fuzzy land cover/land use ontologies – geovocamp 2011 ola ahlqvist, department of geography fuzzy...

Fuzzy Land Cover/Land Use Ontologies – GeoVoCamp 2011 Ola Ahlqvist, Department of Geography

Fuzzy Land Cover/Land Use Ontologies

Ola Ahlqvist, PhD

Department of Geography

The Ohio State [email protected]


Land use and land cover data

• Importance– climate modeling, urban planning, landscape change

assessment, hydrological models, and unknown future issues

– Demand that data can be re-purposed for a variety of end uses

• Initiatives– Standards: National Vegetation Classification Standard

(Vegetation Subcommittee, 1997), the Nordic Landscape Monitoring Project (Groom, 2005), the CORINE Land Cover (CEC, 1995 and 1999; Bossard et al., 2000), the standard classification for land cover of South Africa (Thompson, 1996), GLC2000 (Bartholomé and Belward, 2005) , UNEP/FAO Land Cover Classification System (LCCS) (Di Gregorio and Jansen, 2000)


A traditional land cover taxonomy

• Easily translated to an ontology description language as class-subclass relations and class specific properties

• But no agreement on a unified taxonomy– CORINE– GlobCover– MODIS/IGBP

• Not even our own USGS system could stay the same from one time to the other – National Land Cover Data (NLCD) used slightly different

classes in 1992 and 2001


11 Open Water

21 Low Intensity Residential

22 High Intensity Residential

23 Commercial/Industrial/Transport.

32 Quarries/Strip Mines/Gravel Pits

33 Transitional

41 Deciduous Forest

42 Evergreen Forest

43 Mixed Forest

81 Pasture/Hay

82 Row Crops

85 Urban/Recreational Grasses

91 Woody Wetlands

92 Emergent Herbaceous Wetlands

0 5 10 15 20

Kilometers

t

Land CoverChester County, PA

1992

11 Open Water21 Developed, Open Space22 Developed, Low Intensity23 Developed, Medium Intensity24 Developed, High Intensity31 Barren Land (Rock/Sand/Clay)41 Deciduous Forest42 Evergreen Forest43 Mixed Forest81 Pasture/Hay82 Cultivated Crops90 Woody Wetlands95 Emergent Herbaceous Wetlands

0 5 10 15 20

Kilometers

t

Land CoverChester County, PA

2001

National Land Cover Data (NLCD) change example

Different classification systems create problems!

??


0

2

4

6

8

10

12

14

16

0 20 40 60 80 100Canopy cover (%)

Tre

e h

eig

ht

(m) Sudan

UNESCO

Tanzania

Jamaica

Zimbabwe

Turkey

United States

Estonia

China

Same issues around global definitions of “forest”

after Lund (2006) and Comber et al. (2006)


Solutions?

• The FAO and UNEP Land Cover Classification System (LCCS)– multi-purpose classification system – capable of comparing land cover types across taxonomies– uses diagnostic criteria rather than pre-defined classes

• Now at v.3 - Land Cover Meta Language (LCML)– a predefined set of land cover basic objects are enriched on

their semantic significance with external qualities and attributes

– Use of UML and XML for formal description of an ontology– http://www.glcn.org/ont_2_en.jsp


Example formalization of land cover definitions

CORINE Land Cover Class Name

Corine Code LCCS classifiers LCCS Code

Broadleaved forest 3.1.1. Major land cover class: Natural and Semi-Natural Terrestrial Vegetation (A12) Life form : Trees (A3) Cover: Closed, > (70-60)% (A10), (70-60)%-40% (A12) Height: >30 – 3 m (B2) Leaf Type: Broadleaved (D1) Leaf Phenology: Deciduous (E2)

20090 / 20132-1

Peatbogs 4.1.2. Major land cover class: Natural and Semi-Natural Aquatic Or Regularly Flooded Vegetation (A24)

Life form : Herbaceous (A2) Cover: Closed > (70-60)% Height: >3 – 0.03 m (B4) Water seasonality: On waterlogged soil (C3)

Major land cover class: Natural and Semi-Natural Aquatic Or Regularly Flooded Vegetation (A24)

Life form : Lichens/Mosses (A7) Cover: Open (70-60)-(20-10)% Water quality: Fresh

Major land cover class: Bare Areas (B16) Surface aspect: Bare soil a/o Other Lithology: Organic rock - Peat

40057 / 40985-R1 / 6005-M251


0

2

4

6

8

10

12

14

16

0 20 40 60 80 100

Canopy cover (%)

