers have in a
particular polygon are not stored in the database. Only the
polygon’s resultant class label is stored in the database)
database

Other problems of LCCS (1)
• “Special”, non-standard
“language” and software

Other problems of LCCS (2)
• Doesn’t fulfill ISO 19100 standards
• Not object oriented
• Not UML
• Attributes values not stored in the database: it is still

only a Classification Method !

Outline
1. Introduction
4. New solution:
Parametric Object Oriented databases.
Application to the Land Cover case
6. Conclusions

Back to the example: “People”
Characteristics to be considered:
- nationality
- age
- study level
- work
- residence
- eyes color
- hair color
- diseases
- married ?
- number of sons
- hobbies
- religion
…….…..

Parametric database of “people”
People
- Gender: controlled list (man, women)
- Height (m): real
- Weigh (Kg): real
- Nationality: controlled list (country table)
- Age (years): integer
- Study level: controlled list
- Work: controlled list
- Residence: text
- Eyes color: controlled list
- Hair color: controlled list
- Diseases: controlled list
- Married: boolean
- Number of sons: integer
- Hobbies: controlled list
- Religion: controlled list

One instance of “people”
John Smith
- Gender: man
- Height (m): 1.77
- Weigh (Kg): 82.6
- Nationality: USA
- Age (years): 52
- Study level: University
- Work: Engineer
- Residence: San Diego, CA
- Eyes color: brown
- Hair color: blond
- Diseases: none
- Married: yes
- Number of sons: 2
- Hobbies: golf, sailing
- Religion: protestant

What is a Data Model ?
• A data model is the description of what we are storing in a

database and how.
• It is the “link” between reality and the Database
Data
Data
World
World
Model
Model Database
Database

What is a Object Orientation ?
• “Object Orientation” is the standard in Computer Science today
• Parametric Object Oriented Data Models (POODM) are used

extensively in every type of databases and Information
Systems (airports, hospitals, production facilities,..)….
….including “some” GIS databases

UML (Unified Modeling Language)
The standard for Object Oriented Models

Used by ISO in its standards

Principal relationships between classes:
Inheritance (Generalization / Specialization): A class inherits (or

specializes) the state and behavior of another class
Aggregation: allows to capture the structural relationships

among entities in the real world (part-of)
Association: allows to capture other kinds any of relationships

among entities in the real world

Standards – ISO 19109
Example: Three features (Property Parcel, Building and

Loan)

POODM
• POODM (Parametric Object Oriented Data Models) allow an

unprecedented flexibility and capability in de design and use
of very complex Information Systems

• POODM is the best technology to address the great
complexity of Geographic information:
Represent the “systemic” nature of the world
Define “objects” of different scales (Covers, Elements,…)

and the relations between them
Assign attribute values to each one of these “objects” in a

structured and organized way

9 UML (Universal Modeling Language) lets us express,
store, modify, extend,… this structure easily and make
it understandable by anybody

POODM for Land Cover Information
• Land Cover has been traditionally modeled (Corine, Moland

Andersons,…) using classifications, legends, nomenclatures….
… All of them obsolete techniques
• Up to now (to our knowledge) POODM have not been used for
Land Cover Information

Land Cover Classifications (Moland legend)

Density thresholds
Land Cover Elements
Attributes
Percentage of polygon occupation

Density thresholds:
- somewhat “arbitrary”
- induce class proliferation

Simple components:
- not structured
- not explicit (“hidden” in de definitions of
the classes)
- incomplete
- not “extensible” (one can not subdivide a
component in different types)
Attributes:
- not structured
- not explicit in the database (“hidden” in de definitions of
the classes)
- incomplete
- actual values not stored (only “ranges”)
- not “extensible” (one can not add new attributes)

Percentage of polygon occupation:

- sometimes expressed vaguely (e.g: “predominant”)
- actual values not stored in database
- not explicit (“hidden” in the definition of the classes)

