David R Miller
This paper discusses and illustrates the value of digital photogrammetric techniques combined with automatic characterization of landscape views for mapping the geographic distribution of view types. Digital photogrammetric techniques have been used to derive digital terrain data of areas as an alternative to digitized contour methods and the resultant surface analysed to produce a census of visibility.
National initiatives on the production of digital datasets derived using digital photogrammetric techniques has been gaining ground for the provision of topographic basemaps, map updates or for interpretation of land ocver data (Goebel and Price, 1993; Gunnarsson, 1993; Light, 1993; Skalet et al. 1992). Currently, the most significant of these datasets is the ortho-photograph the value of which is for supplementing line maps, providing backdrop information to fill the `blank' areas on maps and therefore facilitating the user their own interpretation of features in the landscape. Such a source of data provides another input for use in Geographic Information Systems (GIS)(Chapman et al, 1991; Lenzen and Foresman, 1993).
The production of ortho-photography often utilizes elevation data obtained by conventional means, usually analogue or analytical photogrammetry, converted into digital form (Petrie, 1995). This approach has a higher level of productivity, avoiding as it does, the processing overheads associated with deriving the digital elevation model (DEM) directly from the same stereo models (Chapman et al, 1991Hood et al, 1989). In addition there are issues of the reliability of the elevation data which has been derived by such digital techniques (Petrie, 1995). The software for generating ortho-photographs on digital photogrammetric workstations, such as that produced by Intergraph, Zeiss, Autometrics and Helava is being complemented by that of Erdas Digital Ortho or Orthomax, HiView and PCI-Easipace and is becoming more reliable and housed on computing resources of greater power. Therefore, the generation of a digital terrain model directly from remotely sensed data is increasingly common (Konecny, 1979; Miller et al, 1992; Novak, 1992) and presents some of the same challenges of data quality as for contour or height point based products (Robinson, 1994), plus new ones, for example, feature matching (Graff and Usery, 1993; Li, 1993; Zheng, 1993) or deriving heights of feature (Howard, 1991).
One application area employing the use of both elevation data and physical and cultural topographic features is that of landscape planning, involving the perception of landscape and translating descriptors of the physical environment into models of the nature of the environment with which people may relate (Shafer and Brush, 1977; Daniel and Vining, 1983). This may include identifying the content of a view in terms of its vegetation and cultural features. The issue of the observer's ability to discriminate between features when viewed from a particular location means that techniques of distance queuing, hue attenuation and feature size and shape must be applied to that of perspective view geometry (McLaren and Kennie, 1989). The presence and distribution of surface features in the context of the topographic surface contributes to the type of view exposed to an observer (Mayall and Hall, 1994).
To place the calculation of view type into a wider geographic context, a census of visible land (Miller et al, 1995) is calculated, based upon a digital representation of the terrain. This provides a surface of land visibility with which land dominated by particular view types may be mapped (Miller et al, 1994). Land cover effects may be included in this analysis by adjusting the height of each location for example, approximate heights of build up areas or forests. The nature of this data is of a type which would appear to be provided for using digital photogrammetric techniques.
The study area is in the south east of the Grampian Region in Scotland, from Glensaugh to the Cairn O'Mount (Figure 1). This area spans a significant geological feature, the Highland Boundary Fault. It is a low lying, relatively level agricultural land in the south and heather and peatland hill moorland in the north, ranging in altitude from approximately 100 metres to 456 metres above mean sea level.
The land cover within the area is approximately 9% agriculture, 17% forestry and 74% moorland (MLURI, 1993). Of the forestry, over 90% is plantation woodland, comprising mainly Sitka Spruce, Scots Pine and Douglas Fir. The 10% of broadleved trees is distributed around field boundaries plus a few individual trees growing in areas of rough grazings. The location is on a popular tourist route across the mountain, at the top of which there is a developed view point. Therefore, this area provides a range in the physical topography and land use against which to test the techniques for deriving elevation data and an associted census of visibility.
Three stereo-models were derived for the study area. The source photography was 1:24 000, panchromatic, near vertical with a focal length of 152 mm. AN AGFA flatbed scanner was employed in digitizing the photographs at a resolution of 1200 dpi. Topographic map information existed for this area at scales of 1: 10 000 and 1: 2 500 and an existing land survey control network (Miller et al, 1989), all of which could contributed control point data. In addition, field surveyed control data was necessary in the moorland area to the north and west of the model area.
