How Many Volunteers Does It Take To Map An Area Well? Dr Muki Haklay Department of Civil, Environmental and Geomatic Engineering, UCL [email protected].

How Many Volunteers Does It Take To Map An Area Well?

Dr Muki Haklay Department of Civil, Environmental and Geomatic Engineering, UCL [email protected]

Aamer Ather (M.Eng 2009), Sofia Basiouka (MSc GIS 2009) and Naureen Zulfiqar (M.Eng 2008)

Ordnance Survey data was kindly provided by the Ordnance Survey research unit. OSM data was provided by GeoFabrik & CloudMade

Outline

• A bit about quality of geographical information

• Evaluation of OSM with Meridian data set• Evaluation of OSM with MasterMap• Linus’ low –more users: higher quality?

The quality issue

• How good it the data?

– First question: good for what? Subjective quality – fitness for purpose/use

– Second question: how to measure? Objective quality – but need to evaluate it in light of the first question

The quality issue

• How good it the data? – Positional accuracy – the position of features

or geographic objects in either two or three dimensions

– Temporal accuracy – how up to date is the data? Does it presents the existing situation and when will it be updated?

– Thematic/attribute accuracy – for quantitative attributes (width) and qualitative attributes (geographic names)

– Completeness – The presence and absence of objects in a dataset at a particular point in time

– Logical consistency –adherence to the logical rules of the data structure, attribution and relationships

The ‘problem’

“We know little about the people that collect it, their skills, knowledge or patterns of data collection”

“Loose coordination and no top-down quality assurance processes – can’t produce good data”

“It is not complete and comprehensive – there are white areas”

Coverage and completeness

Coverage and completeness

Completeness – difference by user?

Patterns of collaboration

Number of Users Area covered (Sq Km)1 400212 207203 91364 41845 19866 9367 4488 2699 13910 and above 246

Users

• Limited ‘on the ground’ collaboration. Important as this can be the main source of quality assurance - ‘Given enough eyeballs, all bugs are shallow’ (Raymond, 2001)

• Translate to VGI it might mean:“The more users there are per area, the better is the positional and attribute quality”

• But does Linus’ law apply to OSM (and to VGI)?!?

Accuracy and Completeness- Study I

• Comparing OSM to OS Meridian 2 roads layer

• Maridian 2 -Motorways, major and minor roads are... Complex junctions are collapsed to single nodes and multi-carriageways to single links... some minor roads and cul-de-sacs less than 200m are not represented... Private roads and tracks are not included...

• Nodes are derived from 1:1,250-1:2,500 mapping, with 20m filter around centre line generalisation

Positional Accuracy

• Meridian 2 and OSM – Motorway comparison

Goodchild and Hunter (1997), Hunter (1999) method

• Assuming that one dataset is of higher quality

• Create buffer around the dataset with known width

• Calculate the percentage of the evaluated dataset that falls within the buffer

Motorway comparisonMotorway Percentage

OverlapM1 87.36%M2 59.81%M3 71.40%M4 84.09%M4 Spur 88.77%M10 64.05%M11 84.38%M20 87.18%M23 88.78%M25 88.80%M26 83.37%M40 72.78%A1(M) 85.70%A308(M) 78.27%A329(M) 72.11%A404 76.65%

• Buffer of 20m

• Average of 80% - ranging from 59.81% to 88.80%

• Data used for comparison: OS MasterMap Integrated Transport Network (ITN) layer

• ITN consists of road network information

• The most accurate and up-to-date geographic reference for Great Britain’s road structure – Any major real world changes are

updated within 6 months • Used for numerous applications

– e.g. Transport management systems, road routing, emergency planning...

Comparison II – Ordnance Survey Master Map

• Four test locations chosen:

TQ28se TQ38se

TQ17ne

TQ37sw

• Buffer analysis – again based on Goodchild and Hunter (1997) buffer comparison technique:

X

ITNOSM

Buffer width (X):

Comparison methodology

Road Type Buffer size

Motorways 8 metres

A-roads 5.6 metres

B-roads 3.75 metres

• Buffer overlap results:– 109 roads examined covering over 328 km

TileAverage Percentage Overlap

A-roads B-roads Motorways

TQ38se (East London) 87.27% 72.34% -

TQ28se (North/Central London)

88.42% 81.46% -

TQ37sw (South London) 92.62% 77.28% -

TQ17ne (West London) 84.30% 71.52% 98.85%

Results of Master Map comparison

TQ38se (East London) TQ28se (North/Central London)

TQ37sw (South London)

TQ17ne (West London)

0 2000 4000 6000 8000 10000 12000 14000 160000

10

20

30

40

50

60

70

80

90

100

Goodchild-Hunter accuracy estimation of OSM A-roads and B-roads

A-RoadsB-Roads

Road length (m)

Per-

centa

ge o

verl

ap

Road Length of road (km) Percentage overlapA12 15.10 90.54%M4 13.21 98.85%A4 12.00 74.77%A11 9.79 96.46%A13 9.34 86.81%A41 8.95 89.05%A40 8.53 86.24%A400 8.26 92.21%A1203 7.98 75.62%A23 7.96 84.75%A501 7.37 89.12%A215 6.89 93.92%A205 6.80 88.48%A1261 6.03 88.94%A315 5.91 92.44%

Quality not linked to length

Completeness – bulk method

• Assumption: as Meridian 2 is generalised, so for each sq km:If Total length(OSM roads)>Total length(Meridian 2

roads)Than OSM is more complete than Meridian 2

• The comparison can also includes attributes, by testing for the number of objects with complete set of values

1

2

3

5

Methodology

4

Change in completeness Mar 2008 – Mar 2010

England – March 2008

England – March 2009

England – October 2009

England – March 2010

Completeness with attributes

• The test for completeness with attributes checks that roads and streets names have been completed

• Until the release of Ordnance Survey data in 1st April 2010, this was a good indication for ground survey of an area

England – March 2008

England – March 2009

England – October 2009

England – March 2010

Linus’ law and OSM

Conclusions

• OSM quality is high – and it is assumed that the quality is coming from aerial imagery

• Linus’ Law does not seem to apply in a straight forward manner – at least not from 5 and above

• More research is required for lower numbers or participants and different quality of imagery

Further reading • Haklay, M., 2008, How good is OpenStreetMap

information? A comparative study of OpenStreetMap and Ordnance Survey datasets for London and the rest of England, submitted to Environment and Planning B.

• Haklay, M. And Weber, P., 2008, OpenStreetMap – User Generated Street Map, IEEE Pervasive Computing.

• Haklay, M., Singleton, A., and Parker, C., 2008, Web mapping 2.0: the Neogeography of the Geoweb, Geography Compass

• Haklay, M., 2008, Open Knowledge – learning from environmental information, presented at the Open Knowledge Conference (OKCon) 2008, London, 15 March.

• Haklay, M., 2007, OSM and the public - what barriers need to be crossed? presented at State of the Map conference, Manchester, UK, 14-15 July.

• To get a copy, write to [email protected] , or get them on povesham.wordpress.com