Use of Cellular Automata in the Study of Variables Involved in Land Use Changes an Application in the Wine Production Sector

View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by Florence Research Environ Monit Assess DOI 10.1007/s10661-012-2951-z

Use of cellular automata in the study of variables involved in land use changes An application in the wine production sector

Francesco Riccioli & Toufic El Asmar & Jean-Pierre El Asmar & Roberto Fratini

Received: 18 April 2012 /Accepted: 8 October 2012 # Springer Science+Business Media Dordrecht 2012

Abstract The study of changes in land use has been Keywords Cellular automata . Land use change . included lately in territorial processes in order to opti- Markov chains . Territorial analysis . GIS mize future management decisions. The different functions that the territory plays (production, aesthetic, and natural functions, etc.) make planning choices Introduction more difficult. This work focuses on the selection and combination of a set of indicators to analyze the The study of changes in land use is among the new variables through which the changes occur in the land challenges in territorial planning. Investing in such an used in viticulture for wine production. The proposed area of study would be advantageous due to the approach makes use of the Geographic Information resources in the form of public funds and community System (GIS) in the development of a map of land aid that could be available for the creation of a diver- use scenario. It is applied to a case study through a sified picture that takes into account the different model involving cellular automata (CA) implemented territorial functions: functions of production, aes- with maps of suitability for viticulture and Markov thetics, water regulation, territorial management, and chains. The use in this case of the CA is aimed at natural features and the preservation of an identity to validating the scenario map in order to deduce the which the territory is strongly linked. To this end, it is variables and the orientations of the farmers in the of considerable interest to use data management soft- field of wine production. ware tools capable of analyzing the changes occurring during the previous years, relying on the acquired know-how as a starting point for further investigation. F. Riccioli (*) : T. El Asmar : R. Fratini In a particular way, the cellular automata (CA) are Department of Agricultural and Forest Economics, used in almost all land use change models for urban Engineering, Sciences and Technologies, environments (Torrens and O’Sullivan 2001; Ward et University of Florence, Florence, Italy al. 2000; White et al. 1997;Wu1998). Besides urban- e-mail: [email protected] ization, CA-based models now also simulate other processes of land use change; for example, Messina J.-P. El Asmar and Walsh (2001) studied land use and land cover Faculty of Architecture Art and Design, Notre Dame University–Louaize, dynamics in the Ecuadorian Amazon, an area where Zouk Mosbeh, Lebanon tropical forest is converted into agricultural land. Environ Monit Assess

Applications of CA for land use change models in Applied methodology which both urban and rural land uses are considered are provided by Engelen et al. (1995) and White and CA is an instrument belonging to the field of artificial Engelen (2000) (Verburg et al. 2004). intelligence. Through CA, the use of simple behavior- The analysis of changes in land use is coupled with al rules allows simulation of the global behaviors of a the study of the rules of transition, the latter being given system. CA was designed in the 1950s and has characterized by variables that come into play in the recently found use in numerous software applications, evolutionary scenarios. "The definition of the transi- in the field of urban and regional planning primarily tion rules of a CA model is the most essential part to for its interface compatibility with GIS (Engelen et al. obtain realistic simulations of land use and land cover 2002). The classical CA study can be described as change. Land use change is the result of a complicated having the following properties: decision-making process; however, the transition rules 1. The study area is usually divided into cells, parts of of CA models are often defined on an ad hoc basis" a regular two-dimensional grid (typical grid raster). (Verburg et al. 2004). However, more recently three-dimensional and ir- The proposed case focuses on the selection and regular grid CA systems have been used where each combination of a set of indicators to describe the cell is identified by three coordinates forming a transition variables and to create maps of suitabil- cube. ity for viticulture. The CA are directly applied just 2. The state of each cell is determined through attrib- to validate the aforementioned suitability maps utes; in the case of a GIS vector, punctual, linear, which are used indirectly to validate the indicators or polygonal elements from within the cell. In the employed for their creation: this will enable us to case of a GIS raster the whole pixel. understand what may be the reasons currently 3. It is then necessary to define the rules of transition driving or capable of driving farmers in the future from one state of the cell to another. This is based to shift their attention or simply to persevere in on both initial states of the cell and to the adjacent viticulture activities. ones. Computer calculation is due to the fact that In addressing this issue in certain regions of the having k possible states and n cells included Netherlands, Hagoort et al. (2008) have shown that around the identified cell, there are possible rules regional and temporal calibration of neighborhood (e.g., with k02 and n05, 4 billion rules are found). rules is a refinement in the application of CA-based According to these rules it is possible to distin- land use models that pay off by increasing the validity guish various typologies of CA: “deterministic” of the model outcomes. CA using rules of fixed evolution (where evolu- This paper aims to determine a set of indicators tion trends are constant), “stochastic” CA where able to analyze the variables through which land the transition rules are modified during the dy- changes occur in viticulture practices. In this work, namics, and “probabilistic” CA using probability the study area is represented by the southern spot rules. of the province of Florence (Italy), in which the 4. The cells adjacent to the examined one are defined traditional vocation is to foster the growth of vine- as a neighborhood: the best known neighborhoods yards. In this area, agricultural activity is amal- are those defined by J. Von Neumann and Moore gamated with the history and the culture of a (Fig. 1). territory that has invested and is still investing a 5. Definition of a temporal scale: The dynamics gen- lot on grape and wine production. The proposed erally represent temporal rhythms of transition approach applies Geographic Information System rules (evolution) of the cells. This takes place in (GIS) that allows the development of visual sce- “synchronous” and discrete ways, either changing narios of changes in land use combining sophisti- thereby simultaneously the state of each cell at cated mathematical algorithms with local customs each step (in territorial applications the year or and traditions. This generates a model that high- its multiple is matched) or not. lights learning gained from the changes that oc- curred in the past in order to define the variables Practically and based on all of the above conditions that affect viticulture activity. "the cells then react, by a series of rules or relationships, Environ Monit Assess

