You are on page 1of 7

Susan Williams

GEOG 586
October 2013

Project 3: Understanding Random Spatial Processes


Introduction
This lesson dealt with statistical concepts as applied to geospatial patterns,
particularly in regards to deterministic versus stochastic processes. A deterministic
process results in the same outcome whereas a stochastic process may vary from
outcome to outcome in a mathematically-defined way due to the inclusion of a
random element. This is important to us because the study of geography almost
always deals with stochastic processes (OSullivan, 2014).
The main objectives of this project were to simulate three different types of random
point pattern processes (Independent Random Process, Inhomogeneous Poisson
Process, and Processes with Interaction Between Events) and to observe the pattern
of these events in the context of spatial processes.
Independent Random Process
The independent random process (IRP), or complete spatial randomness (CSR),
requires that no
first-order (trends across the study area) or second-order effects (interaction
effects) are present when generating event data points. This means that a point
event has equal probability of existing anywhere and that point has no effect by the
location of other event points (ibid).
In the first exercise, we used the R statistical software package to generate and
plot point patterns according to the independent random process.
As shown in Figure 1 below, there are many cases in which the IRP will seem to
distribute point events in a truly random manner, spreading all points out across the
entire grid and not generating any significant clusters of events. In some instances,
however, there may be some areas of the grid that are emptier than others or in
which points seem to ring around an empty space (such as the upper left corner of
grid p1 in Figure 1 below). This is why it is important to remember that random
refers to the process rather than a pattern created by a process. If this process is
given enough iterations, there will be occurrences when the points seem to leave
large areas empty or cluster points closely together. This does not mean that it is
no longer random, but rather that it is entirely possible for a seemingly non-random
pattern to be produced from a random process.

This reminds me of the multiverse theory in astrophysics, in which many experts


hypothesize that a great many universes must exist in order for there to be one that
fits a pattern such as ours. In fact, many of these experts suggest that, given
enough parallel universes (which we could think of as iterations), there may be
others that exist in a pattern similar to our own (National Public Radio, 2011).

Figure 1: Six different grids of plotted points generated by the independent


random process utilizing the R software package.

Inhomogeneous Poisson Process


The Inhomogeneous Poisson Process is another type of spatial point process but
differs from the previously discussed IRP due to a first-order effect (an underlying
trend across the study area).
Due to the first-order effect that occurs with the Inhomogeneous Poisson Process,
events do NOT have equal probability to be located anywhere. That underlying
trend across the study area causes some locations to be more or less likely to
experience a point event. As shown in Figure 2 below, the six different grids appear
to have more in common than the six IRP grids in Figure 1. Although the point
events shown in all the Inhomogeneous Poisson Process grids below do seem to still
have some randomness, it can also be seen that the points tend to occur more in
the northeast corner of each grid with larger areas of empty space in the
southwestern corner. This pattern can be weighted even more heavily to display
more extreme occurrences of clustering.
Figure 2: Six different grids of plotted points generated by an Inhomogeneous
Poisson Process utilizing the R software package.

Processes with Interaction Between Events


Unlike the IRP, which was free of both first- and second-order effects, and different
from an Inhomogeneous Poisson Process which introduces a first-order effect,
processes with Interaction Between Events are spatial processes that experience a
second-order effect in the data (interaction effects between elements in the
pattern).
I began with a Simple Sequential Inhibition Process in which the inhibition distance
parameter was adjusted to 0.05, 0.01, and 0.07 respectively as shown in Figure 3
below. Just for fun, I attempted to set the parameter at 0.09 and was given a
warning message by the software informing me that it Gave up after 1000
attempts with only 80 points placed out of 100. This makes a lot of sense, of
course, as the method involves discarding an event if it occurs closer to an existing
event than the inhibition distance. A parameter of 0.09 (or larger) in such a small
grid will heavily constrain the number of point events that can be placed, causing
the poor software to throw its virtual hands up in disgust at my ignorance. ;-)

A quick visual inspection of these grids indicate that reducing the inhibition
parameter results in grids that are similar to those produced by the Independent
Random Process. Therefore it may be difficult to spot interactions between events
depending on parameter constraints.
I then performed the Thomas clustering process and experimented with altering
each of the variables. When increasing the second variable, the size of clusters, the
points logically spread farther apart and the grid pattern appears similar to some of
Figure
3: Three
different
plotted
points4generated
the
previous
processes
wegrids
haveof
run
(see Figure
below). by a Simple
Sequential Inhibition Process utilizing the R software package with inhibition
parameters of 0.05, 0.01, and 0.07 respectively.

Figure 4: A Thomas clustering


process with variables set at (10,
1, 10) demonstrated that
increasing the size of clusters as
the standard deviation of a
normally distributed distance can
result in a grid similar to those
produced by other processes.

Decreasing the first variable, the intensity of the process that produces cluster
centers, resulted insubstantially narrowing the range of locations in which point
events occur (see Figure 5 below).

Figure 5: Grid produced with a


Thomas clustering process with
parameters of (5, 0.01, 10).

Decreasing the third variable, the expected number of events in each cluster,
determines the dependence of a new event point based on locations of previous
points. This resulted in obvious clustering, though not yet extreme (see Figure 6
below).

Figure 6: A Tomas cluster


produced using variables (10,
0.1, 5)

Conclusion
This project was a good exercise in the nature of stochastic processes, and how
difficult it can be to determine accurately whether a stochastic process is indeed
completely random or if it contains first- or second-order effects. We verified that
random does not necessarily mean that no clusters or gaps will occur. We also
demonstrated first-order effects through use of the Inhomogeneous Poisson Process,
and second-order effects through use of Interactions Between Events. When
performing a geospatial analysis, knowledge of these effects is important if one is to
truly understand the geographic data being analyzed. It is difficult to accurately
analyze the results without knowledge of the underlying process. It would be easy
to misinterpret many of these resulting grids as having some sort of non-random
pattern, when in reality, all three processes demonstrated were essentially
random. This could be a fine line between detecting a disease outbreak and
crying wolf.

References
National Public Radio [NPR]. ( January 24, 2011) Interview with Brian Green: A
Physicist Explains Why Parallel Universes May Exist. Retrieved October 2013 at
http://www.npr.org/2011/01/24/132932268/a-physicist-explains-why-paralleluniverses-may-exist
O'Sullivan, David. (2014). GEOG 586: Geographic Information Analysis, Lesson 3:
Classical Spatial Analysis. The Pennsylvania State University World Campus.
Accessed October 2013 at https://www.e-education.psu.edu/geog586/l4.html

O'Sullivan, D., & Unwin, D. J. (2010). Geographic Information Analysis. (2nd ed.).
Hoboken, New Jersey: John Wiley & Sons, Inc.

You might also like