An overview of the recent approaches to terroir functional modelling, footprinting and zoning

. Notions of terroir and their conceptualization through agro-environmental sciences have become popular in many parts of world. Originally developed for wine, terroir now encompasses many other crops including fruits, vegetables, cheese, olive oil, coffee, cacao and other crops, linking the uniqueness and quality of both beverages and foods to the environment where they are produced, giving the consumer a sense of place. Climate, geology, geomorphology and soil are the main environmental factors which make up the terroir effect on different scales. Often considered immutable culturally, the natural components of terroir are actually a set of processes, which together create a delicate equilibrium and regulation of its effect on products in both space and time. Due to both a greater need to better understand


Introduction
A search using the keyword "terroir" in the Scopus database finds 385 papers published from 1980 to 2014 (September) including a steady rise from 2005 to 2014. This trend provides evidence of the ever-growing interest of the scientific community in understanding the characteristics and relationships between the many factors of terroir. "Terroir" is a French word, meaning delimited areas with homogeneous environmental features that are likely to confer typical wine qualities identified through collective memory and conveyed from generation to generation within a territory marked by social context and cultural technical choices (Vaudour, 2002(Vaudour, , 2003. The tradition of terroir wines has strong cultural connections, referring to tradition of drinking well, of farming and of producing typical wines that are rooted in a region or made from specific places with organoleptic features easily distinguishable from other wines from other regions. In France these are sometimes named "crus", "clos" and, in the case of Burgundy, "climats". In making a typical wine originating from a given terroir unit, fields or subfields are assigned to this unit and grapes of one or several specific varieties are combined in containers within a winery: as such winegrowing terroirs need to be managed across the geographical space. Both Greek and Latin agronomists developed and used recommendations for the spatial management of terroirs, as exemplified by the Amos farming leases dating back the High Hellenistic period. These leases discriminated between vineyards planted on the "plains" and vineyards planted in "rocky terrains", with differing prescriptions for vineyard planting densities (Vaudour and Boulay, 2013). Inherited from medieval times and monasteries, highresolution grapevine selection from small-sized fields of some hundreds of square metres has long been practiced for the making of famous crus in small volumes (barrels of 200 hL) (Dion, 1990;Unwin, 1991). In that sense, the shaping of terroirs results from a long heuristic process (likely hundreds of years) through history, marked by discontinuities due to wars, the spread of plagues and wine market opportunities. For centuries, such heuristic processes have mostly been carried out on less fertile soils (fertile soils being reserved for annual crop cultivation) and without resorting to irrigation, thus accentuating multi-year variability due to vintage weather ("millésime" effect). As an entity distributed over space and time, terroir has cultural aspects that have heritage, landscape and reputation value-added components (Tomasi et al., 2013) that come from historical, empirically derived technical adjustments, the transmission of taste typicity over generations and strong gastronomical traditions. On the other hand, the agro-environmental aspects of terroir are likely to be conceptualized, in order to characterize, delineate and monitor zones with homogeneous or outstanding grape and/or wine, soil, geomorphological, geo-logical, landscape and climate characteristics at a given spatial level and over a given duration. This process may be in the nascent stages of understanding vineyard spatial management in young winegrowing regions or refined in those winegrowing regions with long-lived wine traditions.
On the international scale, a definition focused on the agro-environmental facets of terroir was adopted in 2010 by the International Vine and Wine Organization (resolution OIV/VITI 333/2010). Originally developed for wine, approaches for defining terroirs are now being carried to other specialty crops such as coffee (e.g. De Assis Silva et al., 2014), tea (e.g. Besky, 2014), tequila (Bowen and Zapata, 2009), honey, maple syrup, cacao, olive oil, fruits, vegetables and cheese (Trubek, 2008;Jacobsen, 2010), ultimately linking the uniqueness and quality of both beverages and foods to the environment where they are produced. The trend for providing the consumer a sense of a place has historically developed alongside the legal protection of these products, through either "protected designations of origin" (PDOs) or "protected geographical indications" (PGIs). However, although they both rely on the assumption of a deterministic relationship between food quality and agro-environmental features, legal definitions of PDOs and PGIs are not always spatially representative of the terroirs they come from. In the European Union for instance, PDOs refer to the names of regions, given areas or even countries assigned to agricultural crops or value-added products which are produced, processed or prepared in a region according to traditional methods. In France, Italy and Spain, official PDOs that provide terroir wines with legal protection from falsified wines from other areas date back to 1935 (Vaudour, 2003). However, there have been many historical precursors of PDOs, such as, Chianti in the 17th century (Tomasi et al., 2013) or Jerez (Cabral Chamorro, 1987) and Champagne (Marre, 2004) in the 19th century. Under a PDO, winegrowers, winemakers and experts jointly define those zones that are permitted to produce wines named after their most renowned places, under common producing rules considered as traditional. These zones are generally based on pre-existing boundaries of administrative districts or easily demarcating patterns derived from hydrological networks, roads or railways. In contrast, PGIs refer to the names of areas, with some link to product quality and with at least one of the stages of production, processing or preparation occurring in the area in question. Despite differing definitions, terroir is sometimes confused with PDO or may even be confused with PGI (Barham, 2003), when one or, more likely, several terroir units may constitute the delimitated areas within the PDO, be included in them, or even intersect them.
In whatever winegrowing region in the world, when valuing inherited management zones or attempting to construct them, the so-called "natural" components of terroir actually result from a set of processes, which together with viticul-SOIL, 1, 287-312, 2015 www.soil-journal.net/1/287/2015/ tural practices create a delicate equilibrium and regulate its effect on products in both space and time (Van Leeuwen et al., 2004;Deloire et al., 2005;Van Leeuwen and Seguin, 2006;Costantini and Bucelli, 2014). In its most basic sense, the so-called "concept of terroir" relates the sensory attributes of wine to the environmental conditions in which the grapes are grown (Van Leeuwen and Seguin, 2006;White et al., 2007;Tempesta et al., 2010). Given the economic importance of the wine industry worldwide, there is clearly a need to better understand the spatial and temporal variability of grape composition and which spatial and cultural scales and resolutions are best suited to manage the production of terroir wines that reveal the typical qualities of terroir units across a given territory, together with minimizing the environmental impacts of this production. Underlying notions of such questioning stem from two main research areas: first, the concepts and knowledge of agro-ecosystems, raised and revisited for present-day agriculture, which faces an increasing number of challenges, including ensuring various ecosystem services by means of implementing agro-ecological practices (Doré et al., 2011;Wezel et al., 2014); second, concepts of and approaches to digital soil assessment based on pedometrics and/or proxy remote-sensing techniques (Carré et al., 2007;Minasny et al., 2013;Werban et al., 2013;Hartemink and Minasny, 2014). These challenges go hand in hand with the need for an interdisciplinary approach of soil (Brevik et al., 2015). There is a greater need to understand regional-to-site and site-toregional variations in crop production, and the growth in spatial analytic technologies is likely to facilitate downscaling and upscaling approaches to address these needs. Together with the emergence of precision viticulture, the explosion of spatial data availability and geospatial technologies in the past 15 years has made the within-field and farm scale of research more valuable to the individual grower, resulting in greater adoption and application (Tomasi et al., 2013). Furthermore, the study of terroir has shifted from a largely descriptive regional science back in the 1990s to a more applied, technical research field at the beginning of the 21st century. Confusion between PDOs and terroir, and the strict observance of no irrigation, according to historical heuristic processes of terroir, results in misleading questions such as the possible compatibility of terroir and precision viticulture (Bramley and Hamilton, 2007). However, the long process of terroir identification over time questions those practices that enhance or diminish terroir sustainability, particularly in recent times with the advent of modern viticulture. For example, viticultural soils appear to be exposed to degradation processes, perhaps more than ever, because of unsuitable practices in land management (Blavet et al., 2009;Follain et al., 2012;Costantini and Lorenzetti, 2013). In addition, soil contamination by copper resulting from the cumulated use of Bordeaux mixture and other copper fungicides is increasingly an issue (Pieztrak and McPhail, 2004;Fernández-Calviño et al., 2013;Chopin et al., 2008;Mirlean et al., 2007Mirlean et al., , 2009El Hadri et al., 2012;El Azzi et al., 2013) and appears to result in modifying the spatial distribution and composition of soil microbial communities (Jacobson et al., 2007;Mackie et al., 2013). Because of water scarcity, irrigation may be practiced with saline water or saline effluent, with possible deleterious effects on plant growth (Walker et al., 2002;Paranychianakis and Angelakis, 2008;Stevens et al., 2010Stevens et al., , 2011 and on soil salinity, structure and quality (Crescimanno et al., 2007;Urdanoz and Aragüés, 2009). Thus, an intensive investigation of the microbial and fungal organization and their function is required, as soil microbial and fungal communities are known to be of vital importance for terrestrial processes by driving the major soil geochemical cycles and supporting healthy plant growth (Nannipieri et al., 2003;Bokulich et al., 2014).
Our objective is therefore to present an overview of existing data and analytical and modelling approaches for the footprinting and zoning of terroirs on local and regional scales. This review will focus on two main areas of recent terroir research: (1) new tools for assessing terroir footprints, comprising metabolomics, the metagenomic approach and microbial and chemical fingerprinting; (2) terroir zoning on different scales, using remote-and proxy-sensing technologies to spatially manage the crop system for higher quality and to spatially monitor the soil quality.

