Department of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences, SE-901 83, Umeå, Sweden.

This is a key note from the SNS conference 1999. It is published in: Lundkvist K. (editor). Rapid generation turnover int the breeding population and low-intensity breeding. Department of Forest Genetics, Uppsala, SLU, Sweden. ISSN =0348-565X. Research Notes 55: 37-48. 2000.

High-intensity breeding is well represented in the literature and in formal training programmes. Methods particularly suitable for low-intensity breeding may not therefore be used or used inefficiently as little thought is given to adequately develop, implement and optimise such methods. A number of low-intensity techniques are discussed, in particular: phenotypic selection; absence of testing; plantations with identified plants and forwarding the breeding population by wind pollination instead of crosses. Low-intensity strategies will require rather advanced quantitative genetic considerations to be sustainable and optimal, in particular predictions how relatedness increases.

Low intensity breeding is breeding that is cheap, convenient, simple, robust and has a small total budget. In contrast, high intensity breeding is less constrained by total budget and uses techniques that are relatively expensive, complicated and demanding on competence, but are expected to return higher and more reliable gains. It fills no useful purpose to formulate exact definitions as the difference is not strict, it is partly a semantic problem, it depends on the exact circumstances and it is often somewhat controversial. Examples of common techniques, which may be considered high-intensive, are grafted seed orchards, controlled crosses with full pedigree (known pollen parent) and genetic testing. These measures are usually justified and affordable for conifer species of major economic importance, but may be debatable for programs with small resources. Examples of such minor programs are those supporting planting of trees in urban and rural areas for parks and landscapes, where some low-intensity strategies have been discussed, e.g., by Lagerström & Eriksson (1996). Another case is such exotics, which are used in a small or experimental scale and have an uncertain future. Low-intensity breeding techniques are probably relevant for many breeding activities in developing countries, and in environments where competence is poorly developed and organisation unstable (Namkoong et al. 1980). The techniques involved can play a role as a form of safety device and back-up to high-intensity breeding programs. Genetic management of the best genetic tiers with genetically narrow high-intensity aggressive breeding for high gain may be complemented with low-intensity breeding of lower tiers for wide diversity and large breeding population (e.g. nucleus breeding, cf. Cotterill 1989).

Low-intensity breeding techniques can, if used wisely, be useful for gene conservation, assuring that more threatened genetic resources will be available in the future at a lower effort. Gene conservation should make possible the harvesting of seeds with reasonable gene diversity, genetic variability and adaptation for renewing the gene resource and should preserve options to support future tree planting and to initiate a breeding program. A low-intensity breeding program can meet these goals, with emphasis on high genetic diversity rather than improving economic traits. A high-intensity program with components of gene conservation may be less efficient, as it directs the same resources to fewer parents and a smaller breeding population accompanied by a larger loss of gene diversity. Or resources may be too small to support anything else than a low-intensity program and when the alternative to low-intensity may be no intensity and no intentional gene conservation.

In other situations, there may be a more immediate demand for improved regeneration materials produced in a cost-efficient way, but the projected quantities are small and budget constraints do not justify a more expensive, high-intensity approach. For many species, elements of the biology may not be sufficiently known, e.g. the reproductive biology may be too poorly known to make successful crosses. There may be reasons to avoid methods that depend on a stable access to funding, technical expertise and technical equipment. Costs are scale-dependent; thus a method which is cost-effective in a large program may be less cost-effective in a small program.

