Dynamic Simulation of landscape transformation of

2 downloads 0 Views 9MB Size Report
Secretariado de Publicaciones de la Universidad de Sevilla. Sevilla 2010. Farina 2011. “Ecología del Paisaje”. Publicaciones Universidad de Alicante, España.
Dynamic Simulation of landscape transformation of the French Riviera to 2050 using cellular automata Authors: Guillermo Hinojos Mendoza; Emmanuel Garbolino & Valerie Sanseverino-Godfrin

MINES ParisTech, CRC- Crisis and Risk research Centre, BP 207, 1 Claude Daunesse St, 06904 Sophia Antipolis, France E-mail address of the authors: [email protected], [email protected], [email protected] Abstract This paper presents a synthesis of landscape transformation in the French Riviera up to 2050. We studied the landscape transformation of the last 40 years with the aim of identifying the spatial transition rules which have been introduced in the cellular automata model. The scenario provides information about future urban expansion according to historic behavior and the evolutionary patterns. The model takes into account accessibility via the transportation network, the natural protected areas, geomorphological restrictions, and the spatial relationships between land use categories. The questions of this work are: can the growth of the artificial areas be considered as driving forces on landscape transformation? And could this transformation be projected according to the historical pattern of territorial transformation? The transformation scenario was evaluated to quantify the potential changes of the spatial characteristics of land use processes. The results of this work show that urban surfaces could double by 2050 compared with 2011. Finally this work demonstrates that it is possible to project landscape transformation by studying historical behavior and spatial pattern. Keywords: Simulation, landscape transformation, cellular automata, SpaCelle, transition rules, French Riviera

Page 1 of 32 !

1. Introduction The landscape transformation has ecological implications that are necessary to evaluate, that is why we consider that the simulation models can provide practical information about this problematic. Cellular Automata have been used to model landscape changes over the past years, (Aguilera Benavente et al., 2010; Barredo et al., 2003; Batty et al., 1999; Chuyang He et al, 2008; Clavero et al, 2010; Feng et al., 2011; Gutiérrez Angonese et al., 2010; Sante et al., 2010; Xie et al., 2010; Wenliang Li et al., 2012; Daniel Z et al., 2001; Lin Li et al., 2003, among many others), probably because these models represent an interesting alternative due to the advantages to simulate complex phenomena like urban processes and their interactions with the different land use classes. In this research, we develop a simulation scenario of landscape transformation with a cellular automata model which incorporates the transition rules that have been interpreted from the spatial analysis of study area. The transition rules consider the constraints caused by the geomorphology conditions, by the natural protected areas as well as the accessibility offered by the transport network, the closeness of the population centers and the pattern of historical change between each land use category. The process of evaluation of the landscape transformation in this work consists in identifying the possible dynamics of urbanization using a cellular automata model for simulation (SpaCelle) according to spatial process on the territory concerned. This evaluation is performed using the historic data from 1975 to 2011 in order to understand the historic pattern or growth, and transition spatial rules in order to project the future scenarios of landscape changes for 2050. The aim of this work is to answer the following questions: Are we able to predict the landscape transformation according to the historical pattern? Is the landscape transformation conditioned by the geomorphological factors? Are the changes of land use conditioned by the urban infrastructure or the communication networks? Finally what is the dispersion level of the land use changes across the concerned territory?

Page 2 of 32 !

2. Materials and methods We first introduce a brief description of the studied area. Then, we will present the methodology developed for this research, which is based on the following three main steps: Step 1: Historic land use changes analysis Step 2: Definition of land use transition rules Step 3: Model calibration

Fig. 1. Methodology. Step 1 is dedicated to understand the landscape transformations in the area since 1975. The results of this step allow defining, in step 2, the spatial transition rules that reflect the territorial behavior of the landscape transformation. Step 3 presents the methodology to calibrate our model, which is based on the spatial comparison between the simulated output model vs. the observed land use model in 2011. 2.1. Presentation of the study area The study area covers the territory of the French Riviera. Urban development in this area has increased rapidly over the last 40 years, and with a striking expansion while agriculture areas have decreased systematically during the same period meanwhile the natural and semi natural surfaces are under constant pressure due to urban sprawl. Since 1970, this pressure has increased especially on the coastal strip that has been continuously urbanized and that is today completely occupied by artificial structures. While the population only increased by one third between 1970 and 2000, urbanization on the coastal strip has increased disproportionally on the same period (DTA1, 2008). 2.2. Step 1: Historic Land Use change analysis In this research we take into account three main factors that explain the land use transformation: 1. The interactions between the land use classes: they explain all the relationships and transitions between land use categories; 2. The role of the road network and infrastructures on territorial dynamics; 3. The effect of geomorphology on the landscape transformation. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

2

Direction of territorial development of the French Riviera. The DTA is an urban and territorial planning policy developed by the state that fixes the objectives of development of infrastructure, transport and sustainable development. !

Page 3 of 32 !

This historical analysis of the land use changes takes into account the following parameters: •







The surface of changes for each category expressed in hectares or cells, and the trajectory of the changes (ex. natural to agriculture or agriculture to urban/artificial). The spatial pattern of the changes for each period, to allow identification of the potential trends and the primary morphologic patterns (Gómez Delgado and Rodríguez Espinoza, 2012), to integrate in the transition rules. The distance gradient of the different territorial elements that have an influence on the land change, like transport networks, coastal strip and established population centers. The potential restrictions that represent the geomorphology of the development of urbanization or artificial surfaces.

2.2.1. Interactions between the land use classes Land cover data were obtained from photo interpretation of the Landsat satellite imagery for the scenes of 1975, 1990 and 2011. This information was first of all processed using IDRISI® software to obtain a segmented thematic for each scene, which was classified by photo interpretation techniques with ArcGis® software in the following classes: agricultural areas, natural and semi natural areas, water and aquatic surfaces, and urban/artificial surfaces.