Parametric object oriented data model
(POODM)
Percentage of polygon
Land Cover elements: Attributes:
occupation:
- complete
- complete - explicit
- structured
- structured - expressed rigorously:
- explicit
- explicit (type: real, integer, boolean, list,…)
- exact values measured
- extensible - exact values measured and
and stored in database
stored in database
- extensible
• From the “Conceptual Model” in an UML diagram an expert in
Relational Databases can easily derive a “Physical
Implementation” in any standard RDBMS
• This RDB stores all the objects and attributes in a structured

and robust way, and allows to interact with all these data
through:
• SQL queries
• Different databases “crossing”
• Interaction with an specific “application”

The use of POODM for Land Cover Information has been
developed, tested and is working in the Spanish SIOSE
Project, which is in advanced production phase
(finishing by end of 2008)

A physical implementation (Relational Database and an
Application to fill it) according with data model specifications
have been developed and are in use in SIOSE production

We are presenting here an “evolution” of SIOSE Data Model,
designed to allow for:
Different scales
Different geographical areas
Different needs
Different production methodology

Core Clases
Land
Cover
Elements
Land
Cover
Classes
Basic principles of POODM (1)
We do not “classify” polygons. We describe polygons
Each polygon has one or more “Land covers” (LC)
We do not use “mixed classes”: we store in the database the % of

the polygon occupied by each “Land cover”

Basic principles of POODM (2)
Land Covers are made up of

“Land Cover Elements” (LCE)
We store in the database the % of each land cover occupied by

each element.
Each LC and LCE can have “attributes”
The actual values of all attributes for each LC and LCE are stored
in the database

Examples
Land cover 1: 100% of poligon’s surface

Artificial area. Built-up area. Residential
Land Cover Elements in it:
• 50% Artificial element. Structure. Building
+individual house
• 15% Artificial element. Artificial surface. Road
• 15% Vegetation. Trees
• 15% Vegetation. Herbaceous.
• 5% Artificial liquid

Examples
Land cover 1: 2% of polygon´s surface

Agricultural area. Herbaceous crops
+Greenhouses: yes
• Vegetation. Herbaceous.
Land cover 2: 70% of polygon’s surface

Agricultural area. Herbaceous crops
• Vegetation. Herbaceous. Wheat

Agricultural area. Permanent crop. Fruit trees plantation
• 60% Vegetation. Woody. Tree. Orange trees
• 40% Vegetation. Woody. Tree. Lemon trees
Land cover 4: 8%of polygon’s surface

Artificial area. Infrastructure. Reservoirs
• Natural terrain. Water

Examples

Natural area. Terrestrial. Natural area with woodland crops
• 45% Vegetation. Woody. Tree.
+ Pinus pinaster
• 55% Vegetation. Woody. Tree.
+ Quercus ilex

“Backward” compatibility
From a robust and well designed Parametric Object Oriented Land

Cover Database, land cover classifications and nomenclatures
can be derived by making appropriate SQL queries to the database

“Forward” compatibility
The information in existing Land Cover classifications datasets can

be entered in an adequately designed POO database:
Class definitions attributes intervals
+ attribute_max
+ attribut_min

Adequacy to INSPIRE data flows ?

Adequacy to INSPIRE data flows
Need Land Cover Object Oriented

classifications Data Models
1. Spatial NO YES
generalization When we aggregate polygons into
bigger ones, there is no way to
automatically derive the class of
the resultant polygon
2. Extensibility NO YES
- Proliferation of unusable classes,
due to the multiple “crossings” of
several classification criteria
- It is impossible to add external
information from a specialized field
to a HC database
3. Semantic Yes, but not applicable YES

generalization because of point 2.