The development of tree breeding science is mainly driven by the needs of large programs. Consequently, there is a risk that methods suitable for large programs are uncritically implemented in small programs, even if they are less suitable for that purpose. There is rather little thought given to low-intensity breeding strategies, and not much is found about it in the more established literature. In the few more known examples there are (e.g. Namkoong et al. 1980, Shelbourne 1992, Barnes 1995), the focus is still on rather fancy techniques, and little effort on optimising real low-intensity programs. It seems the case that tree breeders feel more motivated to focus attention on how well things can be done rather than how cheaply they can be done. It is also natural that intensive programs employ more, better trained and more specialised breeders, who work in more affluent and stable environments, and it is these breeders and their supporting scientists who create the literature and common knowledge. It is easier to make science and appear visionary on fancy techniques (like different variants of molecular breeding), rather than on somewhat sloppy, low-intensity methods. Even in less-developed environments, there is a strong will to use "modern" methods in the research frontier, rather than being cost-efficient. Many of the advantages of high-intensity over low-intensity methods are that the materials become better known, the activity more scientific and the actions more predictable; this is attractive, and may over-shadow cost-efficiency, even if the latter ought to be a more important goal of tree improvement. The outcome of planned tree breeding operations or science is based on uncertain predictions, analogies and belief rather than proven facts, this leaves much room for arguing that fancy methods may be economically efficient even when they are not. A robust, low-intensity strategy is more able to survive neglect for some time in an environment with irregular and unreliable funding. That the local management system is irregular and unreliable and that some budget downgrading is acceptable are arguments which breeders will hesitate to publish. The demand on genetic calculations may be higher for low-intensity methods, but these calculations may be more difficult to make and explain than with more accurate methods, thus low-intensity-methods may have more difficult to get acceptance by decision makers. Thus, cheap and low-intensive strategies have few spokesmen or scientific investigators or promoters.

In the following, some techniques will be discussed, which can be implemented in low-intensity breeding strategies. Even if the text is written with low-budget breeders in mind, some techniques may also have relevance for high-intensity situations. It can also be noted that, depending on special circumstances, it may sometimes be wise to use sophisticated techniques in low-budget environments. The suitability of a particular technique depends on many circumstances, and therefore no sharp borderline between high-intensity and low-intensity can be drawn; some concepts described as low-intensity can be implemented in some high-intensity strategies and vice versa.

Phenotypic selection (mass selection) means selection based on the appearance of the individual tree itself, not considering the performance of relatives. In a study of the genetic gains obtainable from a range of alternatives, Shelbourne (1992) found that phenotypic selection in unpedigreed stands, although somewhat lower in genetic gain than more elaborate strategies, still seemed favourable when rapid generation turn-over and lower cost was taken into account. By considering the performance of relatives, greater gain can be achieved at the same selection intensity, but the additional gain is accompanied by a correspondingly higher relatedness among the chosen individuals. E.g. while selecting among the first generation of offspring, selected trees will be sibs to a higher extent, if sib performance is considered. If comparisons consider gene diversity as well as gain, the case for phenotypic selection is strong. When compared at the same gene diversity among the selections and the same selection intensity, phenotypic selection and combined-index selection (with restrictions on the highest number of offspring per parent) produce approximately the same genetic gain (Wei 1995, Spanos et al. 1997, Andersson et al. 1998). Phenotypic selection seems to be competitive for achieving genetic gain at a given level of gene diversity even when repeated for many generations in material with complex and unbalanced pedigrees (Spanos et al. 1997, Andersson 1999). When only the initial offspring are considered, it may be argued that combined-index selection or similar methods are able to produce a higher genetic gain than phenotypic selection, and foresters may not care that this is associated by a large reduction in gene diversity. This can be considered as an argument against phenotypic selection in the short time perspective. Reduction in gene diversity will, however, lead to reduced gain in later generations, so after around five generations the maximum gain obtainable has been found to be about as high for phenotypic selection as for combined-index selection with restrictions (Andersson et al. 1998). These findings concerned with forest tree breeding are supported by similar results by animal geneticists (e.g., Quinton et al. 1992).

Phenotypic selection in a plantation, without knowing or caring about the pedigree of the trees, is about as efficient for long-term breeding as if the pedigree is known and used as an aid in selection (Wei 1995, Andersson 1999). Approximately, and for a given number of selections, this conclusion can be formulated: no alternative set of selections to that obtained by phenotypic selection offers higher genetic gain unless the selections are more related, and selections will be less related unless their genetic gain is lower. Phenotypic selection is rather efficient, but it is just one among a whole array of efficient solutions, which differ in their relative emphasis on gain and diversity. What the breeder loses by using phenotypic selection is the option to make a deliberate choice between gain and diversity. Numerical simulations indicate that phenotypic selection often results in an intuitively appealing compromise and seldom results in a drastic increase in relatedness. Average relatedness can initially be set at a sufficiently low level by selecting a sufficient number of trees. Later selections can be made by phenotypic selection in plantations of descendants from the initial selections, but the number of selections transmitting genes to the following generations must be sufficiently high. What can be considered as sufficient is mainly dependent on the average relatedness, which will build up as a result of the selections.