Urban evolution in the study area Surface(in(hectares((

!60.000!! !50.000!! !40.000!! !30.000!! !20.000!! !10.000!! !"!!!! 1975!

1990!

2000!

2006!

Date(

Fig. 2. Urban evolution in the French Riviera since 1975 to 2011.

Page 4 of 32 !

2011!

The figure 2 shows the spread of artificial surfaces. This figure was created using data about the composition of surfaces of artificial areas in 1975 obtained by the photointerpretation process of Landsat images (figure 3). It was then compared with 1990 and 2011 surfaces obtained by the same procedure. The 1990 and 2011 data were validated with Corine Land Cover and Local Development Plan (PLU) in 2011. The 1975 data cannot be validated because of lack of ancillary data.

1975

1990

Fig. 3: Spread of artificial surface in the French Riviera since 1975. Legend: red = artificial areas; Yellow = agricultural areas; Green = Natural areas. The spread of artificial surfaces across the territory between 1975 and 2011 is shown in red. Natural surfaces are shown in green and agricultural area in yellow color. This map has been made with information provided by photo-interpretation of Landsat images of 1975, 1990 and 2011. The results of this historic evaluation show a trend that is perfectly observable; “the spread of artificial surfaces along the coastal strip takes advantage of the assets provided by the agricultural surfaces. Outside of the coastal strip the trend is similar in such a way that the artificial surfaces were also colonized principally on the agricultural areas”. Est-ce une citation d’un auteur ? Si oui, il faut indiquer la référence bibliographique. For the evaluation of historic land changes, we used the module “Land Change Modeler” of IDRISI, Selva, (Eastman, 2006). This evaluation consisted in identifying the principal change trajectories and the different patterns in the changes between the different spatial components. The aim of this evaluation was to understand the transformation process in the landscape and the potential trends of the future changes. This evaluation was made for two periods: the period of 1975 to 1990 and the period of 1990 to 2011. The LCM allows the analysis of the historical land use changes and the generation of forecasts of future changes. This model also allows the implications of the changes on biodiversity to be analyzed. For Page 5 of 32 !

2011

this work the LCM has only been used for the study of the dynamic of the land use changes and pattern identification. The study of the changes dynamics is based on the crossing of the categories of two temporal scenarios. If the classification of the two scenarios is identical, it is possible to obtain a transformation matrix which provides information of the changes on each land use class (Gómez Delgado and Rodríguez Espinoza, 2012). For each resultant category or class it is possible to identify the magnitude of change as well as the losses and gains in response to territorial transformation process. The data parameters that may be obtained are as follow: • • •

Gains, expressed as the difference in terms of growth of the area of a given category between the initial date of the analysis and the end date. Losses, expressed as the difference in decrease in the size of a given category between the initial date and the end date of the analysis. The net change, expressed as the absolute value of the difference between gains and losses in each category.

In 1975 the agricultural surfaces occupied 40,398 hectares, decreasing to 31,135 hectares by 1990 (table 1 and figure 4). Meanwhile the urban/artificial surfaces more than doubled between 1975 and 1990, going from 12, 484 to 27,887hectares. This change is remarkable for a relatively short period. The natural/semi natural areas occupied 376,548 hectares in 1975 and decreased to 369,811 hectares by 1990. Water and aquatic surfaces occupied 338 hectares in 1975, increasing to 501 hectares in 1990.

1975 - 1990 net change

1975 -1990 changes 277!

aqua1cs(

Aqua1cs( 18314!

ar1ficials( "2812!

"24659! "20182!

agricuture(

15021!

naturals(

14225!

"30000! "20000! "10000!

Structure(

Structure(

"185!

0!

Ar1ficials( "9638!

15502!

Agriculture(

"5957!Naturals(

10000! 20000! 30000!

"15000!"10000! "5000!

Hectares(

Fig. 4. 1975-1990 land cover changes. Page 6 of 32 !

92!

0!

5000! 10000! 15000! 20000!

Hectares(

Table 1. Surface of land cover changes between 1975 and 1990 Structure Agricultural Artificial Natural Aquatic

S1 (1975) 40,398.693 12,484.8391 376,548.753 338.6424

S2 (1990) 31,135.27 27,887.37 369,811.20 501.55

The table 1 and the figure 3 show the non-linear behavior of the changes and consequently a certain level of complexity. Although most of the changes provides the increase of the net artificial surface, these areas exchange surface with other land use class. This analysis shows that the changes are not homogenous throughout the study area and it is possible to interpret that pressure forces are spatially differentiated. In 1990 agricultural areas occupied 31,135 hectares, decreasing to 14,794 hectares by 2011 (table 2 and figure 5); meanwhile the artificial surfaces increased from 27,887 hectares to 46.213 hectares between 1990 and 2011. The natural surfaces decreased from 369,811 hectares to 367,653 hectares in the same period. Aquatic surfaces increased from 501 hectares to 526 hectares in the period 1990 -2011.

1990 - 2011 net changes

"23475!

aqua1cs( "259! ar1ficials( "6012! agricultural(

425(

aqua1cs( 24696( 6937(

17130( naturals( Hectares( "30000! "20000! "10000! 0! 10000! 20000! 30000!

Structure(

Structure( (

1990 -2011 changes

166!

ar1ficials( "16538! agricultural(

18684!

"2312! naturals(

"19442!

"20000!

"10000!

0!

10000!

Hectares(

Fig.5.Landcover Changes1990-2011

Page 7 of 32 !

20000!

30000!

Table 2. Surface changes between 1990 and 2011 Structure Agricultural Artificial Natural Aquatic

S1(1990) S2(2011) 31,135.27 14,794.48 27,887.37 46,213.64 369,811.20 367,653.37 501.55 626.89

In the second period we can see that the behavior corresponds to the historic pattern especially concerning the net change. However, the exchanges between classes are differentiated according to the intensity. In general we can observe that the pattern of changes follows a similar trend in relation to the previous period. According to these first results, we also wanted to understand the trajectories of the changes. In the figure 6 we can observe that the intensity of the changes between the two periods is not the same for all the classes and we can also note the variation of pressure on each land use class.

