Fractalyse 3.

0
User manual

Gilles Vuidel and Cécile Tannier

2022-01-17
Contents

I Generalities 2
1 Introduction 3
1.1 About Fractalyse 3.0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Terms of use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 System requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 How to install Fractalyse on your PC? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Input data 4

3 Fractal analysis methods 5

3.1 Monofractal analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.1.1 Box counting analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1.2 Dilation analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1.3 Correlation analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.1.4 Radial analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Multifractal analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2.1 Box counting multifractal analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2.2 Sandbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 Network analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

II Graphical user interface 13

4 Data 14
4.1 Loading data: File menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.1.1 Load vector data menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.1.2 Load raster data menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.1.3 Load network graph menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.1.4 Load estimation menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.1.5 Other menus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 Data handling: Tools menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2.1 Rasterize menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2.2 Binarize menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2.3 Negative menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2.4 Selection menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2.5 Contextual layer menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5 Fractal analysis in practice 19

5.1 Choice of a scale range for the analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.2 Vector menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.2.1 Box counting menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.2.2 Dilation menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.2.3 Correlation menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.2.4 Radial menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.2.5 Batch menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.2.6 Multifractal menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

1
 5.3 Raster menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.3.1 Box counting menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.3.2 Dilation menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.3.3 Correlation menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.3.4 Radial menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.3.5 Radial all points menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.3.6 Multifractal menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.4 Network menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.4.1 Choice of distance and mass counting . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.4.2 Correlation menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.4.3 Radial menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.4.4 Service menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.4.5 Backbone menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.4.6 Multifractal menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

6 Estimation module 32
6.1 Monofractal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.2 Multifractal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.3 Save/reload estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.4 Estimation layer: scale range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

7 Miscellaneous 39
7.1 Preferences menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7.1.1 Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7.1.2 Processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7.2 Log window menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

III Command line interface 40

8 Prerequisite 41
8.1 Launch Fractalyse in CLI mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2 Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2.2 Character separator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2.3 Optional parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2.4 Command execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

9 Command reference 43
9.1 General command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
9.1.1 --help: display help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
9.2 Data manipulation commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
9.2.1 --rasterize: convert vector to raster . . . . . . . . . . . . . . . . . . . . . . . . . 43
9.2.2 --binarize: convert grayscale raster to binary . . . . . . . . . . . . . . . . . . . . 44
9.3 Monofractal analysis commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
9.3.1 --boxcounting: box counting on vector data . . . . . . . . . . . . . . . . . . . . . 44
9.3.2 --rboxcounting: box counting on raster data . . . . . . . . . . . . . . . . . . . . 44
9.3.3 --dilation: dilation on vector data . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.3.4 --rdilation: dilation on raster data . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.3.5 --correlation: correlation on vector data . . . . . . . . . . . . . . . . . . . . . . 45
9.3.6 --rcorrelation: correlation on raster data . . . . . . . . . . . . . . . . . . . . . . 45
9.3.7 SAMPLING: scale range definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.3.8 Estimation parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9.4 Other commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9.4.1 --clusters: counting clusters for vector layer . . . . . . . . . . . . . . . . . . . . 46
9.5 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9.5.1 -proc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9.5.2 -mpi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2
10 Examples of commands 47
10.1 Data handling commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
10.2 Fractal analysis commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

11 Performance tuning 48
11.1 Parallelism to speed up execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
11.1.1 One computer: threads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
11.1.2 Computer cluster: mpi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
11.2 Memory management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3
 Part I

Generalities

4
Chapter 1

Introduction

1.1 About Fractalyse 3.0

Fractalyse is a software application for the fractal analysis of 2D patterns (sets of points). This version 3
of Fractalyse has been written in Java language, from which results the improvement of data management
with GIS (Geographical Information Systems), graphical user interface and performance with parallelism.
Fractalyse runs on any computer supporting Java Virtual Machine.

1.2 Authors
Fractalyse has been developed by Gilles Vuidel at ThéMA laboratory (CNRS – University of Franche-
Comté, Besançon, France).

1.3 Terms of use

Fractalyse is distributed free-of-charge. Users must cite it (with the number of the version used) in their
publications. The source code is available and licensed in GPLv3.

1.4 System requirements

Fractalyse 3.0 runs on any computer supporting Java 1.8 or later (PC under Linux, Windows, Mac, etc.).
Please note that when dealing with very large data sets, the amount of RAM memory of the computer
can limit the maximum image size that can be processed in a single run with Fractalyse. In addition, for
some methods, processing power (CPU) will determine the speed of computing. For details, see section
7.1 and 11 below.

1.5 How to install Fractalyse on your PC?

Download and install Java 1.8+ from adoptium.net.
Download fractalyse-3.0.jar from sourcesup.renater.fr/www/fractalyse/ then launch it.

5
Chapter 2

Input data

Fractalyse can read vector and raster input data.

Vector data

Fractalyse supports the vector formats geopackage (.gpkg), geojson (.geojson) and shapefile (.shp), and
three types of geometries: points, lines, and polygons.

Raster data

Fractalyse supports the raster formats TIFF and Ascii Grid.

Ascii grid is a text format. An image is described as a matrix preceded by a header of 5 lines:
ncols 5
nrows 5
xllcorner 0
yllcorner 0
cellsize 1
1 0 0 0 1
0 1 0 1 0
0 0 1 0 0
0 1 0 1 0
1 0 0 0 1
TIFF is a binary format commonly used for raster images. It can contain an extension named GeoTIFF,
which adds spatial references to the image. Fractalyse supports TIFF images with or without GeoTIFF
extension. It does not support RGB color images. Consequently, you must convert your color image into
a gray scale image before analysing it with Fractalyse.

6
Chapter 3

Fractal analysis methods

Here we describe the basic principles of the fractal analysis methods integrated in Fractalyse.

