An overview of farming system typology methodologies and its use in the study of pasture-based farming system : a review

The main objective of the paper is to do a critic study of the use of typology methodologies within pasture-based farming systems (PBFS), especially those situated in less favoured areas, showing in each case the more relevant variables or indicators determining the farming system classification. Another objective is to do an overview of the most used farming system typology methodologies in general. First some considerations about the concept of farming system and approaches to its study have been done. Next, the farming system typology methodologies have been showed in general to different farming systems, but addressed preferably to PBFS. The different tools integrated in these methodologies have been considered: sampling methods, sources of data, variables or indicators obtained from available data and techniques of analysis (statistical or not). Methods for farming system classification have been presented (expert methods, analytical methods or a combination of both types). Among the statistical methods, the multivariate analysis has been overall treated, including the principal component analysis and the cluster analysis. Finally, the use of farming system typology methodologies on different pasture-based farming systems has been presented. The most important aspects considered are following: the main objective of the typology, the main animal species, the employed methods of classification and the main variables involved in this classification. Additional key words: livestock systems; grazing; multivariate analysis; expert methods; farming typologies; GIS. * Corresponding author: w.madry@agrobiol.sggw.waw.pl; dariusz_gozdowski@sggw.pl Received: 02-07-12. Accepted: 06-03-13 Abbreviations used: CA (cluster analysis); FADN (Farm Accountancy Data Network); FSS (farm structural survey); FST (farming system typification); GIS (geographical information systems); HNV (high nature value); LFA (less-favoured areas); PBFS (pasture-based farming systems); PC (principal components); PCA (principal component analysis). Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria (INIA) Spanish Journal of Agricultural Research 2013 11(2), 316-326 Available online at www.inia.es/sjar ISSN: 1695-971-X http://dx.doi.org/10.5424/sjar/2013112-3295 eISSN: 2171-9292


Introduction
In animal production, the intensive concept is related to the maximisation of productivity of the scarcest factors as land, labour force or bought feed (Bernués et al., 2011).The pasture-based farming systems (PBFS) are generically termed less intensive because of their link to natural resources and are more complex than the off-land livestock farming systems.These last are more intensive and traditionally studied and evaluated through technical-economic indicators while other kind of indicators should be additionally used in the PBFS study, which have great ecological, landscape and cultural diversity, besides technicaleconomic aspects, both environmental and social aspects have also high relevance.On the other hand, these systems play a central role in the management and conservation of large High Nature Value (HNV) farmland in Europe, and are mostly located in less productive areas, i.e., southern Europe and mountainous regions (Bernués et al., 2011).Bernués et al. (2011) state that PBFS are also called grazing, pastoral, agro-pastoral or agro-silvo pastoral systems.Different kinds of PBFS can be found including grazing in rangelands, grazing in cultivated pastures, grazing in stubbles and zero grazing.A high An overview of farming system typology methodologies and its use in the study of pasture-based farming system: a review W. Ma˛dry 1 *, Y. Mena 2 , B. Roszkowska-Ma˛dra 3 , D. Gozdowski 1 , R. Hryniewski 1 and J. M. Castel 2 proportion of PBFS are situated in the less-favoured areas (LFA), especially when animals graze in rangelands.To evaluate the relative marginality of LFA, an index is computed in a Geographic Information System (GIS).The index combines land use, demographic and socio-economic data (Bertaglia et al., 2007).Different authors have studied on the LFA sustainability.Bertaglia et al. (2007) have contributed to the delimitation and categorisation of LFA.Numerous authors have obtained results in order to promote and stimulate multifunctional and sustainable development of LFA (Meert et al., 2005;Van der Ploeg et al., 2009).The PBFS, which as stated are frequent in these areas, have higher diversity of characteristics and consequently a higher complexity and heterogeneity in relation to the off-land farming systems.This makes very necessary and interesting the farming system typification (FST) of this type of systems.This analysis can contribute to know these systems and increase the global improvement and promotion possibilities of these areas.
The main objective of the paper is to do a critic study of the use of typology methodologies within PBFS, especially those situated in LFA, showing in each case the more relevant variables or indicators determining the farming system classification.Another objective is to do an overview of the most used farming system typology methodologies in general.