(1975(@(1990(trajectories( 1%!

1990(@2011(trajectories( Natural!to! aqua7c!

Natural!to!aqua7c!

0%! 1%!

0%!

Natural!to! ar7ficial!

Natural!to!ar7ficial! 34%!

Natural!to! agricultural! Agricultural!to! ar7ficial!

62%! 2%!

Ar7ficial!to! aqua7c! aqua7c!to! agricultural!

Fig. 6. Trajectories of the changes

Page 8 of 32 !

25%!

40%!

Agricultural!to! aqua7c!

34%!

Agricultural!to! ar7ficial! 1%!

Agricultural!to! natural! Natural!to! agricultural!

This analysis shows the complexity of the land use change. When we observe the net changes of each studied period, we may be led to believe that the changes have the same behavior, which is actually not true. If the analysis is only based on the surface changes, we see that the artificial surfaces have the same trend to increase. However, the net change can be similar between two periods but produced by variations of the behavior. A failure to take this into account may lead to use information that is not the most suitable for generating transition rules. The figure 7 answers the following two questions: What is the spatial pattern of the changes? Can we truly say that it has spread? The analysis of the spatial patterns shows an increase in the spread of the changes, which explains why the pure aggregate pattern has been decreasing in the department. This dispersion does not occur at random; it mainly occurs at the proximity of the routes and along communication networks. This is why the nodal patterns and nodal aggregate have increased.

1990(@2011(

1975@1990( 7%!

0%!

6%! 2%!

1%!

23%!

linear!aggregate! 7%! 0%! nodal!aggregte!

26%!

dispersed!

dispersed! 57%!

linear! nodal!

!nodal!aggregate! pure!aggregate!

pure!aggregate!

62%!

linear!aggregate!!

9%!

Figure

linear!! nodal!

Fig.7. Spatial pattern of the changes

However, even if there is some spatial dispersion of the changes, it is important to note that this is limited by the presence of linear infrastructure, the proximity to population centers, and the topographic factors that prevent the indiscriminate dispersion through the territory.

Page 9 of 32 !

Figure 8 underlines the influence of distance from the population centers to the changes. The proximity of population centers is very important for the generation of a change. Actually, the shorter the distance, the higher the pressure of change.

1990(@(2011( (

((1975(@1990(( ( 0%! 5%!

0%!

1%!

1%!
9%!

6%!

1km!

13%!

2km!

0%!

0%!

1%!

1km!

9%!

2km!

12%!

3!km!

3!km! 41%! 31%!

4km! 39%!

4km! 5!km!

32%!

6!km! 7km!

Fig. 8. Distance of the changes of population centers

Localized areas between 1 and 3 km from a population center or village are the most likely to undergo a change, while localized areas less than 1km and those more than 4 km are less likely to undergo a change. 2.2.2. Role of road network and infrastructures on territorial dynamics In this paragraph we try to answer the following question: “What is the influence of communication network and roads on the changes? Does the distance have an influence?” The road network and the localization of the principal population centers for the study area were obtained from the IGN2, (2009). The thematic information concerning natural protected areas was obtained from Natura 20003 and includes the “Special Protection Areas”. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 2

3

National Institute of Geography !Natura 2000 is an ecological network of protected areas in the territory of the European Union.

!

Page 10 of 32 !


5!km! 6!km! 7km! >8km!

1975(@(1990( 2%!

0%!

1990(@(2011( 2%!

0%!

0%!

0%!


20%!


1!km!

1!km!

2km!

2km!

3km! 78%!

!4!and!5!km!

76%!

3km! 4!and!6!km!

Fig.9. Distance of the changes of the road network The distance from communication networks and nodes of intersections is a factor that has a strong influence on the occurrence of change (figure 9). The farther from the roads the less significant the density of changes is. Most of the changes occur when the distance is less than 1 km from the roads or nodes of intersections. 2.2.3. The effect of geomorphology on the landscape transformation In this part, we answer the question “What is the relationship between the changes and the geomorphologic factor?” The data concerning of the geomorphology are provided by the vertical dissection model that was generated according to the method proposed by Priego-Santander et al., (2003). The vertical dissection provides information about the form, intensity and the type of geomorphologic processes that act on the landscape. (For more information about “vertical dissection” see Bocco et al., 2010). We consider that geomorphology and land use changes are closely related, in such a way that the changes are not random and do not occur everywhere. We also believe that transport network and principal population centers also have a strong influence on the occurrence of changes. Another factor to explain the occurrence of a change is the presence of specific topographic conditions. Even if there is a trend that the changes are spreading, this spread is conditioned by the topography and the geomorphologic factors. The dispersion phenomenon is therefor not random or homogenous.

Page 11 of 32 !

1975%&%1990% 1%!

21%!

1990%&%2011% 1%!

1%! 16%!

8%! 3%! 50%!

!High!hills!Moderately! dissected! High!Mountains! moderately!dissected! Mountains!Slightly! dissected!! Plains!with!hills!slightly! and!strongly!dissected! !Plains!and!hills!Slightly! dissected!

1%! !High!Mountains!strongly! dissected!

High!Mountains!strongly! dissected! Highly!dissected!!high!hills!

19%!

18%!

Highly!dissected!!high! hills! !High!hills!Moderately! dissected!

8%!

High!Mountains! moderately!dissected!

5%! 48%!

Mountains!Slightly! dissected!! Plains!with!hills!slightly! and!strongly!dissected!

Fig.10. Relationship between geomorphologic factor and the changes Figure 10 shows the occurrence of the changes depending on the geomorphologic factor for each evaluated period. This information explains the influence of the geomorphology in such a way that the frequency of the changes can be used like an indicator of the constraints represented by the geomorphology and also shows the evolution of the land use occupation in the areas that have suitable levels to be used for the different activities or classes of soil occupation. This analysis confirms that the dispersion of the changes follows a certain “regularity” depending on the accessibility provided by the geomorphology and that the idea that the dispersion in the concerned territory is a random phenomenon can be rejected.