Adequacy to Inspire data flows

4. Different update NO YES
periods One can update certain “covers”,
elements or attributes of an Object
Each polygon has a single attribute: the Oriented database more frequently.
class label. Is has to be updated at once E.g:
- urban fabric: 1 year
- forest trees: 5 years
5. Allow for NO YES

It is not possible to calculate derived
multilayer Parameters (and agri-environmental
geoprocessing indicators) based on database's
parameter values, because these values
are not stored in the database
6. Integration with NO YES

remote sensing Each polygon has a single attribute: the The mean of continuous value variables for
each polygon´s area can then be input and
class label
automatically derived stored in the OODM database as a
parameter that qualifies each polygon
parameters

Adequacy to Inspire data flows

7. Mosaicking of NO YES
existing Land Cover
databases There is no way to know de label of a One can translate existing
polygon in a DB with a different classifications to a common POODB
Nomenclature

Outline
1. Introduction
case
6. Conclusions

Classifications “hidden” in other GI themes
Road networks:
Main roads
Secondary roads
Local roads
…
Hidrology:
First order rivers
Second order
rivers
………

Outline
1. Introduction
case
6. Conclusions

Conclusions (1)
Geographic Information should not be modeled by

classifications / legends / nomenclatures, as they imply
a great decrease in the quantity and usefulness of
information stored
FAO’s LCCS (Land Cover Classification System) is not

an acceptable solution either
Land Cover, as any other G.I. theme can and should

be modeled in UML using Parametric Object Oriented
technology

Conclusions (2)
The ideas and problems presented here are applicable

to the development of many other Inspire Themes Data
Models
Most of present GI databases are obsolete, because

they were designed only to store information in paper
maps
Many of new data models being built today are also

obsolete, because they are “contaminated” (“poluted”) by
“map centric” “out of date” way of thinking. (E.g: the use
of classifications, legends or nomenclatures)¡

Conclusions (3)
Many GMES data models currently in use suffer from

this same problems

Conclusions (4)
Some of this obsolete “map centric” data models are

even being proposed as international standards (ISO,
INSPIRE,…)
CEN and INSPIRE should be careful not to adopt

obsolete data models as standards

Conclusions (5)
It is imperative to change all this data models and

databases if we want G.I. to achieve the goals that we
expect of it in the 21st century
INSPIRE is designed mainly to permit the interaction

between existing databases…
As most existing G.I. databases are “map based”….

Conclusions (6)
…. INSPIRE is in great danger of implementing

obsolete specifications (Corine Land Cover,
LCCS,…)

Conclusions (7)
We need a new ‘paradigm’ in 21st century

geographic information !

Thank you !
gmvilla@fomento.es

Extra slides…..

Principles
1) The working area must be divided in a set of closed polygons, each one
containing a surface that is as homogeneous as possible.
2) Our aim is not to classify each polygon but to “describe” each one as well
as possible. These descriptions are made associating “Land Covers”,
“Land Cover Elements” and “attributes” for them to each polygon.
3) “Land Covers” are thematic categories. They are defined with conceptual
definitions of biophysical or socio-economic criteria (morphology, structure,
relation with other land cover entities, etc….)

Principles
4) “Land Cover Elements” are the objects found in the terrain that make up
“Land Covers”. Land cover elements (e.g.: buildings, trees, rock, sand,
etc…) are the basic components of land cover.
5) Each “Land Cover” (LC) and “Land Cover Element” (LCE) has its own
attributes. Attributes are observable characteristics (biophysical or
socio-economic) that describe LC or LCE in more detail. These attributes
take different values in each “instance” (appearance of the LC or LCE).

Principles
6) Some attributes are simple variables of the adequate type (e.g.: number of
floors: integer). Other attributes are “Controlled lists” (e.g.: “vegetation
distribution geometry”). Controlled lists are defined as “enumerations” in
UML. Other attributes are more complex (e.g.: “vegetation state”) and are
represented as “UML classes” (white rectangles in the UML diagram).

