Quality Control. of State-of-the-art Digital Elevation Data. Thomas Knudsen, Danish Geodata Agency.

Quality Control.of State-of-the-art Digital Elevation Data.

Thomas Knudsen, Danish Geodata Agency

State of which art?

Admission:

This is really a case study using new Danish elevation data.

Brief detour: History of Danish elevation data.

History of Danish elevation data

1880-1910: Military surveyors, 2 m contours.

2007 - 2008: LiDAR, ½ point / m2

2014 - 2015: LiDAR, 4 points / m2

“Beauty contest”

Spring 2013: Tender - data for a new Danish elevation modelSelection criteria: Minimum requirements (“at least as good as 2007 data”), fixed maximum price (“at most xx MDKK”)Fall 2013: Contract awarded to a consortium led by AeroData Nederland

State of which art?

The art of being in the sweet spot between

1. Data density2. Data quality3. Price

Specification philosophy

1. Do not specify anything that cannot be checked

2. Check everything specified

What are we trying to achieve?

1. Make sure data delivered are up to specs2. Create metadata documenting just how good

data are3. Check consistency with other geodata

getting better impression of the accuracy of the entire geodata stack

Unøjagtig Upræcis

Inaccurate Imprecise

Accurate Imprecise

InaccuratePrecise

AccuratePrecise

Ver

tica

l HorizontalPre cis eAccurate

Koregt klars

D e n s it y

2007 2014

Horizontal Precision ? < 15 cm

Accuracy 60 cm 15 cm

Vertical Precision 5 (8) cm 2 (5) cm

Accuracy < 15 cm 5 cm

How?

● Delivery in blocks < 40 km x 40 km● Blocks consist of up to 1600 1 km x 1 km

tiles● Most checks tile based● Trivially parallellizable

Density - how

Count – check suspicious parts in QGIS

Vertical precision - how?

Internal consistency along road centerlines from the MapDB

Vertical accuracy - how

Comparison with previously surveyed control patches.

Horizontal: Major problem

How do we check that the achieved horizontal accuracy is 15 cm?

Solution

Photogrammetrically derived building outlines

Horizontal accuracy

Comparison with hypothetical roof ridges derived from building outlines from the MapDB.

Point cloud roof reconstruction (3D Hough transform)

Roof ridge is the comparison target.

Horizontal precision

Similar, but flight strip vs flight strip in the overlap zone.

Roof ridge, LiDAR vs. MapDB

Classification check - how

S. R. A. F. A. P(ask your friendly experimental physicist neighbour!)

But they need some material to run with…

● Points classified as buildings outside building outlines

● Building outlines with no building points● Buildings on roads● Vegetation below buildings…

Implementation● Heavy lifting done by OSGeo et al. open

source products (GDAL, PostGIS, NumPy, SciPy, QGIS etc.)

● Tied together using internally developed code (also Open Source)

● LAStools (mostly the open source parts)● i.e. simple to scale: No licensing worries

Examples

Forest

Results

Very nice!

● Things work● Data quality is high● Supports a smooth communication with the

data provider● We work on releasing source code

Kort & Matrikelstyrelsen SIDE 40




Quality Control. of State-of-the-art Digital Elevation Data. Thomas Knudsen, Danish Geodata Agency.

Documents