United Nations projections estimate that the world population will continue to grow from the current 6 billion to about 10 billion by 2050 (FAO, 1996). The increase in population and the subsequent rise in the demand for agricultural produce are expected to be greater in regions where production is already insufficient, in particular in sub-Saharan Africa and south Asia (Pinstrup-Andersen et al., 1999). The necessary increase in agricultural production represents a huge challenge to local farming systems and must come mainly from increased yield per unit area, given the limited scope for extension of cultivated land worldwide (Evans, 1998).
The so-called Green Revolution - based on the introduction of improved varieties with high yield potential, together with technological packages (mineral fertilizers, pesticides, irrigation etc.) designed to significantly improve the cropping environment - has greatly contributed to the increase in agricultural production in several regions worldwide. Further expansion of this high-input model of agriculture is not sustainable, however, due to the high costs entailed and the negative impact on natural resources (Conway, 1998; Singh, 2000). In less favourable areas with poor ecological potential for crop production, where food insecurity also depends on the marked climatic fluctuations from year to year, the strategy has actually produced only low and unstable economic returns (McCown et al., 1992). Local germplasm is, therefore, often preferred to improved varieties because of its greater tolerance to severe biotic and abiotic stresses (Byerlee and Husain, 1993; Byerlee and Morris, 1993; Eyzaguirre and Iwanaga, 1996; Tesemma and Bechere, 1998; Almekinders et al., 1994). Furthermore, ongoing climatic changes may cause these areas to expand in tropical and subtropical regions (Rosenzweig and Hillel, 1998).
Plant breeding can be expected to assume a pivotal role in increasing the availability and stability of agricultural production in the future, particularly insofar as increasing attention will be paid to:
the sustainability of agricultural systems; and
the development of farming systems in less favourable areas (Sleper et al., 1991; Ceccarelli et al., 1992).
However, national breeding programmes, which are primarily concerned with these objectives, may need to modify some elements of their Green Revolution strategy to produce (when socio-economically convenient) improved germplasm capable of maximizing the agricultural potential of specific areas and farming systems, and of minimizing the occurrence of crop failures or very low yields in unfavourable years. Integration of farmers in the selection process, commonly defined as participatory plant breeding, may help fulfil these objectives, in addition to facilitating the adoption of novel germplasm (Eyzaguirre and Iwanaga, 1996; McGuire et al., 1999; Weltzien et al., 1999). In the long term, biotechnology - if combined with forms of international cooperation and intellectual property rights legislation promoting such techniques for national programmes (FAO, 1999) - may help achieve adaptation and yield stability targets, especially by using plant material with increased tolerance to prevailing biotic and abiotic stresses.
Improved adaptation and yield stability may derive in the long term from the definition of an appropriate breeding strategy, and in the short term from the appropriate choice of cultivars (whether indigenous or foreign, and either traditional or released from public or private breeding institutions). Ideally, decisions concerning the breeding strategy and crop varieties should be based on scientific knowledge of the plant material and its relationship with cropping environments within the target region. This publication highlights the contribution that data from multi-environment trials can provide in this respect. Results of studies relative to similar regions worldwide, as well as common sense and practical considerations, may also contribute to the decision-making process, especially where there is a lack of more objective information based on experimental data.
It is well documented (Cooper and Byth, 1996) that the investigation in breeding programmes of adaptation and yield stability have been modest. This shortfall exists, despite the importance of these issues, the substantial investment by public and private institutions in multi-environment testing, and the wide range of statistical methods available. Furthermore, varieties (or other techniques) are rarely recommended on the basis of a thorough assessment of adaptation and yield stability characteristics. An inversion of this trend probably requires that ordinary breeders and agronomists be sought as the main users of these methods. Relatively simple techniques applicable through friendly, inexpensive software have an obvious appeal in this context.
With regard to the comparison of plant material in a set of multi-environment yield trials, the term genotype refers to a cultivar (i.e. with material genetically homogeneous, such as pure lines or clones, or heterogeneous, such as open-pollinated populations) rather than to an individuals genetic make-up. The term environment relates to the set of climatic, soil, biotic (pests and diseases) and management conditions in an individual trial carried out at a given location in one year (in the case of annual crops) or over several years (in the case of perennials). In particular, an environment identifies a given location-year (annuals) or location-crop cycle (perennials) combination in the analysis of trials repeated over time.