Page 12 of 32 !

2.3. Step 2: Transition rules definition The spatial transition rules explain the transformation of one state to another; this transformation depends on the force of life of each individual and the interaction forces, and the competition between the individuals who evolve in a given place with a particular environment. Each individual (cell, state, land use, etc.) is submitted to the environmental interactions, which can provoke their transformation into another state (Patrice Langlois, 2004). The rules are interpreted in accordance with the knowledge base of the territory concerned, compiled from the precedent analysis. These rules will serve to simulate the potential change and territorial transformation. According to Santé, (2010), a cellular automata (CA) is a formal model composed by a set of cells which take determinate values; these values evolve in discrete steps according to a mathematical expression which is sensitive to the value of neighboring cells. A cell state is a description of the cell characteristics and the change in a state occurs according to a transition rule (Mitsova et al., 2011). Transition rules are defined according to a formal language which incorporates all the knowledge of the study phenomenon (Dubos-Paillard et al., 2003). In this way the transition rules must be defined according to a deep analysis of the study system. We used a developer of cellular automata models proposed by Langlois, (2000) called SpaCelle (System of Environmental Cellular Automata Production) based on transition rules. The advantages of this platform are principally that the user completely defines and handles the transition rules and defines all the components of the simulation model. Other advantages are the possibility to integrate the GIS data multilayer in the platform interface; in addition, complex informatic programming is not necessary. The principal limits of the platform are that the platform does not make the mobile objects, the distance factor can be introduced only in terms of the neighborhood parameter, and finally the language of the platform is French, a limiting factor for non-francophone users. However, we consider that the advantages of the platform outweigh any limits, which can be overcome by data preparation in the Geographic information System (GIS). This is why SpaCelle has been chosen for this work. After assessing behavior during the two periods (1975 – 1990 and 1990 – 2011), we selected the 1990 – 2011 period as the base of simulation because we consider that the pressures on agricultural areas presented during the 1975 – 1990 period were a very important referential of the changes that occurred and the conditions of these structures differed greatly from the current condition of the agricultural areas. One of the reasons is that current distribution of this structure is much more fragmented than before, in such a way that the spatial behavior of the changes can not be reproduced by the simulation. By contrast, all of the behavior identified for the second period of evaluation can be reproduced using the cellular automata. Page 13 of 32 !

725 cells 657 cells

+

Natural

+ +

264 cells

320 cells 602cells Agricultural +

+

5 cells

+ Artificial

+

+

+

54 cells 8 cell

7 cell +

1 cells +

Aquatic

6 cells

Fig.11. Causal Relations of the 1990 – 2011 land cover changes The influence of the artificial surfaces on the natural surface is the most remarkable while the agricultural areas are the most affected by the change, because these areas have lost surface from the synergic pressures of the artificial and natural surfaces and the areas that benefited the most from the changes are the artificial ones. At the same time it is possible to understand that the change is a very complex process. The table 3 shows the number of cells that have changed for each trajectory and shows the equivalence of change in hectares. The size of a cell is 18.50 hectares. Trajectories agricultural to natural artificial to natural aquatic to natural natural to!agricultural artificial to agricultural aquatic to!agricultural natural to artificial agricultural to artificial aquatic to!artificial natural to aquatic agricultural to aquatic artificial to!aquatic total

Cells 657 264 5 320 54 1 725 602 8 6 10 7 2,659 Page 14 of 32

!

Hectares 12,153.60 4,883.64 92.49 5,919.56 998.93 18.5 13,411.51 11,136.18 147.99 110.99 184.99 129.49 49,187.87

Table 3: Spatial transition rules interpretation

The transition rules in SpaCelle platform have the following structure:

Figure 12: Syntax of spatial transition rules in SpaCelle language. Table 4: Spatial transition rules in SpaCelle language States definition

Interaction rules

States

Force of life

E1:Natural

300 years

Spatial transition rules E1>E2= AL(2,21)

E2:Aquatic

Undefined

E2>E4= AL(1,21)

E3:Agricultural

100 years

E3>E4= AL(31,1)+PV(E4,1)

Page 15 of 32 !

Interpretation Two natural cells could change to aquatic cell each 21 years if there is a presence of aquatic cells near One aquatic cell could change to artificial every 21 years randomly 31 agricultural cells could change to artificial when the density of artificial cells is stronger in the close distance than the agricultural cell.

E4: Artificial

Undefined

E5: Natural protected

Undefined

E6 : Natural with geomorphologic constraints

Undefined

E0 :Communication network

Undefined

E1>E4= AL(21,1)+PV(E4,2)*Z V(E3,3)*PV(E0,2)

The natural cells can change to artificial when the density of artificial cells is stronger in the neighborhood than the natural cell, when there is no agricultural cell close and when there are routes in the close proximity of the natural neighborhood. E1>E3=AL(15,1)*PV( The natural cells can E3,1) change to agricultural when the density of agricultural cell is stronger in the proximity of the neighborhood than the natural cell. E3>E1=AL(31,1)*ZV( The agricultural cells E4,1) could change to natural cells when the natural density is stronger than agricultural cells and when no agricultural cells are present in the neighborhood of the agricultural cell. E4>E1=AL(12,1)*ZV( 12 artificial cells can E4,1)*ZV(E3,1) change to natural ones every year in a random way if there is no artificial cell in the neighborhood and if there is no agricultural cell in the proximity. E4>E3=AL(3,1)*ZV(E 3 artificial cells can 4,1)*PV(E3,2) change to agriculture ones every year in a random way if there

Page 16 of 32 !

E5>E5=EP(100)

is no artificial cell in the neighborhood and if there is no agricultural cell in the proximity of the neighborhood. The protected areas remain protected areas in a 100-year period.