Fig. 1 Von Neumann and Moore neighborhoods

to the local conditions around the cells, for example, Areas of study the condition of neighboring cells" (Smith and Smith 2007). The traditional agricultural vocation considered in the The CA method is used nowadays in many areas of areas of study is viticulture. Such areas extends along knowledge: in physics, medicine, and chemistry. An the southeast strip of the province of Florence, located area of application of noted interest is the simulation in Central Italy (Fig. 3). The morphology of the area is of urban phenomena, especially in the field of territo- hilly (over 40 % of the territory lies at an altitude rial planning (e.g., Engelen et al. 1999). ranging between 200 and 600 m above sea level), with This paper describes the above mentioned applica- average annual temperature of 18 °C, and rainfall of tions by means of a CA implemented using suitability around 800 mm annually. The areas represent 13 mu- maps for viticulture (Creation of viticulture suitability nicipalities, having about 12,900 ha of vineyards, and maps Section) and Markov chains (Transition matrix can be divided in two main categories based on the Section) to develop a scenario map for year 2006 type of wine produced: the Empolese Valdelsa and the regarding the targeted area of study. Chianti Classico. This synergy between the different tools is For Empolese Valdelsa, five municipalities have used to overcome two main limitations regarding been thoroughly analyzed: Cerreto Guidi, Empoli, the use of CA, as suggested by Bernetti and Vinci, Capraia, Limite and Montelupo Fiorentino. Marinelli (2008): “In a CA model the transitions This area of "Vino Bianco Empolese" is located in are free, each cell can freely change its state; the north of the Empoli Valdelsa area. Bianco whilst in land use changes analysis it is essential Empolese was founded in 1997 following a regional to presume a final query for each typology of law (L.R. 29 May 1997, n. 38), and 4,380 ha of land use” (this issue was addressed by Markov vineyards cover the area, equal to approximately chains); and “the space in which changes in land 34 % of the overall examined surface (12,900 total ha). use occur is isotropic; the transitions depend only In the second area, eight municipalities were exam- on the rules of the model and the state of the ined: Montespertoli, Certaldo, San Casciano in Val di neighborhood cells. Conversely, the transforma- Pesa, Tavarnelle in Val di Pesa, Lastra Signa, tions in the land use of the area of research Scandicci, and Greve in Chianti. The Chianti area depend on the land uses of the surrounding areas has characteristics opposite to those of the Bianco and on their geographic and socioeconomic char- Empolese area. Indeed, the Chianti area is rich in acteristics” (issue implemented by using the indica- tradition, and a conservation policy was initiated many tors for the development of suitability viticulture years ago to preserve and enhance local specialties. maps). Afterward, the created map resulting from This is evident from the existence of the Chianti Wine the CA is compared with an existing land use map. Consortium since 1927. This consortium brings to- A statistical validation would indirectly specify the gether more than 2,500 manufacturers affecting more reliability of the indicators used. The indicators are than 10,000 ha of vineyards, and producing a volume considered as factors that describe the preference of 70,000,000 l of Chianti wine each year. The vine- towards viticulture by farmers. Figure 2 shows the yards of Chianti examined for this work extend over flow chart of this manuscript. 8,600 ha, accounting for 66 % of the total study area. Environ Monit Assess

