Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7,...

21
Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001
  • date post

    21-Dec-2015
  • Category

    Documents

  • view

    214
  • download

    0

Transcript of Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7,...

Page 1: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Measuring Accuracy of Street Centerline Datasets

Donald Cooke

Founder, GDT

CLEM2001

August 6-7, 2001

Page 2: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Accuracy of Centerline Files

• History: NMAS

• History: GDT involvement

• NSSDA, July 1998

• GDT procedure

• Results

• 3 meter accuracy

Page 3: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

History, NMAS

• NMAS: National Map Accuracy Standard

• Promulgated in 1940’s

• Production Standard: “Build to this Spec.”

• Binary (pass-fail) standard

• Tied to scale/paper/”Gutenberg Disease”

• Did they ever test the quads? How?

Page 4: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

History, GDT Involvement

• Geocoding accuracy

Page 5: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.
Page 6: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Tampa, Florida, April 1997Red: DGPS positions of addressesBlack: Interpolated geocoded addressesMost of error budget comes from storing potential address ranges…….

Page 7: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

History, GDT Involvement

• Geocoding accuracy

• GDT inherited spatial accuracy of TIGER

• Need to realign streets and improve accuracy

• Need to qualify sources for realignment

• Need to Q/C results of realignment operation

Page 8: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

NSSDA

• NSSDA: “National Standard for Spatial Data Accuracy”

• July 1998; draft available well in advance

Page 9: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

NSSDA Procedure

• “accuracy testing by an independent source of higher accuracy is the preferred test”

• Compile a test suite of coordinate measurements for well-defined points in the study area

• Extract corresponding points from the dataset being evaluated

• Statistically compare the two samples by computing RMSE and "accuracy" measures

• ("accuracy" = RMSE * 1.7308)

Page 10: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

GDT Procedure

• Well-defined points: “T” or “Cross” intersections

• Park in center of intersection; average for 1 minute

• Post-processed code-phase differential

• ~30, and later ~45 points collected per sample

• Sample skewed to exclude major road intersections

• Sample sometimes adjusted to fall within DOQQ or 7.5 minute quad tile.

Page 11: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Sample point spacing; Deerfield Illinois

Page 12: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Results

• 56 areas surveyed; more in process

• TIGER: 16-100+ meter RMSE

• Dynamap aligned to DOQQ: 4-6 meter RMSE

• See some results...

Page 13: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Test Area Number TIGER Dynamapof Points RMSE RMSE

(meters) (meters)Sarasota, FL 39 26.5 4.3Ann Arbor, MI 29 46.7 4.5Deerfield, IL 29 27.0 4.2Manchester, NH 36 25.5 4.6Morristown, NJ 48 48.2 6.1Greenfield, MA 70 21.3 6.0Colma/Pacifica, CA 51 57.1 6.4W Palm Beach, FL 38 44.2 3.0* Utility GISDenver, CO 37 16.7 3.4 DatasetWarwick, RI 35 37.6 5.2San Diego, CA 31 20.9 4.1

Page 14: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

0

50

100

150

200

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29

Check point #

Error, Meters

Absolute error for 29 check points, Ann Arbor, MI

Page 15: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

0

50

100

150

200

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29

Check point #

Error, Meters

Absolute error for 29 check points, Ann Arbor, MI

RMSE: 46 meters RMSE of “good” pts: 20 meters; bad: 102 meters

Page 16: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.
Page 17: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.
Page 18: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Results

• TIGER spatial errors are not normally distributed

• Large population from 1:100K DLGs

• Many from 1980 GBF/DIME files

• Some streets “cartooned” freehand

• Picking sample is crucial

• Cannot make a blanket statistical statement that describes TIGER spatial error

Page 19: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Results, continued

• 1:24,000 DLGs are spotty; South Carolina exp.

• Some (25%?) 7.5 minute quads don’t pass NMAS

• Impossible to test accuracy until recently

• 1 meter DOQs appear to be boringly reliable

Page 20: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Conclusions on NMAS vs NSSDA

• NSSDA is a workmanlike improvement

• No longer a binary test

• No longer tied to scale of analogue map

• Procedure extensible to testing GPS accuracy

• But still considers “map” to be a monolithic, one-time compilation to a single standard

• Need to carry accuracy metadata on each object

Page 21: Measuring Accuracy of Street Centerline Datasets Donald Cooke Founder, GDT CLEM2001 August 6-7, 2001.

Census TIGER/2010 Accuracy

• 3-meter or better accuracy (=1.7 meter RMSE)

• Cannot achieve this from off-the-shelf DOQs

• Must use new imagery or drive DGPS

• Code-phase differential GPS is marginal for test suite

• We use carrier-phase for test points