Having replied off list to a number of people about this, I have been asked
to pass on a short exercise, aimed at estimating the errors in a small
hand-held GPS receiver - brand and model removed to protect the innocent -
On 12 February a number of GPS points were recorded in and around the
northern part of Leeds University. They were collected using a small
hand-held GPS receiver and, at each point, the GPS national grid reference,
the GPS HoD, a description of the landcover at the GPS point and the time
were recorded in a notebook. The data was transferred onto a notebook
computer and made into a DBase IV file using Microsoft Excel. The DBF file
was then included into an ArcView project along with a 50m resolution DEM,
an Institute for Terrestrial Ecology (ITE) 25m resolution land cover map and
three OS Landline 1:2,500 tiles.
In the ArcView project the best estimate of where the 'real' point actually
lay was recorded. In some case it was not possible to establish the 'real'
point and in these cases the symbol was positioned directly under the GPS
point. From this map the estimated errors in the GPS positions was derived
and is given in the table below. Only at points, 1, 4 and 15 are the GPS
points over 10m away from the 'real' points. Both points 1 and 15 appear to
have been affected by the proximity of tall buildings and are 19m and 17m
respectively in error. The error in position of point 4 is less easy to
explain and it is possible that I have made some error in the estimation of
the 'real' point.
The table also shows the same points with their GPS heights and data from
the OS Panorama 50m resolution DEM and 'real' heights estimated from 5m
contours and the spot heights on Landline mapping. The GPS heights are
consistently higher than the heights indicated by the DEM and it is
interesting to note that points 3, 4, 5 and 6 were taken going up a
reasonable slope and yet GPS heights at 4, 5 and 6 were the same and lower
than 3, which was visibly the lowest position. The greatest difference
between the DEM and the GPS height is 24m (c.30%) at point 3, the error
probably caused by interference from the tree cover and surrounding
buildings. Having said this 24m is only a very small fraction of the
distance from the satellites to the receiver - thus small variations in
reception can make significant differences.
Finally the ITE landcover classes were compared against the classes recorded
at the GPS points. It is clear that outside the urban/suburban classes the
ITE landcover mapping is not particularly accurate at this level of detail.
The managed grass and (fairly open) woodland in both the park and cemetery
have been represented as tilled land, pasture, heath classes and meadow. It
is clear from this that the land cover mapping has been automatically
compiled from remote sensing data and has had little, if any, hand
adjustment or ground truthing in this area or this type of landscape. The
reflectance values used to assign the landcover class are in error here,
which may be due to cloud shadows or other similar causes.
In conclusion it can be seen that all data sets, even those compiled from
'high tech' devices or by national agencies, are prone to error. Working
with these data sets requires an appreciation of the potential errors and,
if at all possible, some form of ground truthing.
Peter Iles
GPS_ID TIME GPS_EAST GPS_NORTH GPS_SITE_DESC
1 13:10 429381 434413 Mixed suburban, outside RS Lecture Theatre,
surrounded by tall buildings
2 13:12 429366 434419 Top of mound surrounded by mixed
vegetation/garden, surrounded by tall buildings
3 13:16 429257 434491 Under trees on grassland, high buildings
surrounding
4 13:18 429228 434515 Grassland with buildings around (uphill from
ID3)
5 13:20 429200 434495 Gravel car park with many cars, buildings
and trees in near vicinity (uphill from ID4)
6 13:23 429138 434558 Gravel car park, fewer cars, building and
trees in near vicinity (uphill from ID5)
7 13:28 429082 434769 Edge of park, mix of grass and concrete,
dense tree cover, road and buildings nearby
8 13:32 429062 434907 Parkland, edge of grassland, very boggy,
trees within 50m
9 13:37 429132 434966 Parkland, not a wet as ID8, tall university
building in sight
10 13:40 429201 435050 In the flower bed, grass, sculpture and tall
buildings close by (all sats in line o'head)
11 13:44 429309 435052 Triangular area of concrete/tarmac at road
Junction of Clarendon/Ragland Rds etc.
12 13:48 429232 434875 Cemetery path, large shiny grave monuments
on both sides through grassland, nearby trees
13 13:50 429256 434813 Grassland in cemetery, medium trees within
40m
14 13:53 429279 434771 Path in cemetery, low trees and grass
adjacent
15 13:56 429256 434722 Car park outside Geog Old Building, tarmac,
medium buildings on 3 sides
GPS_ID REAL_X REAL_Y XY_POSN_ERROR
1 429388 434396 19m
2 429364 434417 3m
3 429257 434491 Unknown
4 429225 434508 ? c.10m
5 429193 434493 ?c.8m
6 429139 434554 <2m
7 429082 434769 <5m
8 429062 434907 Unknown
9 429132 434966 Unknown
10 429200 435051 <5m
11 429310 435053 <2m
12 429235 434874 <2m
13 429256 434813 Unknown
14 429279 434771 c.7m
15 429266 434707 c.17m
GPS_ID GPS HOD (m) REAL HOD (m) DEM_HEIGHT (m)
1 71 64 60
2 76 64 64
3 94 70 70
4 91 77 77
5 91 77 74
6 91 81 81
7 92 86 86
8 92 90 90
9 91 90 90
10 94 91 90
11 95 89 89
12 100 88 88
13 98 86 86
14 93 84 84
15 90 82 82
Pete Iles
********************
This e-mail contains information intended for the addressee only.
It may be confidential and may be the subject of legal and/or
professional privilege.
If you are not the addressee you are not authorised to
disseminate, distribute, copy or use this e-mail or any attachment to it
The content may be personal or contain personal opinions and
unless specifically stated or followed up in writing,
the content cannot be taken to form a contract or to be an expression
of the County Council's position.
LCC reserves the right to monitor all incoming and outgoing email
LCC has taken reasonable steps to ensure that outgoing communications do not
contain malicious software and it is your responsibility to carry out any
checks on this email before accepting the email and opening attachments.
********************
|