Phenotypic selection is the method used by the nature as the driving force for evolution, and has thus existed for millions of years. It may give some comfort to use a "natural" method, even if it also seems unsatisfying that modern science can not do better than nature. The similarity to nature makes it very likely that it is sustainable. It is very simple. It does not require sophisticated training. The selection situation is similar to forestry practice (e.g. similar mixture of genotypes). It minimises or eliminates the work of experimental layout and documentation. It does not use a complicated algorithm that works like a "black box", able to disguise severe mistakes. It is cheap. It is easy to keep track of data. It is easy to explain. Heritability in phenotypic selection is usually rather low (5-20%). That means - in contrast to selections based on genotype testing - that even the characters not regarded as desirable will remain among the trees selected to constitute the parents of the following generation, thus phenotypic selection has attractive properties for gene conservation. In many countries supporting expensive breeding programs, much of the gain deployed to today’s forests originates from phenotypic selection of plus-trees, thus it can be claimed to be the method with which forest tree breeding has the longest experience, and this experience is generally a positive one.

Selection based on the phenotype need not depend on actual measurements; a subjective evaluation of all characters of a tree (including its performance compared to its neighbours) can be made in the field. Such an evaluation may actually be more accurate than an objective measure, as all characters can be jointly considered including the local environment (thus the performance of the neighbours). It does not depend on a long chain of actions on different places and by different people where mistakes or delays may easily occur. In practical selection there is usually a considerable amount of subjectivity at the final stage, this can not be avoided by high-intensity techniques.

Phenotypic thinning for unpedigreed seed production areas and also plus tree selection has been discussed, advocated and applied e.g. by Harwood et al. (1996). As low-intensity breeding must usually rely on phenotypic selection, it must be pleasing for those applying it to know that it is now acknowledged by modern science as an efficient tool (Wei 1995, Andersson 1999). It is also works in multigenerational scenarios (Andersson et al. 1998). Thus, seed production areas can be established, and the resulting forests later creamed for the best unpedigreed trees, whose descendants are used for establishing new seed production areas, making low-intensity breeding sustainable over multiple generations (provided the numbers involved are sufficiently large).

Conventional progeny testing or clonal testing can be seen as a way of boosting the heritability when selecting individuals. It may seldom be place for this type of testing to get more precise breeding values in low-intensity breeding. A program based on testing needs field identities, documentation, long-term planning, long-term co-ordination of activities, and, often, clonal archives. A strategy including testing means a large investment, which seem possible to justify only in situations when it seems certain the results will be utilised and appreciated in some future. There are many studies indicating that it is not a general rule that testing genotypes is an efficient use of resources. E.g. a study by Routsalainen and Lindgren (1998) showed that, with few exceptions, if offspring were generated with pollen from the tested population, forward selection was generally superior to backward selection (i.e., with testing). Libby (1969) optimised gain from selection in a material from a nested mating design and found that the optimal number of ramets per clone for all cases studied was one, so that clonal testing may also not be cost-effective. That, and other theoretical results, led Frampton and Foster (1993) to suggest that the best strategy for clonal testing may be to test only one or a few ramets, which supports that the benefit from clonal testing may often be negligible.

Some species are as easy to propagate and handle as vegetative propagules as they are from seed, or vegetative propagation may actually be the only practical way to produce plants. There are many potential advantages with vegetative propagation (e.g., Ahuja and Libby 1993). Often there are technical difficulties with the propagation method, and under such circumstances investment in development is not worthwhile for low-intensity breeding. If reasonably large, well-adapted and well-formed trees can be selected and vegetatively propagated for practical use, it seems recommendable to use that option. Vegetative propagation is an evolutionary dead end, thus actions must be considered to assure recombination even for cases where the main propagation methods are vegetative. This can be done by establishing "gene resource plantations" (see below). It must be assured that relatedness remains low and gene diversity high, but that can be done with calculations in combination with sufficient numbers.