3.1 Monofractal analysis

Fractalyse proposes four methods to estimate the (mono-)fractal dimension of a 2D object:
ˆ box counting analysis
ˆ dilation analysis
ˆ correlation analysis
ˆ radial analysis

In any case, estimating a fractal dimension involves:

ˆ a counting process,
ˆ an estimation process.

Counting process

The counting process operates step by step following an iterative logic. At each iteration step, the number
of elements (polygons, pixels...) contained in a counting window is counted. From one step to the next,
the size of the counting window is enlarged. Thus two elements vary according to the counting step
(iteration step) (i):
ˆ the number of counted elements (Ni ),
ˆ the size of the counting window (ri ).

The series of points (ri , Ni ) is plotted on a two-dimension graph. The Y-axis corresponds to the number
of counted elements (Ni ) and the X-axis corresponds to the size the counting window (ri ), which increases
from step to step.

Estimation process

The series of points (ri , Ni ) forms a curve (named the empirical curve). This empirical curve is fitted
with another one, the estimated curve. If the empirical curve follows a (mono-)fractal law, the estimated
curve has the form of a power law (parabolic or hyperbolic):

N = rD

7
or
N = r−D
where D is the fractal dimension. Usually, the empirical curve is represented on a log-log plot. Thus the
estimation is a linear regression and D is the slope of the resulting estimated straight line.

log N = log rD = D log r

3.1.1 Box counting analysis

Box counting analysis is the simplest method that can be used to estimate a fractal dimension. The
pattern under analysis is covered by a quadratic grid. The size of the grid cells ri increases from one
iteration step to the next. Considering each grid cell size ri , the number of cells Ni containing at least
one point (black pixel in case of raster data; point, line or polygon in case of vector data) is counted.

r : 9 N = 125 r : 27 N = 25 r : 81 N = 5 r : 243 N = 1

D ∼ 1.465

3.1.2 Dilation analysis

Each point of the pattern (black pixel in case of raster data; polygon in case of vector data) is surrounded
by a square (or a circle) of size ri , the surface of which is considered to be completely occupied. These
squares or circles are gradually enlarged, and the total area A(ri ) covered at each stage i of the dilation
process is measured. As the squares are enlarged, any details smaller than ri are overlooked. More and
more squares overlap and the total occupied area A(ri ) for a particular value ri is lower than it would be
if the same number of points that make up the original shape would have been surrounded individually.
By dividing this total area by the area of a square (ri2 ) or a circle (π r2i 2 ), we obtain an approximation of
the number of elements Ni required to cover the whole pattern.

A(ri ) A(ri )
Ni ≈ or ri 2
ri2 π2

8
 r:1 r:3 r:9 r : 27
r : 81

D ∼ 1.454

3.1.3 Correlation analysis

Each point of the pattern is surrounded by a small square (or circle) window. The number of points
(black pixels in case of raster data) within each window is counted. Then the mean number of points
per window size is calculated (Ni ). The same operation is applied for windows of increasing sizes ri . In
principle, it is possible to choose any shape for the counting window, such as circle, hexagon, etc. Yet in
Fractalyse, the counting window is a square for raster data, which helps us to avoid rounding errors, and
a circle for vector data. Note: because the correlation analysis considers the simultaneous presence of two
points at a certain distance, i.e. the mean distance between a pair of points, the correlation dimension is
a second order fractal dimension.

D ∼ 1.44

3.1.4 Radial analysis

The radial analysis is identical to the correlation analysis but only one counting point is considered. The
counting point is located anywhere in the pattern (most often at the centre of the pattern). A counting
window is drawn around the counting point. Its size grows at each iteration step ri . The number of
points Ni within the window is counted.
As for the correlation analysis, the shape of the counting window is a square for raster data and a circle
for vector data.

9
 √ √ √
r : 9 2 N = 25 r : 27 2 N = 125 r : 81 2 N = 625
√
r : 243 2 N = 3125

D ∼ 1.465

3.2 Multifractal analysis

Whereas monofractal analyses are performed on binary images representing the presence or absence of
a phenomenon (e.g. built-up surfaces), multifractal analyses also take into account the relative amount
of the phenomenon in each place. The intensity of the phenomenon under consideration varies with
the spatial aggregation level. Multifractal analysis can be thought of as an extension of monofractal
analysis by the addition of information about the relative intensity in each place of the phenomenon under
consideration. It integrates not just a series of nested spatial resolutions (i.e. analyzing the phenomenon
using spatial units of nested sizes), as monofractal analysis does [Frankhauser(1998), Stanley et al.(1999)],
but also a series of points of view about the quantity of information contained in each spatial unit
[Sémécurbe et al.(2016)].
Multifractal analysis consists in analysing the spatial distribution of the measure µi,r for a series of
scales r and then to observe the result of the function when r changes. Obviously, a single function is
not enough to fully characterize the spatial distribution of the measure µi,r . Alternatively, a group of
functions, namely the structure functions Mr,q , are suitable for this:

X
Mr,q = µi,r q
i

10
 Figure 3.8: A series of structure functions Mr,q

Multifractal analysis requires us to set the assumption that the structure functions Mr,q follow a scaling
law when the values of r are small:

Mr,q ≈ rτq

τq is a scaling exponent. It can be calculated using the following formula:

log Mr,q
τq = lim
r→0 − log r

Of course, it is impossible to reduce r toward 0 indefinitely. Consequently, the estimation of τq is given

by the slope of the curve of log Mr,q as a function of log r.

Figure 3.9: τq

On the basis of τq , it is possible to define the generalized dimensions Dq :

1
Dq = τq for q 6= 1
1−q

11
 Figure 3.10: Dimension spectrum Dq

Dimension spectrum

Dq with respect to q is called the dimension spectrum. The coefficient q acts as a filter when calculating
the generalized dimensions. For q → +∞, only the cells in which high measures are concentrated are
taken into account. Conversely, for q → −∞ generalized dimensions describe the spatial distribution of
the elements in which low measures are concentrated. For q = 1, all cells have the same importance
whatever their measure. For q = 0, the generalized dimension does not take into account the spatial
variation of the measure; only the fact that the cells are populated or not is taken into account. Because
of this sort of filtering operated by q, generalized dimensions Dq allow a characterization of the spatial
distribution of population that is both multiscale (via the scale r) and multi-viewpoint (via the exponent
q) [Sémécurbe et al.(2016)].