Fig. 2 Methodology flowchart

Cellular automata for the definition of a land use or from identified situations related to pedologic, cli- map in 2006 matic, orographic, historic, etc., characteristics of the study area. In the case of this work the rules of tran- Creation of viticulture suitability maps sition (refer to Introduction Section) of the geographic surfaces surrounding the study areas are considered. Generally, at the territorial level, land use changes can As stated by Hagoort et al. (2008) the rules of transi- be inferred either from what has happened in the past, tion consist of variables capable of driving the process

Fig. 3 Case study Environ Monit Assess of evolution of certain land uses which are determined of 75 m. In Fig. 4, the more gradual slopes are based primarily on review of the literature and of shown in dark red (values close to 255). interviews with industry experts. Furthermore, The study of the primary road network, represent- Hagoort et al. (2008), in view of the possible influence ing the next variable, has shown that the entire area is of the territory on such variables, identify different characterized by a good road network that increases in types of functions relating to the variables on the basis proximity of major urban centers such as Empoli and of their distance from the land use in question. It is, Scandicci. Based on this variable, the areas near the however, an area difficult to determine, and difficulty main roads are considered particularly suitable for increases exponentially according to the land use con- vine-growing, following the same viewpoint men- sidered (Dear 1977). tioned before (economic or fewer operation costs) for In this study, the suitability maps display for each the slope. The adopted road network layer is based on cell a maximum approximate value (255) close to the the “road traffic” maps developed in 2003 by the considered variable. This value decreases (linearly) region of Tuscany. with increasing distance from the variable itself. The suitability map for viticulture was drawn start- These values directly affect the probability of the cell ing from the road network layer by calculating the transformation in the vineyard. distance from main roads to the related fuzzy distance. Based on available data and on the examined liter- In Fig. 5, the dark red areas (values close to 255) show ature (Lombardo et al. 2005; Pijanowski et al. 2002; the ones next to the road network. Al-Ahmadi et al. 2009) two classical variables and “The fuzzy distance decay membership function is two experimental variables were selected because of used to indicate proximity to a given feature" (Al- their capacity to influence the changes in land use: Ahmadi et al. 2009). Rather than having a single crisp threshold that denotes a distance away from a feature, 1. Slope the fuzzy distance decay function is capable of de- 2. Main roads scribing the benefits of the vineyards' proximity to a 3. Ancillary roads (wine roads) particular variable (in this case the network of main 4. Productivity roads). No priority weights are attributed to the analyzed Two experimental variables linked to the typical variables, classical and experimental; they are consid- phenomenon of the study areas were then introduced: ered equally relevant in their involvement throughout the existing distance from the “Strade del Vino” (wine the phases of the work. The two classical variables are roads), and the productivity of the quality vineyards represented by the slope and distance from main roads. (DOC and DOCG). Regarding the ancillary roads the Considering that the climatic characteristics of the work refers to the “wine trails” surveyed and geo- examined areas are particularly suitable for viticulture, referenced by the region of Tuscany. These are paths no other variables normally affecting vineyards, such within the territory characterized by a high vocation to as temperature, humidity, precipitation or altitude, are wine culture, in addition to the vineyards and wineries, measured. and by natural, cultural and historical attractions par- Overall, 88 % of the studied areas slopes are less ticularly significant for a holistic approach to wine than 22 %, 55 % of which is below 10 %: it is tourism. The wine trails are an instrument for promot- effectively an area characterized by a very gentle ing rural territorial development, and are intended to relief. Vineyards can also be found on slopes that encourage Eno-tourism.1 are greater than 20 % (Greve in Chianti) with the The study area considers two among 14 itineraries presence in some cases of terraced slopes. The proposed by the Region of Tuscany: The Chianti wine areas with gradual slopes were considered suitable of the Colli Fiorentini (Florentine Hills) and the for wine-growing; this is justified only from an Chianti of Montespertoli crossing by the four munic- economic perspective and by the fact that the cost ipalities of Montelupo, Lastra a Signa, Scandicci, and of the cultivation and harvest operations increases with the increase of slope. The slope map (layer) 1 Retrieved from website http://www.terreditoscana.regione. was obtained from the Digital Terrain Model toscana.it/stradedelvino/ita/index-ita.html [last accessed (DTM) of Tuscany in raster format and a resolution October 12, 2011]. Environ Monit Assess

