The Fractal Dimension of Wilderness Agnès Patuano PhD Landscape Architecture Edinburgh College of Art Newcastle University 26-28 March 2015
The Fractal Dimension of Wilderness
Agnès Patuano PhD Landscape ArchitectureEdinburgh College of Art Newcastle University
26-28 March 2015
Agnès Patuano – PhD Landscape Architecture March 2015
The Fractal Dimension of Wilderness
The fractal analysis of landscapes• Mathematical theory• In practice: Calculation of the fractal dimension of complex images
• Results of the comparison of methods
The fractal dimension of Wilderness• Landscape preference and naturalness• The preferred dimension• Limitations: The visual differentiation between naturalness and wilderness
Agnès Patuano – PhD Landscape Architecture March 2015
The Fractal Analysis of Landscapes Definition of fractals
Fractal geometry: the geometry of nature
The fractal dimension
• A fractal is "a rough or fragmented geometric shape that can be split into parts, each of which is (at least approximately) a reduced-size copy of the whole.” (Mandelbrot, 1982)
Fractal Dimension• Measures “the object’s degree of irregularity and break” (Mandelbrot, Fractal Objects, 1975)
• Line=1; Plane =2; Volume = 3• Fractal dimensions: non integer, strictly exceeds topological dimension
Sierpinski GasketDimension: log(3)/log(2) = 1.585
Credits: Taylor, 2006
Credits: Taylor, 2006
Agnès Patuano – PhD Landscape Architecture March 2015
The Fractal Analysis of Landscapes Calculating the fractal dimension
The Box-counting methodAmount of the plane filled by the pattern compared at different magnifications• N(d)=1/d D
• N(d) : number of boxes• d: size of the boxes
• logN(d) plotted against logd. • line slope = - D
Richardson plot
Agnès Patuano – PhD Landscape Architecture March 2015
Concerns
Physical fractals do not exist: For natural fractals, D can not be measured, it can only be estimated.• Fractals are complex mathematical constructs (equations) with an infinity of self-similar details on an infinity of scales.
• Natural objects can show fractal-like attributes but cannot behave like real fractals because of their finite quality.
Landscapes are more likely multifractals
Most non-mathematical studies use the Box counting method • Dependent on image resolution and information content
• Even within this method, results can differ for a particular image if the image is turned upside down or rotated
• The location of the offset of the grid influences results
Mathematical fractals ≠ physical fractals
The limitations of the box-counting method
Agnès Patuano – PhD Landscape Architecture March 2015
Methodology
Fractal Analysis• Picture selection• Software tests• Image analysis
- Greyscale- Silhouette outlines- Extracted edges
Images from the Forestry Commission• Comparison between two types of landscape• Similar resolution and size• No building, landmark or water
• Softwares:• HarFA 5.5: (Harmonic and Fractal Image Analyser)
compiled by the Brno University of Technology• BENOITTM 1.3: Trusoft Int’l Inc, most commonly used
commercial fractal analysis software in academic research.
Agnès Patuano – PhD Landscape Architecture March 2015
Preparing the images
Extracted edges
Image Analysis
Thresholded image
Find Edges
Original image Greyscale Removed sky
Agnès Patuano – PhD Landscape Architecture March 2015
Silhouette Outlines
Image Analysis
Method 1: Find Edges
Method 2: Stroke
Agnès Patuano – PhD Landscape Architecture March 2015
Image Analysis Greyscale analysis
Thresholding
Only one of the software carries out greyscale analysis: HarFA
• Fractal spectrum: Fractal dimension over intensity levels
• Thresholding: • i the intensity level chosen• Each pixel displaying an intensity < i goes white• Each pixel displaying an intensity < i goes black
Agnès Patuano – PhD Landscape Architecture March 2015
Image Analysis Each image transformed yields a different value of D
Each software calculates a different value of D for the same pattern
1 image = 2 x 5 x D
Agnès Patuano – PhD Landscape Architecture March 2015
Comparison of the different methods
Software results highly correlated (τ = 0.925, p<0.01)
Results• Correlation between methods
Correlations
DOUTLINE AVG DEDGEAVG DGREYSCALEMINAVG DGREYSCALEMAXAVG DGREYSCALEAVG
Kendall's
tau_b
DOUTLINE AVG
Correlation
Coefficient
1.000 .145 .609** -.151 .127
Sig. (2-tailed) . .123 .000 .107 .174
N 54 54 54 54 54
DEDGEAVG
Correlation
Coefficient
.145 1.000 .141 .202* .324**
Sig. (2-tailed) .123 . .135 .032 .001
N 54 54 54 54 54
DGREYSCALEMINAVG
Correlation
Coefficient
.609** .141 1.000 -.247** .134
Sig. (2-tailed) .000 .135 . .009 .156
N 54 54 54 54 54
DGREYSCALEMAXAVG
Correlation
Coefficient
-.151 .202* -.247** 1.000 .155
Sig. (2-tailed) .107 .032 .009 . .099
N 54 54 54 54 54
DGREYSCALEAVG
Correlation
Coefficient
.127 .324** .134 .155 1.000
Sig. (2-tailed) .174 .001 .156 .099 .