Chemical fingerprint and metabolomics
Wines from distinct countries or regions can be discriminated through their chemical composition and/or sensorial profiles. A recent study quantifying a large number of elements (33) in wine and soil samples analysed by quadrupole inductively coupled plasma mass spectrometry (Q-ICPMS) in addition to 87 Sr / 86 Sr isotopic analysis allowed for differentiation among the wine-producing regions of Argentina (Di Paola-Naranjo et al., 2011). In this case the 87 Sr / 86 Sr ratio was amongst the best discriminators, and its value ranges were similar between soils and wines. This corroborates the results obtained by Marchionni et al. (2013) across six distinct wine appellations in Italy, but in their study they focused on parent material; the ratio matched those observed for homogeneous soil parental material such as volcanic rocks. Studying the whole parent-material-soil-plant-wine chain at a single rainfed vineyard located in the volcanic area of Campi Flegrei (southern Italy) and considering both a set of putative rare-earth elements geotracers and the Sr isotope ratio, Mercurio et al. (2014) found that only this isotopic ratio was "consistently and inherently transferred and maintained from geologic parent material to wine, through soil horizons, branches, leaves, and grapes". However, considering several vintages, Marchionni et al. (2013)  volcanic rocks, unlike wines from vineyards on sedimentary or granitic rocks. As isotopic ratios of geological substrates belonging to different geological districts may partially overlap, Mercurio et al. (2014) recommend assessing the isotopic ratios together with a pertinent soil classification for a reliable assurance of wine provenance. Geographical fingerprinting of wines may also be possible with areas as close to one another as 4 km. Research by Tarr et al. (2013) found that more than a thousand components were identified through the high-performance liquid chromatography of juices from two varieties (Syrah and Grenache) originating from two distinct terroirs roughly 4 km from each other. Hierarchical clustering of data peaks suggested that terroir played a large part in the final composition of the grape berry metabolome (Tarr et al., 2013). Metabolomics seek to identify the chemical fingerprint of a particular cell, organ, organism or tissue type that results from and is thus indicative of the chemical processes that occur within the specimen of analysis (Tarr et al., 2013). Using metabolite and gene transcript profiling, Castellarin et al. (2007) highlighted that the biosynthesis pathway of flavonols related to water deficit and nitrogen nutrition. The first reaction involves the deamination of phenylalanine by the enzyme phenylalanine ammonia lyase (PAL) into cinnamic acid, thus diverting phenylalanine from the pathway that relates carbohydrates to the synthesis of proteins. Metabolomic studies have also been conducted through Proton nuclear magnetic resonance ( 1 H NMR) spectroscopy (Pereira et al., 2006;Son et al., 2009), which revealed that grapes grown in regions with high sun exposure and low rainfall showed higher levels of sugar, proline, Na and Ca together with lower levels of malate, citrate, alanine, threonine and trigonelline than those grown in regions with relatively low sun exposure and high rainfall (Son et al., 2009). The sensitivity of this method has allowed it to be successfully applied to classify wines according to their phenolic profile and allowed distinguishing between wines from different wineries of the same wine-producing zone and between different vintages for wines of the same variety in Greece (Anastasiadi et al., 2009) andSpain (López-Rituerto et al., 2012).

Biological fingerprinting through molecular and "omics" approaches
Despite the fact that the interaction between microbial communities and the vine may be one of the key factors that influence the plant traits, the role of microbes for terroir and vine traits has been largely ignored (Gilbert et al., 2014). However, the recent development of the so-called "omics" techniques (mainly metagenomics, metabolomics, transcriptomics, proteomics, phenomics) has made it possible to explore the soil functionality, microbial diversity (microbiome) and vine-associated microorganisms in greater detail. In fact, the application of omics techniques to soil may enable the determination of rare microbial species and discover new com-pounds or functions (antibiotic, enzymes, etc.) from the expression of genes of unknown microbial species (Myrold and Nannipieri, 2014). Thus, the application of omics techniques in soil microbiology, together with standard techniques for soil science, is promising for understanding the functioning of soil and its effect on terroir. To date few studies have been carried out on this issue. The interface between roots and soil (rhizosphere) is often considered the key point of interaction between a plant and its environment. Microbes colonizing at the root may migrate through the plant to colonize aerial tissues, either internally (endophytes) or externally (epiphytes) (Compant et al., 2011;Bulgarelli et al., 2013). In addition to soil, the exploitation of the commensal microbial flora that coexists with the vine may be one of the key factors that influence the plant traits.
Most studies dealing with biological fingerprinting are related to the microbial communities present on the surface of the grape berry, which are known to be very large and which change according to the stage of grape development (Barata et al., 2012;Pinto et al., 2014). In fact, the microbiological life of wine starts before the reception and fermentation of the grapes at the winery. In particular the yeast population and bacterial and fungal consortia inhabiting grape surfaces could reflect a wine region, as reported in some recent studies (Renouf et al., 2005;Setati et al., 2012;Bokulich et al., 2013). However, determinants of regional wine characteristics have not been identified. Renouf et al. (2007) identified 52 yeast species and 40 bacteria. The majority of the bacterial groups were present in the study, in particular the proteobacteria, which are not commonly described in oenology, while the most common oenological yeast (Saccharomyces cerevisiae, Brettanomyces bruxellensis) and bacteria (Oenococcus oeni, Pediococcus parvulus, Gluconobacter oxydans) were detected on grape skins from the first stages of development. Bokulich et al. (2013) surveyed 273 grape musts from two vintages of Chardonnay, Zinfandel and Cabernet Sauvignon, demonstrating that the grape surface microbial communities present were significantly different between regions (Fig. 1). The authors also showed that the degree of significant differentiation between regions is increased dramatically when they look at the biogeography within a grape variety of a given vintage, and they alluded to "the existence of non-random 'microbial terroir' as a determining factor in regional variation among wine grapes". This finding suggests that other factors also play a significant role, including host genotype, phenotype (grape variety), local and interannual climate variation (vintage), and soil quality. However, the specific role of soil microbes was not determined.
In a recent work Martins et al. (2013) observed the interaction between telluric bacterial communities and the epiphytic bacteria present on the different grapevine parts. Yet, the ecological interactions and the role of such organisms are still not clear. Vega-Avila et al. (2015) reported how soil management (organic vs. conventional) could affect the structure and SOIL, 1, 287-312, 2015 www.soil-journal.net/1/287/2015/  Gilbert et al., 2014), but the functional diversity of such microbiomes is still largely ignored. The combination of different high-throughput culture-independent methods, such as microarrays, metagenomics and microbiome, might elucidate such aspects. For example, considering soils from two sites close to each other in central Tuscany cultivated with the same variety (Sangiovese) but with contrasting wine quality and water stress, Mocali et al. (2013) used functional GeoChip microarrays and high-throughput DNA sequencing of rRNA genes to explore both microbial composition and functions. Preliminary results revealed different amounts of Actinobacteria and Proteobacteria among the two sites and an overrepresentation of sulfur-oxidation genes in samples where both the increased level of sulfates and the abundance of Firmicutes such as Sulfobacillus thermosulfidooxidans occurred. A further step might be to sequence the entire genomic DNA present in a vineyard soil sample and referred to as the "metagenome", which could provide a cultivation-independent assessment of the largely unexplored genetic reservoir of soil microbial communities and their functions (Daniel, 2005;Mocali and Benedetti, 2010). The importance of this approach has led to the establishment of "Terragenome", an international consortium for the exploitation of the soil metagenome (Vogel et al., 2009). While these studies represent an initial examination of these relationships, they represent an essential step which may potentially help to revolutionize how sites for agriculture are chosen or, indeed, how they could be manipulated by probiotics designed to select suitable bacterial species, which could improve soil quality and, hence, crop productivity. Further research is needed to study the degree to which the use of probiotics could enhance some wine characteristics of given ter-roirs in sites a priori not suitable for generating such characteristics.