Fig. 4 Slope map

Montespertoli. From an operational point of view, the Relying upon the database provided by the Region of proximity to these roads is considered for its capacity Tuscany Wine Production for Agricultural Allocations2 to inspire the start of new vineyards, justifying this (ARTEA), 216 quality (production of DOC and DOCG3 phenomenon also through an aesthetic factor. The wine) wine farms scattered in the area were geo-referenced. wine roads are created to promote a particular territory, For each farm an average production per hectare during making viticulture its strong point, and influencing the 2005–2009 was calculated, by applying formula 1. surrounding areas to preserve or move towards the same 2009P vocation. In this case, too, the suitability map for viti- Pj culture was developed by calculating the fuzzy distance 2005 Paj ¼ ð1Þ from the wine trails. In Fig. 6, the areas close to wine N trails are shown in dark red (values close to 255). where Paj is the annual average production per The last variable considered is based on the produc- hectare of the jth farm, Pj is the annual production tivity of the studied areas: specifically, the productions per hectare of the wine were examined and recorded over a period of time corresponding closely to that of the 2 L'Agenzia Regionale Toscana per le Erogazioni in Agricoltura study. To develop suitability maps based on productivity, (ARTEA) is an agency that allocates funds as anticipated by EU a specific index was created capable of selecting the areas communities regulations for the management of the Commune that are particularly productive. The areas located in the Agricultural Politics (CAP). 3 DOC or DOCG or "Controlled Designation of Origin" is an proximity of the previously mentioned areas were con- Italian quality assurance label attributed to wine produced in sidered as the most suitable for the cultivation of vines. small or medium scale areas, as per wine production standards. Environ Monit Assess

Fig. 5 Map indicating distance from main roads per hectare of the jth farm, and N denotes the belonging to a particular group (in this case, it is repre- number of years considered. sented by productivity groups). Using this scale typolo- The productivity was then normalized as illustrated gy, a high productivity degree was identified with a in formula 2. fuzzy value ranging between 0.25 and 1 (Fig. 7). Among the approximately 12,900 ha of vineyard in Pmax P I ¼ a aj ð2Þ the study area, 7 % (about 941 ha) are highly produc- pj Pmax Pmin a a tive: these vineyards are mainly located in the munic- where Ipj is the normalized productivity index of the ipalities of Greve in Chianti (27 %), San Casciano in max ’ jth farm, Pa is the maximum value of annual aver- Val di Pesa (19 %) and Barberino Val d Elsa (10 %). age productions per hectare (refer to the total number As for the variables relative to the primary and min of farms), Pa is the minimum value of annual aver- ancillary roads, the related areas of the map of viticul- age productions per hectare (refer to the total number ture suitability were finally developed by calculating of farms), and Paj is the annual average production per the related fuzzy distance. In Fig. 8, the areas close to hectare of the jth farm. the highly productive vineyards are highlighted in To identify the most productive farms, the analysis dark red (values close to 255). was based on the use of fuzzy logic quantifiers, applying, for conversion purpose verbal fuzzy numbers, an Transition matrix appropriate scale of linguistic terms based on those proposed by Chen and Hwang (1992). The adopted The changes in land use between 1990 and 2000 are scale is based on the terms high and low degree of examined in this phase using the Markov chains. Environ Monit Assess

Fig. 6 Map of the distances from wine roads

Markov chains (Eastman and Toledano 2000), are land use at time (t)+1(1+1). For the calculation it will defined as a stochastic process in order to calculate be necessary to rely on two maps relative to the time the transition probabilities from land use at time (t)to interval selected (t and t+1) allowing therefore, the estimation of the maximum likelihood of the probability of change through Eq. 3 (Bernetti and Marinelli 2008).

Pnl;lþ1 Pl;lþ1 ¼ ð3Þ nl;lþ1 lþ1

where Pl, l+1 is the probability of transition from land use at time t(l) to land use at time (l+1) and nl,l+1 is the total number of transitions. Assuming a future development that maintains the trend occurring in the past, probabilities can then be

displayed like a stochastic matrix n×n of values pij (Table 1). As described by Coquillard and Hill (1997), the matrix obtained expresses a system that changes through discrete increments of time, in which the Fig. 7 Linguistic scale adopted value of each variable at a given time is the sum of Environ Monit Assess