N 54 54 54 54 54
**. Correlation is significant at the 0.01 level (2-tailed).
*. Correlation is significant at the 0.05 level (2-tailed).
Agnès Patuano – PhD Landscape Architecture March 2015
Comparison of the different methods Results
• Ability to distinguish between landscape types
Comparison between extracted edges,
silhouette outlines, and average fractal
dimension for images of forest (a) and
meadows (b)
(a)
(b)
Agnès Patuano – PhD Landscape Architecture March 2015
The Fractal Dimension of Wilderness
Rationalizing the value of the experience of a landscape
Kaplan&Kaplan, 1989, Attention Restoration Theory• the degree to which a scene is natural or manmade• the extent of topographic variation• the presence or absence of water • the scale and openness of the scene
“Perceived naturalness”• Subjective concept• Linked with vegetation (quantity, type, quality) • Linked with perceived complexity
Complexity linked to concepts of mystery, legibility and coherence
Preference• Complex environments (dense forests) less preferred because
of illegibility and vulnerability to predators • Low complexity environments less preferred because
uninteresting
What is naturalness
What is complexity
Agnès Patuano – PhD Landscape Architecture March 2015
The Fractal Dimension of Wilderness The preferred D Main claims and hypothesis:• Universal aesthetic value of fractal patterns
• Preferred value D= [1.3;1.5]
• mid-range D and perceived naturalness
Aks and Sprott, 1996, Quantifying aesthetic preference for chaotic patterns• Most objects in Nature D =1.3
Haggerhall, Purcell & Taylor, 2004, Fractal dimension of landscape silhouette outlines as a predictor of landscape preference • Link between landscape preference and fractal properties
• Use of the silhouette outline as fractal image
Silhouettes used in Hagerhall, Purcell and Taylor's study (2004)Fractal Dimension a) D = 1.14; b) D = 1.32; c) D = 1.51; d) D = 1.70
• Short, 1991, the Aesthetic Value of Fractal Images• Nature -> Art• Resonance to fractals• Universal preference
• Sprott, 1993, Automatic Generation of Strange Attractors• Preferred D = 1.3
Agnès Patuano – PhD Landscape Architecture March 2015
The Fractal Dimension of Wilderness
Fractal Geometry is used for• observing spatial distribution, • calculating available habitat space• modelling behaviour of communities• Etc
Krummel et al, 1987, Landscape patterns in a disturbed environment.• Perimeter-area method• Comparison of patterns of forest patches• Small areas = low D
Low D considered a sign of human intervention
High D characteristic of superior naturalness
Use of fractal geometry in Ecology
Agnès Patuano – PhD Landscape Architecture March 2015
The Fractal Dimension of Wilderness Limitations: Can we visually differentiate between naturalness and wildness?
Conclusion
Naturalness and Wildness: no or little human intervention• Conservation• Biodiversity
Link between D and perceived naturalness but not actual naturalness
The Box Counting method applied to digital images is not sensitive enough to discriminate
D = 1.27
Thank [email protected]
The influence of fractal dimension on landscape preference
Agnès Patuano PhD Landscape ArchitectureEdinburgh College of Art