Purely environmental effects, reflecting the different ecological potential of sites and management conditions, are not of direct concern for the breeding or recommendation of plant varieties. Genotypic main effects (i.e. differences in mean yield between genotypes) provide the only relevant information when genotype × environment (GE) interaction effects are absent or ignored. However, differences between genotypes may vary widely among environments in the presence of GE interaction effects as large as those reported in extensive investigations (e.g. DeLacy et al., 1990; Annicchiarico, 1997a). In general, GE interactions are considered a hindrance to crop improvement in a target region (Kang, 1998). Moreover, such effects may contribute, together with purely environmental effects, to the temporal and spatial instability of crop yields. Temporal instability, in particular, has a negative effect on farmers income and, in the case of staple crops, contributes to food insecurity at national and household level. On the other hand, GE interactions may offer opportunities, especially in the selection and adoption of genotypes showing positive interaction with the location and its prevailing environmental conditions (exploitation of specific adaptation) or of genotypes with low frequency of poor yield or crop failure (exploitation of yield stability) (Simmonds, 1991; Ceccarelli, 1996).
Growing awareness of the importance of GE interactions has led crop genotypes to be ordinarily assessed in multi-environment, regional trials for cultivar recommendation or for the final stages of elite breeding material selection. GE effects should not be ignored, rather analysed using appropriate techniques, in order to explore the potential opportunities and disadvantages. Provided the information from these trials responds to certain prerequisites, it can help breeding programmes to:
better understand the type and size of the GE interactions expected in a given region, and the reasons for their occurrence; and
define, if necessary, a strategy to successfully cope with the effects of interactions.
The most important GE effects for targeting cultivars or for selection of material are the crossover type affecting top-yielding genotypes. Such effects imply a change of ranks between environments rather than a simple variation in the extent of the difference between genotypes (Baker, 1988). However, all GE interaction effects arising from lack of genetic correlation among environments (including those relating to low-yielding material and not necessarily of the crossover type) can be relevant if the results for a given data set are extrapolated to produce information on the GE effects that are likely to be met in breeding for a target region (Muir et al., 1992; Cooper et al., 1996a).
Reasons for the occurrence of GE interactions are thoroughly discussed elsewhere (e.g. Bidinger et al., 1996; Kang, 1998). Major interaction can be expected when there is: on the one hand, wide variation between genotypes for morphophysiological characters conferring resistance to (or avoidance of) one or more stresses, and, on the other, wide variation between environments for incidence of the same stress(es) (as determined by climatic, soil, biotic and management factors). For example, this situation may arise where there is wide variation between material in terms of intrinsic drought resistance or earliness of crop cycle, and between environments in the level of terminal drought stress. Other pertinent examples may concern the differential response of genotypes to variable levels of stress, such as low temperature, soil salinity, nutrient deficiency, pests, diseases, lodging, grazing or interspecific competition in mixed cropping, as a consequence of: genetic variation in tolerance to biotic or abiotic stresses; ability to capture and use nutrient resources; competitive ability etc.
Large GE interactions have frequently been reported between pairs of environments with contrasting levels of one major stress (Ceccarelli, 1989; Bramel-Cox, 1996), defined as favourable when characterized by low stress and high mean yield and unfavourable with high stress and low yield. However, large interactions may also occur between pairs of unfavourable environments and even between pairs of moderately favourable environments possessing similar mean yield but with differing combinations of stresses or patterns of one major stress (Annicchiarico, 1997a). For example, an environmental factor such as soil texture may produce similar mean yield of material but sizeable genotype × soil texture interaction (Koutsos et al., 1992). The level of matching of genetically-based determinants of phenological development (e.g. photoperiod and vernalization requirements) with site characteristics related to the length of the growing season (e.g. day-length and temperature patterns) is another determinant of remarkable GE interaction, especially across relatively large regions (Wallace et al., 1993a, 1993b).
The genetic structure of plant material may also have a bearing on the extent of GE interaction. Variety types characterized by low levels of heterogeneity (e.g. pure line, clone, single-cross hybrid) or heterozigosity (e.g. pure line) tend to interact with the environment more than types with opposite features (e.g. open-pollinated population, mixture of pure lines), because the lower richness in adaptive genes implied by their genetic structure makes them more susceptible to variation in environmental conditions (Becker and Léon, 1988; Brancourt-Hulmel et al., 1997). Indeed, harvest security is associated with diversity between and within traditional varieties for farmers producing near subsistence level (Clawson, 1985).
