Geosos-Flus Manual en
Geosos-Flus Manual en
Geosos-Flus Manual en
Natural Effects
Authors:
Xiaoping Liu
Xia Li
Xun Liang
Contact info:
http://www.geosimulation.cn/GeoSOS/
Funding/Disclaimer:
Citation:
Abstract
Our newly proposed FLUS model is an integrated model for multi-type land use
scenario simulation by coupling human and natural effects. At the same time, the spatial
simulation module of FLUS model is made into GUI software named GeoSOS-FLUS.
The GeoSOS-FLUS is developed as an extension of the previous GeoSOS software, to
facilitate the multiple land use change simulation. The software provides a multiple CA
allocation model for simulating land use change and scenario analysis.
The multiple CA allocation model is based on the analysis of the pattern of latest
period land use instead of the change in two terms land use, this improvement not
only avoid the effect of error accumulation from the disagreement of two terms
land use data, but also make the model have much wider range of applications.
An artificial neural network (ANN) is used for taking both human activities and
natural ecological effects into consideration by finding the complex relationships
between land use pattern and various human and natural driving forces.
Besides, In order to ensure the speed of model, GeoSOS-FLUS was developed purely
in the C++ language. The ANN technique in our model is from a powerful open source
library called Shark 3.1.0 (http://image.diku.dk/shark/). The UI of the software is built
using a famous open source library Qt 4.8.5 (https://www.qt.io/download/). This UI
provides a real-time display of dynamic changes of land use in simulation process.
Moreover, the using of open source library GDAL 1.9.2 (http://www.gdal.org/) allows
our model to directly read and write raster data (.tif, .img, .txt files) that includes
geographical coordinate information.
Therefore, the GeoSOS-FLUS is a powerful tool for making land use change simulation
more convenience and efficient, which can be easily used for several purposes including
1) the establishment of urban construction boundaries (UCBs); 2) the high-resolution
simulation of the change of internal land use change of the city; 3) environmental
management and urban planning; 4) large-scale land use change and its effect on
climate; 5) Regional land suitability analysis; 6) early warning for the lost of natural
and agricultural land cover types; 7) hotspot recognition for land-use change and etc.
When simulate future land use pattern, users need other external models, such as SD
model or Markov chain, to project future land use demands as inputs of GeoSOS-FLUS.
This users manual is intended to provide users of GeoSOS-FLUS a quick start on
how to use the software. All of the necessary data and files for the tutorial have been
provided and these can be used as templates for how to format your own files latter on.
The main interface is a brief image browser which provide some basic function for
image browse, including showing image in in grey scale, zoom in and zoom out,
viewing image's properties.
Once the compressed file GeoSOS-FLUS VX.0.rar has been inflated, the test
dataset are found in the test data folder. Use this dataset to run example model or
compared with your simulation results.
Click the button in the Land Use Data group box and select the start year
land use data dg2001coor.tif in the pop-up dialog
ThenClick the Set NoData Value buttonand the software will open a window
for defining the no data value of land use data. Under the second column, click on the
drop-down menu to change it to Valid Data or NoData Value. Users can only select
one no data value at a simulation. Click Accept to accept the settings.
3.1.3. Setting the training parameters and output path
Users are allowed to set sampling strategy in the ANN Training group box. The
Uniform Sampling option means the same numbers of sampling points for each land
use type. The Random Sampling option means that the sampling points will
randomly scatter across the study area. Then setting the sampling rate and the number
of hidden layers of ANN in the following two spin boxes, respectively. The unit of
sampling number is thousandths of the total number of valid pixels. When Sampling
rate is set to 20, it means about two percent of valid pixels is selected for training. The
number of hidden layer is set to 12 according to the expertise.
Then, click the button in the Save path group box and input the save
path of probability-of- occurrence image in the pop-up dialog. The generated
probability-of- occurrence file is named Probability-of-occurrence.tif.
TheSave path group box gives users the options to select the output data type. If
users choose the Single Accuracy option (the default), the module would finally
generate the FLOAT type probability-of- occurrence from TIF file. Similarly, another
Double Accuracy option will generate a DOUBLE type probability-of- occurrence
TIF file. We usually adopt theSingle Accuracy option, since it is able to achieve the
ordinary precision requirement of simulation and saves lots of memory spaces, although
the Double Accuracy option can generate higher accuracy data.
Click the button in the Driving Data group box and select multiple driving
factors in the pop-up dialog. Our example data provide eight driving factors for example
run of FLUS model.
Attribute information of driving factors are displayed in a list box, including file name,
data type, maximum value and maximum value, etc. Click button for
adding one new driving factor to list box, or click button for removing one
driving factor from list box.
Besidesif users choose the Normalization option (the default)all the driving factors
will be normalized to a range of 0-1.0. If you driving factors have previously been
normalized, you can choose No Normalization option to skip this step.
Once the setup mentioned above has completed, click the button
for running the module. There will come out a message box as below when the model
process run is complete.
The text box at the right of the interface will output the training velocity and accuracy
of the network after a successful run of ANN-based probability-of- occurrence
estimation module. Three indicators are the RMSE, average error and average relative
error, for measuring the training accuracy of the ANN, respectively.
Users are able to use the image browser provided by GeoSOS-FLUS to examine the
output result of ANN-based probability-of- occurrence estimation module. You can see
a multi-band image, which each band corresponds to the probability-of- occurrence of
one land use type.
Click the Self-adaptive inertia and competition mechanism CA item to start the
self-adaptive inertia and competition mechanism CA for simulation.