Singularity spectrum f (αq ) (also called multifractal spectrum)

The multifractal formalism enables the determination of the singularity spectrum through a Legendre
transform of the scaling exponent τq [Jaffard et al.(2007)]. f (α0 ) gives the maximum value of the singu-
larity spectrum. It indicates the strength of the irregularities of the measure. Complementarily, the range
of the spectrum informs us about the multifractality of the spatial distribution of the measure (more or
less strong, even non-existent in the case of a monofractal distribution) [Wendt et al.(2007)].

Figure 3.11: Singularity spectrum f (αq )

12
Case studies in spatial analysis are highly dependent on the size of the spatial units used for the analysis.
This has been expressed by [Openshaw(1983)] as the Modifiable Areal Unit Problem (MAUP): whatever
the phenomenon under consideration, it is impossible to identify a single spatial partition that would
be most appropriate for analyzing it. Interestingly, both spectrums Dq and f (αq ) enlighten the MAUP
[Sémécurbe et al.(2016)].
ˆ The absence of MAUP means that the spatial distribution of population is almost uniform. In this
case, the dimension spectrum is a horizontal line with the y-intercept equal to 2 and the singularity
spectrum is the point of coordinates (2, 2).
ˆ When the measure is monofractal then the singularity spectrum f (αq ) is only a point of coordinates
αq = f (αq ) = D0 . In this case, the MAUP only results from the spatial distribution of the
populated places. Basically, the study area contains identical buildings all having the same number
of inhabitants and being spatially distributed in a monofractal manner. In this case, local differences
in population density come only from the spatial distribution of the buildings.
ˆ A singularity spectrum having a range of αq values indicates a multifractal behavior. In this case,
the spatial footprint of the buildings varies (e.g. small single-family houses, multi-family houses,
high and large collective buildings) as well as their number of inhabitants: the MAUP results from
both.

3.2.1 Box counting multifractal analysis

The counting method follows the same principle as the monofractal box counting method (3.1.1) except
that µi,r measures here the proportion of elements in the grid cell i of size r and not just the fact that
the grid cell is occupied or not.

For q = 1 Dq is: P
log i µi,r log µi,r
D1 = lim
r→0 log r

When q = 0, the function Mr,q corresponds to the mono-fractal box counting function, i.e. the number
of occupied cells is simply counted whatever the proportion of elements within each grid cell.

3.2.2 Sandbox

The counting method follows the same principle as the monofractal correlation method (3.1.3).
Since the sandbox dimension is a second order fractal dimension, q exponent in Mr,q is shifted:
n
1X
Mr,q = µi,r q−1 = hµi,r q−1 i
n i=1

When q = 2 the function Mr,q is equivalent to the mono-fractal correlation dimension.

For q = 1 Dq is:
hlog µi,r i
D1 = lim
r→0 log r

3.3 Network analysis

Correlation (3.1.3), radial (3.1.4) and sandbox (3.2.2) analyses have been adapted to deal with network
data. In this case, the size ri measured is not a euclidean distance but a network distance (topological
distance, metric length or any cumulative cost). The counted elements Ni can be the edges or the nodes
of the graph as chosen by the user.

13
Bibliography

[Frankhauser(1998)] Pierre Frankhauser, 1998. The fractal approach. a new tool for the spatial analysis
of urban agglomerations. Population, 10(1) 205–240.
[Jaffard et al.(2007)] Stéphane Jaffard, Bruno Lashermes, and Patrice Abry, 2007. Wavelet leaders in
multifractal analysis. In Tao Qian, MangI Vai, and Yuesheng Xu, editors, Wavelet Analysis and
Applications, Applied and Numerical Harmonic Analysis, pages 201–246. Birkhäuser Basel.
[Openshaw(1983)] Stan Openshaw, 1983. The modifiable areal unit problem. Geo Books, 38.
[Stanley et al.(1999)] HE Stanley, LA Nunes Amaral, AL Goldberger, S Havlin, P Ch Ivanov, and C-K
Peng, 1999. Statistical physics and physiology: monofractal and multifractal approaches. Physica A:
Statistical Mechanics and its Applications, 270(1) 309–324.
[Sémécurbe et al.(2016)] François Sémécurbe, Cécile Tannier, and Stéphane G. Roux, 2016. Spatial distri-
bution of human population in france: Exploring the modifiable areal unit problem using multifractal
analysis. Geographical Analysis, 48(3) 292–313.
[Wendt et al.(2007)] Herwig Wendt, Patrice Abry, G. Roux, Stéphane, and Stéphane Jaffard, 2007. Anal-
yse multifractale d’image : l’apport des coefficients dominants. In GRETSI 2007. Université de tech-
nologie de Troyes (UTT) ; Université de Reims Champagne-Ardennes (URCA), GRETSI, Troyes,
France.

14
 Part II

Graphical user interface

15
Chapter 4

Data

4.1 Loading data: File menu

4.1.1 Load vector data menu

Fractalyse can read three types of vector data from Geographic information systems (GIS): Geopackage
(.gpkg), GeoJSON (.geojson) and Shapefile (.shp).
Three types of geometries are supported: points, lines, and polygons.
When a shapefile is loaded with the menu Load vector data, a new layer is created and displayed in
the GUI (graphical user interface).

Figure 4.1: Vector data loaded in Fractalyse

Network vector data

If the vector layer contains only simple linear geometries, a graph representation of the network is auto-
matically created. This allows the user to perform network analysis in addition to vector analysis.

16
 Figure 4.2: Vector network data loaded in Fractalyse

4.1.2 Load raster data menu

Fractalyse can read two types of raster data: TIFF and Ascii Grid.
When the input data file is loaded with the menu Load raster data menu, a new layer is created and
the image is displayed. If the image contains only 0 and 1 values, Fractalyse classifies the layer as a
binary layer and sets white color for 0 and black color for 1. Otherwise, it sets a gray scale color ramp
for the image. Color images (RGB) are not supported.

