The pages linked above describe how to obtain and use the individual products. The purpose of
this page is to note some of the major pros and cons of each product, and to compare them numerically.
To begin:
SRTM data is held in binary format whereas OS data is held in text format. In most
situations this makes SRTM data both smaller to store and faster to load and process.
OS data at 50m resolution is intermediate between the two versions of SRTM, it doesn't have
quite as good resolution as SRTM1 at ~30m but better than the more commonly used SRTM3 at ~90m.
OS covers only the narrower UK, whereas SRTM covers the wider British Isles in particular, and
most of the inhabited world in general.
All of these data sets have practical problems in use. LP is of limited usefulness in the
neighbourhood of man-made holes such as quarries and open-cast mines, because the data reflects
the land contours as they are assumed to have been before being altered by man, not as they
actually were at the time of survey. Many T50 coastal tiles have sea-levels that are at
different non-zero heights, causing stepwise changes in sea-level between neighbouring tiles,
and this can make scanning from tile to tile problematic in some applications. SRTM,
particularly at the extremes of its range near 60°N, has many data voids which must be
interpolated over, but for use in the UK a more significant failing is lack of coverage for
the northernmost parts of The Shetlands.
This lack of complete SRTM coverage of The Shetlands can be overcome by remaking the relevant
tiles from OS data. However, this raises questions of which OS data set to use and what
procedure to follow in conversion. In particular:
Which data set should be used as the source, LP or T50?
If using T50, should the sea be excluded or set to zero?
Given that cell coordinates in SRTM and LP refer to the SW
corner of each cell, while T50 cell coordinates refer to the centre of each cell, numerically
how should such differences be reconciled? Should …
SRTM and LP be used directly, and T50 adjusted (referred to as normal below)?
SRTM and LP be adjusted, and T50 used directly (referred to as alternative below)?
How should coordinates be converted? The OSGB36 projection used in the UK is distorted by
historical cumulative errors in surveying up to about 5m horizontally and / or
vertically. Rather than update OSGB36 to a new standard with greater accuracy, OS chose to
keep it as the UK standard projection, instead introducing a rubber-sheet transformation called
OSTN02™
which converts accurately between ETRS89 and OSGB36 allowing for the distortions in the
latter, thus giving a transformation path between the UK and the rest of the world via ETRS89.
There are freely available OSTN02 implementations which can batch convert
- including OS' own on-line
Coordinate Transformation Tool
and the downloadable Windows program
Grid InQuest™.
However, although documentation and a DLL from the latter are available, implementing seamless
conversions in one of the common programming or scripting languages is at best very difficult,
while alternative libraries - such as
Proj4js-
do not allow for the distortions of OSGB36, so will have errors up to 5m, but, perhaps more
significantly, being JavaScript will run very slowly, particularly on a Windows machine when
interpreted by Windows Scripting Host. A further consideration is that OS heights are
referred to local Ordnance Datums such as Newlyn, while SRTM is referred to the geoid EGM96, and the
only way known to this author of converting between the two is OSTN02.
This author decided to resolve these questions by trialing all the possibilities and choosing the
combination that resulted in data that was least different from the original SRTM, where the latter
existed. It should be noted that this is not necessarily the same thing as choosing the
'best' method, in the sense of giving the most accurate representation of actual UK terrain, but
in the absence of independent absolute data against which to measure accuracy, it's probably the
next best thing. Given this target, it is obvious enough that when using T50 data using a
mask to exclude the sea will give better agreement, because fewer points will be changed, and if
doing so for T50, it needs to be done for LP as well, so that the comparison between the two is
like for like. The results of the trials of other possibilities are listed below.
Although no one trial is best by every criterion, Terrain 50 without any of the possible
adjustments (bold) gives the best overall agreement with SRTM. A surprise is that it seems
to be better not to attempt to correct for Geoid height, whereas to do so would seem to be the
logically correct thing to do.