Fig. 8 Map of distances from highly productive vineyards percentages of the values in the instant before. The Standards (ENV 12657) is the reference source sum of the fractions along a row of the matrix is equal adopted for data availability and consistency of to one, while the diagonal in the matrix contains the analysis. percentages of pixels that do not change between The analysis that adopted a proportional error of initial and final dates. 0.10 of the input maps4 has shown that, in the process From an operational point of view, a transition of conversion to vineyards, mainly heterogeneous ag- matrix has been created at year 2006 (Table 2) ricultural areas and forests are involved. based on changes between 1990 and 2000. Corine Land Cover European Geographic Information Implementation of the suitability maps and of the transition matrix in a cellular automata

Table 1 Probabilities matrix The suitability maps for viticulture and the transi- State i (t+1) tion matrix derived from the Markov process were

State j (t) p11 p12 p1n p21 p22 p2n 4 "The proportional error: a measure of uncertainty that can be pn1 pn2 pnn assigned to the transition probability matrix according to each land cover class. This takes into account the error in the classi- where n is the number of discrete states in the Markov chain, fication of the land use maps." (Tattoni et al. 2011). It is often set and pij is the transition probabilities (between 0 and 1) from state at 0.15-0.10 because a common value of accuracy for a land use j to state i in time interval between t and t+1. map is 85 %–90 %. Environ Monit Assess

Table 2 Transitions matrix of 2006 land use (in the present case the areas suitable for Land use at time t+1 (2006) viticulture) as to understand if the variables used (slope, primary roads, ancillary roads, and productiv- Vineyards Agr. mix. Forest Other ity) can affect changes in land use. (%) areas (%) (%) (%) The objective of the next section is the comparison between the 2006 map developed through the CA and Land use Vineyards 90 0 0 0 a referenced land use map of the same year. If the at time t Agr. mix. 48900 (2000) areas statistical analysis validates the derived map, then it is Forest 2 2 90 2 possible to assert that the variables introduced in the Other 2 2 2 90 model are an active part of the change scenario affecting the vineyards. used to implement a CA able to compute, through a Moore neighborhood of 5×5 pixels, a land use Validation of results at year 2006; based on that, calculation of the vineyards’ development can be highlighted. From The time prospect is the year 2006, since it is the most an operational point of view the final result (Eq. 4; recent year for which a land use map can be accessed Engelen et al. 2002) is represented by a determin- and is comparable to the one obtained from the CA: istic map of the land use scenario of year 2006 the source was therefore based on Corine Land Cover (derived map). as this was previously used to obtain the transitions matrix. Pj ¼ vSjZjNj ð4Þ This map allows the comparison and validation of where Pj is the land use at year x of the jth pixel, the process that led to the creation of a land use for the 5 v is a scalable random disturbance term, Sj is the year 2006, and through the use of Markov chains, to intrinsic suitability of the jth pixel (the result of a the maps of land use suitability (Fig. 9). This process weighted sum of suitability maps), Zj is the prob- has indirectly revealed the appropriateness of the se- able land use of the jth pixels deriving from lected variables involved in the land use changes in the Markov transition matrix, and Nj is the neighbor- vineyards. To this end, the Cohen (Cohen’s Kappa) hood effect assigning to the jth pixel the prevalent concordance index was used with success for its sim- land use of the neighborhood in consideration. plicity and clarity of interpretation (Carletta 1996; The final prediction map (derivation map) was de- Hagoortetal.2008; Pontius 2000) through which veloped using the information previously calculated the two compared maps have revealed a fair degree (transition matrix and suitability maps for viticulture) of conformity from both quantitative (number of pix- and by balancing the possible states of land use for els in the vineyard) and spatial points of view (i.e., each pixel. checking the exact location of each pixel in both As evoked by Berger et al. (2001a, b), all this is maps): two maps can have the same percentage of done through an iterative process that assigns each pixels for each class of land use, but different spatial pixel (with high probability of belonging) to the re- allocations of these quantities. The study of spatial spective classes of land use, in order to make use of all inconsistency is complex; it derives from a shift in the pixels under each category. A pixel that has a high position between pixels of different categories probability of realization for multiple classes is (Pontius and Schneider 2001). Index K was calculated assigned to the category having the highest value. to be equal to 0.9728. As mentioned above, the map derived is the point Moreover, the statistics about vineyards surfaces of departure to emphasize not so much the changes in confirm the good prediction of CA: derived map estimates about 13,128 ha of vineyards while reference map estimates about 12,990 ha (138 ha are probably 5 "The deterministic value is given a stochastic perturbation missing due to the rounding of the model data). There (using a modified extreme value distribution), such that most values are changed very littlebutafewarechanged are not differences in vineyards surface (between de- significantly"(Engelen et al. 2002) rived and reference map) in Greve in Chianti and Environ Monit Assess