Concept of farming system and approaches to its study
A farming system is def ined as a population of individual farm systems that have broadly similar resource bases, enterprise patterns, household livelihoods and constraints, and for which similar development strategies and interventions would be appropriate (Dixon et al., 2001;Köbrich et al., 2003).Depending on the scale of the analysis, a farming system can encompass a few dozen or many millions of households.A farming system is seen as comprising the totality of production and consumption decisions taken by a farm-household, including the choice of crop, livestock and non-agricultural both on-farm and off-farm enterprises, and food consumed by the household and each farm has its own, unique farming system (Köbrich et al., 2003).
There is a wide diversity of farming systems among a group of farms, not only on a large-scale in terms of geographic space but also within restricted rural areas or more oriented types of these systems (Castel et al., 2003(Castel et al., , 2010;;Carmona et al., 2010;Van de Steeg et al., 2010).Then, a farming system includes the conjoint use of crops, grass and livestock (Engledow et al., 1978) and in order to synthesize, it can be defined as an economic and agricultural concept holistically describing (as a whole, based on a set of many variables and indicators) a farm household in terms of agricultural land use, i.e., the systems of crop and livestock production, non-agricultural economic activities of farm household members (on-farm and off-farm activities), the income generated and the structure.Moreover, the natural, social, economic, infrastructural and institutional resources and environments that determine all economic activities occurring in the farming system should be considered (Kostrowicki, 1977;Keating & McCown, 2001;Köbrich et al., 2003;Bertaglia et al., 2007).As a level of organization, farming systems mediate the relationship between farmers and the landscape by bringing together biophysical, economic, social, cultural, and political factors that constrain or promote the land use and land cover change (Carmona et al., 2010;van de Steeg et al., 2010).
Every farming system is different, if not unique, facing distinctive decision-making problems, whose solutions could also be unique.Consequently, it becomes necessary to classify or group farms in some way (typif ication) - (Lesschen et al., 2005).Such groups (clusters) are constituted by roughly homogenous farmers with similar circumstances for whom we can make more or less the same recommendation (Köbrich et al., 2003).The farming system typification (FST) helps to know the characteristics of different farming systems found on a farmland.The knowledge of the strengths and weaknesses of each system type (diagnosis) allows for some improvement strategies.
To carry out a FST, previously objectives of them should be established and elements or subsystems which integrate systems should be found.In this sense, Spedding & Brockington (1976) stated that the kind of information needed to construct a FST model could only be specified by examining their role or purpose.Each subsystem is related to farm structure, technology, social relations, interrelation between values and objectives of people integrated in the system and planning, organization and control of the system.Keating & McCown (2001) recognized two key components of farming systems, namely the biophysical 'Production System' of crops, pastures, animals, soil and climate, together with certain physical inputs and outputs, and the 'Management System', made up of people, values, goals, knowledge, resources, monitoring opportunities, and decision making.These authors stated that six types of farming systems analysis and intervention had been reviewed, which have evolved over the last 40 years, namely: (1) economic decision analysis based on production functions, (2) dynamic simulation of production processes, (3) economic decision analysis linked to biophysical simulation, (4) decision support systems, (5) expert systems, and (6) simulation-aided discussions about management in an action research paradigm.These different modes of study and analysis of the systems are destined to better know them.After this analysis, if farm systems are suff iciently heterogeneous, a classification can be done, looking for weaknesses and strengths in order to obtain a diagnostic which allows establishing some system changes or improvements (Köbrich et al., 2003).
Also, the principal components analysis can be added like a method for farming systems study (Lesschen et al., 2005).This method which later is more detailed is a part of multivariate analysis, being a preliminary step before the farming systems classification (cluster analysis).Principal components include diverse variables or indicators that determine the more important differences between the different groups which go to be done through the cluster analysis.

Farming system typology methodologies
The farming system typology methodologies showed in this section in general could be addressed to particular more general (major) kinds of farming systems (Caballero, 2001;Castel et al., 2003;2010;Pardos et al., 2008;Gaspar et al., 2011).However, references have been addressed preferably to pasturebased farming systems, overall those situated in less favoured areas.Before the study of farming system typology the selection of the aim of such a study should be made.Next, different tools should be considered: sampling methods, sources of data, variables or indicators obtained from available data and analysis techniques (statistical or not).