A possible application is to use phenotypic selection in a "gene resource plantation" as a first step. In a second step, vegetative propagules of the best selections could be placed together in a stand. A stand with some type of selected clones would actually be equivalent to a grafted seed orchard. In that way, the pollen would come only from good phenotypes and the selected phenotypes would originate from an improved population. If clones were placed in row plots, the worst-looking clones could be removed. Clonal identity and field maps would not be required for this operation; whether or not it is a good clone can be judged by evaluating the row plot, without knowing the identity of the clone. Seeds from the better clones could be used for the next cycle of gene resource plantations. Coancestry and inbreeding can be kept under control by calculations in the same way as for phenotypic selection, clonal selection can be considered equivalent with phenotypic selection with high heritability. To use clone plantations for seed harvest has been suggested as a cheaper alternative to seed orchards of Norway spruce in Sweden (Lindgren and Karlsson 1993).

Low-intensity programs must be concerned with inbreeding, relatedness and diversity. These factors must be predicted to manage gene resource plantations and to plan seed collection. In high-intensity programs, control can be kept by known pedigrees, individual identification and selection algorithms that utilise this knowledge. In low-intensity programs, the actual operational control may be relaxed, and thus it becomes more important to forecast what will happen by appropriate use of theoretical predictions and by reasonable estimates of key factors. What happens depends mainly on the gene pool of the population and variations in the contributions of individuals to the next generation. To predict what happens is an advanced operation. The likely consequences may be forecasted by simple heuristic rules, tables and instructions for less-advanced users, but as future low-intensity breeders will very likely have access to competence and computers, the need of predictions will seldom be bottlenecks.

The gene pool of the offspring is the same as the gene pool of the successful gametes of the parents; this connection links generations. It seems natural to link fertility to successful gametes, but it is not known what gametes will be successful in advance and there will be stochastic variation, which is high if low-intensity measures are used. It is thus more useful to define fertility as a characteristic of the paternal genotype. Fertility is defined as "a parent's ability to produce successful gametes". The true number of successful gametes per parent is both technically and principally difficult or impossible to estimate (e.g. there is no unequivocal definition of "successful"). Quantitative estimation of variation among trees in female or male reproductive structures can be made even for a low-intensity program, and it seems likely that such counts will be sufficiently accurate for most situations. Anyway, the differences in fertility among considered objects or years are likely to be more important that the inaccuracies of the estimation method.

A quantification of fertility differences among a group of parents in probabilistic terms can be made as a basis for predictions and theoretical development. The parameter A refers to the probability that two gametes, chosen randomly from the gene pool of gametes, originate from the same parent, compared to that in the gene pool of the parents. It could be called "sibling coefficient" as it refers to the probability that individuals share the same parent, and thus are sibs. Mathematically, A can be defined as A=NS p_i², there N is the number of individuals and p_i is the relative fertility of individual i. Relevant theory is developed in papers by Bila and Kang, e.g., Bila & Lindgren (1998) and Kang & Lindgren (1998 and 1999). "The effective number of parents" can be expressed as N/A, which often seems to be identical to the classical "variance effective number". A is a function of the coefficient of variation for fertility and A=2 corresponds to a coefficient of variation of 100%. A for a forest stand may typically be 2 (Bila & Lindgren 1998). A=2 means that there will be twice as many sibs among the seeds as expected if all matings were equally frequent. A=2 means that relatedness and later inbreeding will build up over generations twice as fast as if there were random mating. Reasons to suggest "sibling coefficient" to be useful in predictions of generation shifts are that it is independent of the number of members in the population, that it focuses on probabilities and that it has transparent interpretations as mentioned above. An example of this calculation technique is demonstrated in Bila et al. (1999). The loss over generations is predicted when seeds are collected and a small sample of these seeds is used to replace the stand (a gene conservation stand). If a limited number of offspring is considered, the successful gametes can be seen as obtained by sampling from all gametes. An option to increase the effective number, thus to reduce group coancestry, loss of gene diversity and subsequent inbreeding, is to keep female fertility constant by mixing the same amount of seeds from all trees. This is a rather effective measure. The technique can also be used to trade off gain and gene diversity in the seed crop or gain and gene diversity in the stand itself (low-intensity breeding). The idea is that inbreeding and the associated phenomenon can be kept manageable and balanced against cost, gain and other desiderata by management technique and numbers, rather than keeping exact pedigrees. Such calculations fit well in a low-intensity program.

