Chandler, Jim 
  
time required to aaquire imagery was just 13 hours, mostly taken by the time required to climb up to the camera 
stations. 
Photogrammetric data processing was caried out back in the UK, using OrthoMAX to produce DEMs and OrthoBASE 
for orthophoto production. It is not possble for OrthoMAX to generate DEMs automatically from oblique imagery and 
using standard, documented procedures. However, previous reseach work (Pyle et al., 199% Chandler, 1999 has 
shown that if the control is rotated so that the average vedor between the two camera axes is nea verticd, then it is 
possble to use the automated DEM extradion tools successfully. This $rategy was adopted, and a reverse rotation 
applied to transform the automaticaly derived data bac into the original coordinate system. DEMs were aeded at a 
resolution of 0.20m, for those days where change was experienced (Figure 3). Examination of the DEM, which consists 
of 250000 points, reveds that the distinctive acuate patterns within the bedforms are represented. These are consistent 
with the sequence of erosion and sedi ment deposition seen in flume experiments. It is also clea that small topagraphicd 
feaures are represented, including minor bar-tops and channels. There ae aeæ where inacarate data has been 
generated, but these «incide with those regions inundated by flowing water at the time of photo-aœquisition. 
Comparison of DEM elevations with heights extraded from the profiles measured by field survey reved that the DEM 
acauracy is -0.044m, a satisfadory result considering that median gai nsize was 0.04m and that ground coverage of 
ead pixel was 0.025m or greder. Orthophotos (Figure 4) could be produced using the OrthoBASE product without 
recourse to rotating the control coordinates. 
  
  
4 DISCUSSON 
The pradicd work cited in sedion 3.0 
demonstrates that digital photogrammetry 
can usefully be used to derive 
morphologicd data necessary to describe 
water worked surfaces. There ae various xn 
controls upon the quality of such data. 
The scde ad configuration of the 
photographs aaquired has a dired and highly 
predictable impad upon the predsion of 
DEMs generated. One of the unique and 
well-known advantages of photogrammetry 1600 
is the positive relationship between 
predsion and photo-scde and this has again 
been illustrated during this Study. 
Applicaion of the methods for the Flood 
Channel Fadlity utilized a camera 
configuration that can be more dficient than 
the conventional 60% overlap and vertica 950 ue 
imagery. The difficult sinusoidal wave “| 
shape of the FCF demanded a radicd | 
approach, with the most effedive means of Figure 4, Orthophoto df braided system, Sunwapta- 30-07-99 
  
  
  
  
  
  
  
  
obtaining stereo-coverage being to oktain 
  
discrete overlapping stereo-pairs. Moreover, 
by orientating the aes of each camera slightly inwards, a 90-9596 overlap can be achieved, again improving efficiency 
and ease of obtaining appropriate stereo coverage of complex objeds. 
Accurate camera cdibration is clealy critica for deriving acarate DEMs, although the aiticdity of parameter groups 
used to model the inner geometry of the camera varies (Chandler e al., in pres3. It has been found that it is particularly 
important to acarately model lens distortion, which arises from using the cheaper classof digital cameras designed for 
professonal photography, rather than photogrammetry. Other parameters are less $gnificant. Focd length and principal 
paint offset, which can be important for convergent high predsion networks used for deformation monitoring, were 
found to be insensitive to acaracy tests. It is suggested that for verticd imagery used to extrad DEMs of surfaoes 
which display minimal relief variation in relation to camera/objed distance that these two sets of parameters need to be 
estimated approximately only, (Chandler et al., in press. 
It is pertinent to compare the acurades achieved in the three caes, but particularly between the FCF and the Tilting 
flume, where acarades are similar but the camera devation was over twice & low. This paradox can be explained by 
considering the extent and likelihood d poar matching due to dead ground in regions between particles. The li kelihood 
of dead ground is dependent upon the aomparative size of the particles (median grain size) with the ground coverage of 
pixels comprising the images. In the FCF the median grains ze was 0.9mm and ead pixel covered an areaof 1.4mm, 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 255