Tre

e h

eig

ht

(m) Sudan

UNESCO Virgin Islands

Tanzania

Ethiopia

South Africa

Jamaica

Zimbabw e

Morocco

MozambiqueTurkey

United States

Mexico

Netherlands

Cambodia

Kenya

Estonia

Kyrgyzstan

China

B1

2-7

m

B2

> 3

m

B7

3-7

m

B6

7-1

4 m

B5

> 1

4 m

A16 1-4 %

A15 4-15 %

A13 15-40%

A12 40-65%

A10 > 65% LCCSClassifiers

Parameterization using FAO’s Land Cover Classification System v.2

• Unnecessarily crude where detail is actually available


With v.3 - Land Cover Meta Language (LCML)

Domain Scale Range

waterCov.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

waterPhase.owl Nominal {Ice, Water}

imperviousPct.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

vegetationCov.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

development.owl Nominal {Residental, Commercial, Mining}

surfaceType.owl Nominal {Earthen material, Constructed}

treeCov.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

treeHeight.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

deciduousPct.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

evergreenPct.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

shrubCoverPct.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

woodyTenure.owl Nominal {(Semi)Natural, Cultivated/Planted}

grassHerbTenure.owl Nominal {(Semi)Natural, Cultivated/Planted}

grassHerbCoverPct.owl Ratio [0 10 20 30 40 50 60 70 80 90 100]

crop.owl Nominal {RowCrops, SmallGrains, Fallow, Hay, Grass}

waterPersistence.owl Nominal {Permanent, Periodically, Waterlogged}

U.S. NLCD 1992 Low Intensity Residential


0 20 40 60 80 100 Canopy cover (%)

Woodland (Hyytiäinen, 1995)

Woodland (USDA F.S., 1997)

Overlap

Distance

Attributes values allow for evaluation of category semantics

• Two metrics of semantic relations– Distance– Overlap

Domain Scale Range

















Domain Scale Range

















U.S. NLCD 1992 Low Intensity Residential U.S. NLCD 2001 Developed, Low Intensity


0 20 40 60 80 100 Canopy cover (%)

Woodland (Hyytiäinen, 1995)

Woodland (USDA F.S., 1997)

Overlap

Distance

Attributes values allow for evaluation of category semantics

Ov

erl

ap

-1Distance

Very similar classes

Similar but Disjoint

classes

Class/subclass

relationship

Very differentclasses

• Two metrics of semantic relations– Distance– Overlap

• Bivariate color scheme– Different types of change


But land cover/use concepts are rarely clear cut…

• Fuzzy set theory and fuzzy numbers allow for vagueness

“Closed Tree Cover”

“Open Shrub Cover”

“Sparse vegetation”


…and they typcially have many attribute dimensions

Domain Scale Range

















Domain Scale Range

















Comparison in 16D can be summarized

123456789

10111213141516

123456789

10111213141516

U.S. NLCD 1992 Low Intensity Residential U.S. NLCD 2001 Developed, Low Intensity


1992/11

1992/21

1992/22

1992/23

1992/32

1992/33

1992/41

1992/42

1992/43

1992/81

1992/82

1992/85

1992/91

1992/92

2001Open water 5.6 0 0 0.4 0.1 0 0.5 0.2 0.1 0.4 0.1 0 0 0

Developed, Open Space 0.3 51.6 0.8 4.8 0.3 0.5 19.9 4 10.5 32.0 5.3 2.8 0 0.2

Developed, Low Intensity 0.3 23.5 1.6 6.8 0.6 1.3 12.1 2.1 3.6 21.3 5.6 0.1 0 0.3

Developed, Med. Intensity 0.3 9.3 1.7 9.4 1.3 1.1 3.5 0.4 0.5 5.9 2.7 0 0 0.2

Developed, High Intensity 0.1 1.9 0.6 4.7 0.7 0.4 0.5 0 0.1 0.9 0.7 0 0 0

Barren Land (Rock/Sand/ Clay) 0.1 2.8 0 0.3 0.3 0.3 7.8 1.7 2.1 4.1 0.8 0 0 0.1

Deciduous Forest 1.2 10.5 0.1 1.4 0.1 0.8 408.7 8.2 22.6 73.9 3.8 0 2.4 1.3

Evergreen Forest 0.1 1.4 0 0.1 0 0 7.3 4.8 3 2.8 0.1 0 0 0.1

Mixed Forest 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Pasture/ Hay 1.1 19.4 0.3 2.0 0.1 12.8 114.5 26.2 40.8 476.3 54.1 0.5 0.8 1.3