17
 Figure 4.3: Raster binary data loaded in Fractalyse

Figure 4.4: Raster gray scale layer of data loaded in Fractalyse

4.1.3 Load network graph menu

If your network data is defined as a point layer and a table of links, you can load your network from the
menu Load network graph.

18
 Figure 4.5: Dialog box that opens with the Load network graph menu

4.1.4 Load estimation menu

A previously saved estimation can be reloaded from this menu entry. For more details, see 6.3.

4.1.5 Other menus

Other File menu entries are explained in Miscellaneous chapter.

4.2 Data handling: Tools menu

In case of vector analyses, some types of data are not supported (for example, lines or polygons can not
be used for a correlation analysis). When raster data are used, most analyses require binary input data.
Fractalyse integrates functions in the Tools menu that enable the conversion of vector data into raster
data (Rasterize menu) and the conversion of gray scale raster data into binary data (Binarize menu).

4.2.1 Rasterize menu

The Rasterize menu converts vector data into raster data. Any vector layer can be rasterized. If Field
is not filled in, the result is a binary raster layer and a gray scale raster otherwise. If the vector layer
contains polygons, the rasterization mode may be changed with Poly mode option.

Figure 4.6: Rasterization of a vector layer

19
4.2.2 Binarize menu

The Binarize menu converts gray scale raster to black and white raster.

Figure 4.7: Binarization of a gray scale layer

4.2.3 Negative menu

The Negative menu reverses binary raster data (i.e. black and white images). The user has to right click
on the layer and select the Negative menu. Black pixels become white and white pixels become black.

4.2.4 Selection menu

Selection menu allows to select a sub-part of a layer to analyse it separately. After selecting the region of
interest by a rectangle, click on Create button to create a new layer containing only the selected region.

4.2.5 Contextual layer menu

By right-clicking on a layer, user can access a contextual menu allowing to :

ˆ Remove: remove the current layer from Fractalyse
ˆ Style: change the drawing style of the layer
ˆ Export: save the current layer to a new file

20
Chapter 5

Fractal analysis in practice

Fractalyse offers three analysis menus, one for each type of input data: Vector, Raster, Network.

5.1 Choice of a scale range for the analysis

All fractal analysis methods require to define the range of scales that will be considered for the analysis.
This sequence is given in a panel contained in each analysis frame.

Figure 5.1: Choice of a scale range

The scale range is comprised between Min size and Max size. By default, the sequence of scales is
geometric, which means that each scale value is multiplied by the value Coef (starting from Min size).
When the sequence chosen is arithmetic, the value Coef is added to each scale value. For geometric
sequence, Coef must be strictly greater than 1, for arithmetic sequence Coef must be strictly greater
than 0.
Min size and Max size values are by default automatically estimated from the input data. For raster
data, Min size is the pixel size. For vector data, Min size is the size of the smallest element (polygon
or line segment). The value of Max size is set by default as the half of the minimum extent of the data
layer.
For raster data, the scales have to be rounded to the pixel unit. This may result in slight deviations from
the defined sequence.

21
5.2 Vector menu
Fractal analysis for vector data.

5.2.1 Box counting menu

The method is described in 3.1.1. The user first needs to set the vector layer under consideration and
defines the scale range for the analysis. She/he also has to set the value of two specific parameters:
Number of grid positions and View boxes. At the end of the computation, the estimated fractal
dimension is displayed in the Estimation frame (see 6.1) and a new layer displaying the scale range at
the bottom of the map is added.

Figure 5.2: Box counting dialog box

Number of grid positions

At each iteration step, a chosen Number of grid positions are tested and the number of elements Ni
is counted for each grid position. The appropriate grid position corresponds to the lowest value of Ni .
Note that, by the way, the position of the grid may vary from one iteration step to the next.

View boxes

If you check the View boxes option, all the boxes containing data are saved at each iteration step and
can be displayed on the map; one layer for each scale. Note that this option is memory intensive since
the boxes kept in memory can be numerous.

22
 Box size: 256 Box size: 512

Box size: 1024 Box size: 2048

Figure 5.3: Box counting - View boxes

5.2.2 Dilation menu

The method is described in 3.1.2. The user first needs to set the vector layer under consideration and
defines the scale range for the analysis. She/he also has to set the value of two specific parameters:
Stop at one cluster and View dilated patterns. At the end of the computation, the estimated
fractal dimension is displayed in the Estimation frame (see 6.1) and a new layer displaying the scale
range at the bottom of the map is added.
In the estimation window, the curve of the number of clusters at each dilation size can also be displayed
in the Other curve panel.

23
 Figure 5.4: Dilation dialog box

Stop at one cluster

If the user selects the Stop at one cluster option, the dilation stops when the dilated pattern is made
up of a single polygon. Look at the pattern dilation size 2048 on figure 5.5.

View dilated patterns

If the user selects the View dilated patterns option, the dilated geometries are saved at each iteration
step and can be displayed on the map; one layer for each scale. Note that this option can require a large
amount of memory since all dilated geometries are kept in memory.

24
 Dilation size: 256 Dilation size: 512

Dilation size: 1024 Dilation size: 2048

Figure 5.5: Dilation - View dilated patterns

5.2.3 Correlation menu

The method is described in 3.1.3. Only point data can be analyzed with the vector correlation. The
analysis of lines or polygons requires first to rasterize (see 4.2.1) the data, and then to analyse them with
the raster correlation (5.3.3).
The user first needs to set the vector layer under consideration and defines the scale range for the analysis.
At the end of the computation, the estimated fractal dimension is displayed in the Estimation frame
(see 6.1) and a new layer displaying the scale range at the bottom of the map is added.