2.4. Step 3: Model Calibration Before making the simulation of the transformation scenario, a model calibration was done (figure 13). It consisted in obtaining a model of 2011 scenario from the simulation with cellular automata using as baseline of simulation the land use thematic of 1990.

Fig.13. Model calibration diagram A combination of different methods of validation was used for the calibration. First of all we used the information provided by the visual interpretation that consists in observing the most representative differences between both maps (real map vs simulated map). Then in the second method of calibration we used the statistical approach provided by VALIDATE module of IDRISI SELVA. Then we have evaluated the spatial pattern followed by the simulation by cellular automata that is why landscape ecology metrics were used. These metrics can provide a vision of the similitude between both maps relating to the spatial pattern, the surface pattern of the patches, the shape and the distribution of the elements of the map (Gomez Delgado et al., 2012). This evaluation can provide more important Page 17 of 32 !

information than the kappa index application alone, because it enables us to know whether the transition rules follow the spatial pattern observed in reality. 3. Results and discussion We present in this part three main results: 1. The model calibration: it is based on the spatial comparison between the observed land use model (2011) and the simulated land use from the 1990 situation; 2. The output map of the landscape transformation for 2050 from 2011 land use model; 3. The validation of the 2050 landscape transformation simulation: it consists in verifying if the simulated model retains the observed behavior. 3.1. Results of the model calibration Figure 14 shows three maps. The first map represents the base of the simulation: it is the land use model for 1990 based on a Landsat image. The second map represents the observed situation for 2011. The third map shows the simulated land use for 2011 from the 1990 land use model and the integration of the spatial transition rules. This simulated model was performed using 21 iterations in the SpaCelle cellular automata platform.

Fig.14. Results of model calibration

Page 18 of 32 !

The visual interpretation allows confirming the coherence of the simulation by the spatial transition rules. The simulation reproduces the observed spatial patterns in a similar way of the reality. The main transformations are related to the increase of artificial areas while the agricultural areas decrease. These transformations are mainly localized closed to the coastal strip as it is underlined in the real situation in 2011. To complement these results we have used the statistical approach to examine the quantitative agreement between both maps. The VALIDATE module offers one comprehensive statistical analysis that answers two questions 1).- How well do a pair of maps agree in terms of the quantity of cells in each category? And 2).- How well do a pair of maps agree in terms of the location of cells in each category? (Pontius, 2006). The levels of information provided by VALIDATE are defined by the next statistical expressions (We include above the exact explanation given by Pontius, 2006 for each level of information provided by VALIDATE): Expression N(n) “N(n) is the agreement due to chance, which is the agreement between the reference and a map that has a membership of 1/J to each category in every grid cell, where J is the total number of categories in the analysis. If the reference map shows that each cell is in exactly one category at the resolution of the raw data, then the agreement N(n) is 1/J at that resolution, because each cell is 1/J correct. This level of agreement is equivalent to the expected agreement between the reference map and a map in which every raw grid cell is assigned to one of the J categories with a probability of 1/J.” Expression N(m) “N(m) is the agreement between the reference map and a map that has a distribution of m among the various categories in every cell, where m denotes a vector of the distribution of categories in the comparison map. N(m) is equivalent to the agreement between the reference map and a modified comparison map, where the modification is to randomize the locations of the raw cells within the comparison map.” Expression H(m) “H(m) is the agreement between the reference map and a modified comparison map, where the modification is to randomize the locations of the cells within each stratum of the comparison map. When the modification randomizes the location of the grid cells, each cell remains within its stratum (i.e. no grid cells move across stratum boundaries).”

Page 19 of 32 !

Expression M(m) “M(m) is the agreement between the reference map and the unmodified comparison map. It is the proportion of grid cells classified correctly, which is the most commonly used measure of agreement between maps. However, M(m) is tricky to interpret because a surprisingly large proportion of the landscape can be correct due to chance. Furthermore, M(m) confounds agreement due to quantity and agreement due to location.” Expression K(m) “K(m) is the agreement between the reference map and a modified comparison map, where the modification is to rearrange as perfectly as possible the locations of cells within each stratum of the comparison map in order to maximize the agreement between the modified comparison map and the reference map. The method of modification swaps the location of the grid cells, with the constraint that each cell remains within its stratum (i.e., no grid cells move across stratum boundaries).” Expression P(m) “P(m) is the agreement between the reference map and a modified comparison map, where the modification is to rearrange as perfectly as possible the locations of cells within the entire comparison map in order to maximize the agreement between the modified comparison map and the reference map. The method of modification swaps the location of the grid cells anywhere within the comparison map (i.e. rearrangement of location of grid cells is allowed to occur across stratum boundaries). Therefore P(m) = 1 if and only if the distribution of proportions m in the comparison map is the same as the distribution of proportions p in the reference map.” Expression P(p) “P(p) is perfect agreement, which is the agreement between the reference map and a map that has perfect information of both quantity and location. Therefore, P(p) is always 1.” “VALIDATE gives also the traditional Kappa Index of Agreement (KIA). KIA is denoted also by Kstandard: [1] Kappa for no information (denoted Kno), [2] Kappa for grid-cell level location (denoted Klocation), and [3] Kappa for stratum-level location (denoted KlocationStrata). All of these statistics are linear functions of points in the VALIDATE output. Specifically, Kno = {M(m)-N(n)}/{P(p)-N(n)}. Klocation = {M(m)-N(m)}/{P(m)N(m)}. KlocationStrata = {M(m)-H(m)}/{K(m)-H(m)}. Klocation indicates how well the grid cells are located on the landscape. KlocationStrata indicates how well the grid cells are located within the strata (Pontius, 2006).”

Page 20 of 32 !

Table. 5. Kappa Interpretation: Kappa

Agreement Less than chance agreement Slight agreement Fair agreement Moderate agreement Substantial agreement Almost perfect

< 0.00 0.00 – 0.20 0.21 – 0.40 0.41 – 0.60 0.61 – 0.80 0.81 – 1.00

Table 6. Results of statistical validation of the quantitative agreement between both maps. Classification!agreement/disagreement!according!to!ability!to!specify!accurately! quantity!and!allocation! !!