General perspectives
The differentiation and mapping of regions of grape and wine quality require comprehensive spatial modelling of climatic, soil and agronomical properties, including their changes through time (Vaudour, 2002;Costantini and Bucelli, 2014). The construction of such spatial models for demarcating terroir units and predicting their viticultural and wine response is undergoing a methodological revolution as new technologies and analytical methods enable the capturing of detailed spatial and temporal variability of grapevines according to functional properties in the soil. Recent developments in the combined use of several geospatial technologies including geographic information systems (GIS), global positioning systems (GPS), remote sensing (RS) and direct measurements in the field (proxy sensing) are likely to open many new areas in the spatial modelling of winegrowing terroirs. A considerable amount of research dealing with terroir zoning (> 120 journal papers, book chapters and books) has been published since 2002 (Fig. 2). Most of the published research has been carried out on the within-field scale in the context of the so-called "site-specific" or "precision viticulture" (Bramley and Hamilton, 2004;Bramley, 2005;Bramley et al., 2011a, b, c, d, e;Tisseyre and McBratney, 2008;Roudier et al., 2008Roudier et al., , 2011Arnó et al., 2009Arnó et al., , 2012Pedroso et al., 2010), and its median study area covered only 0.12 ha, while between-field studies covering more than 0.11 km 2 only represented a quarter of the total (Fig. 3). There is therefore a gap to fill regarding farm (≤ 0.1 to 1 km 2 ) and district (≤ some tens of square kilometres) to regional scales (≥ tens to thousands of square kilometres). The number of map units tends to increase with the log of study area; however, its variation is higher for larger study extents than for within-field studies, and regional studies focus on a larger range of target properties. Twenty or so viticultural, environmental or oenological target properties have been considered over the last decade (Table 1), mainly focusing on grape and canopy characteristics and yield or biomass and secondarily on soil properties on the within-field scale.  field sites, new techniques are promising both for capturing the detailed spatial variability of vineyard areas and for collecting a large amount of soil, vine and grape data. A number of terroir-related studies of the last decade have relied upon a large amount of yield data collected by means of on-the-go yield sensors mounted on mechanical harvesters, which were made commercially available beginning with the 1999 harvest. Pioneering research by Bramley and Hamilton (2004) and Bramley (2005) not only highlighted the magnitude and extent of the spatial variability of yield in some Australian vineyards (of between 4.5 and 7.3 ha) but also emphasized the significant influence of soils in driving yield differences. The need to assess spatial variations in soil properties has driven the development and application of direct (so-called "proximal") geophysical sensing, particularly for measuring soil apparent electrical conductivity by means of either electrical resistivity surveys and/or electro-magnetic induction scans (EMI) (Lamb et al., 2005;Morari et al., 2009;Taylor et al., 2009;Trought and Bramley, 2011;Fulton et al., 2011;André et al., 2012;Martini et al., 2013;  Moreover, the need to assess spatial variation in grapevine biomass and canopy properties, or to map terroir units or to identify vines, has driven the development, acquisition and processing of remote-sensing data. Figure 4 shows that the development of very high spatial resolution airborne acquisitions for the purpose of characterizing grapevine physiological status has represented about one-third of the terroirrelated studies over the last decade (Hall et al., 2002(Hall et al., , 2003(Hall et al., , 2008Dobrowski et al., 2003;Lamb et al., 2004;Zarco-Tejada et al., 2005;Martín et al., 2007;Gil-Pérez et al., 2010;Meggio et al., 2010;Hall and Wilson, 2013). Most approaches combine several data sources, methods (geostatistical, statistical, image processing, computer vision, mechanistical models) and remote or proxy sensors (Table 2). All approaches use geopositioning devices (not detailed in Table 2) the error positioning requirements of which need to be compatible with the study objectives (i.e. accurate positioning of individual sampled vines) and the spatial resolution of the acquired imagery.
where R NIR is reflectance in the near infrared spectral band and R R is reflectance in the red spectral band. Providing a number between −1 and +1, quantifying the relative difference between the near infrared reflectance "peak" of vegetation tissues and the red reflectance "trough" due to chlorophyll and carotenoids pigment absorption, the NDVI is the most widely used indicator of plant canopy vigour and relates to the leaf area index (LAI: the ratio of leaf surface area to ground area), fractional cover, biomass, shaded area (Hall et al., 2002(Hall et al., , 2003Johnson, 2003;Johnson et al., 2003;Dobrowski et al., 2008) and grape quality . In very high spatial resolution (VHSR) vineyard imagery, canopy-only pixels tend to have very high NDVI values, commonly ranging between 0.75 and 0.85 (Hall et al., 2002(Hall et al., , 2003(Hall et al., , 2008. Interpretation of NDVI threshold values requires adaptation to each case study region. This is because typical multispectral images with coarser resolution, such as 4 m IKONOS, may have values that integrate mixed signals from both vine row and interrow and from either bare soil or soil vegetated with an interrow crop. However, many studies do not give sufficient detail on the retrieval of NDVI, which basically requires that the digital numbers of the raw images be atmospherically corrected to surface reflectance values prior to NDVI computation in order to minimize differences in light conditions (Vaudour et al., 2014b). In an attempt to address the issue of variable environmental conditions when acquiring multitemporal imagery, the digital numbers are either normalized (e.g. Vaudour et al., 2010) or alternatively sliced into high, medium and low values for each image (e.g. Dobrowski et al., 2003). However, more often than not the relationships between NDVI and grapevine vegetation and terroir parameters are assessed on an individual image basis, using raw digital numbers (e.g. Johnson et al., 2003;Lamb et al., 2004) or even ignoring atmospheric effects. Using airborne hyperspectral sensors, such as the Compact Airborne Spectrographic Imager (CASI), the Reflective Optics Imaging Spectrometer (ROSIS) and the Digital Airborne Imaging Spectrometer (DAIS-7915), and atmospherically corrected images, Zarco-Tejada et al. (2005) demonstrated that narrowband hyperspectral indices in the 700-750 nm spectral region performed better than NDVI for the purpose of estimating the chlorophyll content of leaves and then detecting iron deficiency chlorosis. Using the airborne hyperspectral scanner (AHS) at a 1000 m altitude flight providing a spatial resolution of 2.5 m, Meggio et al. (2010) devised other physiological indices for predicting carotenoid and anthocyanin leaf contents, which were related to grape quality in a previous study .