Relatedness and gain are the major outcomes of breeding, and there are optimal combinations of them in the sense that under given constraints and gene diversity, there exists an optimal strategy that maximises gain. An optimal strategy may be conservative or aggressive depending on the demand of gene diversity. It is a good breeding practice to allow the better trees to be over-represented in both breeding populations and production populations. Truncation selection is the common way of achieving this; it draws a strict border between pass and fail. A more gradual differentiation in treatment of materials relative to their goodness is more optimal. Sophisticated algorithms for identifying such strategies have been developed for deployment (linear deployment, e.g., Bondesson & Lindgren 1993; Wei & Lindgren 1995) and selection (group-merit selection, Lindgren & Mullin 1997, Andersson 1999). While paternal fertility may be unknown, it is still possible to trade off against female fertility, and thus pick more seeds and plant more plants from the best female parents. There seems to be a need to develop techniques that are simple to handle, but still close to optimal for low-intensity breeding. Many optimisation techniques require that plants be identified by family within gene resource plantations; this is an argument for family identification in the field. These techniques will contribute a better guarantee that diversity is preserved and make the balance between gain and diversity more of a deliberate choice. It may, however, be doubted if it is worth keeping identifications in a low-intensity program for this reason, but if this is done for other reasons, then the possibility could be used. Another application could be to consider the family value for low heritability characters or characters such as survival, as a guide to how many to select from each family, but to select phenotypically within family (preferable for other characters).

Professional breeders spend much attention on situations where large investments are made to establish installations for seed production. The idea to use grafted superior individuals as parents is attractive. If superior trees are grafted, all their genes are utilised, while if just their open pollinated offspring is used only half of the genes will be from those trees. In the grafted seed orchard not only the seed parents will be plus trees, but also the pollen parents. The intensively managed grafted pine seed orchard has worked rather well for many of the most influential breeding programmes and has thus become the standard model. Subsequently, breeders acting in low cost environments often also see the grafted orchard as the ideal because of influences by training and literature. However, the use of grafts demands organisation, planning and development of practical skills. Grafting demands skilful experienced workers and this is complicated if there is not some scale and continuity. Grafts are sensitive to damage, sensitive to competition from other vegetation, expensive, not easy to keep in backup reserve, and need continuous tending by experienced professionals. The intensive management at the start of the seed orchard means a large investment, which is justified only when the advantages are large and certain; one prerequisite for that is stability over time. They are not likely to remain operative over long periods if funding for their management is withdrawn. They constitute a type of land management, which is not typical and may therefore cause irritation. The large investment, the need for attention and perhaps also need for internal pollination may force grafted seed orchards to be concentrated in large expensive units, where they can be conveniently managed.

There may be a place for low-intensity seed production units even in high-intensity programs using grafted seed orchards. Lindgren & Karlsson (1993) suggested for Norway spruce in Sweden that the two projections of the need for seed are made: the lowest estimate could be met by grafted seed orchards, and low intensity gene resource plantations could be used to meet the difference between the high and the low estimate. The economic loss by investments, which never get a return, will be low if the low estimate will be realised. If the reality hits the high estimate, the loss in genetic quality will be limited, in particular considering the possibilities to expand on the best of the grafts (e.g. by cutting propagation).

"Gene resource plantations" are suggested as the solution to some of the problems discussed above. This concept is often similar to or may be interpreted as a low-intensity variant of "breeding seedling orchard (BPO)" (e.g. Barnes 1995) or "unpedigreed (or half-pedigreed) seed production areas" (Harwood et al. 1996). Often these may be established as a mixture of half-sib families obtained from open pollination on selected trees (from a previous cycle of gene resource plantations). Similar plantations constitutes a part in many breeding programs, but when documented, they still often appear without focus on minimising costs and maximising simplicity or optimising the procedures used. The gene resource plantations may be established in a similar fashion to regular plantations. If the gene resource plantation is similar to - and is managed in ways not too different from - an ordinary stand, it does not require much special competence. It can be managed with the existing competence and organisation of forest caretakers. There are less constraints on the location. These features may make it easier to find localities where flowering, seed production, seed quality, land access, representavity for plantations, difference for desirable characters and pollen isolation will be favourably combined. A gene resource plantation can be small and cheap, and the same concept can be used for many species and sub-populations. A gene resource plantation can perhaps sometimes be called a seedling seed orchard, but seed production may not be the only aim. It is comparatively easy to arrange for multiple use so that it serves concurrently all or some of the functions of conservation of the species; conservation of the genetic diversity of the species; seed source for plantations; production of desirable commercial products; and as a local demonstration plot, while retaining options to initiate a more regular improvement program. If there are subsides for commercial plantations, these can be applicable for the stand, making it an economically attractive alternative. The idea to harvest commercial seeds in objects that look like a forest or progeny test has been used, for example in improvement of coastal Douglas-fir in Oregon and Washington (Silen and Wheat 1979).