A detailed description and discussion of various aspects of GE interaction analysis is available in numerous review articles (Freeman, 1973; Hill, 1975; Denis and Vincourt, 1982; Westcott, 1986; Lin et al., 1986; Becker and Léon, 1988; Crossa, 1990; Romagosa and Fox, 1993; Cooper and DeLacy, 1994; van Eeuwijk, 1995; Brancourt-Hulmel et al., 1997; Kang, 1998), in papers included in the books edited by Williams (1976a), Kang (1990), Kang and Gauch (1996), Cooper and Hammer (1996a) and Kang (2002), and in the monographs by Gauch (1992), Prabhakaran and Jain (1994) and Basford and Tukey (2000). This publication focuses mainly on a limited number of analytical techniques that, owing to the quality of information, the ease of application (also in relation to the recommended software) and the limited amount of input data required, can be considered of major interest for:
breeding programmes, especially national programmes in less developed countries, in order to increase knowledge of GE interactions and, possibly, modify the strategy accordingly;
breeding programmes, as well as national institutions committed to testing and recommending crop varieties, in order to assess adaptation and yield stability patterns of the available germplasm and exploit the information for more effective selection and targeting of material.
The second point may be extended to the assessment of the adaptability and yield stability of different agricultural techniques (e.g. soil tillage, weed control, sowing date) compared on a multi-environment basis (Gauch and Zobel, 1996a; Piepho, 1998).
Genetic improvement is a basic component of national policies for raising agricultural production in many countries. Although breeding programmes run at international centres of the CGIAR (Consultative Group on International Agricultural Research) system have made an important contribution to major food crops, national breeding programmes (public or private) maintain a fundamental role for crop improvement in their target region (usually applicable to the whole country), particularly insofar as the exploitation of specific adaptation and yield stability characteristics are concerned, owing to their better knowledge of, and easier access to, local germplasm and cropping environments. These programmes aim to make correct decisions on a number of issues which comprise a breeding strategy (Simmonds, 1979). Decisions may concern, in particular:
adaptation strategies, yield stability and other (e.g. crop quality) targets;
genetic resources forming the genetic base (indigenous or exotic, from traditional or improved varieties);
techniques for the recombination and introgression of useful genetic variation;
variety type (e.g. single-cross hybrid, double-cross hybrid, improved population or synthetic variety, with regard to outbred species); and
breeding plan and selection procedures (selection environments, indirect selection criteria, presence and extent of participatory breeding, experimental designs etc.).
The definition of a strategy with respect to GE interactions may require decisions on most of these elements, namely: adaptation strategy and stability targets, genetic resources, variety type, breeding plan and selection procedures. Initial decisions may change with time as a consequence of new opportunities offered by scientific progress, experimental evidence, available funding, food security policies, changes in national seed systems, international cooperation etc., but they should remain consistent with the breeding objective. For example, the inconsistency between targeting also unfavourable areas and adopting genetic resources and selection procedures producing material specifically adapted to favourable environments has contributed to the partial failure of a number of breeding programmes carried out in the Green Revolution context (Simmonds, 1979; Ceccarelli, 1994).
It is worth noting that the potentially extensive application of genetic engineering techniques does not eliminate the need for breeding programmes to cope with GE interactions, because almost no cultivar can assemble genes conferring superior performance in all environment types within a relatively large region. This derives from genetically based trade-offs between yield potential and tolerance to major stresses, e.g. drought (Ludlow and Muchow, 1990; Acevedo and Fereres, 1993), as well as from the need to choose between incompatible levels of a key adaptive trait, such as earliness of flowering (Wallace et al., 1993a). Also the possible selection for yield based on molecular markers may require a preliminary definition of adaptation and yield stability targets, since a remarkable portion of useful markers are environment-specific (Paterson et al., 1991; Hayes et al., 1993). Indeed, the potential benefits of exploitation of environment-specific markers may justify, in the long run, greater emphasis on specific adaptation strategies. In a wide adaptation prospect, marker-assisted selection may prove distinctly less effective than multi-environment, phenotypic selection for yield in the presence of relatively large GE interactions, especially when epistatic effects contribute significantly to genetic control (Cooper et al., 1999).
 Ibid., p.
 Ibid., p. 358.