Identification and/or characterization of vineyards
Some researchers have developed change detection classifications in order to extract regional land use changes including vineyards from multidate, multispectral images with a medium spatial resolution, such as 30 m Landsat data (Lanjeri et al., 2004;Rodriguez-Pérez et al., 2007;Manandhar et al., 2010). However, much of the work in this area over the last decade is especially influenced by the advent of VHSR images, which have favoured innovative approaches for retrieving specific patterns of vineyard arrangements from helicopter colour images with ∼ 0.25 m resolution (Wassenaar et al., 2002), airborne multispectral images with ∼ 2 m resolution (Gong et al., 2003) or 0.5 m resolution (Rabatel et al., 2006), satellite panchromatic 1 m IKONOS images (Warner and Steinmaus, 2005), satellite panchromatic 0.6 m Quickbird images (Rabatel et al., 2006), and ultra-light motorized (ULM) colour 0.5 m images Delenne et al., 2010). Approaches for vineyard identification include grapevine field detection (Wassenaar et al., 2002;Rabatel et al., 2006), grapevine field delineation (Da Costa et al., 2007), grapevine row extraction (Hall et al., 2003;Delenne et al., 2010;Matthews and Jensen, 2013;Puletti et al., 2014) and the detection of missing plants (Chanussot et al., 2005;Delenne et al., 2010). These approaches have mostly used greylevel images (often the red band) and either relied on frequency analysis (Wassenaar et al., 2002;Rabatel et al., 2006;Delenne et al., 2010) or developed textural analysis, a branch of image processing focused on the spatial statistics of the grey levels of images, the variations of which are perceived as homogeneous areas by the human eye (Haralick et al., 1973). The textural analysis is based on co-occurrence matrices, i.e. "the histogram, in a given neighbourhood for each pixel (e.g. 7 × 7), of the grey-level transitions when considering a given translation in a given direction, from which various parameters can be computed", for instance energy, correlation, directivity, entropy and contrast (Rabatel et al., 2006). Provided that the boundaries of each field are available, one can apply a local Fourier transform to extract information on the type of vineyard planting as well as crop spacing and orientation (Wassenaar et al., 2002) or isolate each individual field by selecting the corresponding frequencies in the Fourier spectrum using a specific Gabor filter applied recursively Delenne et al., 2010). Another approach is that of spatial autocorrelation, requiring that grapevines be equally spaced and that the spatial resolution be fine enough that individual grapevines are distinguished from one another (Warner and Steinmaus, 2005).

Characterization of soil types, soil properties, soil surface condition and erosion
As vineyard inter-rows are frequently left bare, particularly in Mediterranean regions, remote-sensing satellite images acquired in spring, before budburst, have been used for the purpose of mapping soil surface type and condition. For instance, high-resolution multispectral 20 m SPOT images during spring have been processed using a supervised Bayesian maximum likelihood classifier to map red Mediterranean soils originating from Plio-Pleistocene fluvial deposits in the Southern Rhône Valley, with an overall accuracy of 60-70 % (Vaudour, 2008). Being important factors in runoff and infiltration processes, soil surface characteristics (SSCs) have been mapped on the within-field scale using VHSR images with 0.1 m resolution acquired by means of unmanned aerial vehicles (UAVs) equipped with a colour camera: Pixy ® (Corbane et al., 2008a, b) or DRELIO ® . Depending on SSC classes and surface conditions, overall accuracy ranged from 63 to 84 % (Corbane et al., 2008b). The clay content of viticultural soils of the La Peyne catchment in Languedoc (southern Mediterranean France) has been successfully estimated from the 2206 nm wavelength of an airborne hyperspectral HyMap ® image with 5 m spatial resolution and then spatially predicted from a co-kriging model using this image as a co-variable .