Principles
7) Homogeneous polygons (at the scale of the database) have one Land Cover.
8) When homogeneous surfaces have an area smaller than the Minimum Mapping
Unit, the photointerpreter must draw a non-homogeneous polygon that
encloses areas with different characteristics. In this case, the
photointerpreter must measure (or estimate), and store in the database,
the percentage of surface in which each “Land Cover” is present in the
polygon. The sum of all percentages of each polygon must be 100 %.
9) For each “Land Cover” found in a polygon, the photointerpreter must study its
“inner composition”, and measure and store in the database:
- The exact values for each of the attributes in this “Land Cover”
- The “Land Cover Elements” present in this “Land Cover”, and the percentage of
the surface that each occupies.
- The exact values for each of the attributes in each “Land Cover Element”. (e.g.:
trunk diameter=0.40m; Number of floors = 4)
CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 100100

Principles
10) All the information of each polygon (percentage of surface of each LC

present, average values of each parameter affecting each LC or LCE,…) is
stored in an alphanumerical relational database (RDB). This RDB has
been designed with two objectives in mind:
- Materialize as exactly as possible the Parametric Object Oriented Data Model
represented in the UML diagram.
- Allow for unlimited future extensions of the model, making it as easy as possible
to add more classes, parameters, conditions, etc…

Murbandy/Moland legend (1/9)


1.2. Industrial, commercial and 1.2.1 Industrial, commercial, public and private units: 1.2.1.1. Industrial areas
transport units Artificially surface areas devoid of vegetation, occupy more then Surfaces occupied by industrial activities, including their related areas.
50% of the area in question, which also contains buildings and /or 1.2.1.2. Commercial areas
vegetated areas. Surfaces basically occupied by commercial activities, including their related areas.
1.2.1.3 Public and private services not related to the transport system
Surfaces occupied by general government, semi-public or private administrations,
including their related areas.
1.2.1.4 Technological infrastructures for public services
1.2.1.5 Archaeological sites
1.2.1.6 Places of worship
1.2.1.7 Non-vegetated cemeteries
1.2.1.8 Hospitals
1.2.1.9 Restricted access services
1.2.1.10 Agro-industrial complexes
1.2.1.11 Surface pipelines
1.2.2 Road and rail networks and associated land 1.2.2.1 Fast transit roads and associated land
Motorways, railways, including associated installations . Minimum Motorways, by-pass roads, toll-ways, etc
width to include: 25 m 1.2.2.2 Other roads and associated land
1.2.2.3 Railways and associated land
1.2.2.4 Other railways
1.2.2.5 Additional transport structures
1.2.2.6 Parking sites for private vehicles
1.2.2.7 Parking sites for public vehicles
1.2.3 Port areas
Infrastructure of port areas, including quays, dockyards and
marinas
1.2.4 Airports
Airport installations: runways, buildings and associated land

1.3.Mine, dump and construction sites 1.3.1 Mineral extraction sites

Areas with open-pit extraction of industrial minerals or other
minerals (opencast mines). Included flooded gravel pits, except for
river-bed extraction.
1.3.2 Dump sites
Landfill or mine dump sites, industrial or public.
1.3.3 Construction sites
Spaces under construction development, soil or bedrock
excavations, earthworks.
1.3.4 Abandoned land 1.3.4.1 Bombed areas
1.4.Artificial non-agricultural vegetated 1.4.1 Green urban areas 1.4.1.1. Vegetated cemeteries
areas Areas with vegetation within urban fabric. Includes parks (and
cemeteries with vegetation)
1.4.2 Sport and leisure facilities
Camping grounds, sport grounds, leisure parks, golf courses,
racecourses, etc. Includes formal parks not surrounded by urban
zones.

2. AGRICULTURAL 2.1 Arable land 2.1.1 Non-irrigated arable land 2.1.1.1 Arable land without dispersed vegetation
AREAS Cereals, legumes, fodder crops, root crops and fallow land,
flowers, vegetables, nurseries of fruit trees, whether open field, 2.1.1.2 Arable land with scattered vegetation
under plastic or glass. Includes other annually harvested plants 2.1.1.3 Greenhouses
with more than 75% of the area under a rotation system.
2.1.1.4 Drained arable land
2.1.2 Permanently irrigated land

Crops irrigated permanently or periodically, using a permanently
infrastructure. Most of these crops cannot be cultivated without an
artificial water supply.
2.1.3 Rice fields
Land prepared for rice cultivation. Flat surfaces with irrigated
channels. Surfaces periodically flooded.
2.2 Permanent crops 2.2.1 Vineyards
Areas planted with vines, when the vineyards parcels exceed the
50% of the area and/or they determine the land use of the area.
2.2.2 Fruit trees and berry plantations

Parcels planted with fruit trees or shrubs: single or mixed fruit
species, fruit trees mixed with permanent grass surfaces.
2.2.3 Olive groves

Areas planted with olive tress, including mixed occurrence of olive
trees and vines on the same parcel.