To provide the additional control a Trimble GPS was used in differential mode with a base station set-up at an Ordnance Survey triangulation pillar (tertiary control point in the national co-ordinated reference system of the United Kingdom). Fifteen control points were identified and their co-ordinates recorded. Nine of the additional points were identified in the upland, moorland areas of the study area, where no field control existed and no features are mapped on the base maps. The other six points were selected at features already mapped at 1:2 500 or 1:10 000 scales to clarify ambiguities of what the nature of features represented on the ground (such as the intersection of a fence or a ditch).
Scanner distortions were quantified using the measured co-ordinates of points on the scanned test grid compared to a computed grid intersections, from which a thrid order polynomial transformation was derived and applied to the original image (Burnside, 1979). This produced a scale distortion of 0.29 of a pixel at 1200dpi. The focal length, image coordinates and ground coordinates of ground control points were then used to determine the exterior orientation parameters of the image.
Control points, were selected for the model orientation. The sources were selected on the basis of clarity on the scanned photographs, relocation on the map or in the field and reliability of the control co-ordinates. Table 1 contains a summary of the source of control data and the accuracy with which it is recorded (measured or reported). Figures 2 and 3 show extracts of the the derived elevation model and the ortho-photograph for the area.
| Control Source | Planimetric Accuracy | Height Accuracy |
| 1:10 000 | +-3 m | +-1 m |
| 1:2 500 | +-0.625 m | +-0.5 m |
| Ground Survey | +-0.1 m | +-0.05 to 0.2 m |
Check points were used to assess the planimetric and elevation accuracy of the derived products. The 11 check points comprised locations for which the co-ordinates were derived using traditional land survey techniques of theodolite and distance measuring equipment or global positioning systems (GPS). Table 2 contains a summary of the observations at the check points.
| Number of Check points | Control Method | X-Coordinate RMS | Y-Coordinate RMS | Z-Coordinate RMS | 6 | Land Survey | +-0.13 m | +-0.11 m | +-0.15 m | 5 | GPS | +-0.36 m | +-0.45 m | +-0.83 m |
Observations to trees within stands were restricted to those of which both the base and the top were visible. In certain plantations this necessitated the selection of sample points to be determined by the visibility of the trees. Thus, in plantations of narrowly spaced Sitka Spruce, which had not been thinned, there were fewer observations taken within the plantation compared to those taken for other species.
The height of individual trees was observed using two different optical or electro-optical devices: theodolite and electronic distance measuring equipment or inclinometer and tape measure (Philip, 1994). In each case the vertical angles to the base and top of the tree were observed and the slant distance to the base of the tree. From these observations, the height of each tree was derived, examples of which is included in Table 3.
The observations were made six years after the original photography, thus a correction was required for their growth during the intervening period. The correction was for the growth during this period an indication of which was taken to be the number and spacing of the whorls at the top of the tree. Observations were made of the the height of the tree to the sixth whorl from the top and from this an estimate of the height of the tree at the time of the photograph was derived (Table 3).
The theodolite was used to calculate the difference in height between the bottom of the trees observed in the field and the observations to ground level of the open location. This ensured that the absolute difference in height between the point taken from the digitally derived elevation model and the tree and the top of the tree was known to an estimated accuracy of better than 0.25m. That is, to the order of magnitude with which one can ascertain what and where the base of the tree actually is.
The observations in Table 3 contain two examples of trees the height of which was not resolvable with reliability. One is an Ash and the second a Scot's Pine. In each case the tree was found in a narrow line of trees which were not adequately discriminated on the photography from the surrounding vegetation. As a consequence, the pattern matching algorithm was not able to remove the X-parallax in the local area and the derived heights were not reliable.
A linear regression between 80% (50 observations) of the field and terrain model observations on tree height give the following equation:
DEM_Tree_Height = 1.02 x Field_Measured_Tree_Height - 0.768
The RMS error for the regression is +-0.27 m.
The other 20% of observations have been used to validate the regression model. The standard deviation of the residuals from these additional points was 0.31 m.