Size and characteristic of farm samples
The size and characteristics of the samples of farms must be related to the number of farms and characteristics of a study area.Samples should include a number of farms of each geographical area (Caballero, 2001;Usai et al., 2006;Castel et al., 2010;Marey-Pérez & Rodríguez-Vicente, 2011) and/or each economic activity structure, in the case where the farms are multifunctional (Gaspar et al., 2008), and each type of farm, in relation to aspects considered important in the classif ication (size, productive aptitude, gender of workers, etc.) should be proportionally represented (Ruiz et al., 2008).Regarding the size of the sample, this should be greater when studying large areas, regions or countries (Riveiro-Valiño et al., 2009;Acosta-Alba et al., 2012) that when studying specific areas (Pardos et al., 2008;Gaspar et al., 2008).In addition, the sample should have a larger size in a basic study of an area or region (Castel et al., 2003) than in a specif ic study of an already characterized area, especially when in this last area the used variables are quantitative and measurements are reliable (Ruiz et al., 2008).
The data can come from the regional (Riveiro-Valiño et al., 2008, in Galicia-Spain), national or supranational administration [Andersen et al. (2007) in the EU-15;Caballero et al. (2008) in 27 cases of European large scale grazing systems belonging to different countries; Caballero et al. (2009) doing a spatial distribution of large scale grazing systems in three Mediterranean countries].Data can be more explicit when the study is referred to one particular system; that occurs for instance in a study of typology of cereal-sheep farming systems conducted by Caballero (2001) in Castile-La Mancha, wherein the average results are expressed for the entire region and according to each province and the evolution of farming system characteristics between 1989 and 1997 can be observed.Kempen et al. (2011) conclude that at this moment the European wide farm information is only available at a rather aggregated administrative level, for example Farm Accountancy Data Network (FADN) and Farm Structural Survey (FSS) or other state or local administration organisations (Keating & McCown, 2001;Lesschen et al., 2005;Clavel et al., 2011).Additionally, geographical information systems (GIS) can deliver geographic, natural and ecological data that are appropriate for the typology of landscape complexity and farming systems at a spatial area unit level (Bertaglia et al., 2007;Jellema et al., 2009).But this kind of data cannot provide all the data on farm management that are needed for the modeling at farm level in EU.Examples of data that are not available in these European datasets are fertilizer and livestock management practices.It is therefore envisaged that this kind of information will have to be collected by consulting farm advisors and researchers at regional or national level (Andersen et al., 2007).Likewise, a lack of differentiated data about the real situation of farms is cited by Riveiro-Valiño et al. (2009) in order to use a program for working in the development of a decision support system for Agricultural Production Planning in Galicia, Spain.Furthermore, the data sometimes have different units of measure when coming from different countries.In some cases it is possible to transform the variables to achieve results comparing different countries (Ruiz et al., 2009).
In the case of livestock farming systems, Gibon et al. (1999) suggested that conventionally recorded information about the structure of livestock farms and the usual technical-economic indicators do not provide suff icient information for sustainable livestock development purposes and by consequence, researchers usually include a characterization of the diversity of the organization and operation of farming systems by identifying objectives and practices in family farms.Some examples of data collection in livestock farming systems using a structured direct questionnaire for indepth interviews, including questions properly selected to obtain a general description of farm characteristics and overall management practices are the following: Daskalopoulou & Petrou (2002) and Gelasakis et al. (2012) in Greece; Castel et al. (2003) in Andalusia, Spain; Gaspar et al. (2008) in Extremadura, Spain, Marey-Pérez et al. (2011) in Galicia, Spain;Paz et al. (2005) in Argentina; Usai et al. (2006) in Italy; Milán et al. (2006), Castel et al. (2010) in Poland and Carmona et al. (2010) in Chile.

Variables or indicators
Variables representing all the essential inputs and outputs combined with all others, representing social, operational, production and structural attributes of agriculture, provide a basis for identif ication of agricultural (or farming) type systems (Kostrowicki, 1977).This author situated variables related to the characteristics of agriculture in the following groups: (i) social and ownership characteristics (provides answers to such questions as who is the landowner, the holding operator or the decision-maker and what is the scale of operation); (ii) operational (organizational and technical) characteristics (explains what the labour and capital inputs are and how the holding is operated); (iii) production characteristics (discloses how much is produced and for what purpose); and (iv) structural characteristics (answer questions about the proportion of land used for different purposes, about the proportion in which various farm animals are raised, and about how much is sold or delivered off the farm).