Information of Quantity

Information of Allocation Perfect[P(x)] PerfectStratum[K(x)] MediumGrid[M(x)] MediumStratum[H(x)] No[N(x)]

No[n] P(n) = 0.0012 K(n) = 0.0012 M(n) = 0.0008 H(n) = 0.0002 N(n) = 0.0002

Medium[m] P(m) = 0.9935 K(m) = 0.9935 M(m) = 0.9466 H(m) = 0.4377 N(m) = 0.4377

Perfect[p] P(p) = 1.0000 K(p) = 1.0000 M(p) = 0.9457 H(p) = 0.4370 N(p) = 0.4370

AgreementChance = AgreementQuantity =

0.0002 0.4375

0.0468 0

AgreementStrata = AgreementGridcell =

0 0.5089

DisagreeGridcell = DisagreeStrata = DisagreeQuantity =

!!

!

!! !

Klocation = KlocationStrata = Kno = Kstandard =

0.0065 !! !

0.9157 0.9157 0,9466 0,9051

In order to evaluate the level of similitude between maps in terms of morphological conditions, we have used the Landscape ecology metrics (LEM). The LEM can be used to

Page 21 of 32 !

obtain information about the proximity, form and surface. In this work we used the follow metrics: Patches Index (Shape Index): area / perimeter ratio adjusted to a square or a circle. When the form is compact, the index has a value of 1. This value increases with irregular shapes. Fractal dimension of the average patch (FD) measures the complexity of shapes; its value is between 1 and 2. Values close to 1 correspond to very regular perimeters, and values close to 2 correspond to very complex shapes. Shanon DiversityIndex: The “Shannon diversity" index determines in a neighborhood, the number of types of land present in approximately equal proportion. This index reflects the diversity of the landscape: the higher the index, the greater the number of types of land present in the same proportion. Edge Density: "Edge density” index calculates the length in a neighborhood of contours per hectare. This index reflects the complexity of the forms in the landscape: the higher the density contours, more complex is the landscape and the units are overlapped (Farina, 2011) Table. 7. Landscape ecology metrics Map

FD

Real Simulated Global difference

1.44 1.43

Patches Index 7.64 7.03

-0.01

-0.61

Shannon Diversity

Edge density

0.48 0.49

0.19 0.20

0.005

0.0003

From the results of the visual interpretation, statistical approach and landscape metrics allow us to deduce that the transition rules applied within the cellular automata have an acceptable approximation degree with the real situation observed in the real model of 2011 and also show that the behavior follows an acceptable spatial and morphologic trend during the simulation. It is thus possible to make a territorial transformation model from simulation by cellular automata. 3.1. Landscape transformation for 2050 scenario The iteration represents one year. 2011 is the year of the beginning of the simulation and after completing 39 iterations the territorial transformation scenario is produced for 2050.

Page 22 of 32 !

Landscape transformation scenario

Fig.15. Landscape transformation scenario for 2050

The results show that the territorial transformation occurs in different ways. While densification is constant near the coastal strip, in other parts of the territory, dispersion occurs, which can be explained by the influence of different territorial elements such as routes and road network, village and population centers that together act as an asset of expansion. The results presented by Fusco et al., (2008), support our results given that they mentioned the possibility of the urban expansion in the next decades because the territory has the suitability to host the population expected for 2030 by the demographic previsions with the difference that our work gives a potential map of future expansion that was not presented earlier.!! !

Page 23 of 32 !

The simulation shows a possibility of duplication of artificial surfaces for 2050, which corresponds to 21% of total territory in detriment of the natural surfaces and agricultural areas.

2050$

2011$ 0%! 3%!

1%!

11%! AGRICULTURAL!

21%$

ARTIFICIAL!

0%!

NATURAL! 86%!

ACUATIC!

78%!

Fig.16. Potential land occupation by artificial surfaces From this results it is possible to observe that the principal affectation of the spread of artificial surfaces occurs on natural structures in consequence is possible that biodiversity could be affected by this phenomenon of landscape transformation. These affectations will be potentially in the interior of the territory (middle and highlands) because the littoral strip is very impacted by the urbanization since 1975. 3.5. Validation In this case the method of validation consists in verifying that the model retains the behavior observed in the analysis of historical dynamic department in a logical and coherent form. Therefore, validation is to ensure that land use classes have a tendency to change in a "similar" way, or at least with a certain level of closeness to the process observed in the analysis of historical change and that the transition spatial rules used for simulation have complied with the spatial behavior and morphologic trend during the transformation process.

Page 24 of 32 !

Do the transition rules reproduce the observed spatial pattern?

1990$%2011$

1975%1990$ linear! aggregate! nodal!aggregte!

6%! 2%!

7%! 0%!

23%!

62%!

7%!

dispersed!

linear!aggregate!!

9%! 1%!

0%! 26%!

pure!aggregate!

2011$%$2050$ linear! aggregate!

2%!

!nodal! aggregate! pure!aggregate!

28%!

14%!

pure! aggregate!

dispersed! 57%!

linear!!

linear!

nodal!

nodal!

7%!

!nodal!!

49%!

linear!! nodal! aggregate!

Figure 17: Validation by spatial pattern The results of the comparison of spatial trend show that the change follows a continuous tendency that enables us to validate that transition rules have complied with the behavior observed in the past. The exception concerns the dispersed pattern because the model does not consider a transition rule that takes into account only the random aspect that in reality happens with some changes. This omission is considered as a weakness of the model.

Page 25 of 32 !

Do the transition rules reproduce the influence of the distance of communication network on the changes?

2%!

1975$%$1990$

0%!

0%!

2%!

0%!

1990$%$2011$

1!km!

78%!

2011$%$2050$

1!km!

2km!

2km!

3km!

