Georeferencing:Theory and Challenges

Dr Michael RigbyAAH-DARIAH-ARDCThursday 28 March Knowledge Exchange Session 2 GeoHumanities

rigby@unimelb.edu.au

22

Entities

33

ST Representation

<algorithm>

<references>

4

GIScience is an information science(Kemp, 2008)

Refers to the scientific study of geographic information (GI)

Requires understanding of▪ Fields of inquiry▪ Knowledge creation methods▪ Shared expertise across fields

(Duckham, 2017)

Information science

55

Perspectives

EpistemologyTeleology

6

Definition:

The linking between an entity and a spatial footprint

Entity must have spatial grounding

Georeferencing

7

SemioticsO

gden

an

d R

ich

ard

s (1

92

3)

8

Georeferencing process

Input

Reference data

ParsingFeature

MatchingFeature

InterpolationOutput

General components (Goldberg, 2017)

9

1. TextCharactera) Structured (e.g. address)b) Unstructured (e.g. toponym)

Integerc) Structured (e.g. joining IDs)

2. Rasterd) Grid (e.g. image transformation)

Example input

10

a) Address geocoding

Assigning an address a spatial footprint

Input: Text (char, structured)

Example: “64 Lincoln Ave Melbourne Australia”

Goldberg, Wilson, and Knoblock (2007); Hill (2006)

11

a) Address geocoding

Reference data:▪ Address file

▪ G-NAF Live ▪ G-NAF Open (3 months)

Feature Matching

Reference data

https://www.psma.com.au/products/g-naf

12

a) Example geocoding levels

“64 Lincoln Ave Melbourne Australia”

“Lincoln Ave Melbourne Australia”

“Melbourne Australia”

“Australia”

Point

Line

Polygon

Polygon

Address Input Geocoding Output

13

But existing tools are a black box

GoogleBingMapboxPSMAGisgraphyHEREGeocode.xyzLocationIQTomtomGeocode.farmYahoo BOSSgeocode.earthSmartyStreets…

Feature matching algorithm?Reference datasets?

14

Repurposing / Tool Making

Digital methods –

How do we know that one approach is appropriate for another’s purpose?

15

Example gazetteer

Observatory Hill, SA

How might this place be represented?

What else might we need to consider?

Location vs place

16

b) Toponym resolution

Assigning a toponym (place name) a spatial footprint

Input: Text (char, unstructured)

Example: “Lake Macquarie”

17

b) Toponym resolution

Input ParsingFeature

MatchingFeature

InterpolationOutput

Adapted from Goldberg (2017)

Training Corpora

NERAmbiguity Resolution

Reference data

18

b) Toponym examplePlace description: NSW 1881 CensusSource: http://hccda.ada.edu.au/

Country?

Admin or Topographic?

Topographic?

tinker.edu.au

19

b) Toponym resolution

Reference data:▪ Gazetteers

▪ National (e.g. GA)▪ State (e.g. VICNAMES)

▪ Other▪ GeoNames (http://www.geonames.org/)▪ DBpedia (https://wiki.dbpedia.org/)▪ ANPS (http://www.anps.org.au/)▪ …

20

Multiple candidates“Lake Macquarie” – “Awaba”

http://www.geonames.org/search.html?q=lake+macquarie&country=AU

Bas

e la

yer:

Go

ogl

e Ea

rth

(2

01

9)

What ST representation?

21

Ambiguity resolution ST

The process of identifying a single spatial footprint from multiple candidates

This process can be assisted using:▪ Confidence scores▪ Ontologies▪ ST extents▪ Previous research (feedback)

22

NER and chunking

Context is critical

“Lake Macquarie”

Topological relations

“Kahibah at the entrance of … Lake Macquarie … thence by”

Places can change…The Electoral District of Kahibah was created in 1894 … It was abolished in 1920 with the introduction of proportional representation. It was recreated in 1927.It was abolished and partly replaced by Waratah in 1930. It was recreated in 1950It was abolished again in 1971 and replaced by Charlestown.

Source: DBpedia.org

23

Sands & McDougall Directories

Sou

rce

: h

ttp

://w

ww

.kin

gsto

n.v

ic.g

ov.

au/l

ibra

ry

Yellow Pages

Source: http://www.abc.net.au/news

Historical challenges

What about bias?

24

External world knowledge cannot be derived from linguistic principles alone

Leidner (2017)

Note

25

Thought experiment

Imagine we had a complete repository of

reference data

for the entire world…

for all time…

Could we identify a location?

Could we identify a place?

26

c) Joining IDsSo

ftw

are:

ESR

I Arc

Map

10

.x

ABS SA2 Boundaries 2016

27

d) Image transformation

htt

p:/

/des

kto

p.a

rcgi

s.co

m/e

n/a

rcm

ap/1

0.3

/man

age

-dat

a/ra

ster

-an

d-

imag

es/f

un

dam

enta

ls-f

or-

geo

refe

ren

cin

g-a-

rast

er-d

atas

et.h

tm

Soft

war

e: E

SRI A

rcM

ap 1

0.x

Funding for AURIN has been provided by the Australian Government under the National Collaborative Research Infrastructure Strategy (NCRIS) and associated programmes.

AURIN Administrative OfficeThomas Cherry BuildingCorner Swanston and Elgin Street, Carlton(entrance through Level 2, McCoy Building, The University of Melbourne VIC 3010T: +61 3 8344 3212E: admin@aurin.org.auwww.aurin.org.au

@aurin_org_au

Thank you

Steve Bennett, Steve McEachern, Steve Cassidy, Rob Hutton and members of the HASS DeVL team

Contact:

Dr Michael RigbyAURIN, The University of Melbournerigby@unimelb.edu.au