3
<em>r.random.surface</em> generates a spatially dependent random surface.
4
The random surface is composed of values representing the deviation from the
5
mean of the initial random values driving the algorithm. The initial random
6
values are independent Gaussian random deviates with a mean of 0 and
7
standard deviation of 1. The initial values are spread over each output map
8
using filter(s) of diameter distance. The influence of each random value on
9
nearby cells is determined by a distance decay function based on exponent.
10
If multiple filters are passed over the output maps, each filter is given a
11
weight based on the weight inputs. The resulting random surface can have
12
<em>any</em> mean and variance, but the theoretical mean of an infinitely
13
large map is 0.0 and a variance of 1.0. Description of the algorithm is in
14
the <b>NOTES</b> section.
17
The random surface generated are composed of floating point numbers, and
18
saved in the category description files of the output map(s). Cell values
19
are uniformly or normally distributed between 1 and high values inclusive
20
(determined by whether the <b>-u</b> flag is used). The category names
21
indicate the average floating point value and the range of floating point
22
values that each cell value represents.
25
<em>r.random.surface's</em> original goal is to generate random fields for
26
spatial error modeling. A procedure to use <em>r.random.surface</em> in
27
spatial error modeling is given in the <b>NOTES</b> section.
29
<h3>Detailed parameter description</h3>
31
<dt><b>output</b></dt>
32
<dd>Random surface(s). The cell values are a random distribution
33
between the low and high values inclusive. The category values of the
34
output map(s) are in the form <em>#.# #.# to #.#</em> where each #.#
35
is a floating point number. The first number is the average of the
36
random values the cell value represents. The other two numbers are the
37
range of random values for that cell value. The <em>average</em> mean
38
value of generated <tt>output</tt> map(s) is 0. The <em>average</em>
39
variance of map(s) generated is 1. The random values represent the
40
standard deviation from the mean of that random surface.</dd>
42
<dt><b>distance</b></dt>
43
<dd>Distance determines the spatial dependence of the output
44
map(s). The distance value indicates the minimum distance at which two
45
map cells have no relationship to each other. A distance value of 0.0
46
indicates that there is no spatial dependence (i.e., adjacent cell
47
values have no relationship to each other). As the distance value
48
increases, adjacent cell values will have values closer to each
49
other. But the range and distribution of cell values over the output
50
map(s) will remain the same. Visually, the clumps of lower and higher
51
values gets larger as distance increases. If multiple values are
52
given, each output map will have multiple filters, one for each set of
53
distance, exponent, and weight values.</dd>
55
<dt><b>exponent</b></dt>
56
<dd>Exponent determines the distance decay exponent for a particular
57
filter. The exponent value(s) have the property of determining
58
the <em>texture</em> of the random surface. Texture will decrease as
59
the exponent value(s) get closer to 1.0. Normally, exponent will be
60
1.0 or less. If there are no exponent values given, each filter will
61
be given an exponent value of 1.0. If there is at least one exponent
62
value given, there must be one exponent value for each distance value.</dd>
65
<dd>Flat determines the distance at which the filter.</dd>
67
<dt><b>weight</b></dt>
68
<dd>Weight determines the relative importance of each filter. For
69
example, if there were two filters driving the algorithm and
70
weight=1.0, 2.0 was given in the command line: The second filter would
71
be twice as important as the first filter. If no weight values are
72
given, each filter will be just as important as the other filters
73
defining the random field. If weight values exist, there must be a
74
weight value for each filter of the random field.</dd>
77
<dd>Specifies the high end of the range of cell values in the output
78
map(s). Specifying a very large high value will minimize
79
the <em>errors</em> caused by the random surface's discretization. The
80
word errors is in quotes because errors in discretization are often
81
going to cancel each other out and the spatial statistics are far more
82
sensitive to the initial independent random deviates than any
83
potential discretization errors.</dd>
86
<dd>Specifies the random seed(s), one for each map,
87
that <em>r.random.surface</em> will use to generate the initial set of
88
random values that the resulting map is based on. If the random seed
89
is not given, <em>r.random.surface</em> will get a seed from the
90
process ID number.</dd>
96
While most literature uses the term random field instead of random surface,
97
this algorithm always generates a surface. Thus, its use of random surface.
100
<em>r.random.surface</em> builds the random surface using a filter algorithm
101
smoothing a map of independent random deviates. The size of the filter is
102
determined by the largest distance of spatial dependence. The shape of the
103
filter is determined by the distance decay exponent(s), and the various
104
weights if different sets of spatial parameters are used. The map of
105
independent random deviates will be as large as the current region PLUS the
106
extent of the filter. This will eliminate edge effects caused by the
107
reduction of degrees of freedom. The map of independent random deviates will
108
ignore the current mask for the same reason.
111
One of the most important uses for <em>r.random.surface</em> is to determine
112
how the error inherent in raster maps might effect the analyses done with
116
Random Field Software for GRASS by Chuck Ehlschlaeger
119
As part of my dissertation, I put together several programs that help
120
GRASS (4.1 and beyond) develop uncertainty models of spatial data. I hope
121
you find it useful and dependable. The following papers might clarify their
125
<li> Ehlschlaeger, C.R., Shortridge, A.M., Goodchild, M.F., 1997.
126
Visualizing spatial data uncertainty using animation.
127
Computers & Geosciences 23, 387-395. doi:10.1016/S0098-3004(97)00005-8</li>
129
<li><a href="http://www.geo.hunter.cuny.edu/~chuck/paper.html">Modeling
130
Uncertainty in Elevation Data for Geographical Analysis</a>, by
131
Charles R. Ehlschlaeger, and Ashton M. Shortridge. Proceedings of the
132
7th International Symposium on Spatial Data Handling, Delft,
133
Netherlands, August 1996.</li>
135
<li><a href="http://www.geo.hunter.cuny.edu/~chuck/acm/paper.html">Dealing
136
with Uncertainty in Categorical Coverage Maps: Defining, Visualizing,
137
and Managing Data Errors</a>, by Charles Ehlschlaeger and Michael
138
Goodchild. Proceedings, Workshop on Geographic Information Systems at
139
the Conference on Information and Knowledge Management, Gaithersburg
142
<li><a href="http://www.geo.hunter.cuny.edu/~chuck/gislis/gislis.html">Uncertainty
143
in Spatial Data: Defining, Visualizing, and Managing Data
144
Errors</a>, by Charles Ehlschlaeger and Michael
145
Goodchild. Proceedings, GIS/LIS'94, pp. 246-253, Phoenix AZ,
152
<a href="r.random.html">r.random</a>,
153
<a href="r.random.cell.html">r.random.cell</a>,
154
<a href="r.mapcalc.html">r.mapcalc</a>
158
Charles Ehlschlaeger, Michael Goodchild, and Chih-chang Lin; National Center
159
for Geographic Information and Analysis, University of California, Santa
163
<i>Last changed: $Date: 2011-10-07 21:53:04 +0200 (Fri, 07 Oct 2011) $</i>