The heterogeneity of production system and the difference of difficulty level to collect data in different areas made it necessary structuring indicators in different levels (Toussaint, 2002): level 1 which considers the minimum descriptive elements to identify the production system structure, level 2 which considers quantitative data, level 3, which includes more detailed measures, both quantitative and qualitative, which explain some results of previous levels, and level 4 which includes certain necessary data in the special studies.Toussaint (2002) classified variables in structural and operational.The latter are more important to study the most developed systems, while the former are more important in most primary systems.The variable selection for characterization depends of production system management and objectives, as other factors, but depends over all of information characteristics that can be found in farms and in the farm environment.When the information is scarce qualitative variables are mainly used, being that frequent in level 1 stated by Toussaint (2002), but it occurs also in any cases in the level 2 when qualitative variables are important because of different options are found in farms, for instance in farm activities, characteristics of rangelands, type of feed supplied, presence of infrastructures, animals performance, type of reproduction management (Castel et al., 2003).In some cases, the qualitative variables can be transformed into quantitative expressed in percentages, for example Castel et al. (2010) do this to express the quality of soil in each farm.However, whenever possible quantitative variables are used because they are more accurate.In fact, quantitative variables are the most used by of researchers for making multivariate Typology methodologies to study pasture-based farming systems: a review analysis: Milán et al. (2006) and Gaspar et al. (2008) in dehesa agro-ecosystems, Ruiz et al. (2008) in dairy goat systems and Pardos et al. (2008) in meat sheep systems.The more frequent quantitative variables used in livestock production system multivariate analysis are related to the following aspects: surface size, flock size, animal density, workforce, feed supply and productivity, among technical variables, and incomes, expenses and margins, among economical aspects (Milán et al., 2006;Gaspar et al., 2008;Ruiz et al., 2008).Variables with a medium-high variability coeff icient should be selected for the multivariate analysis.However, when some of them are statistically correlated, only some of the correlated being more important, taking into account the expert criteria, should be used in the analysis (Hair et al., 1998;Lesschen et al., 2005).

Techniques of typology classification analysis
According to Engledow et al. (1978), ideally, a classification should be easy to make, the basic data should be readily obtainable and easily manipulated and analysed and the degree of precision should be adjustable to meet the criteria required.The best typology of farms will have to show a maximum amount of heterogeneity between the types, while obtaining maximum homogeneity within particular types or categories (Escobar & Berdegué, 1990).
Assessment of the diversity of farming systems and their typology in rural areas can be performed using expert methods (Clavel et al., 2011) or analytical methods (Riveiro-Valiño et al., 2008;Carmona et al., 2010).The expert methods were the first to be used as tools in order to def ine concrete areas (Escobar & Berdegué, 1990) and are based on skilled knowledge supported by land cover maps, which guide researchers or agricultural extension experts, and all available official synthetic information collected by state and local administrations (Clavel et al., 2011).An example of this kind of method was presented by Daskaloupolou & Petrou (2002) who used an empirically method to identify the main types of Greek farms in terms of their structure, evolution and adjustment strategies.Another example of expert-based method was presented by Acosta-Alba et al. (2012) who made a classification of dairy farms in Brittany (France) as a function of mode of production, quantity of milk production, and fodder-crop and grass area.
Taken into account that a classification should be made from tests of the data rather than by f irst arbitrarily def ining the groups and then f itting the individual farms to them (Escobar & Berdegué, 1990), the expert methods have frequently been replaced by more formal and reliable analytical (statistical) methods, referred to as analytically based farming system typology or statistical farming system typology (Köbrich et al., 2003;Lesschen et al., 2005) or has been used as complementary tools together with analytical methods (Carmona et al., 2010;van de Steeg et al., 2010;Clavel et al., 2011).However, sometimes, an expert method can also be the subject of statistical analysis (Caballero et al., 2008).