3km!

!4!and!5!km!

0%!

0%!


20%!

3%! 0%!

76%!

4!and!6!km!

22%!

1!km! 2km! 75%!

3km! 4!and!6!km!

Figure 18: Validation by communication network distance measure

The results of validation by the distance of communication network confirm that the model reproduced the observed behavior. Also we can observe that even if the dispersion phenomenon is progressively present most of the changes occurred close of the roads remaining the harder pressure in the maximal proximity of 1 km.

Page 26 of 32 !

Do the transition rules reproduce the influence of the distance of population centers on the changes?

$$1975$%1990$$ $ 0%! 0%!

5%! 1%!

1990$%$2011$ $

0%! 1%!
6%!

1km!

9%! 13%!

2km!

0%!

1%!

9%!

41%!

4km! 5!km!

1%!
12%!

2km!

3!km! 31%!

2011$%$2050$

39%! 32%!

6!km!

0%!

0%!

7%!

2!km!

13%!

3!km! 37%!

32%!


9%!

3!km! 4km!

1%!

4km! 5!km!

5!km!

6!km!

6!km!

7km!

Figure 19: Validation by population centers distance measure

In the same logic, the validation of the results concerning the distance of the changes to the population centers confirms that the transition rules have complied with the tendencies of the observed behaviors. It confirms also that we have mentioned along this work about the non-randomly influence on urbanization given that the different elements of territory have a strong influence on this phenomenon.

Page 27 of 32 !

4. Conclusion

This work demonstrates that it is possible to project landscape transformation by studying historical behavior and spatial pattern. We have also been able to prove the strong influence of geomorphological factors and communication networks above the dispersion of the land use changes and we can reject the idea that this dispersion is random and homogenous across the concerned territory. We also have an image of the possible future of landscape that enables us to make decisions about land use and ecological planning which is considered as the originality of this work given that it was not presented earlier for the studied territory. The landscape transformation of the concerned territory has ecological implications that must be evaluated. First of all we consider that the changes may cause the biodiversity affectation because of the landscape fragmentation. This is why it is necessary to evaluate the alternative policies for territorial and biodiversity governance that enable the control of the urban expansion and dispersion as a measure of sustainable urban planning. The results have a degree of approach that is considered as a limit for the research. However they allowed the model to meet its objectives (calculate and project the trends of change in the landscape of French Riviera). Even if the method proposed in this work has a high degree of complexity and that there are technical difficulties to obtain the optimal results, we consider that it can be used as a tool for the ecological and urban management and for decision-taking. Conclusions about the main results of simulation model: As it is possible to observe, the territorial transformation occurs in a different way across the territory because near the coastal strip and the littoral, urbanization is consolidated and compact. This is due to the infrastructure and the strong attraction of the zone. In addition, the model shows the strong influence exerted on the transformation roads and lines of communication that provokes a significant dispersion in the urbanization, especially in medium and high country of the department. Moreover, the analysis of changes shows that the transformation of the territory can provoke changes in the spatial organization and it is also possible to note that the changes follow a trend to dispersion influenced by the communication and transport networks. This dispersion could be a major change in the model of historical urbanization in the department. Finally these results can provide information about the needs of planning measures and about the most sensitive areas to change.

Page 28 of 32 !

The principal limits of this work: • • • • •

• •

The spatial rules are stationary and depend on the level of knowledge of the observed phenomenon. The emergence of the new artificial areas is conditioned by the deterministic conditions of the spatial rules and the elements of the model. The transition rules are identifiable only on relatively long periods of time. The model has been tested only in a landscape scale, and it is very possible for the model not to be optimal for intra-urban simulation. The model is first of all an ecological model that seeks to understand the phenomenon rather than taking decisions (where is possible the artificial dispersion? From what conditions? What are the factors involved?) The model is limited to intuitively understand the consequences of artificial on biodiversity. The model is not able to answer questions related to infrastructure planning, population growth, etc, because the model is not designed with these goals.

Page 29 of 32 !

5. References Aguilera Benavente, F.; Plata Rocha, W.; Bosque Sendra, J. y Gómez Delgado, M. 2009. “Diseño y simulación de escenarios de demanda de suelo urbano en ámbitos metropolitanos”. Revista internacional de Sostenibilidad Tecnología y Humanismo. España. Aguilera Benavente, L. Valenzuela Montes 2010. “Simulacion de escenarios futuros en la aglomeracion urbana de Granada a traves de modelos basados en automatas celulares”. Boletin de la Asociacion de Geografos Españoles No. 54 págs. 271-300. Barredo, Marjo Kasanko, Naill Mc Cormick, Carlo Lavalle 2003. “Modeling dynamic spatial processes: simulation of urban future scenarios trough cellular automata”. Landscape and urban Planning, vol. 64 págs. 145-160. Batty, M., Xie, Y. and Sun, Z. (1999), “Modelling Urban Dynamics through GIS-Based Cellular Automata', Computers, Environment and Urban Systems”, 23(3): 205-233. Bocco, G., Priego, A. & H. Cotler. 2010. “The contribution of physical geography to environmental public policy in México”. Singapore Journal ofTropical Geography 31: 215223. Caglioni. M. “The SLEUTH Urban CA-Based Model: an evaluation”. Università di Pisa Dipartimento di Ingegneria Civil. Castillo Romano, J. Brena Zepeda. “MODELACIÓN PROSPECTIVA DEL PAISAJE DE UNA CUENCA”. Instituto Mexicano de Tecnología del Agua Especialistas en Hidráulica. Chuyang He, Norio Okada, Qiaofeng Zhang, Peijun Shi, Jingshui Zhang. 2008. “Modeling urban expansion scenarios by coupling cellular automata model and system dynamic model in Beijing, China”. Landscape and Urban Planning, vol. 86 (2008) p. 79 - 91. Clavero, M. Santos, R. Navarro, J.J. Guerrero, F. Cáceres1, J. M. Moreira 2010. “IMPLEMENTACIÓN DE UN SISTEMA DE ESCENARIOS FUTUROS SOBRE EL MAPA DE USOS DE SUELO DE ANDALUCÍA”. Secretariado de Publicaciones de la Universidad de Sevilla. Sevilla 2010. Farina 2011. “Ecología del Paisaje”. Publicaciones Universidad de Alicante, España. 688 págs. Feng, Yan Liu, X. Tong, M. Liu,Susu Deng 2011. “Modeling dynamic urban growth using cellular automata and practical swarm optimization rules”. Landscape and urban planning, vol.102 Pages. 188-196. Page 30 of 32 !