Fig. 9 Comparison between the derived map and the reference map

Cerreto Guidi municipalities, while the main differ- This filter was applied to both the derived map of ences in vineyards surface are mainly located in land use and the reference map. This was followed by Montespertoli and Certaldo municipalities. comparing the vineyard areas with Cohen’s concor- Once a good correlation between the pixel–pixel dance index (as done previously): in this case an maps was found trough the analysis, the next step is to appropriate conformity was obtained as shown by determine a possible correlation to the geographic coefficient K00.9699. This analysis allowed us to analysis of the neighborhood of a given pixel: this validate the test variables (mostly experimental), and would allow understanding which would be the ap- confirm that the changes in the vineyards are highly propriate spatial location of each land use, and of its dependent on the slope, their proximity to the main neighborhoods. The study of the geographic neighbor- road, their proximity to the wine roads, and their hood is a delicate theme. It was approached by proximity to highly productive vineyards. Verburg et al. (2004) through a methodology enabling The results obtained from this research could be the analysis of the characteristics. used to support territorial planning decision making, The method used for this purpose is based on a and together with other GIS oriented methodologies it filter floating window (moving window) through is able to determine environmental and landscaping which the value of a given pixel is the result of critical situations. As for example the variables used analysis of the adjacent pixels (geographic neighbor- could be integrated in the study undertaken by hood). A filter of 5×5 pixels was then applied in Martinuzzi et al. (2007) that developed a classification which the central pixel is calculated as the average of schema to categorize the relative tendency to sprawl of the geographical area or adjacent neighborhood. urban developments by integrating land consumption Environ Monit Assess

(population statistics) and land development (high and appropriate tools for tackling integrated spatial policy low density development). The applied methodology problems. However, policy models come with a set of could also be an integral part of the analysis conducted requirements of their own distinguishing them rather by Lopez et al. (2001), in which land cover and land clearly from research models. Policy makers are most use change was projected forward in the future using served by models in which the time horizon, the Markov chains and regression analyses. The target spatial and the temporal resolution are policy problem could be the analysis of relationships between urban oriented and not so much process oriented as in re- growth and landscape change, and between urban search models. They need adequate rather than accu- growth and population growth. Another GIS study rate representations of the processes modeled and and complementary to the presented model is the one sketchy but integral rather than in depth and sector developed by Bernetti and Fagarazzi (2002), in which models. While research models are as complicated as they have proposed the use of multiple criteria deci- necessary and scientifically innovative, the policy sion analysis in Geographic information to evaluate maker is better served with an instrument that is as the possibilities of local development of a territory. simple as possible and scientifically proven" (Engelen The aim being to solve possible conflicts generated et al. 2002). from the alternative use of resources: in this case, our Integrated spatial models are increasingly becom- methodology brings forward a valid contribution to ing entrenched; accordingly, it is appropriate to em- analysis of possible and alternative land uses. phasize how the results depend strongly on the input It is should be stressed that, although the model data used. The study of the variables involved in the focuses on a double statistical validation, unfortunate- changes of land use is a critical point that often guides ly it does not allow overcoming other limitations on these types of analysis. Given the complexity of envi- CA. One example is the synchronous update of the ronmental systems, these variables cannot be consid- cells whereas the events that occur at the territorial ered alone, but multiple variables must be included level have different times from each other, different simultaneously in a given geographical area. speed variation and evolution and therefore require Choosing to use and validate experimental varia- different mechanisms for asynchronous update. In ad- bles involved in land use allows us to study the factors dition, the model presents a regular CA, which implies and dynamics of future evolution. The main purpose is that symmetry in the structure of a determined area is not to plan the direction and quantity of expansion (or rather unlikely to occur, given that the territory is contraction) of determined crops. made up of elements having irregular boundaries and CAs were not used to predict future scenarios, neighborhoods (Cecchini and Rizzi 2003; Blecic et al. bypassing one of the limitations that accompanies the 2005). use of the CA or the lack of statistical validation of results. The main purpose is rather to investigate the reasons behind the occurrence of such expansions (or Conclusions contractions) in the past, present and future. The selected case study represents a symbolic and traditional The study of evolutionary dynamics of the territory Italian agricultural and economic activity. The vine- cannot be done without relying on the use of spatial yard has always been regarded as one of the driving software and statistical methodologies. These, com- forces behind the primary, secondary, and tertiary eco- bined with GIS offer the opportunity to evaluate and nomic sectors. The examined area of study is one of examine the complex relationships characterizing the the most important at the national level in the field of changes in land use. To this end, one of the most viticulture, offering, therefore, the possibility to gen- recently used methods involves the use of CA to study eralize the research. Through the variables analyzed, these phenomena of transition (Candau 2000; this paper focused on the economic, social, and envi- Jenerette and Wu 2001; Sui and Zeng 2001). CA ronmental dimensions, by examining vines from dif- models can be the core of successful integrated spatial ferent aspects. models. "Model integration is a deep scientific prob- In conclusion, these variables were validated and can lem but even so a pragmatic multi-criteria, multi- be used successively to create plausible scenarios for objective problem. Integrated spatial models can be directing the actors involved in decision making Environ Monit Assess