The correlation between the measured and derived heights suggests that the methodology will be reliable for remotely measuring tree heights across large areas. The regression constant is likely to be attributable to the vertical resolution of the DEM being only 0.5 metre, the error in estimating the growth in the trees over the six year period and the observation being made at the edge of the stand or in gaps in the canopy.
The set of all locations visible from any specified location has been termed the isovist by Benedikt (1979) who described the importance of the observer's location as central because it is `representing the position of the observer whose spatial experience we are trying to explore'. The objective identification of the boundaries of this `spatial experience' is the basis for the following analysis of the digital terrain data. A census of the total area visible from all locations within the study area (each location is a pixel in the raster dataset) was calculated providing a relative intervisibility of land within the study area.
Figure 4 illustrates a perspective of the view up the mountain, draping the ortho-photograph across the derived DEM. The view is approximately north-west, with a range in altitude between the lowest (bottom right) and highest (top left) points of 300 metres. A road follows the shoulder of the hill across the centre of the illustration. The darker tones to the edges of the extract are predominantly heather moorland and the brighter tones on the valley sides are grasslands and bracken (Pteridium aquilinum).
A calculation of intervisibility was undertaken for each cell and the total area visible was attributed to the cell. Each cell in the elevation model is counted only once, although the calculation could be weighted according to the inverse of the distance from the cell. The analysis considers a complete 360 degree rotation around each location for a radius of up to two kilometres on the raster DEM derived from the stereo-models. The output of this calculation is a cell-by-cell scoring of the visibility with a two km radius, based upon the topography derived from the three stereo-models, is restricted to a central area of 1.3 km x 5.5 km.
Figures 5 and 6 are detailed views of two extracts of the area, showing the elevation data and the derived visibility census. Figure 6 has two areas highlighted, approximately 200 metres apart, with contrasting visibility levels and topographic features. On the left the land rises, with convex slopes, to a low hill top. The level hill top has a low visibility, with few cells having an open view of the land. On the right, there is a small stand of woodland, approximately 100 metres by 50 metres in size, the edges of which are highlighted as open to the surrounding land. The land to the south west has a low visibility level, on rising land which is hidden from the rest of the area by the forest stand.
Figure 6 shows two extracts from the south west of the area. At the bottom left there is an open, west facing slope which levels out and leaves the forest stand edge exposed compared to the surrounding terrain. The forest canopy is also measured as having as higher visibility than the surrounding terrain but lower than the stand edges. The example to the top right is open moorland where the variation in visibility levels is due to small, shallow gullies or low hummocks.
Overall, the visibility levels are influenced by:
1. extend to which the land is exposed to surrounding land; 2. the shape of the surrounding land;The visibility of forestry is determined by the topographic context plus the height of the trees and their distribution across the terrain.
Combining the graphical summaries of the visibility in an area together with the observed values provides a means for identifying those areas of greatest prominance to an observer within an area and differences between the relative visibility of similar land cover features such as forestry.
Despite over 25 years of the derivation of ortho-photographs they have played very little part in the increased use of GIS. Similarly, the increasing availability of digital photogrammetry and the potential for deriving digital elevation models has yet to be realized by a substantial percentage of the relevant parts of the GIS community (Brown and Bara, 1994). An increasing volume of data will become available across a continuum of scales, resolutions and precisional accuracies. Improvements in the quality of aerial photographic equipment and materials coupled with reduced costs of computer disk space and processing time will permit the expansion of a role of photogrammetry in GIS (Zilberstein, 1994) if the quality and flexibility of the products are matched to user need.
Further work is being undertaken on the measurement of within stand variation of tree heights using aerial photography from different years, and photography with and without trees present. Traditional methods of plotting contours using photogrammetric instruments necessitates a combination of interpolation between spot heights in areas of open canopy, and allowing a vertical offset for the heights of the trees and contouring the canopy top. Therefore, the contours in forested areas will necessarily be of lower accuracy than those in open terrain. Two opportunities may be taken to address this issue. Either the photography flown before planting for forest planning may be used or photography taken after felling so that if GIS were to be used in future planning an accurate DEM would be available. Thereafter, the monitoring of forest growth can be undertaken remotely.