2.3 Pastures 2.3.1 Pastures 2.3.1.1 Pastures without trees and shrubs
Dense grass cover, of floral composition, dominated by
graminaceas, not under a rotation system. Mainly for grazing , but 2.3.1.2 Pastures with trees and shrubs
the fodder can be harvested mechanically.
2.4 Heterogeneous agricultural areas 2.4.1 Annual crops associated with permanent crops
Non permanent crops (arable land or pastures) associated with
permanent crops on the same parcel
2.4.2 Complex cultivation patterns 2.4.2.1 Complex cultivation patterns without settlement
Juxtaposition of small parcels of diverse annual crops, pastures
and/or permanent crops. 2.4.2.2 Complex cultivation patterns with scattered settlement
2.4.3 Land principally occupied by agriculture, with significant 2.4.3.1 Prevalence of arable land and significant areas of natural vegetation
areas of natural vegetation
Areas principally occupied by agriculture, interspersed with
significant natural areas, including wetlands or water bodies, out 2.4.3.2 Prevalence of pastures and significant areas of natural vegetation
crops
2.4.4 Agro-forestry areas
Annual crops or grazing land, covering less than 50% of the
surface, under the wooded cover of forestry species.

3. FOREST AND 3.1 Forest 3.1.1. Broad-leaved forest 3.1.1.1 Deciduous forest with continuous canopy
SEMI-NATURAL Areas occupied by forest or woodland Vegetated formation composed principally of trees, including 3.1.1.2 Deciduous forest with discontinuous canopy
AREAS with a vegetation pattern composed of shrub and bush understoreys, where broad-leaved species
native or exotic trees and which can be predominate (more than 75% of the formation) 3.1.1.3 Evergreen forest with continuous canopy
used for the production of timber or 3.1.1.4 Evergreen forest with discontinuous canopy
other forest products. The forest trees
3.1.2 Coniferous forest 3.1.2.1 Coniferous forest with continuous canopy
are under normal climatic conditions
Vegetated formation composed principally of trees, including
higher than 5 m with a canopy closure
shrub and bush understoreys, where coniferous species 3.1.2.2 Coniferous forest with discontinuous canopy
of 30% at least.
predominate (more than 75% of the formation)
3.1.3 Mixed forest 3.1.3.1 Forest mixed by alternancy of single trees with continuous canopy
Vegetated formation composed principally of trees, including 3.1.3.2 Forest mixed by alternancy of single trees with discontinuous canopy
shrub and bush understoreys, where neither broad-leaved nor
3.1.3.3 Forest mixed by alternancy of stand of trees with continuous canopy
coniferous species predominate.
3.1.3.4 Forest mixed by alternancy of stand of trees with discontinuous canopy
3.2 Shrubs and/or herbaceous 3.2.1 Natural grassland 3.2.1.1 Coarse permanent grassland / Tall herbs without trees and shrubs
vegetation associations Low productivity grassland (at least 75% of the surface). Often
3.2.1.2 Coarse permanent grassland / Tall herbs with trees and shrubs
situated in areas of rough, uneven ground. Frequently included
rocky areas, briars and heathland. 3.2.1.3 Coastal and floodplain meadows
3.2.2 Moors and heathland 3.2.2.1 Heath land

