| Associated entities to this dataset: |
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
| Attribute Name -- Attribute Label |
Unit |
Attribute definition |
Land_use
--
land use
|
no unit |
Land use |
|
Method
(Sampling/processing description of the values of
this attribute by one ore more steps.)
|
|
--- Step
1
---
|
|
Title
:
|
Modelling the amount of change |
|
Description
:
|
The future amount of change was predicted using a discrete-time second order Markov chain (MCA) which is implemented in IDRISI Andes (Eastman 2006). MCA is based on LUCC observed by the two past LUC classifications 1987 and 2001 (Meyer et al. 2011). The output of this procedure was the expected amount of pixels to change from one LUC to another. This step was accomplished once with the 1987 and 2001 classification to gain the expected amount of deforestated pixels for the next ten years. The results were fixed and applied for all further recalculation steps. In contrast to this, the amount of all other LUC changes was expected to depend on the current state of the LUC. Therefore, MCA was applied in each dynamic recalculation step using the respective previous period as input information, in order to ensure an adaptation to the current state of LUC.
Dataset citation:
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 1987. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1038) from DFG-FOR816dw.
Meyer H. , Thies B. , and Bendix J. (2011): Land use/land cover map of South Ecuador 2001. Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=1039) from DFG-FOR816dw. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
2
---
|
|
Title
:
|
Creating input variables |
|
Description
:
|
The initial variable ?distance from forest edges? was produced from a boolean map of the location of forest which is based on the 1987 classification. The forest edges were then identified by calculating ?center versus neighbors? using the PATTERN module in IDRISI (Eastman 2006) with an area size of 3x3 cells. The Euclidean distance from Forest edges, roads, rivers and villages was calculated for each cell. Aspect and slope were calculated from SRTM 90m elevation data. In order to prepare the variables for the modelling procedure, the variables were finally normalized by stretching the values into byte values from 0 to 255. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
3
---
|
|
Title
:
|
multi-layer perceptron training |
|
Description
:
|
The multi-layer perceptron (MLP) is implemented in IDRISI Andes (Eastman 2006). MLP learning was accomplished for each transition being modelled. The basic design of the network followed an 8-4-2 topology, with the 8 explanatory variables presenting the neurons in the input layer and 4 neurons in the hidden layer. The training sites specification image contained two classes: Class one depicted the change of the relevant transition between 1987 and 2001. Class two consisted of those cells which could have undergone the transition but did not. This means that if the transition from forest to grassland was modelled, all forest cells from 1987 which did not change until 2001, belonged to this class of training sites. A dynamic learning rate was chosen with momentum factor 0.5 and a sigmoid constant of 1.0. The training pixels per category varied according to the maximal availability of training pixels for each transition. Each learning procedure contained 10,000 iterations. The relevant output of each MPL learning process was a weight matrix which contained the computed weights for each variable. There were fixed and applied in each recalculation step. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
--- Step
4
---
|
|
Title
:
|
Creating maps of transition potential |
|
Description
:
|
The dynamic variable ?distance from forest edges? was updated to the information of 2001 like explained in step 2. Since the computed weights were fixed in the MLP learning process, dynamic variables could easily be exchanged in the network without the need of a further learning procedure. The results were transition potential maps for each transition, showing the potential for change at the time 2001. Finally, the transition potential maps were overlaid by a mask which contained all classes which did not belong to the previous LUC class of the investigated transition, since therefore they do not have a potential for change. Further, cells belonging to the Podocarpus national park were masked out since according to the scenario, they also should not have potential for change. This step was accomplished every 10 years using the latest LUC image as input for forest edge calculation (Step 2). |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
|
--- Step
5
---
|
|
Title
:
|
Multi-objective land allocation |
|
Description
:
|
Multi objective land allocation (MOLA) allows one not only to allocate land according to its potential but also to solve conflicts if two objectives own the same potential for change. MOLA is implemented in IDRISI Andes (Eastman 2006). The transition potential maps first had to be ranked in descending order according to their potential for change. MOLA then allocated the amount of change for each transition modelled with MCA according to the ranked order of its spatial potential. |
|
Instrumentation:
|
Instrument:
Description:
|
Instrument: Idrisi
|
Vendor: Clark Labs
|
|
|
Citation:
|
Eastman, J
.
(2006):
IDRISI Andes guide to GIS and image processing
. Available online
(http://tropicalmountainforest.org/publications.do?citid=1037) from DFG-FOR816dw.
|
|
|
|
|
|
|