Tiles Corrected
Points Corrected
Data Voids Found
Data Voids Filled
Minimum Correction
Maximum Correction
Average Correction
Landform Panorama
-L
71
38,989,647
7,375
7,375
1
447
3.745
-L -A
71
39,570,832
7,382
7,382
1
447
4.019
-L -G
71
39,268,640
7,375
7,375
1
447
3.804
-L -A -G
71
39,791,822
7,382
7,382
1
446
4.073
Terrain 50
71
35,912,171
7,375
7,375
1
489
3.268
-A
71
41,175,533
7,382
7,382
1
490
7.039
-G
71
36,458,095
7,375
7,375
1
489
3.318
-A -G
71
41,344,491
7,382
7,382
1
490
7.064
Command-Line Switch Meanings
-A
Use alternative cell centring as defined above
-G
Correct for EGM96 geoid height
-L
Use Landform Panorama heights
Although in the numerical comparisons above the amount of maximum disagreement may seem alarming,
with the exception of the point of maximum correction discussed below, these are mostly in The
Shetlands where SRTM data is either non-existent or of low quality. Further, it should be
remembered that while indeed there are a significant number of significant differences, all of
which would require explanation in an ideal world, mostly there are a large number of comparatively
small differences. This can be demonstrated by visual comparison between the original SRTM
tiles and those made as per the best result above, but images for the entire UK would be too large
to display here (but are included in the OS2SRTM-All.zip below), so instead only
a test section across the approximate centre of the UK mainland at 55°N is examined here.
The first image is a simple comparison where colour intensity is proportional to differences in
height between the sources; the second is graded so as to emphasise pixels where the sources
disagree by more than 10m (half intensity) and 100m (full intensity); the third shows colour at
full intensity wherever the sources disagree by 1m or more. While the last image is heavily
coloured, showing that disagreement is widespread, the first is predominantly greyscale, showing
that this disagreement is mostly small in magnitude. The middle image shows that nevertheless
there are enough significant disagreements to merit caution in the use of both sets of data.
Note: In all cases greyscale intensity is roughly proportional to terrain height, while
colour intensity is determined by differences in height between the sources, as described above,
with areas where SRTM is higher tending to blue and T50 higher tending to green, except for data
voids in the original SRTM, which are shown in magenta.
Although such comparison images for the entire UK cannot be displayed here, when examined they
throw up enough of interest to merit further discussion, to understand the significance of which
it is necessary to know that SRTM radar data was collected by the shuttle orbiting roughly NW-SE
and NE-SW, while OS data is combined from land surveys and aerial stereoscopic photographic imaging.
In the full saturation image …
Areas such as Ireland and The Channel Islands which are covered only by SRTM appear
as all blue.
In the three Shetland tiles at 60°N, the lack of SRTM data north of 60.385°N
and the noisy sea both stand out as swathes of respectively solid magenta and
blue/green speckles.
The shuttle's orbital paths appear as an obvious diagonal hatch-pattern across much of
the UK, most obviously over England, while it can be deduced from horizontal striping
over the central Scottish Highlands east of The Great Glen that the OS' flights in this
area were W-E. Additionally, many OS tile boundaries can be seen as vertical and
horizontal edges.
In the graded saturation image …
The artifacts described in the last bullet point above almost completely disappear, and
from this comparison we can deduce that the capturing processes for both sets of data
were subject to systematic error between 1m and 10m.
It becomes very obvious that the colours tend to follow geographic features.
That is to say, high ground tends to be greener while low ground tends to be bluer, so
it seems likely that one or other or both capture processes were subject to further
systematic error of 10m or more that depended on the height of terrain being captured.
There is a bright green artifact at col 3882 row 3068 in the NW Scottish
mainland corresponding to the maximum correction in the results table above, at an
apparent mound or hill at 58.4425°N 4.765°W or col 175 row 124 of
Terrain 50 tile NC35, but which does not appear to have any basis in reality.
It does not appear in OS Maps, Landform Panorama, SRTM, or either of the following … Bing Maps Google Maps
… so it must be concluded that either this is a blunder by OS, or they have
put it in deliberately as a copyright test point against people trying to steal their
data. Whether there are any other such artifacts, I do not know, but, as there
are plenty of other points where the two sets of data differ by more than 100m, it
would be wise to assume that there are.
In the greyscale image …
As with the test image above, the lack of coloration shows that overall agreement is
fairly good.
Shetland Conversion
The programs that made the above comparisons will also remake just the corrupted / missing
SRTM tiles covering The Shetlands, by giving them the -S command-line switch.
If SRTM1 rather than SRTM3 tiles are required, also give the -1 switch, though
this has been less well tested. Ready-made SRTM3 Shetland tiles are available via the link
SRTM3-OST50-Shetlands.zip below.
Downloads
For those that wish to explore this subject further, these downloads may prove useful:
The above programs and a complete set of SRTM3 data generated by them (warning 2.5GB):
OS2SRTM-All.zip
Note: No breach of copyright of source data from SRTM or OS is intended, such copyright
remains with the original sources. Additional material created by Charles Macfarlane is
available under the Creative Commons terms relating to my site.