Concerning to analytical methods, statistical ones can be applied at the level of administrative units such as rural sub-districts (Bertaglia et al., 2007) or to other area units covering a certain rural area (Jellema et al., 2009;van de Steeg et al., 2010).But, these methods are especially important at the farm level for identifying the diversity of farming systems and their typology (Köbrich et al., 2003;van der Ploeg et al., 2009;Castel et al., 2010).Some examples for the PBFS are the following: Castel et al. (2003Castel et al. ( , 2010)) The more used statistical typology classification methods are the multivariate ones which include two steps: the principal component analysis (PCA) and the cluster analysis (CA) - (Escobar & Berdegué, 1990;Paz et al., 2005).The main objective of the first method (PCA) is to reduce a usually fairly large number of diagnostic variables included in an analysis to a considerably more limited number of formal variables, referred to as principal components (Lesschen et al., 2005).Through the PCA data for the examined diagnostic variables (input variables) are transformed into a small set of new synthetic variables -dimensions, principal components (PC) or principal coordinates-with little loss of information (Escobar & Berdegué, 1990).In many cases, it is possible to reduce a large number of farming system variables to two or three PC with relatively little loss of information (accounting for the total variation among the studied entries to a high degree with the first two or three PC).These f irst PC usually contain the most essential information characterising the diversity of entries in the studied set, they are often used as variables in CA (Lesschen et al., 2005;Castel et al., 2010).Three kinds of PCA can be applied: correspondence analysis, optimal scaling analysis and factorial analysis.The correspondence analysis is used to reduce the dimensionality of the variation system when diagnostic variables are categorical (also referred to as qualitative), including nominal, ordinal and interval variables (Benzécri, 1992;Castel et al., 2003;Asselin & Anh, 2005;Moreno-Pérez et al., 2011).The optimal scaling analysis introduced by Bock (1960) is used when the variables analyzed are both qualitative and quantitative.In the analysis process, the quantitative variables were f irst stratif ied to convert them into qualitative variables.The more used principal component analysis is the factorial analysis which need that all the input variables are quantitative.In order to identify the initial indicators with extracted factors (components) a Varimax rotation was done, which allows original indicators to be easily located in the extracted values (Escobar & Berdegué, 1990;Paz et al., 2005;Ruiz et al., 2008).
The CA (Everitt, 1980) is a more used classification analysis method and has been used by numerous authors (Kostrowicki, 1977;Castel et al., 2003Castel et al., , 2010;;Köbrich et al., 2003;Carmona et al., 2010).CA allows distinguishing homogenous groups (clusters) of entries with respect to many quantitative and categorical variables.For quantitative continuous and some discrete variables, the Euclidean distance or squared Euclidean distance has most commonly been used.In CA of entries to identify farming system typology, a hierarchical procedure known as Ward's method has frequently been used.Ward's method is a hierarchical clustering method that usually provides a clear division of entries into homogenous groups (Köbrich et al., 2003;Lesschen et al., 2005;Castel et al., 2010).The CA can be made from input variables but if a PCA has been made before, the obtained dimensions in this first analysis should be used as variables (Hair et al., 1998;Lesschen et al., 2005).The number of groups distinguished should not be too large.Researchers usually restrict themselves to distinguishing a few (typically 3 to 6) groups (clusters) of the studied entries that are homogeneous in terms of the multi-variable criterion used for discrimination of the types of these systems (Castel et al., 2003(Castel et al., , 2010;;Köbrich et al., 2003;Milán et al., 2006;Gaspar et al., 2008).Therefore, the number of distinguished groups of entries in one research case would not be optimal in other circumstances (Köbrich et al., 2003;Lesschen et al., 2005).However, the number of groups can be predefined by the scientist (k-means cluster analysis) - (Castel et al., 2003(Castel et al., , 2010(Castel et al., , 2011;;Ruiz et al., 2009) or undefined (hierarchical cluster analysis) - (Paz et al., 2005;Usai et al., 2006).The graphical result of grouping entries via hierarchical methods of agglomeration is a dendrogram and, in this case, scientist decides the proper level of similarity to identify groups of entries showing suff iciently discriminated types of farming systems in a studied area.The hierarchical procedure allows much more flexibility in a CA: it is possible to use any of a number of distance or similarity measures, including options for binary and count data, and there is no need to specify the number of clusters a priori (Escobar & Berdegué, 1990).Other statistical methods can also be used to classify farming systems, such as the discriminant analysis or the combinatorial method.The discriminant analysis is a multivariate procedure used to model the value of a criterion factor (dependent variable) on the basis of its relationship with one or more predictors (independent variables) when these relationships and their nature are not completely expected a priori (Marey-Pérez & Rodríguez-Vicente, 2011).The combinatorial method used by Riveiro-Valiño et al. (2008) in Galicia (Spain) to determine the different types, production sizes, production systems and location of farms based on data from the agricultural census.