(municipal, provincial, or other public or private levels) Eastman, J. R., & Toledano, J. (2000). Markov chain and to correct planning choices. cellular automata approaches to land cover change modeling II, workshop presentation (invited), proceeding of 4th International Conference on Integrating GIS and Environ- References mental Modeling, Banff, Alberta, Sept. 2–8, 2000. Engelen, G., White, R., Uljee, I., & Drazan, P. (1995). Using cellular automata for integrated modelling of socio- Al-Ahmadi, K., See, L., Heppenstall, A., & Hogg, J. (2009). environmental system. Environmental Monitoring and Calibration of a fuzzy cellular automata model of urban Assessment, 34,203–214. dynamics in Saudi Arabia. Ecological Complexity, 6,80– Engelen, G., Geertman, S., Smits, P., & Wessels, C. (1999). 101. Dynamic GIS and strategic physical planning: a practical Berger, T., Goodchild, M., Janssen, M. A., Manson, S. M., Najlis, application. In J. Stillwell, S. Geertman, & S. Openshaw R., & Parker, D. C. (2001a) Part 2: Methodological consider- (Eds.), Geographical information and planning (pp. 87– ation for agent-based modeling of LUCC — agent based 111). Berlin: Advances in Spatial Science. Springer. models of LUCC, Report and Review of an International Engelen, G., White, R., Van der Meulen, M., & Hahn, B. Workshop October 4–7, 2001, Irvine, California, USA, LUCC (2002). Sustainable developments of islands: a policy sup- Report series No. 6, (pp. 7–26). On line available http:// port framework for the integrated assessment of socio- www.globallandproject.org/arquivos/LUCC_No_6.pdf. economic and environmental development. In H.-H. M. Accessed 15 Oct 2012. Hsiao, C.-H. Liu, & H.-M. Tsai (Eds.), Sustainable devel- Berger,T.,Couclelis,H.,Manson,S.M.,&Parker,D.C. opment for island societies: Taiwan and the world (p. 251). (2001b) Part 1: Introduction and conceptual overview — Taipei: Asia-Pacific Research Program, Academia Sinica agent based models of LUCC, Report and Review of an and SARCS Secretariat. International Workshop October 4–7, 2001, Irvine, Califor- Hagoort, M., Geertman, S., & Ottens, H. (2008). Spatial exter- nia, USA, LUCC Report series No. 6, (pp. 1–6). On line nalities, neighbourhood rules and CA land-use modelling. available http://www.globallandproject.org/arquivos/ Annals of Regional Science, 42,39–56. doi:10.1007/ LUCC_No_6.pdf. Accessed 15 Oct 2012. s00168-007-0140-8. Special issue paper. Bernetti, I., & Fagarazzi, C. (2002). L’impiego dei modelli Jenerette, G. D., & Wu, J. (2001). Analysis and simulation of multicriteriali geografici nella pianificazione territoriale. land-use change in the central Arizona, Phoenix region, Aestimum, 41,1–26. USA. Landscape Ecology, 16,611–626. Bernetti, I., & Marinelli, N. (2008). L'impiego degli automi Lombardo, S., Pecori, S., & Petri, M. (2005) Investigating cellulari per la costruzione di scenari di cambiamento del- territorial dynamic using decision tree. Atti del convegno l'uso del suolo. Aestimum, 52,1–30. Cupum 2005, paper 90, (pp. 1–16). On line available http:// Blecic, I., Cecchini, A., & Tronfio, G. A. (2005) Costruire scenari 128.40.111.250/cupum/searchpapers/papers/paper90.pdf. futuri per la pianificazione territoriale strategica: Metodo- Accessed 15 Oct 2012. logie e strumenti. In Cecchini A., Plaisant A. (a cura di), Lopez, E., Bocco, G., Mendoza, M., & Duhau, E. (2001). Analisi e modelli per la pianificazione – Teoria e pratica: Predicting land-cover and land-use change in the urban lo stato dell’arte. FrancoAngeli editore, (pp. 259–270). fringe. A case in Morelia City, Mexico. Landscape and Candau, J. (2000). Calibrating a cellular automaton model of Urban Planning, 55, 271–285. urban growth in a timely manner. In B. O. Parks, K. M. Martinuzzi, S., Gould, W., & Ramos, G. O. (2007). Land de- Clarke, & M. P. Crane (Eds.), Proceedings of the 4th velopment, land use, and urban sprawl in Puerto Rico International Conference on Integrating Geographic Infor- integrating remote sensing and population census data. mation Systems and Environmental Modeling: Problems, Landscape and Urban Planning, 79(2007), 288–297. Prospects, and Needs for Research. Boulder: University of Messina, J. P., & Walsh, S. J. (2001). 2.5D morphogenesis: Colorado. modeling land use and landcover dynamics in the Ecua- Carletta, J. (1996). Assessing agreement on classification tasks: dorian Amazon. Plant Ecology, 156,75–88. the kappa statistic. Computational Linguistics, 22, 249– Pijanowski, B. B., Brown, D. G., Shellito, B. A., & Manik, 254. G. A. (2002). Using neural networks and GIS to fore- Cecchini, A., & Rizzi, P. (2003) Perché gli automi cellulari cast land use changes: a Land Transformation Model. sono uno strumento utile della scatola degli attrezzi per Computers, Environment and urban systems, vol. 26, il governo del territorio dell’urbanista del nuovo mil- n. 6, pp. 553–575, Elsevier. lennio. In S. Lombardo (a cura di) Ingegneria del Pontius, R. G., Jr. (2000). Quantification error versus location territorio ed ingegneria della conoscenza. Alinea error in comparison of categorical maps. Photogrammetric editore. Engineering and Remote Sensing, 66(8), 1011–1016. Chen,S.J.,&Hwang,C.(1992).Fuzzy multiple attribute Pontius, R. G. J., & Schneider, L. C. (2001). Land cover change decision making. Berlin: Springer-Verlag. model validation by an ROC method for the Ipswich wa- Coquillard, P., & Hill, D. R. C. (1997). Modélisation et simula- tershed, Massachusetts, USA. Agriculture, Ecosystems & tion d’écosystèmes. Paris: Masson. Environment, 85(I.1–3), 239–248. Dear, M. (1977). Spatial externalities and locational conflict. In Smith, J., & Smith, P. (2007). Environmental modelling, an D. B. Massey & P. W. J. Batey (Eds.), Alternative frame- introduction. Oxford: Oxford University Press. works for analysis (pp. 152–167). London: Pion (Papers in Sui, D. Z., & Zeng, H. (2001). Modeling the dynamics of Regional Science 7). landscape structure in Asia’s emerging desakota regions: Environ Monit Assess