3.2.2. Input land use data
Click the button in the Land Use Data group box and select the start year
land use data dg2001coor.tif in the pop-up dialog
ThenClick the Set Land Use Type, Color Display and NoData Value button
and the software will open a window for defining the no data value of land use data and
setting display Color . The first column of window list shows the land use code of each
land use type. Under the second column NoData Option, users need to click on
the drop-down menu of one of the rows to change it to Valid Data or NoData Value
to define it as NoData Value, only one no data value is allowed. The third column
Land Use Type is for users to name the tag of each land use type. The GeoSOS-
FLUS software will default the names of various land use types as Landuse1,
Landuse2, etc.
ClickSet Colorbutton of each row in the list box to set an RGB value (such as
rgb(150,200,50)) for all land use types.
Input the save path of simulation result in the Save Simulation Result group box.
Some spatial policies restrict all land use change in designated areas, such as open water
and some strictly protected nature reserves area. When spatial policies need to be
considered into simulation, users should prepare a binary restricted area image that all
the pixel are either 0 or 1. The value 0 means that grid cells with one land use type can
never be converted to another type; 1 means that land use change is allowed in this
region.
3.2.6. Setting the simulation parameters
Simulation parameters mainly include the land use demands, number of iterations,
neighborhood influence, neighborhood weights and acceleration factor. The land use
demands are parameters of our model, which need to be firstly set according to the
actual situation of study region. So users need to firstly use external model to determine
the land use demands of future scenario. For our example run, the time span of the land
use change simulation is from 2001 to 2006, so the land use demands are set to be the
same as the actual area of various land use types in 2006. The number of iterations is
set to a large value (maximum number of iteration, default is 300), the model will stop
when the allocated area equal the demanded area for all land use types or reach the
maximum number of iteration. The neighborhood effect of GeoSOS-FLUS is similar
to traditional CA, the default value is 3. The acceleration factor is a parameter for
improving the speed of reaching land use demands, ranging from 0 and 1.0. When
simulating land use change for large area, users may need to set a larger value of
acceleration factor for shortening the time consumption in simulation process.
In the tab widget, the second row Future Pixel Number in the Land Use Demand
page is for inputting the future land use demand of each land use type. The initial pixel
number will be automatically loaded after land use pattern data is input.
Then, switch the tab widget to the Cost Matrix page. The columns of this matrix
indicate the current land use types, and the rows indicate the future land use types. A
value of 1 means the conversion is allowed while a value of 0 indicates that the
conversion is not possible. A Cost Matrix is a collection of a series of prior knowledge
of experimenter about study area. An example of cost matrix is provided below:
And in the Weight of Neighborhood page, users need to set the neighborhood weight
of different land use types, because the neighborhood effects may be different for
different land use types in a study region. The value of the neighborhood weight for
each land use type is determined according to expert knowledge and a series of model
tests, ranging from 0-1. At this example, we set the parameter as below:
Finally, click Accept for confirming what you have set up before.
Switch the tab widget to the Show page, click button to start the
simulation. The GeoSOS-FLUS is able to show the process of land use change
dynamically, including the spatial change, change in value and change curve of each
land use type.
The model will stop and save the simulation result when the allocated area equal the
demanded area for all land use types or reach the maximum number of iteration. Besides,
user can manually stop the simulation process and save the result by clicking the
button.
3.3. Precision validation of simulation
GeoSOS-FLUS provides a kappa statistic tool and a FoM statistic tool for measuring
agreement between simulation result and actual land use pattern.
Click the Validation Precision ValidationKappa item on the main menu to start
the kappa statistic tool.
3.3.1.1. Load data
Click the button in the Ground Truth rowto input the actual land use
GeoSOS-FLUS supports two strategies for sampling: Random Sampling and Uniform
Sampling. The Uniform Sampling option means the same numbers of sampling
points for each land use type. The Random Sampling option means that the sampling
points will randomly scatter across the study area. At this example run, we choose
Random Sampling strategy and set the sampling rate to 10%.
Click the button to start calculating Kappa coefficient. A message box will
pop up when the program completes the calculation.
The results include Kappa coefficient, overall accuracy and confusion matrix, which
will be saved in a file named Kappa.csv.
Click the Validation Precision ValidationFoM item on the main menu to start
the kappa statistic tool.
Click the button in the Start Map rowto input the start land use pattern
Click the button to start calculating FoM coefficient. A message box will pop
up when the program completes the calculation.
The results include FoM statistics, producers accuracy, which will be saved in a file
named FoM.csv.
4. Some notes in using GeoSOS-FLUS
1) The valid land use codes of land use data must be a tolerance of 1 arithmetic
progression beginning with 1(e.g. 1,2,3,4,5,). NoData Value can be any other
value except these valid land use codes.
2) All the input image data need to have the same number of rows and columns,
including the land use pattern, restricted area, driving factors, and probability-of-
occurrence data.
3) The Self-adaptive inertia and competition mechanism CA module will save all the
parameters to two configuration files: config_color.log, config_mp.log. The
module will automatically loads the parameters that are recorded in the configuration
files including the information about show colors and names of land use types, which
is convenient for the user to use to repeat their experiments. But users must firstly
delete these two configuration files when changing to a new study region, for
resetting simulation parameters.
5. Generated files
After simulation, GeoSOS-FLUS creates two kinds of files: configuration files and
recorded files.
config_color.log: Record the RGB value, name of land use types, number of pixels of
each land sue types in base year in last simulation.
config_mp.log: Record the land use demands, cost matrix, maximum number of
iteration, acceleration factor, neighborhood weights of last simulation.
output.logSave the number of pixels of each land use type in every iteration.