Transcript of "Land Use Prediction Using Land Transformation Model (LTM)"
Maryam Adel SAHARKHIZ
October 31st 2006
LAND TRANSFORMATION MODEL (LTM)
FOR SEMENIYH BASIN
HUMANS ARE CHANGING THE LANDSCAPE AT AN
WHAT CAN WE EXPECT OUR FUTURE
LANDSCAPES TO LOOK LIKE?
TOPIC OF TUTORIAL
Run the Land Transformation Model starting from
land use maps and different drivers in GIS form.
a model run for Semeniyh Basin in Selangor
Predict future LandUse layout based on past land
use data. (2006 and 2010)
Basin land use
in 2006 (left)
and 2010 (right)
We will model LandUse expansion in Semeniyh Basin
using 2 land Use maps, one from 2006 and the other
After going through the Model we will be able to run
the LTM on our study area, to forecast future land use
changes in 2014.
CREATE DRIVERS = PREDICTOR VARIABLES
Driver layers represent phenomena that influence what
are trying to model.
In this study, we assume that the following 6 drivers will
influence urbanization an agriculture expansion in
Proximity to urban in 2006, to highways, to roads, to
rivers, to Lake of Semeniyh and to inland lakes.
Drivers was created using Euclidean Distance of
ArcGIS. It calculates, for each cell, the Euclidean
distance to the closest source.
FORMAT LAND USE LAYERS
After Diver’s creation two land use layers were
reclassified to zeros and ones, ones being the class
wanted to model. In this case we are modeling
urbanization and agriculture expansion so we reclass
all urban and agricultures pixels to 1 rest to zero.
2006 (left) and
PREPARE EXCLUSIONARY LAYER
Exclusionary cells are cells which we don’t want to
include in the analysis, i.e. cells which the LTM will
In our dataset we excluded water pixels, Agricultures
and urban in 2006 as we did not want urban and
Agriculture to expand to those locations
exclusionary cells Reclassed as 4, rest of the data as 0.
All data layers need to be exported to ascii files which
will be readable by the Neural Network.
STEP 1) CREATE INPUTFILE.TXT
At first step we tells the NN which files it needs to get
information from for the predictor variables
STEP 2) CREATE NETWORK FILE
Gives the structure of the NN by following syntax:
Createnet 6 6 1 ltm.net
STEP 3) CREATE PATTERN FILES
Keeps track of which cell has what values in the
various base and driver layers as well as the output
Createpattern.6.5 inputfile.txt v
STEP 4) BATCHMAN _ TRAINING
Different cycles are as Outputs, and learns from the
patterns in the data.
It run by bellow comment
Batchman –q –f train.bat > traincycles.csv
The rms for each of these cycles is recorded in the
CREATE REAL CHANGE MAP
After running step 4 the number of new urban cells between
2006& 2010 was calculated and saved in Real Change raster
Record # of 1s
STEP 5) TESTING FIRST STEP:
First RERUN createpattern Syntax this time with
createpattern.6.5 inputfile-test.txt v
to 1 in your
and save it as
STEP 5) TESTING SECOND STEP:
BATCHMAN _ TESTING
Another step in order to Testing process
Based on batchman –f batch-test.bat at the
res_10000.asc and ts_10000.asc are results of
CALCULATE PERCENT CORRECT METRIC
To estimate Spatial Accuracy, file0123 layer was created from
ts10000 and RealChange layers as follow. The numbers 0,1,2,3
represent the following:
0 = no real change and no predicted change
= True Negative
1 = no real change but change predicted by the model
= False N
2 = real change but not predicted by the model
= False Positive
3 = real change and predicted change
= True Positive
CALCULATE PERCENT CORRECT
The Percent Correct Metric (PCM) is just the number of 3’s divided by the
number of cells that transition (here 207551)
PCM = (144933/ 207551) * 100 = 69.83% spatial accuracy
Kappa = 0.658229
Sixty to 80% accuracy is
considered an exceptional model.
40% to 60% is acceptable.
LTM_stats.txt is including of PCM
for all training files.
STEP 6: FORECASTING
After Testing step, using inputfile-forecast.txt as well as following
comments forecast layer has been created
Syntax: Createpattern.6.5 inputfile-forecast.txt
Then: asciits2.3 fullreference.txt res_10000 landusefinal.asc
ts_10000F.asc 1 12072
Result of forecasting saved in ts_10000F file into