The derivation of tree heights using digital photogrammetry is restricted by the same conditions that apply to aerial photographic interpretation and any photogrammetric measurements. For example, photographs flown in autumn, at a small scale, will not be of any value for deriving heights of broadleaved trees. Tree species will be further significant in that the shape of the crown will determine what is represented in the elevation model. Limitations will be:
1. Tree spacing too wide; 2. Tree height too small.In each case, the trees (either individually or as a stand) have not been resolved in the derivation of the elevation model. Further analysis may be undertaken to present the uncertainty associated with the DEM based upon the check point observations. These observations, taken together with the patterns of residual parallax in the stereo-model may be used to derive a model of the uncertainty of the DEM with which one could modify the derived calulations of visibility (Fisher, 1994).
The calculation of terrain visibility has been artificially restricted by the use of a maximum scan of two km. In practice, the visibility of the terrain will extend significantly further. However, low resolution DEMs are inappropriate for the representation of surface features close to the observer, which will have a greater visual impact than those further away (McLaren and Kennie, 1989). Therefore, the resolution and detail should be tuned to match the distance from the observer, that is, the further from the observer the lower the resolution and the nearer the observer, the higher the resolution. The approach taken in this paper utilizes high resolution data, the analysis of which demands significantly higher computing resources than lower resolution data for the same area.
It would appear that further work is required on the representation of surface features for the analysis of visibility. Such work could utilize a data model which facilitates the accessing of records of features (such as trees or buildings) the visibility "shadow" of which could be computed with respect to the observer. Access to such records would be screened according to the likeliehood of being able to see such a feature at the location based upon the intervening topography.
The results from the analysis of visibility provides a method of comparing the visibility of locations on a comparable basis. If the analysis if undertaken over a sufficiently large area and with sufficient detail, the result is an absolute score for each location based upon the level of visibility of the land. Further refinement would require account to be taken of the nature of an object at the location. That is, the level of contrast of the object with its surroundings, the lighting conditions and its shape. An additional aspect would be any associated information that the feature and the prevailing conditions provided the observer which may enhance or reduce the level of significance of its presence, such as the casting of shadows.
In conclusion, the example presented indicates the extent to which digital photogrammetric techniques can provide additional data for use in landscape management and planning. The use of aerial photography as a source of the data permits updating of the estimates of height of the forest stands and thus monitoring of the changes in visual impact of the forestry within an area over time. Considerable additional work requires to be undertaken to translate the techniques and processing into an operational facility but complementary work in the fields of design and psychology will contribute to a more robust means of translating level of visibility into an appraisal of visual impact.
Benedikt, M. L. (1979) To take hold of space: isovists and isovist fields, Environment and Planning B, 6, pp 47-65.
Brown, D G and Bara, T J (1994), Recognition and reduction of systematic error in elevation and derivative surfaces from 7.5 minute DEMs. Photogrammetric Engineering and Remote Sensing, Vol. 60(2), pp. 189 - 194.
Burnside, C D (1979), Mapping from aerial photographs. Fletcher and Sons, Norwich. pp. 304.
Chapman, D, Dowman, I and Muller, J-P (1991), Digital photogrammetry - interfaces with GIS. In: Proceedings of the AGI'91 Conference, Birmingham, November 1991.
Daniel, T.C. and Vining, J. (1983) Methodological Issues in the Assessment of Landscape Quality. In: Behaviour and the Natural Environment (eds. Altman, I. and Wohwill, J.),Chapter 2, 39-83, Plenum Press.
ERDAS Inc. (1992), Digital Ortho Manual, ERDAS Inc, Atlanta.
Fisher, P.F (1994), Probable and fuzzy models of the viewshed operation, In: Innovations in GIS 1 (ed. M.F. Worboys), Taylor and Francis, pp 161 - 176.
Goebel, J J and Price, L C (1993), National resources inventory and digital ortho-photography databases: characteristics, availability and use. In: The Proceedings of the 2nd International Conference on Environmental Modeling and Geographic Information Systems, Breckenridge, Colorado, September 1993.
Graff, L H and Usery, E L (1993), Automated classification of generic terrain features in Digital Elevation Models. Photogrammetric Engineering and Remote Sensing, Vol. 59(9), pp. 1409 - 1417.