Incorporation of remote-sensing information into the spatial modelling of terroirs
Remote-sensing images have been used for the purpose of mapping terroir units on regional scales, facilitating the selection of plant material, the assemblage of harvest and the monitoring of vine phenology and status across a number of individual fields spread over a regional viticultural area. A terroir concept adapted to spatial modelling from remote sensing considers soil landscape units as base elements for defining terroir units, jointly with climate series and grape composition series (Vaudour, 2003). A soil landscape, also referred to as a soilscape (Hole, 1978) or "pédopaysage" (Girard, 1983), can be defined as a landscape unit including a limited number of soil classes that are geographically distributed according to an identifiable pattern (Lagacherie et al., 2001). Soil landscape units were also defined as "a set of pedological horizons and landscape features (vegetation, effects of human activities, geomorphology, hydrology and parent material) whose spatial organization allows for defining a soil mantle (or a subset of it)" (Girard, 1983;Carré and Girard, 2002). Their identifiable pattern can be retrieved from the visual interpretation of several geographical data layers including image classification results and stereoscopic photograph examination (Vaudour et al., 1998;Vaudour, 2003). The visual interpretation process follows a set of rules describing a conceptual model of soil landscape organization, which relies on the assumption that soil landscape units may be inferred from the geomorphological identification of surficial formations (e.g. mainly glacio-fluvial terraces in the Rhône Valley), the age and relative elevation of which correspond to distinct durations of pedogenesis ("fersiallitization") and thus distinct soil layer depths and properties, including soil surface stoniness and colour (Vaudour, 2008). In this pioneering approach, visual interpretation was digitally performed within a GIS along with the description and recording of several soil landscape attributes at each delineated polygon area. Potential terroir units were then obtained from the Ward's clustering of these soil landscape attributes and were further validated against a considerable set of grape composition data spanning 17 years. In an attempt to reduce the time-consuming stage of visual interpretation, alternate approaches which solely relied on the automatic processing of remote-sensing and/or morphometric data were proposed, either based on combining per-pixel and textural classifiers (Vaudour, 2003) or on bootstrapped regression trees (Vaudour et al., 2010). In these studies, the resulting map units were termed "terroir" and "viticultural" because they were tested against a considerable set of grape composition data over a long-term period (Vaudour et al., 1998;Vaudour, 2003) or relied on ∼ 50 reference vineyards, the oenological properties of which were known from previous research (Vaudour et al., 2010). In another study, the spatial units of which were termed "terrons", because they were not tested again viticultural data (Hughes et al., 2012;Malone et al., 2014), Landsat bands and several Landsat band ratios including NDVI were included as covariates of soil profile data in a number of geostatistical models in order to spatially predict several soil properties. These properties include soil pH, clay percentage, soil mineralogy (clay types and presence of iron oxides), continuous soil classes and the presence or absence of marl, the predicted soil properties were then combined with landscape attributes (derived from a digital elevation model) through fuzzy k-means in order to predict 10 non-marl "terron units" and 2 marl "terron units" depending on the presence of marl (i.e. active lime) at 0.5 m depth. Such an approach to defining "terrons" was initially proposed by Carré and McBratney (2005) and is meant as an initial stage prior to defining viticultural terroirs. A similar study using the term "natural terroir units" and derived from the geostatistical methods described by  was carried out in Tuscany but did not include remote-sensing layers . "Natural terroir units" (NTU) were first proposed by Laville (1990), as "a volume of the Earth's biosphere that is characterized by a stable configuration and values of the environmental factors", and built from morphometric data and lithological units. Because of the data availability, topography, climate, substrate and soil are the most commonly used land features in digital terroir zoning (e.g. Carey et al., 2008;Herrera-Nuñez et al., 2011). In fact, there is a conceptual similarity, if not a direct relationship, between the NTU as constructed by Priori On the within-field scale, "management zones" originating from a set of soil and/or vegetation proxy and remotely sensed attributes, typically NDVI, were also obtained using either fuzzy k-means (Bramley and Hamilton, 2004;Pedroso et al., 2010;Taylor et al., 2013;Tagarakis et al., 2013;Priori et al., 2013b;Urretavizcaya et al., 2013) or Ward's clustering . Similar clustering approaches have been performed using a multivariate set of spatial layers, including apparent electrical conductivity but without remote-sensing images (e.g. Martini et al., 2013).
In order to be applicable in an operational manner, management zones need (Bramley and Hamilton, 2004) (i) to provide stable, constant patterns from year to year; (ii) be related to yield; (iii) be manageable; and (iv) be more economically beneficial than conventional uniform management. An objection to using remote-sensing images for delineating management zones is that they "provide only a within-season snapshot and may not relate to final crop yield" (Taylor et al., 2007). According to , later confirmed by Trought and Bramley (2011), "yield or vigour (pruning weight, size of the canopy) maps of the previous years are relevant in designing site-specific management strategies in the year "n+1" and subsequent years" and conversely in producing maps of quality parameters (sugar content, titratable acidity, pH) which present no temporal stability of withinfield variability. This is in compliance with the observations made previously on the regional scale, in the terroir units mapped in the Southern Rhône Valley, for which a long-term frequency analysis of Grenache berry composition had highlighted a strong vintage interaction and no temporal stability of berry composition groups (Vaudour, 2003). When acquired over bare soils in spring, multispectral SPOT images showed a high temporal stability for deriving homogeneous terroir zones (Vaudour, 2008) which matched those obtained from the supervised support vector machine classifier of a single-year within-season SPOT4 Take 5 time series acquired from February to June and accounting for vegetation vigour and phenology (Vaudour et al., 2014a). In order to account for the between-terroir variation in grass vegetation across seasons in vineyards with grass intercrops, regional terroir units may be retrieved from multitemporal multi-seasonal images (Vaudour et al., 2010).
On the within-field scale, while unsupervised clustering algorithms are commonly used for defining viticultural management zones, other methods have been proposed for cerealgrowing systems (e.g. Roudier et al., 2008) in order to address the problem of the manageability of the mapped zones. These authors emphasized the distinction between classification methods, which define classes, such as groups of individual pixels presenting similar properties, and segmentation methods, which define regions or the expression of those groups in space and time and form individual patches. They proposed using a segmentation method stemming from mathematical morphology and applying the watershed algorithm initially proposed by Beucher and Lantuéjoul (1979) and later formalized by Vincent and Soille (1991). They suggested an approach for reducing the over-segmentation of zones, the number of which was selected following Lark's parsimony principle (2001). According to this principle, the most suitable number of clusters is that "after which the vegetation parameter (such as biomass) variance reduction remains more or less constant or declines more slowly" (Roudier et al., 2008). Generally, the number of within-field zones effectively chosen is empirically defined, comprises between 3 and 5 zones and, when illustrated through raster monovariate maps, is eventually described through up to 20 map units based on equal-distance or equal-density intervals (Fig. 3). Pedroso et al. (2010) proposed another withinfield segmentation approach based on a contextual colour and shape criterion and performed on an NDVI airborne ADS40 image with 5 m resolution; they aimed at optimizing variance partitioning of vine circumference data and removing small unmanageable polygons covering less than 0.1 ha.
Zoning based on automatic procedures should also enable us to assess prediction error and/or error uncertainty along with predictions. In the approach devised by Pedroso et al. (2010), the effectiveness of the management unit delineations was determined from the adjusted R 2 from ANOVA using the management units as the independent variable and grapevine circumference data as the dependent variable. Most terroir units or management zones are validated following a similar variance testing procedure, i.e. demonstrating how well delineation explains a key growth, grape or berry composition parameter (e.g. Priori et al., 2013b;Santesteban et al., 2013). However, while the agronomical and soil property model is verified and/or validated, its spatial prediction error is not, or only seldom, assessed except for geostatistical approaches Lagacherie et al., 2012) and for the bootstrapped approach by Vaudour et al. (2010).

Prediction of evapotranspiration and management of irrigation on the basis of remote-sensing information
Recent developments of remote-sensing applications in terroir-related studies have dealt with the prediction of evapotranspiration using multispectral Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) images through the thermal infrared bands with 90 m resolution (Galleguillos et al., 2011a, b). Using the approach developed by Galleguillos et al. (2011a, b) and a linear downscaling based on land use at smaller pixel sizes, Taylor et al. (2013) demonstrated that the ASTER-derived evapotranspirationbased covariates were of particular significance in the soil depth modelling while water table depth was better explained by models that used digital terrain attributes at smaller pixel sizes.

Contribution of lidar and UAV remote sensing
The use of UAVs in precision viticulture is very recent and promising, as the time of acquisition is tightly controlled and adapted to the user's needs. In particular, promising approaches to mapping vine water stress were presented by Baluja et al. (2012a) and by Bellvert et al. (2014), based on 0.3 m thermal UAV images acquired around noon (solar time), the pixels of which were significantly linked to midday water potential (MWP) measurements. UAV equipped with lidar sensors also enabled the detection of rows, the 3-D reconstruction of a vine plantation (Llorens et al., 2011) and the quantifying of vineyard canopy (Llorens et al., 2011;Matthews and Jensen, 2012), while airborne lidar images enabled the mapping of landscape linear features (ditches) in a viticultural catchment (Bailly et al., , 2011 and also the hydro-geomorphological analysis of terraced vineyards (Tarolli et al., 2015). UAVs equipped with a visible camera and taking multiangular images using the computer vision technique of structure from motion also enabled the 3-D reconstruction of a vine plantation and were promising in LAI estimation (Matthews and Jensen, 2013). UAVs allow for the acquisition of spectral or thermal measurements that are very comparable to proxy measurements and are generally carried out along with field proxy measurements. Their use, however, requires a perfect mastering of a chain consisting of image series acquisition, acute georeferencing, spectral calibration, mosaicking and processing, which is the subject of ongoing technical developments, as shown by Verger et al. (2014) for predicting the green area index of annual crops.

Geophysical proxy measurements
Observations should account for the entire depth of the root systems of vines, which may explore the soil parental material, often being a surficial formation (Vaudour, 2002(Vaudour, , 2003Costantini et al., 2012). Geophysical techniques applied to soil (Samouëlian et al., 2005;Doolittle and Brevik, 2014) offer a unique opportunity to explore deep horizons, and it may be expected that key soil properties related to the soil-vine water balance be retrieved from EMI or ground-penetrating radar (GPR) measurements, in possible conjunction with remote-sensing images. However, as in the case of remotesensing images, these techniques require local calibration as the measured signal is a bulk signal. Sensing results are often limited to qualitative information and geophysical sensing results are ambiguous, making reliable quantification of sensing information still a major challenge (Werban et al., 2013). Indeed, geophysical surveys have mainly resulted in delineating within-field zones (Lamb et al., 2005;Taylor et al., 2009;Costantini et al., 2010;André et al., 2012;Mar-tini et al., 2013;Priori et al., 2013a, b), rather than predicting soil properties, such as clay content (Rodríguez-Pérez et al., 2011;Andrenelli et al., 2013), extractable Na + and Mg 2+ contents (Rodríguez-Pérez et al., 2011), or soil moisture (Brillante et al., 2014). Apparent electrical conductivity (ECa) values are "affected by various soil properties in a complex manner and it is difficult to discriminate the weight that each soil parameter has on the final apparent measured ECa" (Martini et al., 2013) so that the Pearson's correlation coefficient is often not significantly high between ECa and soil parameters such as clay content or gravimetric water content. However, Brillante et al. (2014) state that, when soil characteristics are available, it is possible to take them into account in a multiple adaptive regression spline model to build a pedotransfer function for predicting soil moisture from ECa with reduced error (±2 % vol.).