After a farming system classification a validation it is necessary.For instance, if a CA has been made, the typology can be validated by statistical methods such as discriminant analysis, from which gain functions that calculate the probability of the farms belonging to the groups obtained.Whatever the statistical method used to classify, it can be validated using an empirical method by checking if the classification is adapted to the actual conditions.In any case, the results of the classification of farms must be similar regardless of the methods used to make the typology (Escobar & Berdegué, 1990;Riveiro-Valiño et al., 2009).
Once the different clusters are obtained, they can be described and afterward compared using one-way ANOVA or a chi-squared test.In this way, Clavel et al. (2011) and Kempen et al. (2011) stated that the obtained outcomes allow characterization of the distinguished groups of entries exhibiting similar farming systems and determination of the spatial locations of the groups.This information is particularly important because it allows one to evaluate the spatial distribution of farms with particular type of system.To evaluate the existence of spatial autocorrelation (i.e., whether farms with the same farming system occur in geographic proximity), spatial statistics are necessary, e.g., analysis based on Moran's I or Geary's c for univariate data or Mantel correlograms for multivariate data (Lesschen et al., 2005).The combination of classical methods of statistical analysis and spatial statistics allows a comprehensive inference to be obtained and should be recommended for research on farming system typology.
Next, a brief of some works made using different farming typology methods is made.Hazeu et al. (2011) in order to know the state and trends of the European environments exposed five new European typologies using the spatial environmental datasets.Different methods have been used in each typology: multivariate statistical clustering, multi-scale segmentation (that recognizes objects based on spatial characteristics, the most often used technique in the interpretation of satellite imagery) and f inally, methods based on geographical coherence and on expert knowledge.Jellema et al. (2009) presented a methodology for landscape character assessment using the pattern of landscape features as stored in a GIS to delineate, characterize and evaluate landscapes using a region growing algorithm.They applied this methodology in the north of the Netherlands and stated that the obtained classification was more consistent than the expert classif ication of the same region.This methodology was proposed as support for spatial planning processes and policy development for landscape conservation.Carmona et al. (2010) constructed a spatial typology of farming systems and assessed their influence on the extent and spatial distribution of deforestation, forest re-growth, and agriculture expansion in southern Chile between 1999 and 2007.Kempen et al. (2011) presented an approach to spatially allocate farm information to a specific environmental context which makes possible to aggregate farm types both to natural and to lower scale administrative regions.These authors used multiple sources of information and obtained three allocations of farm types: according to farm size, according to farm intensity and according to farm specialization.
To finish, an example of styles of farming typology is showed (Sturaro et al., 2009).These authors studied the relationships between livestock systems, landscape maintenance and farming styles in the Eastern Italian Alps.Farm information was collected on technical and productive aspects, on landscape features of land managed, which was identified by aerial photographs and digitized in a GIS environment, and on the farmers' background, attitudes and approach to farming.Six different livestock systems have obtained according intensification level (expert method).After, four different farming styles have been obtained through a hierarchical CA.The different farming styles are distributed across all the livestock systems, and authors stated that this lack of a linkage between the livestock systems and the way the farms are run has the important implication that informative knowledge to address policy decisions needs to integrate the definition of livestock systems with the assessment of farming styles.

Use of farming system typology methodologies on different pasture-based farming system
This section presents different farming system typologies, considered at farm level, studied by different authors.For each one, the main objective, the main animal species and his aptitude, the employed method of classif ication and the main variables involved in the classification are showed (Table 1).
As said, in general an analytical methods are used for obtain a farming system typology.However, in some cases an expert based method is used.This kind of method is use overall when farming systems are less developed (Perovolotsky, 1990) or when instead farming systems are highly evolved and consequently they are well known by experts (Acosta-Alba et al., 2012. Perevolotsky (1990) stated that moreover of ecological and agricultural factors, the type of land tenure and the human factors (workforce, cooperation, etc.) determine the characteristics of livestock operations.Also, the production and the economy being basically of subsistence, the more or less commercial activity of farmers is an important issue.In the case of Acosta-Alba et al. (2012), authors modeled the land use configuration for a sustainable dairy sector in Brittany (France) and explored sustainable farming scenarios at a regional scale.