Vegetation with low and closed cover, dominated by bushes,
shrubs and herbaceous plants (heather, briars, broom, gorse, 3.2.2.2 Dwarf pine
laburnum, etc).
3.2.3 Sclerophyllous vegetation
Bushy sclerophyllous vegetation, includes maquis and garriga. In
case of shrub vegetation areas composed of sclerophyllous
species and heathland species with no visible dominance (each
species occupy about 50% of the area), priority will be given to
sclerophyllous vegetation and the whole class will be assigned
class 323.
3.2.4 Transitional woodland/shrub 3.2.4.1 Artificial young stands

Bushy or herbaceous vegetation with scattered trees. Can 3.2.4.2 Natural young deciduous stands
represent either woodland degradation or forest
3.2.4.3 Natural young coniferous stands
regeneration/recolonisation.
3.2.4.4 Wooded fens, bogs and wooded transitional bog

3.3 Open spaces with little or no 3.3.1 Beaches, dunes and sand planes 3.3.1.1 Dunes
vegetation Beaches, dunes and expanses of sand or pebbles in coastal or
3.3.1.2 Beaches
continental locations, including beds of stream channels with
torrential regime. 3.3.1.3 Inland sand
3.3.2 Bare rock 3.3.2.1 Littoral/sub-littoral rocks
Scree, cliffs, rock outcrops, including active erosion, rocks and
3.3.2.2 Coastal cliffs
reef flats situated above the high-water mark (75% of the land
surface is covered by rocks). 3.3.2.3 Inland cliffs/ bare rocks/ volcanic debris
3.3.3 Sparsely vegetated areas 3.3.3.1 Sparse vegetation on sand
Includes steppes, tundra and badlands. Scattered high-altitude
vegetation. Vegetation layer covers between 15% and 50% of the 3.3.3.2 Sparse vegetation on bare rocks
surface.
3.3.4 Burnt and damaged by disaster areas 3.3.4.1 Burnt areas
Areas affected by recent fires, still mainly black.
3.3.5 Glaciers and perpetual snow
Land cover by glaciers or permanent snowfields (glaciers and
perpetual snow more than 50%)

4. WETLANDS 4.1 Inland wetlands 4.1.1 Inland marshes 4.1.1.1 Marshes with reeds
Low-lying land usually flooded in winter, and more or less 4.1.1.2 Marshes without reeds
saturated by water all year around.
4.1.1.3 Open fen and transitional bog
4.1.2 Peat bogs 4.1.2.1 Exploited peat bog with lawn communities
Peatland consisting mainly of decomposed moss and vegetable 4.1.2.2 Unexploited peat bog with lawn communities
matter. May or may not be exploited.
4.1.2.3 Peat bog with pool communities
4.2 Coastal wetlands 4.2.1 Salt marshes 4.2.1.1 Salt marshes with reeds
Vegetated low-lying areas, above the high-tide line, susceptible to
flooding by seawater. Often in the process of filling in, gradually 4.2.1.2 Salt marshes without reeds
being colonized by halophilic plants.
4.2.2 Salines
Salt-pans, active or in process of abandonment. Sections of salt
marsh exploited for the production of salt by evaporation. They are
clearly distinguishable from the rest of the marsh by their
parcellation and embankment systems.
4.2.3 Intertidal flats

Generally unvegetated expanses of mud, sand or rock lying
between high and low water marks. 0 m contour on maps.

5. WATER BODIES 5.1 Inland waters 5.1.1 Water courses 5.1.1.1 Canals
Natural or artificial water-courses serving as water drainage
channels. Include canals. Minimum width for inclusion: 25 m 5.1.1.2 Rivers
5.1.2 Water bodies 5.1.2.1 Natural standing water

Natural or artificial stretches of water. 5.1.2.2 Artificial reservoirs
5.2 Marine waters 5.2.1 Coastal lagoons
Stretches of salt or brackish water in coastal areas which are
separated from the sea by a tongue of land or other similar
topography. These water bodies can be connected to the sea at
limited points, either permanently or for parts of the year only.
5.2.2 Estuaries
The mouth of a river within which the tide ebbs and flows.
5.2.3 Sea and ocean

Zone seaward of the lowest tide limit.