Gunnarsson, R J (1993), Digital photogrammetry: the foundation for Iceland's geographical database. GIS Europe, November 1993, pp. 37 - 39.
Hood, J, Ladner, L and Champion, P (1989). Image processing techniques for digital ortho production. Photogrammetric Engineering and Remote Sensing, Vol.55(9), pp. 1323 - 1329.
Howard, J A (1991), Remote Sensing of Forest Resources. Chapman and Hall, London. pp.
Konecny, G (1979), Methods and possibilities for digital differential rectification. Photogrammetric Engineering and Remote Sensing, 45 (6), pp. 727 - 734.
Lenzen, T W and Foresman, T W (1993), Digital image databases support GIS operations. GIS World, November 1993, 36 - 38.
Li, Z (1993), Mathematical models of the accuracy of digital terrain model surfaces linearly constructed from least square gridded data. Photogrammetric Record, 14 (82), pp. 661 - 674.
Light, D L (1993), The national aerial photography program as a Geographical Information System resource. Photogrammetric Engineering and Remote Sensing, 59 (1), pp. 61 - 65.
Mayall, K. and Hall, G.B. (1994) Information Systems and 3-D Modeling in Landscape Visualization. In: Urban and Regional Information Systems Association Annual Conference Proceedings, Vol 1, 796-804.
McLaren, R. A. and Kennie, T. J. M. (1989) Visualisation of digital terrain models: techniques and applications. In: Three dimensional applications in geographic information systems, (ed. J. Raper), pp. 79-98. Taylor and Francis: London.
Miller, A B, Helava, U V and Helava, K D (1992), Softcopy photogrammetric workstations, Photogrammetric Engineering and Remote Sensing, 58 (1), pp. 77 - 83.
Miller, D R, Morrice, J G and Whitworth, P L (1989), The bracken problem in Scotland. In: Bracken'89: biology, control and management. (eds J A Thomson and R T Smith), Australian Institute of Agricultural Science Occasional Publication No. 40. pp 121 - 132.
Miller, D R, Morrice, J G, Whitworth, P L and Aspinall, R J (1994), The use of GIS for the analysis of scenery in the Cairngorm Mountains, GIS in Mountainous Regions, (eds. I Heywood and M Price), Taylor and Francis.
Miller, D R, Law, A N R and Brooker N A (1995), Calculation of a census of visibility. In: Procceedings of the 15th ESRI User Conference. ESRI Inc, Redlands, Ca. USA. 1995.
MLURI, 1993, The Land Cover of Scotland by aerial photographic interpretation. A report to the Scottish Office Environment Department. The Macaulay Land Use Research Institute, Aberdeen.
Novak, K (1992), Rectification of digital imagery, Photogrammetric Engineering and Remote Sensing. 58 (3), pp. 339 - 344.
Petrie, G P (1995), Air photo inputs to GIS for environmentl and strategic planning - technologies, procedures and products. In: Proceedings of Aerial photographs for strategic planning, Remote Sensing Society GIS Special Interest Group meeting, November 1995.
Philip, M S (1994), Measuring trees and forests. 2nd Edition, CAB International.
Robinson, G J (1994), The accuracy of digital elevation models derived from contour data. Photogrammetric Record, Vol.(14)83, pp. 805 - 814.
Skalet, C D, Lee, G Y G and Ladner, L J (1992), Implementation of softcopy photogrammetric workstations at the U.S. Geological Survey. Photogrammetric Engineering and Remote Sensing, 58(1), pp. 57 - 63.
Shafer, E.L. and Brush, R.O. (1977) How to measure preferences for photographs of natural landscapes. Landscape Planning, 4, 237-256.
Tilley, G B (1994), Digital ortho-photography for natural resource management, In: The Proceedings of the GIS'94 Symposium, Vancouver, February 1994.
Zheng, Yong-Jian (1993), Digital photogrammetric inversion: theory and application to surface reconstruction. Photogrammetric Engineering and Remote Sensing, Vol LIX (4), pp. 489 - 498.
Zilberstein, O (1994), National GIS from aerial photographs using a softcopy photogrammetric workstation. In: The Proceedings of the 4th European Conference on geographical Information Systems, Paris, March 1994.