Modules for Land Use Change Evaluation/ LULC Prediction

GIS & RS Solution YouTube Channel

YouTube Link: https://www.youtube.com/channel/UCs1fLqqz84QbQ89DnRmMEVQ/videos?view_as=subscriber
Facebook PAGE Link: https://www.facebook.com/GISRSITSOLUTION
Facebook Group Link: https://www.facebook.com/groups/292078775381372

Welcome everyone. Today I’ll describe how to prepare Land use and land cover prediction map using
MOLUSCE Plugin of QGIS (Modules for Land Use Change Evaluation). Before doing that, I have already
prepared Land Use/land cover map for the year of 2010, 2014 and 2020 using ArcGIS. You can do it using
another GIS/RS software like QGIS. In this tutorial I will do prediction map for the year of 2030.
For doing this, along with LULC map, you need to create some variable map like DEM, Distance from
Road, Distance from River, Slope, distance from settlement and so on that can be influence to change the
future land use and land cover. Here I am using Distance from road and DEM as variables.
Remember that, All Maps need to be prepared and saved in same spatial resolution and geographical
references.
You can use Euclidean distance from spatial analyst tools to prepare Distance from Road Map.
After completing all essential map in raster format, Just open QGIS software. Add all maps in QGIS. (Watch
my video and follow the below instruction to Prepare the LULC Prediction map)

Inputs
Evaluating correlation
The evaluating correlation module contains three techniques for performing correlation analysis:
❖ Pearson's correlation
❖ Cramer's coefficient
❖ Joint information uncertainty
You can choose between a two-way raster comparison by selecting first raster and second raster or check
all rasters loaded into MOLUSCE. The user can run correlation by pressing the check button located at the
bottom of the window.
Note: The Cramer's coefficient and joint information uncertainty work only with categorical data.

I prefer to use Pearson's correlation.

Area Changes

The update tables button produces class statistics and transition matrix tables.

The class statistics table shows the initial and final land use/cover change (LUC) areas.

The transition matrix shows the proportions of pixels changing from one land use/cover to another.
The create change map button will generate a map of change classes. This will be added automatically to
QGIS and saved as a GeoTiff.

Note: Data from tables can be copied and pasted directly into spreadsheet programs, simply by selecting
the desired rows/columns and by pressing the “Ctrl + C” keyboard combination.
Transition potential modelling
MOLUSCE uses Artificial Neural Network (ANN), Multi Criteria Evaluation (MCE), Weights of Evidence
(WoE) and Logistic Regression (LR) methods to model land use/cover transition potential. You can select
a method from the drop-down menu. I will recommend you to use ANN methods in this regard.
Artificial Neural Network (ANN)
The define samples function specify number of samples and sampling mode. In addition, the sampling
points created can be saved and displayed.
Five inputs are used to customize the ANN modelling:

• The neighborhood defines count of neighbor pixels around current pixel. Size=1 means 9 pixels
(3x3 region), size=2 means 25 pixels (5x5), etc.
• Learning rate, momentum and max iterations number define parameters of learning. Big learning
rate and momentum allow fast learning, but the learning process can be unstable (spikes on the
graph). Small learning rate and momentum means stable but slow learning.
• Hidden layers input string takes a list of numbers: N1 N2 … Nk, where N1 is number of neurons
in 1 st hidden layer, N2 is number of neurons in 2nd hidden layer and so on, Nk is the number of
neurons of the last hidden layer (kth layer). For example, if the user types in the input string “2”
then a network with 1 hidden layer and 2 neurons will be created. In order to create a network with
2 hidden layers the user should insert 2 numbers, such as “10 2” which will create a network with
10 neurons in the first hidden layer and 2 neurons in the second.
The following outputs are proposed (for the current learning iteration):
❖ The min validation overall error contains information about min reached error on validation set
of samples.
 ❖ The delta overall accuracy contains difference between min reached error and current error.
❖ The current validation kappa shows the kappa value.
❖ The graph area. Contains errors of training and validation sets. It is the main information about
learning process. The graph can be edited and saved as image.
The process can be started by pressing on the train neural network button and stopped at any time using the
stop button. Learning algorithm analyses the reached accuracy on training and validation sets of samples
and stores the best neural net in memory. The training process finishes when the best accuracy is reached.

Cellular Automata Simulation

Once one method has been chosen from the transition potential modelling tab, the user can then access to
the cellular automata simulation tab. Be aware that MOLUSCE will keep in memory the latest method
processed, if for example the user runs first the ANN and then the LR methods, the cellular automata
simulation tab will retain the results from the LR.
Three type of output maps are produced. A check box at the beginning of each output is provided to allow
the user to enable only what it is necessary. A browse... button located at the end of each output allows to
save each map.
❖ The prefix of transition potential maps button allows to select prefix of the names of transition
potential maps. Transition potential map shows the probability or potential to change from one land
use/cover class to another. Transition potential values range from 0 (low transition potential of
change) 100 (high transition potential). Transition potential maps will be produced from the
corresponding land use/cover changes (e.g., “forest to unstocked forest” transition potential, “forest
to non-forest transition potential).
❖ The certainty function (As mentioned earlier, this is an experimental function). (In the video
tutorial, I did not use this function)
 ❖ The simulation result produces a simulated land use/cover map. (This function is very much
important)

Validation
The validation tab allows the user to check, validate and compare the simulation results. Reference and
simulated land use/cover maps must be loaded in order to start the validation process. The former
indicates a land use/cover map (T3). T1 refers to the initial land use/cover, while T2 refers to the final land
use/cover used in the model). A browse. button located at the end of each output allows to load the desired
map. A two-way map comparison is performed from reference land use/cover (T3) and simulated land
use/cover maps.

In order to perform a three-way map comparison, the user can check the risk class validation map check-
box. Although it is not shown explicitly, the three-way map comparison uses the initial land use/cover map
(T1), the reference land use/cover map (T3) and the simulated land use/cover map.

The multiple-resolution budget method is also used in MOLUSCE. The user can decide the number of
validation iterations (Generally 5 iterations is default) and can start the process by pressing the start
validation button. The graph can be edited and saved as image.

the overall accuracy (% of correctness), kappa (overall), kappa (histo) and kappa (loc) can be executed by
clicking the calculate kappa button.
Watch my video along with the instruction and then try to make your LULC prediction map. If you don’t
know how to prepare LULC map, check my videos. Also, if you don’t have country shapefile, road shapefile
etc. just check my videos how to download shapefiles for any country. I think Any kind of GIS and Remote
Sensing solution you will find in my channel.
This manual has prepared with the help of QGIS MOLUSCE Plugin module. If you Like my videos, you
can consider SUBSCRIBING this channel to get more GIS and Remote Sensing tutorials. Best wishes for
you.

Thank you.