Figure 5.6: Correlation dialog box

25
5.2.4 Radial menu

The method is described in 3.1.4. The user first needs to set the vector layer under consideration and
defines the scale range for the analysis. She/he also has to set the value of one specific parameter: Centre
that corresponds to the starting point of the counting process. The coordinates can be entered directly
or set by clicking on the map.
At the end of the computation, the estimated fractal dimension is displayed in the Estimation frame
(see 6.1) and a new layer displaying the scale range on the map is added.
The counting unit depends on geometry type of the vector layer:
ˆ puntal: number of points
ˆ lineal: length
ˆ polygonal: area

Figure 5.7: Radial dialog box

5.2.5 Batch menu

Vector methods can be executed in batch mode on each subarea of a vector layer. The subareas can be
defined by a regular grid or another polygon layer.

Figure 5.8: Batch dialog box

26
Each resulting polygon contains the fractal dimension in its attributes and also the confidence interval
and the r2.

5.2.6 Multifractal menu

The principles of multifractal analysis are described in 3.2. Fractalyse implements 2 multifractal analysis:
boxcounting and sandbox.

Box counting menu

The method is described in 3.2.1. The settings are exactly the same as the monofractal boxcounting
analysis (see 5.2.1).
At the end of the computation, the estimated fractal dimension spectrum is displayed in the Multi-fractal
estimation frame (see 6.2) and a new layer displaying the scale range at the bottom of the map is added.

Sandbox menu

The method is described in 3.2.2. The settings are exactly the same as the correlation analysis (see 5.2.3).
At the end of the computation, the estimated fractal dimension spectrum is displayed in the Multi-fractal
estimation frame (see 6.2) and a new layer displaying the scale range at the bottom of the map is added.

5.3 Raster menu

Fractal analysis for raster data. Most analysis only work with binary raster except radial and multi-fractal
sandbox which support binary and grayscale raster layers.

5.3.1 Box counting menu

The method is described in 3.1.1. The user first needs to set the binary raster layer under consideration
and defines the scale range for the analysis. At the end of the computation, the estimated fractal
dimension is displayed in the Estimation frame (see 6.1) and a new layer displaying the scale range at
the bottom of the map is added.

Figure 5.9: Box counting dialog box

5.3.2 Dilation menu

The method is described in 3.1.2. The user first needs to set the binary raster layer under consideration
and defines the scale range for the analysis. At the end of the computation, the estimated fractal

27
dimension is displayed in the Estimation frame (see 6.1) and a new layer displaying the scale range at
the bottom of the map is added.

Figure 5.10: Dilation dialog box

5.3.3 Correlation menu

The method is described in 3.1.3. The user first needs to set the binary raster layer under consideration
and defines the scale range for the analysis. At the end of the computation, the estimated fractal
dimension is displayed in the Estimation frame (see 6.1) and a new layer displaying the scale range at
the bottom of the map is added.

Figure 5.11: Correlation dialog box

5.3.4 Radial menu

The method is described in 3.1.4. The user first needs to set the raster layer (binary or grayscale) under
consideration and defines the scale range for the analysis. She/he also has to set the value of one specific
parameter: Centre that corresponds to the starting point of the counting process. The coordinates can
be entered directly or set by clicking on the map.
At the end of the computation, the estimated fractal dimension is displayed in the Estimation frame
(see 6.1) and a new layer displaying the scale range on the map is added.

28
 Figure 5.12: Radial dialog box

5.3.5 Radial all points menu

This analysis applies radial method centered on each black pixel of the raster and estimates for each, the
fractal dimension. The result is a raster where each black pixel contains the local fractal dimension. Addi-
tional rasters contain the r2 quality of estimation and the confidence interval if Compute confidence interval
is checked. The default estimation model is axd and can be changed. The sampling sequence is arithmetic
from size 1 to Max size. The max size can be estimated automatically by checking Auto threshold. In
this case a minimal max size must be given. The max size is estimated by finding the main inflexion point
in the estimated curve between min and Max size parameters. If there is no inflexion point Max size is
used. With this option, an additional raster is added containing the estimated max size.

Figure 5.13: Radial all points dialog box

29
 Figure 5.14: Radial all points result

5.3.6 Multifractal menu

The principles of multifractal analysis are described in 3.2. Fractalyse implements 2 multifractal analysis:
boxcounting and sandbox.

Box counting menu

The method is described in 3.2.1. The settings are exactly the same as the monofractal boxcounting
analysis (see 5.3.1).
At the end of the computation, the estimated fractal dimension spectrum is displayed in the Multi-fractal
estimation frame (see 6.2) and a new layer displaying the scale range at the bottom of the map is added.

Sandbox menu

The method is described in 3.2.2. The settings are exactly the same as the correlation analysis (see 5.3.3).
At the end of the computation, the estimated fractal dimension spectrum is displayed in the Multi-fractal
estimation frame (see 6.2) and a new layer displaying the scale range at the bottom of the map is added.

5.4 Network menu

This menu contains fractal analysis for network data. For each method, before defining the sampling as
previous methods, user have to define the distance type and counting mass data to be used.

30
5.4.1 Choice of distance and mass counting

Figure 5.15: Network panel

First the distance measure must be set. It can be of 3 types:

ˆ (topology): topological distance ie. number of edges
ˆ (length): length in spatial units of the edges
ˆ layer attribute: the edge distance corresponds to its attribute value

After, the counted mass must be set. This mass corresponds to:
ˆ Distance: the total length of the network in the selected distance unit.
ˆ Node attribute: the sum of the selected node attribute.
ˆ Edge attribute: the sum of the selected edge attribute.

After setting these 2 parameters, the sampling can be set manually or automatically by clicking on
Estimate sampling.

5.4.2 Correlation menu

The method is described in 3.1.3. The user first needs to set the network layer under consideration,
defines the distance type and the counted mass, and defines the scale range for the analysis. At the end
of the computation, the estimated fractal dimension is displayed in the Estimation dialog (see 6.1).