Canopy and grape proxy measurements
In addition to field reflectance measurements on leaves to define spectral indices of water status (e.g. Rodríguez-Pérez et al., 2007), several field sensors have been developed in the last decade not only for characterizing canopy and vigour -for instance, with Crop Circle ® passive reflectance sensors and active sensors (e.g. Stamatiadis et al., 2010) and Greenseeker ® reflectance sensors computing NDVI values in real time (e.g. Mazetto et al., 2010) or ground-based lidar scanner for LAI estimation (Arnó et al., 2013) -but also for measuring grape quality parameters including using the Multiplex ® portable sensor (Ben Ghozlen et al., 2010;Bramley et al., 2011b;Baluja et al., 2012;Agati et al., 2013). In particular, both fluorescence-based anthocyanin and flavonol indices originating from this sensor showed a high potential for monitoring technological maturity, according to the recent findings by Agati et al. (2013), in Sangiovese and Vermentino varieties respectively.

The issue of big-data handling and the statistical processing of the varied spatial data collected
In addition to remote and proxy data collected from distinct sensors, the use of mobile devices with multitag technologies (Cunha et al., 2010;Luvisi et al., 2011) facilitates the recording of a great wealth of data. These numerous spatial data have stimulated new developments in both software and hardware, jointly with statistical processing which stems from geostatistics, image pattern recognition and satellite image processing and which includes machine learning. However, the most common pattern adopted for within-field spatial data and observed for ∼ 40 of the studies in terroirrelated research carried out in the last decade (Table 3) relied on geostatistical analysis, as emphasized by Baveye and Laba (2015). Considering the target parameter as a random property following a random process with an assumption of stationarity (i.e. there is the same degree of variation from SOIL, 1, 287-312, 2015 www.soil-journal.net/1/287/2015/  Fig. 2).

Use of geostatistics N
No geostatistics 87 Ordinary kriging 15 Geostatistics but unexplained 8 Block kriging 6 Further use of variogram attributes 5 Factorial, co-kriging, block co-kriging 2 Legend: N -number of papers place to place), Matheron (1962Matheron ( , 1965 formalized the approach to predict target properties from spatially correlated sample data through the computation of the semi-empirical variogram, to which are fitted a number of standard parametric models (Oliver and Webster, 2014). "Kriging is a generic term for a range of least-squares methods to provide the best linear unbiased predictions in the sense of minimum variance", through solving a set of linear equations (the kriging system) from the fitted variogram function and the available data (Oliver and Webster, 2014). Ordinary kriging based on primary spatial information such as yield or ECa is the most popular method used in terroir-related studies (Table 3), though more sophisticated spatial models using ancillary spatial information have also been built, such as block co-kriging with a hyperspectral image  or factorial kriging of several soil variables and ECa (Morari et al., 2009). To single out NTU-based viticultural terroir units on the province scale (1 : 125 000 scale), Priori et al. (2014) combined a multivariate and geostatistical approach showing the variability of the soilscapes within the DOC and DOCG territories in Italy. A first principal component analysis (PCA) was performed to relate climate, pedoclimate and morphometric features to the viticultural data, and a second PCA linked the main soil features with viticultural data. The two PCAs revealed which environmental features were better related to the viticultural parameters of the experimental farms. Several geostatistical models  were then used to spatialize the selected environmental features: (i) regression kriging to interpolate rooting depth; (ii) simple kriging with varying local means, to interpolate coarse fragments and redoximorphic mottle depth; and (iii) multicollocated simple co-kriging with a varying local mean to interpolate AWC, clay and sand content. Finally, a k-means cluster analysis was performed in the viticultural areas, using the selected variable maps (mean annual temperature, mean annual precipitation, mean annual soil temperature, elevation, clay, sand, coarse fragment content, available water capacity, rooting and redoximorphic mottle depths) to determine the NTU-based viticultural terroir units. The map results obtained from digital terroir zoning performed at regional scales over some thousands of square kilometres lead to the need for a better understanding of the issue of the relevant scales for defining and managing terroirs. It is important to address the usefulness of both spatial and temporal resolutions that provide insights into terroir zoning according to the aim of the research or management practice. Studies need to examine the nature of within-field zones and the regional terroir units as the relationship between the number of map units and the surveyed area is log-linear (Fig. 5).
The underlying motivations for detailed assessments of the within-field spatial variation of yield, biomass and soil properties are related to the variable-rate application of inputs and selective harvesting at parcel level Unamunzaga et al., 2014). However, considering the withinfield scale only may be questionable in the case of highly parcelled vineyards of less than 1 ha each, such as in European countries with secular viticulture. More than ever, the issue of information synthesis and spatial scale is at the very heart of the spatial modelling of terroir (Vaudour, 2002). Considering a whole 90 ha vineyard composed of 27 contiguous fields, Santesteban et al. (2013) showed that a per-field, within-field study would have resulted in missing the spatial trend due to slope. Therefore, the usefulness of the multiplefield approach becomes one of a "whole-vineyard" or regional approach. Such issues similarly arise when attempting to predict the vine water status from a set of leaf and stem predawn water potential (PLWP/SWP) or midday (MWP) measurements, which are very time-consuming, require accurate methods to collect with a pressure chamber (Scholander et al., 1965) and should be made at several fields managed by the vineyard operation. Using PLWP (Acevedo-Opazo et al., 2008a, b, 2010b or MWP (Acevedo-Opazo et al., 2010a) measurements, Acevedo-Opazo et al. (2008a, b, 2010a) devised a linear model applicable at the within-field level but requiring a time-consuming calibration set; in the case of PLWP, it corresponded to rainfed vineyards with high water stress, while in the case of MWP, it www.soil-journal.net/1/287/2015/ SOIL, 1, 287-312, 2015 was aimed at scheduling irrigation. A similar approach using δ 13 C measurements as ancillary variables enabled the extrapolation of stem water potentials for a rainfed vineyard with moderate water stress (Herrero-Langreo et al., 2013). These approaches may be transferable to a whole-vineyard scale (∼ 29 km 2 ), according to Baralon et al. (2012), who constructed an important measurement database (58 fields monitored in all) over the course of three consecutive vintages. This approach considered a stratified sampling based on soil types, which is most effective when water restriction is high . However, several limitations hinder its practical application, namely those due to spatial sampling optimization and the improvement of the temporal resolution of the model, using real-time monitoring sensors (such as sap flow sensors) (Herrero-Langreo et al., 2013). According to farmers, a model should be capable of predicting water stress at least 2 weeks before severe irreversible water stress damage occurs (J. C. Viaud, personal communication, 2014). Arnó et al. (2012) stated that overall within-field variability of grape yield and quality raises important questions concerning whether site-specific crop management could be used in vineyards. Problems arise, in particular, when looking for causes of this variability, especially those related to the presence of soil carbonates, which may lead to Mn deficiencies, with deleterious effects on grape colour. Another problem pointed out by Baveye and Laba (2015) and related to N or P manuring is the frequent possibility that P or N deficiency in a management zone may be due to a higher leaching rate in the area, leading to an increased rather than decreased risk of groundwater contamination if more fertilizer is applied in this management zone.