Plantation practices commonly employed in many less developed regions are often less uniform than industrial plantations. Seedling quality is often low, planting quality poor, replanting common, survival low and stocking patchy; species are sometimes mixed and agroforestry is often practised. Gene resource plantations are likely to function much better than regular progeny tests or grafted seed orchards under these conditions, as they are robust and less sensitive to irregularities. However, the gain by phenotypic selection will be larger if selection is practised in uniform conditions, and the risk for loss of whole families is larger if conditions are sloppy, thus improved plantation techniques probably combine well with many variants of low-intensity breeding.

If a gene resource stand is to function as a seed source, some investment should be made in thinning and managing the stand for that purpose. Seed collection may use methods that damage the production of the forest, e.g., the most economic way of harvesting the seeds may be to fell trees. These costs appear close to seed harvest, when it is known that the seeds are needed and the investment worthwhile. It is not like a grafted seed orchard, which depends on a large investment long time before the benefit appears, and the investment make it doubtful to let seed collection harm future seed production.

A plus tree in a plantation represents a good tree while a seedling from a plus tree just has a good seed parent. The idea of clonal seed orchards is that the pollen parents are also selected. However, this expectation of pollination with plus tree pollen parents has not been realised as expected in many seed orchards, we must often accept that up to half of the pollen parents are often found outside the orchard (Lindgren 1991). Authorities (like EU, OECD and national boards of forestry) feel an interest to keep genetic diversity under control, but the only factor that can be conveniently and objectively audited, is census numbers. Therefore, much legal attention and restrictions go to numbers, and these numbers tend to be high. This attitude by authorities will probably seriously constrain innovations and gain. Seedlings are genetically different and it is therefore likely that legal worries will be less severe for plants from seedling seed orchards (seedling stands). Seeds are much easier to handle than vegetative propagules. The gain from the initial step of selecting seed parents is only half of that of selecting the whole genotype, which is a severe draw back. However, half-sib families have four times the effective size (status number) as clones from the same plus tree (Lindgren et al. 1996). Thus, some of the gain lost in the first step can be recovered by using the excess diversity to make more radical selection later. Another way of phrasing it is that a plantation obtained from open pollination of 100 trees can harbour four times as much gene diversity as a seed orchard with 100 clones.

The "gene resource plantations" can initially be established by selecting desirable trees in the forest, harvesting seeds from them, use a number of seeds from each mother tree to produce plants, and establish one or (mostly) several plantations from that. Desires on the site may include accessible, not very small, and appropriate for the species (more factors are discussed above). There is a minor advantage if there is some isolation from possible pollen sources outside the stand. It can be advantageous to establish a gene resource plantation similar to a commercial plantation. That means that selection effects will make the genetic of the materials more suitable for plantation forestry than natural seed sources, that management will fit better into forest operations and that it is more likely there is a commercial use for the production besides its genetic functions.

The planting should probably use a rather close spacing to allow later thinning. Identity of trees need not necessarily be known (thus no records on the family identity are required), as calculations can predict that the number of trees and harvested mothers is large enough. To keep known identities requires: 1) keeping family identity known at plantation; 2) mapping the area after plantation; 3) placing identification markers in the field; 4) maintaining identifications; and 5) keeping maps available for long time. Thus, known tree identities require considerable cost, fails sometimes, is more requiring on competence and means there must be organisational stability.