31
 Figure 5.16: Network correlation dialog box

5.4.3 Radial menu

The method is described in 3.1.4. The user first needs to set the network layer under consideration,
defines the distance type and the counted mass, and defines the scale range for the analysis. She/he also
has to set the value of one specific parameter: Starting point that corresponds to the starting point of
the counting process. The coordinates can be entered directly or set by clicking on the map. The starting
point will be automatically snapped to the nearest node of the network. At the end of the computation,
the estimated fractal dimension is displayed in the Estimation dialog (see 6.1).

32
 Figure 5.17: Network radial dialog box

5.4.4 Service menu

Not documented yet

5.4.5 Backbone menu

Not documented yet

5.4.6 Multifractal menu

The principles of multifractal analysis are described in 3.2. Fractalyse implements one multifractal
analysis for network: sandbox.

Sandbox menu

The method is described in 3.2.2. The settings are exactly the same as the correlation analysis (see 5.4.2).
At the end of the computation, the estimated fractal dimension spectrum is displayed in the Multi-fractal
estimation dialog (see 6.2).

33
Chapter 6

Estimation module

After the counting process, the fractal dimension(s) can be estimated with this module. There is 2 main
types of estimation depending of the counting method: monofractal and multifractal.

6.1 Monofractal
The principles of the monofractal estimation is described in 3.1. The empirical curve coming from
the monofractal method is estimated by an OLS (ordinary least square) regression. The estimated D
coefficient corresponds to the fractal dimension. There is 2 types of regression in Fractalyse:
ˆ LOG: the curve is logged to use a linear model log y = D log x + b. This type is used by default
with geometric sequence.
ˆ DIRECT: the curve is estimated with a power law function of type y = axD . This type is used by
default with arithmetic sequence.

It is strongly recommended to not change the Type and the Model.

34
 Figure 6.1: Estimation window

The graph part shows the empirical curve as little circles and the estimated curve (regression model) as
red line.

Numerical results

ˆ Dimension: the estimated fractal dimension (D coefficient of the model)

ˆ a,b,c: other coefficients depending of the chosen model
ˆ R2: the determination coefficient of the regression r2
ˆ p-value: the significance of the regression, must be lower than 0.001
ˆ Confidence (95%): the confidence interval of the fractal dimension at 95%. Only computed for
linear regression.
ˆ Bootstrap confidence: the confidence interval of the fractal dimension at 95% estimated by boot-
strap. Computed for all models.

Bounds

By default the regression is calculated on all points of the empirical curve. User can remove some points
at the start (left) or end (right) of the curve. The regression is automatically updated and also the scale
range on the map.

35
 Figure 6.2: Estimation with user defined bounds

Scaling behaviour

The scaling behaviour corresponds to the first derivative of the logged empirical curve. It shows the local
estimates of the fractal dimension.

Figure 6.3: Scaling behaviour

Other curve

For some methods, there is additional curve. For the moment only dilation method produce additional
curve: the number of clusters for each dilation size.
The Export button allows to export all numerical results in a text file or export the graphical curves in
SVG format.

36
The Save button allows to save the estimation in XML format to reload it after. See Save/reload
estimation.

6.2 Multifractal
The principles of the multifractal estimation is described in 3.2. The logged empirical curves (Mq )
coming from the multifractal method are estimated by a linear OLS (ordinary least square) regression.
The estimated slopes correspond to the τq curve.

Figure 6.4: Multifractal estimation window - Mq

q set

The set of q can be changed with min, max and increment step. User have to click on Update button
after, to update all curves (Mq , τq , ...) and numerical results.

Numerical results

The numerical results consist of the min/max fractal dimension spectrum Dq and D0 corresponding to
the classical monofractal dimension.

Bounds

As monofractal estimation, the regression on Mq curves can be only computed on a subset of scale sizes.

Curves

By default, the graph shows the Mq curves. User can displayed the other curves : τq , Dq and f (αq ) by
selecting it in the dropdown list.

37
View estim

Since there is many linear regressions (one for each Mq ), they are not showed directly in the frame.
Estimation of one Mq curve can be shown by selecting the wanted q and click on View estim button. A
monofractal estimation window is shown with the estimation of the selected Mq curve.
The Export button allows to export all numerical results in a text file or export the current graphical
curves in SVG format.
The Save button allows to save the estimation in XML format to reload it after. See Save/reload
estimation. Be careful, the q-set cannot be changed after reloading it.

Multifractal curves

The τq curve must be monotonic decreasing.

Figure 6.5: Multifractal estimation window - τq

The dimension spectrum Dq must be monotonic decreasing. If not, the results cannot be used.

38
 Figure 6.6: Multifractal estimation window - Dq

Figure 6.7: Multifractal estimation window - f (αq )

6.3 Save/reload estimation

Estimations can be saved and reloaded later. To save an estimation, use the Save button on the estimation
frame. The estimation is stored in XML format. If you have already closed the estimation frame
you can reopen it from the layer panel of the main frame. To reload an estimation, use the menu
File - Load estimation. After selecting the XML file, the estimation window will appear with the
saved estimation.
For multifractal estimation, the q-range cannot be changed after reloading it.

39
6.4 Estimation layer: scale range
For each estimation frame, a new layer showing the scale range on the map is created. This scale range
is updated when user changes the bounds of the estimation.
The contextual menu of this layer allows to:
ˆ Remove: remove this estimation and the scale range layer
ˆ Estimation: reload the estimation frame if you have already closed it
ˆ Save: save the estimation in a XML file

40
Chapter 7

Miscellaneous

7.1 Preferences menu

7.1.1 Memory

The memory used by Fractalyse plays an important role. If there is not enough RAM, computation will
be slower or may fail (OutOfMemoryError or GC Overhead message). This panel can be used to adjust
the memory allocated to Fractalyse. If you have a 32-bit version of Java, Fractalyse will be limited to
about 2 Go (2000 Mo) of memory. If your computer has more than 2 Go of RAM memory, it is highly
recommended you install the 64-bit version of Java to use the available memory beyond 2 Go. After
changing the amount of memory, you have to restart Fractalyse.

