Objective

 Using Markov Automata for Land use Change Modelling.

Requirement
 IDRISI TAIGA Software
 Land use data for Westboro from two different time periods, 1971 and 1985

Introduction
 Markov Chain Analysis is a convenient tool for modelling land use change when
changes and processes in the landscape are difficult to describe. A Markovian process is
simply one in which the future state of a system can be modelled purely on the basis of
the immediately preceding state. Markov Chain Analysis will describe land use change
from one period to another and use this as the basis to project future changes. This is
accomplished by developing a transition probability matrix of land use change from time
one to time two, which will be the basis for projecting to a later time period.
 One inherent problem with the Markov Analysis is that it provides no sense of geography.
The transition probabilities may be accurate on a per category basis, but there is no
knowledge of the spatial distribution of occurrences within each land use category, i.e.,
there is no spatial component in the modelling outcome.
 Cellular Automata (CA) will be used to add spatial character to the model.
CA_MARKOV module is an extension of the MCE procedures discussed in the earlier
exercises on Decision Making and combines the CA and Markov Chain land cover
prediction procedures.
 Using the outputs from the Markov Chain Analysis, specifically the Transitions Area file,
CA_MARKOV will apply a contiguity filter to ‘grow out’ land use from time two to a
later time period.
 In essence, the CA filter will develop a spatially-explicit weighting factor which will be
applied to each of the suitability, weighing more heavily areas that proximate to existing
land uses. This will ensure that land use change occurs proximate to existing like land use
classes, and not wholly random. (The actual procedure will apply the contiguity filter to
the masked land use category then multiply this result to the original suitability map to
derive a new suitability map for input into MOLA. If more than one iteration is specified,
 i.e., n iterations, each MOLA run will allocate 1/n of the desired areal goal to the solution
and add 1/n to each successive run. At the end of each MOLA run, each land use is
masked and the contiguity filter run in each, then multiplied to each original suitability
map for another new suitability map for yet another MOLA run.)

Markov Chain Analysis

 Although it is difficult to see the changes between the time periods at first glance, the
biggest areas of change are occurring along transportation networks, especially the major
transportation route running east-west, Route 9. Westboro is situated very near the rapidly
 expanding technology corridors west of Boston. Continual expansion of this industry has
resulted in rapid expansion to outlying areas such as Westboro. To get a better sense of
the change that is taking place, we will do a cross-tabulation.
 Run CROSSTAB with LAND USE71 as the first image and LAND USE85 as the second
image. Specify to output both the image and the table. Call the output image
CROSS7185.

 Using the legend as a guide in the resulting image, you can see the changes of any
particular land use to any other land use.
 By clicking any of the legend categories, you can toggle to a Boolean display of that
category.
 Move the cursor to the legend. Find the legend category 9|3. Click on and hold down the
legend box for that category.

 What you are seeing are areas that were once grassland in 1971 have become industrial\
commercial land in 1985.
 Below is the first output Crosstab table
 Run the module MARKOV and specify LAND USE71 as the earlier land cover image,
LAND USE85 as the later land cover image. Give the prefix 7185 for the conditional
probability images. Specify 14 for both time periods between the land cover images and
the time periods to project forward. Assign 0.0 to the background cells. Assign a
Proportional Error of 0.15 (it is typical that most land use maps are 85% accurate). Hit
OK.

 The transition probabilities matrix (stored with a name derived from a combination of the
prefix and the phrase “transition_probabilities.txt”) records the probability that each land
cover category will change to every other category. This matrix is the result of cross-
tabulation of the two images adjusted by the proportional error. The transition areas
matrix (stored with a name derived from a combination of the prefix and the phrase
“transition_areas.txt”) records the number of pixels that are expected to change from each
land cover type to each other
land cover type over the next
time period. This matrix is
produced by multiplication of
each column in the transition
probability matrix by the
number of cells of
corresponding land use in the
later image.
 In both of these files, the rows represent the older land cover categories and the columns
represent the newer categories. MARKOV also outputs a set of conditional probability
images. Taken from the transition probability matrix, the images report the probability
that each land cover type would be found at each location, in the next future phase, as a
projection from the later of the two land cover images. These can be used as direct input
for specification of the prior probabilities in Maximum Likelihood Classification of
remotely sensed imagery (such as with the MAXLIKE module). But for our purposes, we
will use these files to predict the land use for the specified period, 1999.
 Each conditional probability image shows the likelihood of transitioning to another
category. Although there are alternatives for aggregating these images to predict land use
in 1999, in IDRISI we will use STCHOICE. Using the conditional probability images as
input, STCHOICE will create a stochastic land cover map by evaluating the conditional
probabilitiesthat each land cover can exist at each pixel against a uniform random
distribution of probabilities.
 Run the module STCHOICE. Specify 7185 as the group file. Enter ST1999 as the output
image .STCHOICE generates a random value between 0.0 and 1.0 for each pixel from a
uniform distribution.
 The result from STCHOICE ideally illustrates the problem of the strictly stochastic model
of Markov. The salt and pepper result shows that, although the transition probabilities are
accurate on a per category basis, there is no knowledge of the spatial distribution of the
occurrences within each category. Thus, the stochastic Markov model alone lacks
knowledge of spatial dependency. In the following exercise we will explore the use of the
module CA_MARKOV to give a more spatially dependent result.

CA_MARKOV
 By definition, a cellular automaton is an agent or object that has the ability to change its
state based upon the application of a rule that relates the new state to its previous state
and those of its neighbours.
 We will use a CA filter to develop a spatially explicit contiguity-weighting factor to
change the state of cells based on its neighbours, thus giving geography more importance
in the solution. The filter we will use is a 5 by 5 contiguity filter:
00100
01110
11111
01110
00100
 The contiguity filter will be applied to a series of suitability maps already identified for
each land cover class.
 Display the suitability maps: HDRESSUIT, LDRESSUIT, INDCMSUIT, ROADSUIT,
WATER85, CROPSUIT, FORESTSUIT, WETSUIT, and GRASSSUIT.
 Each map was empirically derived according to such criteria as proximity to roads, water
bodies, protected lands, or existing land cover. The major difference in developing the
suit abilities for this exercise as opposed to the earlier Decision Making exercises is that
the factors used were not developed in association with constituent groups, but developed
empirically in relation to the underlying land use change dynamics between the years
1971 and 1985. This will be explained further as we go along. The production of these
 images, although empirically derived, follows the same procedures outlined in the
Decision Making exercises on MCE.
 Thus, CA_MARKOV combines both the concept of a CA filter and Markov change
procedure. After running MARKOV,CA_MARKOV will use the transition areas table
and the conditional probability images to predict land cover change overthe period
specified in Markov chain analysis. In our case, over a 14 year period to 1999.
 Run CA_MARKOV. Specify the basis land cover image, LAND USE85,
7185TRANSITIONS_AREAS file, TRANSSUIT as the transition suitabilities image
group, and the output image of LAND USE99. Specify 14 as the number of CA iterations
and hit OK. This module will take time to run.

 With each pass each land cover suitability image is re-weighted as a result of the
contiguity filter on each existing land use. Once re-weighted, the revised suitability maps
are then run through MOLA to allocate 1/14 of the required land in the first run, and 2/14
the second run, and so on, until the full allocation of land for each land cover class is
obtained. Recall, that the transition areas file will determine how much land is allocated
to each land cover class over the 14-year period.
 Display LAND USE99 and ST1999 side by side.Notice that LAND USE99 is a much
better result geographically.
 Using the contiguity filter, those areas likely to change will do so proximate to existing
land cover classes.