APES 
n-making, 
velopment 
e data as 
duced and 
ention has 
S exists to 
inicated to 
and other 
ptions for 
work first 
y and data 
t and most 
etzt, um 
at werden 
isierungen 
Szenarien 
Idaten und 
bisher der 
llung von 
gbar, dass 
existieren 
's erfolgte 
‚stehenden 
ese Arbeit 
thoden in 
ehmungen 
nd hôchst 
scape and 
ding more 
deal with 
ntal states. 
enario due 
including 
cnowledge 
ell, 2003: 
t; to this is 
an actions 
scale, and 
precision. 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
Multiplication of these error sources means that any visualised 
landscape can only ever be said to be a possible future, one of a 
great number of alternatives. 
Although there is a desire to reduce uncertainty, stemming from 
3 end-users' and decision-makers' demands for information that 
is certain (Foody & Atkinson, 2002), it is clear that it can never 
be eliminated. We must therefore find ways to communicate 
scientific uncertainty in a clear and honest way, particularly 
when dealing with the public, so that the most appropriate 
decisions are made. However, care must also be taken not to 
overwhelm the viewer by adding excessive or overly 
complicated information about uncertainty — the purpose of the 
visualisation is, after all, to increase the accessibility of 
environmental information. To date, little research has been 
done to address this issue. 
Previous work with planning professionals (Appleton & Lovett, 
in press) has suggested that uncertainty should be carefully 
considered when visualising a scene for the environmental 
decision-making process, in order that viewers' (and managers") 
time is not wasted in commenting on uncertain or representative 
aspects of a landscape future as if they were certain and 
specific. For example, showing an visualisation of a future rural 
scene which includes wind turbines may elicit detailed 
comments on their size and location, when in fact they were 
only included as an indication of increased reliance of 
renewable energy in the scenario shown. Furthermore, futures 
which appear more certain than they really are may be accepted 
by the audience as a foregone conclusion rather than questioned, 
explored, and commented on. 
One potential barrier to finding a simple and effective way to 
represent uncertainty is that uncertainty itself is very difficult to 
define. For example, we could attach particular level of 
uncertainty to a landscape as a whole, or to the various elements 
within that landscape; we could be trying to show that one 
outcome is simply more or less likely than another given certain 
conditions, or we may have more quantitative information on 
the overall probability of different futures. We may be asking 
for preferences or suggestions relating to choices about the 
landscape, or trying to communicate the consequences of 
different courses of action that the viewer could take. Exactly 
what is to be communicated will necessarily affect the 
suitability of the various methods detailed in the following 
section. 
It may even be that the passive presentation of uncertainty is not 
enough for some uses. Couclelis (1992) urges that data quality 
information should grab the attention, forcing the viewer to 
assess quality, although such methods could prove confusing for 
the type of non-expert audiences that visualisation is usually 
intended to bring into the decision-making process. 
It is worth noting that to some extent, anecdotally at least, 
where visualisation technology is capable of creating highly 
detailed images these capabilities are used without question in 
the belief that “more realistic” visualisations are “better”. 
However, it is arguable that the “best” visualisations are in fact 
those that give the viewer the truest picture of the future in 
question — scientific uncertainty included. It may be, therefore, 
that detail and visual realism should not be the overriding 
concerns, and careful attention should be paid to the overall 
message conveyed by a visualisation. One thing is clear — a 
visualisation cannot simply be presented as a fait accompli if the 
audience is to make meaningful decisions based upon it. 
Somehow, the idea that a future landscape is just one possibility 
4 
7 
must be communicated if more participatory environmental 
decision-making is to fulfil its potential. 
3. PREVIOUS WORK 
Perhaps the simplest ways of expressing uncertainty relate to 
numerical information, where it can be conveyed using error 
margins and ranges rather than single figures (although these 
too have their problems (Gigerenzer, 2002)). Such methods are, 
however, largely inappropriate for visual and/or spatial data 
where there is a large variation in uncertainty over the dataset. 
Research has therefore also examined ways of showing data 
quality and error when using 2D or 3D cartographic methods; 
indeed, uncertainty as a whole has been an important research 
topic in geographical information science for the past decade 
(Atkinson & Foody, 2002). There are broadly three strands of 
research relating to uncertainty in geographical information: 
reducing uncertainty, characterising sources of uncertainty, and 
providing information to users about the uncertainty of a 
particular dataset. It is the latter of these that provides 
information relevant to the work described here, but as stated by 
Hunter & Goodchild (1996), that investigation of the 
presentation of uncertainty could still be said to be sparse. This 
is particularly true for the field of landscape visualisation, due 
to its relative newness. 
Existing literature mostly focuses on error in spatial databases, 
and usually includes separate maps of error (for example a 
colour gradation) to overlay or combine with the original data in 
some way. A common example is 3D draping, whereby a 
surface is generated to represent a variable over a certain space, 
with that surface then coloured according to the error at each 
location. Cartographic techniques lend themselves to variation 
in representation, e.g. point clouds, thick or braided lines, or 
fuzzy features to represent uncertainty of location (Hunter & 
Goodchild, 1996). As well as visual methods, there are 
suggestions of animation and sound-based solutions, but they 
seem to be little used in practice so far (Beard et al., 1991; 
Krygier, 1994). The applicability of these tools to current 
landscape visualisation methods is discussed in Section 5. 
It is also important to note that there is still little guidance as to 
how a given degree of uncertainty might be inferred from the 
use of a particular technique. Visual techniques are generally 
indicative rather than quantitative, giving users information 
quickly, but not necessarily clearly (Hunter & Goodchild, 
1996). 
4. CURRENT TOOLS FOR VISUALISATION 
The way in which uncertainty is represented in any landscape 
visualisation is governed to some extent by the choice of 
visualisation method and output type. 
Still images offer the greatest levels of realism, as there is no 
limit on the time taken to draw a scene — it may take several 
hours to render, but once completed it can be shown many 
times. One example of software that produces this sort of output 
is Visual Nature Studio (3DNature, 2004) Interactive methods 
are still behind in terms of the detail that can be shown 
(although the gap is narrowing as available computing power 
increases), as their content needs to be rendered many times à 
second to keep up the illusion of movement through a scene. 
For instance, the terrain modelling software TerraVista (Terrex, 
2004) can be used to produce models which can be explored in