7.1.2 Processors

Most of Fractalyse calculations are parallelized. This means that Fractalyse can use several cores/processors
to speed up computations, a quad-core processor being theoretically four times faster than a single-core
processor. By default Fractalyse sets the number of cores to the number of computer cores minus 1.
After changing the parameter, you have to restart Fractalyse.

7.2 Log window menu

This window show some details when error occurred. These messages are useful for debugging purpose.

41
 Part III

Command line interface

42
Chapter 8

Prerequisite

Fractalyse can be used in command line interface (CLI). It is useful for executing Fractalyse on a distant
computer without a graphical interface, or batching some processes that are not available in the graphical
user interface (GUI).

8.1 Launch Fractalyse in CLI mode

First you have to open a terminal window. Then, go to the directory of the Fractalyse program with cd
command. Finally, type the following command to display the Fractalyse help screen:
java -jar fractalyse-3.0.jar --help
Result
Usage :
java -jar fractalyse.jar [-proc n]
...
...
You’re ready to use Fractalyse in CLI mode !

8.2 Syntax

8.2.1 Definition

Commands always start with a double dash (--).

A global option starts with only one dash (-).
A parameter does not have a dash.

8.2.2 Character separator

Blank spaces are used to separate commands and parameters. You cannot have a name containing blank
spaces.

8.2.3 Optional parameter

Parameters enclosed in brackets are optional. Therefore, parameters not in brackets are mandatory.

43
8.2.4 Command execution

Fractalyse can execute only one command.

java -jar fractalyse-3.0.jar ...

44
Chapter 9

Command reference

9.1 General command

9.1.1 --help: display help

Command :
java -jar fractalyse-3.0.jar --help
Result :
Usage :
java -jar fractalyse.jar [-mpi | -proc n] COMMAND
COMMAND:
--rasterize [neg] res=val file_1.shp [... file_n.shp]
--binarize min=val max=val file_1.tif [... file_n.tif]
--boxcounting [gliding=val] SAMPLING [estim=log|direct] file_1.shp [... file_n.shp]
--rboxcounting SAMPLING [estim=log|direct] file_1.tif [... file_n.tif]
--dilation SAMPLING [estim=log|direct] file_1.shp [... file_n.shp]
--rdilation SAMPLING [estim=log|direct] file_1.tif [... file_n.tif]
--correlation SAMPLING [estim=log|direct] file_1.shp [... file_n.shp]
--rcorrelation [border=val] SAMPLING [estim=log|direct] file_1.tif [... file_n.tif]
--clusters [buf=radius]file_1.shp [... file_n.shp]
SAMPLING:
[coef=val] [min=val] [max=val] [seq=arith|geom]

9.2 Data manipulation commands

9.2.1 --rasterize: convert vector to raster

--rasterize [neg] res=val file_1.shp [... file_n.shp ]

This command converts vector files into raster tif file. Shapefile and Geopackage vector layer can be
rasterized. The result is a binary raster layer. The new files have the same name. Only the file extension
is changed.

Parameters

ˆ res=val: the pixel resolution of the resulting raster

ˆ neg: if set pixel values are reverse, black pixels become white and white pixels become black.

45
9.2.2 --binarize: convert grayscale raster to binary

--binarize min=val max=val file_1.tif [... file_n.tif ]

This command converts gray scale raster to black and white raster. The new files name are suffixed with
-bin.

Parameters

ˆ min=val: the minimal value for a pixel to become black

ˆ max=val: the maximal value for a pixel to become black.

9.3 Monofractal analysis commands

Each fractal method can be applied to one or more spatial data files. For each file, the counting process and
the fractal dimension estimation are calculated and saved in a text file with the same name. Another text
file named with the method and its parameters is also created containing the summary of all estimations.
The parameters are globally the same for all methods :
ˆ method: the methods name prefix with r for raster data
ˆ optional parameters if any
ˆ sampling: the scale sampling (optional)
ˆ estimation type: logarithmic or direct (optional)
ˆ one or more files corresponding to vector or raster layers

9.3.1 --boxcounting: box counting on vector data

--boxcounting [gliding=val ] SAMPLING [estim=log|direct] file_1.shp [... file_n.shp ]

This command apply the vector boxcounting method to one or more vector layers. The method is
described in 3.1.1. The sampling configuration is described in 9.3.7 and the estimation type in 9.3.8.

Optional parameter

ˆ gliding=val: number of gliding grid tested. For more details see Number of grid positions in
section 3.1.1.

For each vector file a text file with the same name is created containing the detail of each estimation.
One text file is also created containing the summary of all fractal estimations.

9.3.2 --rboxcounting: box counting on raster data

--rboxcounting SAMPLING [estim=log|direct] file_1.tif [... file_n.tif ]

This command apply the raster boxcounting method to one or more raster layers. The method is described
in 3.1.1. The sampling configuration is described in 9.3.7 and the estimation type in 9.3.8.
For each raster file a text file with the same name is created containing the detail of each estimation.
One text file is also created containing the summary of all fractal estimations.

46
9.3.3 --dilation: dilation on vector data

--dilation SAMPLING [estim=log|direct] file_1.shp [... file_n.shp ]

This command apply the vector dilation method to one or more vector layers. The method is described
in 3.1.2. The sampling configuration is described in 9.3.7 and the estimation type in 9.3.8.
For each vector file a text file with the same name is created containing the detail of each estimation.
One text file is also created containing the summary of all fractal estimations.

9.3.4 --rdilation: dilation on raster data

--rdilation SAMPLING [estim=log|direct] file_1.tif [... file_n.tif ]

This command apply the raster dilation method to one or more raster layers. The method is described
in 3.1.2. The sampling configuration is described in 9.3.7 and the estimation type in 9.3.8.
For each raster file a text file with the same name is created containing the detail of each estimation.
One text file is also created containing the summary of all fractal estimations.

9.3.5 --correlation: correlation on vector data

--correlation SAMPLING [estim=log|direct] file_1.shp [... file_n.shp ]

This command apply the vector correlation method to one or more point vector layers. The method is
described in 3.1.3. The sampling configuration is described in 9.3.7 and the estimation type in 9.3.8.
For each vector file a text file with the same name is created containing the detail of each estimation.
One text file is also created containing the summary of all fractal estimations.