a case study in Shenzhen. Landscape and Urban Planning, Ward, D. P., Murray, A. T., & Phinn, S. R. (2000). A stochas- 53,37–52. tically constrained cellular model of urban growth. Com- Tattoni, C., Ciolli, M., & Ferretti, F. (2011). The fate of priority puters, Environment and Urban Systems, 24, 539–558. areas for conservation in protected areas: a fine-scale Mar- White, R., & Engelen, G. (2000). High-resolution integrated mod- kov chain approach. Environmental Management, 47, 263– elling of the spatial dynamics of urban and regional systems. 278. doi:10.1007/s00267-010-9601-4. Computers, Environment and Urban Systems, 24,383–400. Torrens, P. M., & O’Sullivan, D. (2001). Editorial. Cellular White, R., Engelen, G., & Uijee, I. (1997). The use of con- automata and urban simulation: where do we go from here? strained cellular automata for high-resolution modelling of Environment and Planning B: Planning and Design, 28, urban land-use dynamics. Environment and Planning B: 163–168. Planning and Design, 24, 323–343. Verburg, P. H., De Nijs Ton, C. M., Van Eck, J. R., Visser, H., & Wu, F. (1998). Simulating urban encroachment on rural land De Jong, K. (2004). A method to analyse neighbourhood with fuzzy-logic-controlled cellular automata in a geo- characteristics of land use patterns. Computers, Environ- graphical information system. Journal of Environmental ment and Urban Systems, 28(2004), 667–690. Management, 53, 293–308.