Cultivated Crops 0.6 6.7 0.2 2.4 0.4 2.3 73.5 5.5 10.7 169.3 59.6 0.1 0.6 1.1

Woody Wetlands 1.2 1.6 0 0.6 0.1 0.1 19.9 1.4 1.3 5.2 0.6 0 1.7 0.7

Emergent Herbaceous Wetlands 0.8 0.4 0 0.4 0.1 0 2.6 0.4 0.3 2.2 0.6 0 0 0.3

1992

Op

en W

ater

Low

inte

nsity

resi

den

tial

Hig

h in

ten

sity

resi

den

tial

Co

mm

erci

al/

Indu

stri

al/

Tra

nsp

orta

tion

Qu

arri

es/

Str

ip M

ines

/

Gra

vel P

its

Tra

nsiti

onal

De

cid

uous

For

est

Eve

rgre

en

For

est

Mix

ed

For

est

Pas

ture

/Ha

y

Ro

w C

rops

Urb

an/

Rec

reat

iona

l

Gra

sse

s

Wo

ody

Wet

land

s

Em

erg

ent

He

rba

ceo

us

We

tland

s

Semantic relations summarized in matrix form

Ove

rla

p-1

Dissemblance


Water

Developed

Barren

Vegetated (Forest)

Shrub land

Non-natural woody

Grasslands

Herbaceous planted

Wetlands

National Land Cover Data

MDS of class definitions

1992 2001

… with the semantic similarity metrics we can compare within and across conceptual spaces

Developed, High Intensity

Developed, Medium Intensity

Developed, Low Intensity

Developed, Open Space

Commercial/Industrial/ Transportation

High Intensity Residential

Low Intensity Residential

Transitional

Bare Rock/ Sand/Clay

Quarries/Strip Mines/Gravel Pits

Barren Land (Rock/Sand/Clay)

Fallow

Orchards/ Vineyards/ Other

Open Water

Perennial Ice/ Snow

ShrublandShrub/Scrub

Deciduous Forest

Evergreen Forest

Mixed Forest

Grasslands/ Herbaceous

Emergent Herbaceous Wetlands

Woody Wetlands

Row CropsSmall GrainsPasture/Hay

Urban/Recreational Grasses

Cultivated Crops

Pasture/Hay

Developed, High Intensity

Developed, Medium Intensity

Developed, Low Intensity

Developed, Open Space

Commercial/Industrial/ Transportation

High Intensity Residential

Low Intensity Residential

Transitional

Bare Rock/ Sand/Clay

Quarries/Strip Mines/Gravel Pits

Barren Land (Rock/Sand/Clay)

Fallow

Orchards/ Vineyards/ Other

Open Water

Perennial Ice/ Snow

ShrublandShrub/Scrub

Deciduous Forest

Evergreen Forest

Mixed Forest

Grasslands/ Herbaceous

Emergent Herbaceous Wetlands

Woody Wetlands

Row CropsSmall GrainsPasture/Hay

Urban/Recreational Grasses

Cultivated Crops

Pasture/Hay


Semantic change image

• Summarize landscape change from a cognitive perspective

• An overall, spatially explicit evaluation of land cover change throughout the study area

• Nuanced assessments of graded changes even for heterogeneous, nominal land cover types


Some references

Ahlqvist, O., 2004, A parameterized representation of uncertain conceptual spaces, Transactions in GIS, 8(4), 493-514.

Ahlqvist, O., 2008, Extending post classification change detection using semantic similarity metrics to overcome class heterogeneity: a study of 1992 and 2001 National land Cover Database changes, Remote Sensing of Environment, 112(3):1226-1241

Comber, A., Fisher, P., and Wadsworth, R., 2006, What is land cover?, Environment and Planning B, 32: 199-209

Gärdenfors, 2000, Conceptual Spaces: The geometry of thought, MIT press.Kaufman A and Gupta M M, 1985, Introduction to fuzzy arithmetic. New York, Van

Nostrand Reinhold CompanyLund, H.; Gyde (coord.) 2006. Definitions of Forest, Deforestation, Afforestation, and

Reforestation. [Online] Gainesville, VA: Forest Information Services. http://home.comcast.net/~gyde/DEFpaper.htm.

Global Land Cover Network (GLCN) - http://www.glcn.org/index_en.jsp- Land cover Ontology - http://www.glcn.org/ont_0_en.jsp- ISO standard - Part 1 & 2 http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=32562http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=44342

U.S. National Land Cover Database - http://www.mrlc.gov/

fuzzy land cover/land use ontologies – geovocamp 2011 ola ahlqvist, department of geography fuzzy...

Documents

land cover types

corine land cover cec

barren land rocksandclay

land coveruse concepts

land cover meta language

phd department of geography

low intensity slide

fuzzy numbers