9.3.6 --rcorrelation: correlation on raster data

--rcorrelation [border=val] SAMPLING [estim=log|direct] file_1.tif [... file_n.tif ]

This command apply the raster correlation method to one or more raster layers. The method is described
in 3.1.3. The sampling configuration is described in 9.3.7 and the estimation type in 9.3.8.

Optional parameter

ˆ border=val: exclude pixels as counting centre too near from the raster border to avoid border
effect. The value given corresponds to the distance from the raster border in pixel. To avoid
completely border effect, user have to give the same distance in maxsize sampling parameter with
respect to spatial unit.

For each raster file a text file with the same name is created containing the detail of each estimation.
One text file is also created containing the summary of all fractal estimations.

9.3.7 SAMPLING: scale range definition

[coef=val ] [min=val ] [max=val ] [seq=arith|geom]

All fractal analysis methods require to define the range of scales that will be considered for the analysis.
If not set the sampling is automatically estimated from the data layer.

47
Optional parameters

ˆ seq=arith|geom: arithmetic or geometric sequence. By default the sequence sampling is set to

geometric.
ˆ coef=val: coefficient for increasing scales from min to max. By default coef = 2 for geometric
sequence and coef = 1 for arithmetic sequence.
ˆ min=val: the minimal scale size
ˆ max=val: the maximal scale size

The scale range is comprised between min and max. By default, the sequence of scales is geometric, which
means that each scale value is multiplied by the value coef (starting from min). When the sequence
chosen is arithmetic, the value coef is added to each scale value. For geometric sequence, coef must be
strictly greater than 1, for arithmetic sequence coef must be strictly greater than 0.
The parameters min and max are by default automatically estimated from the input data. For raster data,
min is the pixel size. For vector data, min is the size of the smallest element (polygon or line segment).
The max parameter is set by default as the half of the minimum extent of the data layer. If a fractal
method is applied on several data layer the sampling range (min and max size) may differ for each layer.
For raster data, the scales have to be rounded to the pixel unit. This may result in slight deviations from
the defined sequence.

9.3.8 Estimation parameter

[estim=log|direct]
By default, the estimation type (log or direct) is set depending on the sequence sampling: for geometric
sequence the estimation is logarithmic and for arithmetic sequence the direct type is used. User can
change this behaviour by adding estim parameter. For more details on estimation method, see 6.1.

9.4 Other commands

9.4.1 --clusters: counting clusters for vector layer

--clusters [buf=radius] file_1.shp [... file_n.shp ]

9.5 Options

9.5.1 -proc

Defines the number of processors (or cores) used by Fractalyse. By default, CLI mode uses the value
defined in the preferences window. See the Performance tuning section for more details.

9.5.2 -mpi

This option is used to execute commands on several computers in the MPI environment. See the paral-
lelism section of Performance tuning for more details.

48
Chapter 10

Examples of commands

10.1 Data handling commands

–rasterize

The following command rasterizes all vector geopackage files contained in the current directory. For each
.gpkg file, it creates a new raster .tif file with a resolution of 5 meters. Each new file has the same name
with the suffix: _res5.tif.
java -jar fractalyse-3.0.jar --rasterize res=5 *.gpkg

–binarize

The following command binarizes all tif images contained in the current directory. For each .tif file, it
creates a new file in which each pixel having a value between 1 and 10 is set to 1 and to 0 otherwise.
Each new file has a name with the suffix: _bin1-10.tif.
java -jar fractalyse-3.0.jar --binarize min=1 max=10 *.tif

10.2 Fractal analysis commands

The following command calculates the fractal dimension with box counting method for each raster pre-
viously binarized.
java -jar fractalyse-3.0.jar --rboxcounting *_bin*.tif
The following command calculates the fractal dimension with box counting method for each raster previ-
ously binarized and the scale sampling range is set between 10 and 1000. The sequence is left unchanged
(geometric) and also the coefficient (coef = 2).
java -jar fractalyse-3.0.jar --rboxcounting min=10 max=1000 *_bin*.tif
The same principle can be applied for all monofractal methods.

49
Chapter 11

Performance tuning

11.1 Parallelism to speed up execution

11.1.1 One computer: threads

If your computer has more than one core (most of them), you can take advantage of parallelism. Most
Fractalyse commands are parallelized. You can speed up command execution by defining the number of
cores (or processors) used by Fractalyse with the option -proc after the project command:
java -jar fractalyse-3.0.jar -proc 8 ...
By default, CLI mode uses the number of processors defined in the Preferences window of the GUI.

11.1.2 Computer cluster: mpi

Fractalyse can be run on computer clusters wich support Java for OpenMPI.
mpirun java -jar fractalyse-3.0.jar -mpi ...
Only some commands can be used in mpi environments: –boxcounting and –rcorrelation

11.2 Memory management

In CLI mode, the memory configuration defined in the Preferences window cannot be used. By default,
the amount of memory available for Fractalyse is system dependent. It can vary from 128 Mb to several
Gb. In most cases, Fractalyse runs normally. But if a large image is analysed, some commands would
be slow or even crash due to memory limitation. If Fractalyse execution ends up with the message
OutOfMemoryError, GC overhead or Java Heap space, you need to increase the memory allocated to
Fractalyse.
To define manually the maximum amount of memory allocated to Fractalyse, use Java option -Xmx :
java -Xmx2g -jar fractalyse-3.0.jar ... # 2Gb allocated
java -Xmx1500m -jar fractalyse-3.0.jar ... # 1500 Mb -> 1.5Gb allocated
If you cannot allocate more than 1Gb or 1.5G although your computer has more memory available, you
have probably a 32-bit version of Java, which is limited to less than 2Gb of memory. Check your Java
version:
java -version
If it is a 32-bit version, install a 64-bit Java version to handle all your computer memory.

50