Modelling and depicting climate on the region-to-vineyard scale
Another growth area in terroir zoning studies is the development of spatial climate-data products. Historically, a region's climate and suitability for viticulture were assessed via climate station analyses, which seldom depict the spatial variation of climate at actual or prospective vineyard sites within wine-producing regions . As a result, reference vineyard networks were developed within regions to better capture the spatial climate characteristics (e.g. Jones and Davis, 2000). However, the low network density even in reference vineyard networks does not account for the range in mesoclimates found within regions. To overcome these issues, spatial climate-data products providing robust, validated and more spatially appropriate climate data have been developed through the interpolation of existing longterm, quality-controlled data sources. Numerous techniques, such as kriging and smoothing splines, have been used to produce interpolated surfaces from hundreds to thousands of stations containing valuable meteorological inputs. The results are spatial climate products on daily or monthly timescales and on a range of spatial and temporal scales such as Daymet (Thornton et al., 1997), PRISM (Parameter-elevation Relationships on Independent Slopes Model) (Daly et al., 2008) and WorldClim (Hijmans et al., 2005). Most of these approaches use elevation data (digital elevation data) and station climate data to calculate a climate-elevation regression for each grid cell, and stations entering the regression are assigned weights based primarily on the physiographic similarity of the station to the grid cell (Daly et al., 2008). These models attempt to account for location, elevation, coastal proximity, aspect, vertical differences in atmospheric layers and orographic effects, although they can vary in the number and complexity of the factors involved. Validation procedures have shown that these products are generally robust on the scales for which they have been developed (Daly et al., 2008) and even at sub-grid resolutions, showing accuracy in regions characterized by sparse station data coverage, large elevation gradients, rain shadows, inversions, cold air drainage and coastal effects. Using these new gridded climate-data sets, previous studies have examined viticulture region climate characteristics at various resolutions, including in 18 wine regions in Europe (1 km; Jones et al., 2009), 50 PDOs and sub-PDOs in Portugal (1 km; Jones and Alves, 2012), 35 PGIs and PDOs in Greece (1 km; Anderson et al., 2014), 135 American Viticultural Areas in the western United States (400 m; , 63 Geographical Indications in Australia (500 m; Hall and Jones, 2010) and 21 wine regions in New Zealand (500 m; Anderson et al., 2012) and providing more holistic measures to help understand the range of climates within viticulture regions. Various climate parameters such as heat accumulation indices, frost timing, evapotranspiration and dryness indices are often used along with mapping and spatial summaries over delimited winegrowing regions, ultimately helping to define the climate component of terroir over time and space.
Much work has been done examining the likely impacts of climate change on water and/or nitrogen dynamics through models, such as Lebon's (adapted from the models in Riou et al. 1989Riou et al. , 1994Lebon et al., 2003) Lin and Host's , Soil and Water Assessment Tool (SWAT) (Martínez-Casasnovas et al., 2013;Ramos and Martínez-Casasnovas, 2014) and soil-water-atmosphere-plant model (SWAP) (Bonfante et al., 2011) hydropedological models that aim to simulate the dynamics in seasonal soil water balance. Kersebaum's model was proposed to simulate nitrogen dynamics (Nendel and Kersebaum, 2004), while the Simulateur mulTIdisciplinaire pour les Cultures Standard (STICS) model simulates crop growth, soil water and nitrogen balances driven by daily climatic data (Brisson et al., 2009); the Water baLance for Intercropped Systems (WaLIS) model simulates water partitioning in intercropped vineyards . WaLIS was also used to quantify the effects of water deficit and nitrogen stress on yield components (Guilpart et al., 2014). To model the seasonal pattern of evaporation from a grassed Mediterranean vineyard, Montes  (Costantini et al., 2006). Bonfante et al. (2011) have integrated the outputs of the SWAP model with a set of regional spatial data in order to map crop water stress indices together with soil map units, bioclimatic indices and potential radiation onto terroirs with simulated crop water status. However, scale issues need to be further addressed for both terroir zoning and other applications, such as the mapping of water stress. In vineyards, micro-variations in weather and climate often produce the greatest risk (e.g. frost or freeze zones, heat stress areas), and to truly address climate change we will need finer scales to assess the potential impacts (Quénol and Bonnardot, 2014).