The gene resource plantation should probably be subject to rather intense silviculture and thinning so that desirable, well-adapted phenotypes are favoured. Establishment and management depends partly on the aims; if the goal is more to support a good seed source, then close spacing with subsequent thinning is desirable, while if the aim is just conservation, close spacing is not desirable. Even if the purpose is gene conservation, it should be remembered that the healthy well-adapted trees remaining after thinning and selected for harvesting would probably also be favoured by natural selection. Thus, a gene conservation goal is probably better met by a larger number of trees and more sites than by very low selection intensity. A close initial spacing and subsequent thinning increases the cost. This is more motivated the more certain is the need for improved seeds; if it seems too uneconomic it could be cancelled. There could be specific considerations that need to be made, e.g., if females and males are different trees, or if it is predominantly a selfing species, or if consideration must be given to pollen vectors. If seed production is an important aim, spacing could be rather wide compared with a stand where wood production is the main aim. If the seed requirement continues, it is probably a good idea to replace the gene resource stand by a more improved gene resource stand, rather than heading for using the initial stand as a seed source as long as physically possible. The gene resource plantation must be renewed at some interval, and it may be practical to combine cutting with seed harvest.

The effect of the different actions could be evaluated if certain steps were taken. If seeds are stored from different stages of the program (seeds from the initial selections in the forest, seeds from the resulting improved stand, and seeds with which the next-generation stand will be established), the option to establish comparative trials exists. Some long-term storage of seeds is recommended when it is possible and simple. If this is not the case, it may not be worthwhile or even practically possible to conserve the different stages.

The provenance question is always somewhat uncertain. Something may be known or guessed and a gene resource plantation, planned to serve as a local seed source, should not contain genetic material that is likely to be inferior. On the other hand, it should cover gene mass from some range and have diversity in characters. It may often be wise to arrange a gene resource plantation with known identities and a layout identical to a combined provenance - half sib - trial, like what was suggested by Nanson (1972). Small plots (single-tree plots or five-tree row plots may be best) could be used on the provenance level, so as not to leave unproductive gaps if unsuitable provenances are thinned out. If there is interest in and funding for recovery of more scientific information, the trial could be measured before first thinning and the results evaluated. In practice, the major cost with provenance trials is probably not their establishment, but rather their measurement, evaluation and presentation. Their establishment could be justified as gene resource plantations, even if they are never measured. It is important, however, that thinning is not delayed. It may be noted that (provided tree identification remains) even a phenotypically thinned trial can provide provenance information. If there is a provenance variation in a gene resource stand, and the stand is phenotypically thinned and the best looking trees harvested for seeds, this is a way to utilise the well-adapted suitable provenances, as such provenances will be over-represented among trees selected for harvest. If much of the gene resource stand turns out to be comprised of inferior provenance material, the next cycle of gene resource stand could contain fresh collections, including more material from the areas where the good origins grow. While, if much of the provenance material seems acceptable and the range satisfactorily covered, the stand could serve as the seed source for the next cycle. Actually, this type of gene resource plantation could be more efficient than the currently most used procedure for non-priority species, that is to first identify the best provenances and then initiate seed collections based on that knowledge.

In the foregoing, a number of actions were suggested, which reduce the cost and relax the control. At first sight, it may seem that this makes breeding simpler and thus less demanding of qualified breeders and quantitative genetic competence. Actually, the situation may be the opposite. That the tools are less controlled makes it more difficult to optimise the system, and less control means a larger need for predictions. The breeding strategy and tactics still need to be optimised (numbers, consequence analysis, consideration of specific factors, etc.). The competence and number of people who do the field work can be reduced (grafting, crossing), but the breeders and scientists who do the planning will need more and not less skills in quantitative genetics. The bottlenecks where low-intensity breeding will be practised are unlikely to be in the competence of the top people or their advisers. Low-intensity breeding may often be practised today in environments with limited resources, but it is little discussed. Even low-intensity breeding needs to be optimised and that is a tough and competence-demanding job. A good thing is that by focusing more attention on it, what is already practised in low-intensity breeding will be more efficiently done.

Many people have contributed to the discussion around this manuscript and thoughts. Among those who have commented or contributed I would like to mention Tim Mullin in particular. An unknown reviewer has made many useful suggestions. Among other persons who have commented the manuscript are Thúy Olsson, Leopoldo Sanchez-Rodrigues, Ola Rosvall, Adolfo Bila, Rowland Burdon, Huan-Lin Lai, Yongqi Zheng, Matti Haapanen, Tore Skrøppa, Per Ståhl and Dag Rudin.