Sometimes, in the less developed farming systems analysis, when these are a few more known by the scientist, a PCA method is used for the typification PCA (1) and CA (2)  (89 farms) Numerous variables related to all the system (2) k-means cluster analysis.
analysis (previous to a CA).In this case, the more proper PCA is a correspondence analysis (Castel et al., 2003) or an optimal scaling method (Castel et al., 2011).But, whatever are the method (expert o analytical) used for the less developed farming systems classif ication, it is necessary to be account of qualitative variables are important, for instance the importance of each farmer activity, the animal milk aptitude, the reproductive behavior of animals, the production destination, etc. (Perovolotsky, 1990;Castel et al., 2003Castel et al., , 2011)).
In the case of more developed livestock systems, the more used typology method is a multivariate analysis that includes a factorial analysis (a kind of PCA) and a CA (Table 1).The more is the heterogeneity of farming systems (which are less known by the scientist) more input (observed) variables should be chosen to the typology of these systems and more of the variables contribute substantially to main principal components.For instance Castel et al. (2003), Usai et al. (2006), Milán et al. (2006) and Gelasakis et al. (2012) use numerous variables in PCA, mainly due to their structural nature, while Ruiz et al. (2008) and Castel et al. (2011) use lesser number and they refer to the farm dimension, farm management and animal productivity.At the same time, it occurs that the more is the heterogeneity of farming systems more should be the farm sample size.When data come from a monitoring process (for instead monthly), this size can be lower (for instance in the work of Ruiz et al., 2008).Anyway, quantitative variables included in PC are frequently related to the farm size (surface and flock) - (Castel et al., 2003, Pardos et al., 2008;Milán et al., 2006;Ruíz et al., 2008Ruíz et al., , 2009)).The variables related to the presence of different animal species (indicating farmer pluri-activity) are also frequently used in the PCA (Usai et al., 2006;Nahed et al., 2006;Gaspar et al., 2008;Castel et al., 2010).
Many authors in the farming system typology studies at the farm level consider technical and also economic indicators (Serrano et al., 2004b;Nahed et al., 2006;Pardos et al., 2006), but in some cases authors give more importance to the economic aspects (Gaspar et al., 2008;Ruiz et al., 2008).On the other hand some authors give more importance to sociological (Maseda et al., 2004) or to environmental aspects.Nahed et al. (2006), through a MESMIS analysis after applying a multivariate analysis, take into account the overall sustainability.Many authors after presenting the strengths and weaknesses of the different obtained groups, provide strategies for improving production systems.For example, Ruiz et al. (2008) proposed improvements in the grazing use and labor management as well as the increase of farmer's advices and Gelasakis et al. (2012) established four different farmer attitudes related to the on-farm feed production and the extent of pasture which aim to achieve the decrease of the production costs.However, some authors present the strengths and weaknesses of different obtained clusters (diagnosis) but propose making more studies to draw up such strategies (Castel et al., 2010).To finish, it should be noted that more specific aspects related to the livestock production system typology such as the production of cheeses (Gaspar et al., 2011) or the organic production in dairy goat Spanish have been recently considered (Mena et al., 2012).

Conclusions
The first step in the application of farming system typology methodologies should be to analyze the objectives of the study.They may relate to different areas: a region, a country, the EU or simply at farm level.The studies at farm level are rather specific and they use data taken from surveys or data collection systems (annual or periodic).
Studies at farm level can refer to all elements of the system (sub-systems) or only some of them.The more primary systems are in general more unknown by scientists and more parts thereof should be considered and more variables should be engaged to establish a classification of farms.On the other hand, qualitative variable are of special importance in this kind of systems.
In typology studies at farm level for PBFS, often methods that take into account the statistical analysis are used.The most common used is the multivariate analysis (a CA after a PCA).
Further progress in the use of new technologies (GIS, etc.) to support the typology farming system methodologies should be done.These, together with eff icient collaboration of scientists from different countries in the sharing of the indicators used, contribute to the results of the analysis are more useful for everyone.When making large-scale studies (national or supranational) a special care should be taken in the results application.It is necessary to distinguish between different situations.In general, to farming system classif ication, we should try to integrate some environmental or sociological variables, as well than the technical and economic, in the study of typologies.

Table 1 .
Examples of use of farming system typology methodologies on different pasture-based farming systems