Fusco G., Scarella F. 2008. “L’evolution de l’habitat dans les Alpes-Maritimes et dans l’Est Var”. UMR ESPACE, equipe de Nice dans le cadre du PREDAT des AlpesMaritimes. Gomez-Delgado, Rodriguez Espinoza 2012. “Analisis de la dinámica urbana y simulación de escenarios de desarrollo futuro con tecnologías de la información geográfica”. Editorial Ra-MA., Madrid, España. 350 págs. Gutiérrez Angonese, Jorgelina; Gómez Delgado y Bosque Sendra 2010. “Simulación de crecimiento urbano mediante evaluación multicriterio y sig en el gran san miguel de tucuman (ARGENTINA)”. Secretariado de Publicaciones de la Universidad de Sevilla. Sevilla. Pp. 873-888. ISBN: 978-84- 472-1294-1. Langlois, Pierrick Tranouez, E. Dubos-Paillard, D. Provitol, C. Tannier – CR ( U. Besançon) – CR ( U. Besançon) – Mcf ( U. Besançon), – IR ( U. Rouen) – McF/HdR ( U. Rouen), F. Amblard – Mcf ( U. Toulouse) “Tours des Plateformes” . Programme Jeunes Chercheuses – Jeunes Chercheurs. Lin Li, Yohei Sato, Haihong Zhu 2003. “Simulating spatial urban expansion based on a physical process”. Landscape and Urban Planning 64 (2003) 67–76. Mcdonald, Richard T.T. Forman, Peter Kareiva, Rachel Neugarten, Dan Salzer, Jon Fisher 2009. “Urban effects, distance, and protected areas in an urbanizing world”. Landscape and Urban Planning, vol. 93 pages. 63–75. Mcdonaldad, Peter Kareiva, Richard T.T. Forman 2008. “The implications of current and future urbanization for global protected areas and biodiversity conservation”. BIOLOGICAL CONSERVATION,vol.141 pages. 1695 –1703. Mitsova Diana, Shuster William, Wang Xinhao 2011. “A cellular automata model of land cover change to integrate urban growth with open space conservation” Landscape and Urban Planning 99 (2011) 141–153. Pointius G.Jr and Laura C. Schneider 2001. “Land-cover change model validation by an ROC method for the Ipswich watershed, Massachusetts”. Agriculture Ecosystems and Environment, vol. 85. Pontius Robert G.JR and Hao Chen 2006. “Land Change Modeling with GEOMOD” Clark University 2006. Pontius Robert G.JR 2000. “Quantification Error versus Localtion Error in Comarison of Categorical Maps” Photogrammetric Enginnering & Remote sensing. Vol. 66 No.8, pp. 1011-1016. Pontius Robert G.JR and MILLONES M. 2011. “Death to Kappa: birth of quantity disagreement and allocation disagreement for accuracy assessment”. International Journal of Remote Sensing Vol. 32, No. 15, 10 August 2011, 4407–4429.

Page 31 of 32 !

Priego -Santander, Jose Luis Palacio-Prieto, Patricia Moreno-Casasola, Jorge LopezPortillo, Daniel Geissert Kientz 2004. “Heterogeneidad del paisaje y su riqueza de flora: su relacion en el Archipielago de Cmaguey, Cuba”. INTERCIENCIA Vol.29. Priego-Santander,E-Isunza-Vera, N. Luna-González y J. L. Pérez-Damián 2003. “Método para realizar mapa de diseccion vertical.” Instituto Nacional de Ecología, SEMARNAT.http://mapas.ine.gob.mx/website/ metadato/cuencas/diseccion.html. Puliafito 2004. “La evolución urbana desde el punto de vista de un modelo espacio temporal: caso gran Mendoza”. Mecánica Computacional Vol. XXIII G.Buscaglia, E.Dari, O.Zamonsky (Eds.) Bariloche, Argentina, November 2004. Recio, J.E. Pardo Pascual, L.A. Ruiz Fdez., A.Fdez. Sarría y P.Córcoles Tamarit. “Detección y cartografiado de los procesos de expansión urbana mediante técnicas combinadas de teledetección y S.I.G”. Dept. Ingeniería Cartográfica, Geodesia y Fotogrametría Universidad Politécnica de Valencia [email protected]. Rodríguez Álvarez, W. Plata Rocha, M. J. Salado García, M. Gómez Delgado, J. Bosque Sendra 2009. “HERRAMIENTA PARA LA ASIGNACIÓN ÓPTIMA DE USOS DEL SUELO”. Departamento de Geografia de la Universidad de Alcala Calle Colegios, 2, 28801 Alcalá de Henares, Madrid, España Web: http://www.geogra.uah.es. Sante, A. M. García, D. Miranda, R. Crecente 2010. “Cellular automata models for the simulation of real-world urban processes: A review and analysis”. Landscape and urban planing, vol. 96 (2010) Pages. 108-122. Xie, Y. and Batty, M. (2003). “Integrated Urban Evolutionary Modeling, Centre for Advanced Spatial Analysis” (University College London): Working Paper 68, London, England. Zavala, R. Diaz -Sierra, D. Purves, G.E. Zea, I.R. Urbieta 2006. “Modelos espacialmente explícitos”. Revista científica ECOSISTEMAS, vol.15 (3) págs.: 8889.

Page 32 of 32 !