Open issues

Open issues (1)
What is the best way to model land use: new dm, complex attributes
to land cover classes, methods,…?
More “levels” in the data model (e.g: polygon > land cover >
patch/stand > population/set > individual > part > material)
Hierarchical polygons ?
Attributes values by mean + standard deviation of the

population/set?
Temporal existence/presence of elements?
“Multiple fuzzy membership”?

Open issues (2)
Model and measure “mobile” elements presence ?:
Artificial: cars, trains,..
Natural:
Animals:
Wild
Cattle
People

Open issues (3)
Buildings
Agriculture
Geology Aquaculture
Other Vegetation Soils
artificial
elements
Transport
Industrial Natural areas:
Energy Seas
Mineral Biotopes
Utilities Hydrography
Open issues (4)
Buildings
Agriculture
Geology Aquaculture
Other Vegetation Soils
artificial
elements
Land Cover =
Transport
(part of) the Core of the
Industrial Natural areas:
Inspire Consolidated Energy Seas
UML model ? Mineral Biotopes
Utilities Hydrography

Image Classification Taxonomy

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Image Classification Taxonomy

Uploaded by

Copyright:

Available Formats

From Classifications, Legends and

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008

We apologize for the annoyances it may cause

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 2 2

(*) in a certain historical time

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 3 3

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 4 4

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 6 6

 Our world is infinitely complex, and it is made by multiple

 It is constantly changing (plant fenology, agricultural cycles,

 G.I. Information uses concepts and terms:

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 7 7

 G.I. needs to “mix” concepts and terms from many different

Geology, Geomorphology, Edaphology

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 8 8

• Historically, the information has been stored in paper maps

• So it was useless to retrieve information

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 9 9

• The problem is that information you can store in a paper map is

Map-centric data model

• The first information that was introduced in digital databases

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 11 11

• Databases that would be able to store all the information

 That most of present GI databases are obsolete, because they

 That it is imperative to change all this data models and databases

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 13 13

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 14 14

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 15 15

Characteristic Possible values Number of

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 16 16

Characteristic Possible values Number of

Classification Number of classes = 2 x 3 x 3 = 18

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 17 17

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 18 18

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 19 19

250 * 100 * 4 * 100 * 250 * 5 * 4 * 20 * 4 * 10 * 20 * 10 =

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 20 20

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 21 21

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 22 22

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 23 23

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 24 24

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 25 25

(*) paradigm= unquestioned ‘true’ in a certain technical domain

December 2007 INSPIRE D2.6 workshop Ispra 26 26

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 27 27

LEVEL 1 LEVEL 2 LEVEL 3 LEVEL 4

This is the same methodology used in Land Cover

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 28 28

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 29 29

Proliferation of “difficult to use” classes, due to the multiple

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 30 30

Complex definitions, in which several “classification rules” apply

- E.g.: Moland 1.1.1.1. Residential continuous dense urban fabric:

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 31 31

Information stored in the database is much less than information

E.g.: The photointerpreter evaluates a certain polygon’s trees percentage

… But the user only receives the information

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 32 32

Need to create “mixed classes”, due to the obligation to assign a

- Mixed classes provide little information to user:

- Mixed classes lead to erroneous conclusions in the use of the

CEN TC 287 Mediterranean Workshop. Athens, 14 March 2008 33 33

Important spatial variations in certain parameters values do not

Our world is infinitely complex, and it is made by multiple

It is constantly changing (plant fenology, agricultural cycles,

G.I. Information uses concepts and terms:

G.I. needs to “mix” concepts and terms from many different

That most of present GI databases are obsolete, because they

That it is imperative to change all this data models and databases

Land Cover Classification System (LLCS) has been developed

LCCS establishes a standardized comprehensive system that

LCCS correctly exposes some of the problems of Land

In particular, it addresses and solves quite well

LCCS makes the same mistake than traditional HC LU/LC

The standard for Object Oriented Models

Example: Three features (Property Parcel, Building and