Outlook: terroir sustainability assessment and the design of new preservation practices
The above-mentioned modelling approaches of terroir or terroir components with the aim of both spatially and temporally updating information lead to certain considerations in an emerging and complex study area: the sustainability assessment of terroirs. Some recent zoning studies using geospatial technologies have addressed this issue in terms of erosion (Table 2). Terroir sustainability started to be accounted for in the late 1990s, through some growers unions' initiatives in both California and Champagne and also through the growing awareness of soil contamination by copper due to the cumulated use of Bordeaux mixture (Vaudour, 2003). Mediterranean viticultural areas are amongst those most exposed and aware of the sustainability issue because of the soil losses due to intense erosional processes (e.g. Le Bissonnais et al., 2007;Martínez-Casasnovas et al., 2009Novara et al., 2011), which may lead to the destruction of a vineyard field (Fig. 6). Another issue is that of abandonment and land use change due to either urban pressure, aging of farmers or declines in profitability (Fig. 7). Depletion of soil fertility in general, along with the concomitant problems of weeds, pests and diseases, is the fundamental root cause of low agricultural production on a global level (Tan et al., 2005). Even if grapevines are not demanding with regard to nutrients, they nevertheless require an adequate supply, which may no longer exist in terroirs where the soils are undergoing degradation processes, especially soil losses (Le Bissonnais et al., 2007;Martínez-Casasnovas et al., 2009Paroissien et al., 2010;Novara et al., 2011;Quiquerez et al., 2014;Chevigny et al., 2014;Lieskovsky and Kenderessy, 2014), nutrient depletion (Ramos and Martínez-Casanovas, 2006), compaction (Lagacherie et al., 2006), salinization and sodization (Clark et al., 2002;Crescimanno and Garofalo, 2006; Costantini and Lorenzetti, 2013), pesticide runoff and deposition (Landry et al., 2005;Louchart and Voltz, 2007;Lacas et al., 2012;Daouk et al., 2013;Lefrancq et al., 2013) and copper contamination (Pieztrak and McPhail, 2004;Chopin et al., 2008;Komárek et al., 2008;Wightwick et al., 2008;Mirlean et al., 2007Mirlean et al., , 2009Rusjan et al., 2007;El Hadri et al., 2012;Fernández-Calviño et al., 2013;El Azzi et al., 2013;Lai et al., 2013). In Burgundy, using vine-stock unearthing-burying measurements (Brenot et al., 2008), Chevigny et al. (2014) estimated that the erosion rate had increased significantly over the last decade and also that the spatial distribution of erosion had changed and was now basically controlled by slope steepness and present-day vineyard chemical weeding and no-tillage management instead of past surface tillage. Using the same method in Languedoc, Paroissien et al. (2010) estimated that the average soil loss reached 10.5 t ha −1 year −1 and was much higher than the average erosion rates established around 3 t ha −1 year −1 for the other cultivated soils. The high vulnerability of vineyard soil to erosion may be partly explained by the steep slopes where vineyards are established, but also by the generally low content in organic matter (< 2 %) and the low microbial activity of soils, which leads to a reduced aggregate stability that increases soil crusting and soil erosion (Le Bissonnais et al., 2007). Very poor eroded soils can show low or very low nitrogen content, as a consequence of little soil organic matter. In the Pénédes viticultural region, Ramos and Martínez-Casanovas (2006) estimated that runoff processes exported significant amounts of nutrients, which represented as much as 8 kg ha −1 year −1 of N and 6.5 kg ha −1 year −1 of P. Moreover, the nitrogen deficiency can be enhanced in moderate to severe water stress conditions . According to Blavet et al. (2009), chemically weeded vineyards result in the highest runoff rates and soil losses, but the losses can be reduced when the prunings are left on the soil, when straw mulching is used, when rock fragments are left, and when grass intercrops are used. The corollary issue of runoff www.soil-journal.net/1/287/2015/ SOIL, 1, 287-312, 2015 is that, in addition to soil erosion and soil nutrient loss, it also leads to fertilizer and pesticide residue loss to surface waters, depending on the timing of the applied pesticide (Louchart and Voltz, 2007). In order to mitigate environmental damages and foster soil conservation and restoration, studies over the past decade have focused on the specific effect of one or several soil surface management techniques, such as cover cropping vs chemical weeding or tillage and/or mulching, on soil structural stability (Goulet et al., 2004;Ruiz-Colmenero et al., 2011, soil loss (Ruiz-Colmenero et al., 2011Novara et al., 2011;Lieskovsky and Kenderessy, 2014), pesticide leaching (Landry et al., 2005), soil nutrient and water management, vine growth and yield Belina, 2008, 2010;Ripoche et al., 2010Ripoche et al., , 2011Novara et al., 2013), soil C dynamics , and grape quality (Lee and Steenwerth, 2013). As recently described by Ruiz-Colmenero et al. (2013) and Agnelli et al. (2014), a greater organic-matter accumulation is fostered by the presence of the grass cover and the absence of tillage. In the Vosne-Romanée area in Burgundy, Landry et al. (2005) showed that glyphosate and its metabolite (AMPA) leached in greater amounts through a chemically treated bare Calcosol than through a vegetated Calcosol. In an upslope vineyard in the Beaujolais area, Lacas et al. (2012) demonstrated the usefulness of a grass buffer strip on a coarsely textured soil to limit the dispersal of diuron losses by runoff towards surface and subsurface water. Jacobson et al. (2005) studied the leaching of the herbicide diuron jointly with that of copper, through vineyard soils contaminated with copper, and found no direct interaction between the metal and herbicide but interpreted their findings as meaning that Cu was possibly affecting microbial activity, resulting in slight increases in diuron persistence.
Despite the varied environmental problems that viticultural terroirs are facing currently, putting their functioning at risk, the design of new preservation and mitigation practices has just begun to be addressed in the literature of the past decade. Although some methods to globally assess the sustainability of agricultural systems have emerged (Bockstaller et al., 2009), they have seldom addressed the sustainability of the viticultural system (Abbona et al., 2007), not only from the environmental point of view but also with regard to the attractiveness of the landscape, which, in many viticultural areas, adds considerable value to the wine produced and the region in question (Tempesta et al., 2010;Costantini and Barbetti, 2008).
The global water crisis, particularly water scarcity (Hanjra and Qureshi, 2010), that threatens those viticultural areas under semi-arid or arid climates, particularly in the Mediterranean area (Iglesias et al., 2007;Plan Bleu, 2013), questions the relevancy of irrigating, in addition to the degradative effects that this practise may have on soil properties over a long-term period and its high remediation costs (Hajkowitcz and Young, 2005). To address the issue of terroir sustainability in the years to come, one of the greatest challenges is the design of efficient soil restoration practices along with crop and/or intercrop management plans, taking into account the possible effects of climate change. This implies a complex multicriteria decision analysis, as attempted by Ripoche et al. (2010), in order to evaluate a range of intercrop management plans. The monitoring of soil quality as potentially obtainable from remote or proxy techniques is likely to identify when soil degradation is occurring and to allow management intervention. Amongst the key biological indicators are soil organic carbon, potentially mineralisable nitrogen and microbial biomass (Riches et al., 2013;Salome et al., 2014;Zornoza et al., 2015), although further research is still needed to identify a suite of biological indicators for viticultural soils. The application of organic amendments is likely to improve soil quality, but its effects are seldom studied, particularly during long-term experiments (Tatti et al., 2012). In rainfed Marchesi Antinori vineyards observed over two consecutive growing seasons, Baronti et al. (2014) suggested that biochar amendment could be used to improve soil water content, but other possible negative effects of changes in surface albedo or the accumulation of polycyclic aromatic hydrocarbons still need to be studied. In contrast, considering a 3-year period in a Valais vineyard (Switzerland), Schmidt et al. (2014) observed only small and mostly non-significant effects of either biochar or biochar-compost amendments. However, over the same duration in a vineyard in Jumilla (Spain), other organic inputs, such as winery and distillery waste composts, induced "an increase in the activity of the soil microorganisms and in the soil macro and micronutrient contents, as well as a slow release of inorganic N" (Bustamante et al., 2011) in a soil characterized by a highly calcareous sandy loam (Torriorthent). When soils are biologically very poor, even organic farming may be not enough to restore soil functionality, at least over a short time period . Coll et al. (2011) evaluated the longterm effect of organic viticulture on soil quality in commer-SOIL, 1, 287-312, 2015 www.soil-journal.net/1/287/2015/ cial vineyards of Languedoc, where plots which had been organically managed for 7, 11 and 17 years where compared to conventionally managed plots. The results emphasized that a transition period of 7-11 years, depending on the considered indicator, was needed to clearly separate conventional and organic farming. The overall benefits of organic farming were an increase in soil organic matter, potassium content, soil microbial biomass, plant feeding, fungal feeding and nematode densities, while its drawbacks were increased soil compaction and decreased endogeic earthworm density (due to reduced soil porosity), both consequences of the increase in the traffic for tillage and phytosanitary treatments in organic management. The grapevines studied by these authors were not intercropped with grass, and knowledge is actually scarce about the joint long-term environmental effects of the various possible sets of practices, such as cover or intercrop, mulching, mouldboard ploughing and/or tillage depth and frequency, type and quantity of applied organic amendments, and the use of chemical fertilizer. Further studies should also address the possible reintroduction of agroforestry systems jointly with vineyards, as traditionally practised in Mediterranean regions, such as in Ancient Greece (Vaudour and Boulay, 2013) or in Italy and Provence till the 19th century. Such systems are known to limit water consumption, fertilizers and pest diseases, particularly Botrytis cinerea bunch rot in Portugal (Altieri and Nicholls, 2002), and have positive impacts on phytoseid mite species, known for their ability to control mite pests, as shown in the last decade for vines co-planted with Sorbus domestica L. or Pinus pinea L. in Languedoc (Liguori et al., 2011).

Conclusions
Recent studies based either on metabolomics or on the Sr isotopic ratio lead to a strengthening of the assumption that geographical origin does leave a footprint in wines and that both soil and substrate, in interaction with climate and cultural choices, influence the shaping of grapevine phenology and grape and wine quality. Furthermore, the use of the current "omics" technologies seems to confirm the existence of a "microbial terroir" as a key factor in regional variation among wine grapes. Despite the role of soil microbial communities on terroir still being unclear, in the near future the combination of the omics techniques and traditional approaches could give further insights into activity and composition of vine-associated microbes, especially those living on the grape or leaf surface (phyllosphere) and root surfaces (rhizosphere) but also those within the plant tissues (endophytes), and their interactions with the plant and soil.
The differentiation and mapping of viticultural terroirs, in the sense of homogeneous regions of grape and wine quality, need comprehensive spatial modelling of soil, agronomical and climatic properties, including their changes through time. Because of this, a myriad of either remote-or proxysensing techniques develop, along with the corollary challenge of processing large quantities of data acquired at a very fine spatial resolution and/or at several spatial resolutions, scales and organizational levels. These techniques in data collection and processing are needed to produce easy-toupdate decision maps with associated uncertainties that allow users to make appropriate and timely management decisions. This is a revolution in the spatial management of terroir units, as the managed zones will be updatable and the effects of viticultural and/or soil management practices might be easier to control. The prospect of facilitated terroir spatial monitoring makes it possible to address another great challenge in the years to come: the issue of terroir sustainability and the construction of efficient strategies for assessing and applying adequate viticultural and/or soil practices across numerous scales. These include the design of efficient soil restoration practices along with crop and/or intercrop management plans and/or agroforestry viticultural systems that take into account the possible effects of climate change. Therefore, terroirs are more and more likely to be considered as part of the concept of ecosystem services, as viticultural agro-ecosystems, the services of which need to be constantly evaluated and rationalized.