OPTIMAL NUMBER OF RAMETS PER CLONE IN 2 C LONAL TESTS

Many forest tree breeding programmes are now based on clonal tests wit h the purpose of selecting improved genotypes for clonal forestry. Of course, manpower, budget and space for tests are limited and it is imp ortant to obtain the best selection efficiency within these constraint s. The present paper reports a study of optimization of statistical ef ficiency in clonal tests. The goal is to maximize genetic gain. We exa mined the consequences of trade-offs between the number of ramets per clone (ie, accuracy of evaluating gene type means) and the number of c lones tested (ie, selection intensity), when the total number of trees tested is held constant. The data originated from a test of 32 clones of wild cherry (Prunus avium) aged 7 years and repeated on three site s, and from a test of 96 clones of hybrid larch (Larix x eurolepis), a ged 8 years and planted on only one site. Characteristics studied were height increments, girth, branching parameters, wood quality and form notations. Clonal variability was significant at a 0.1% level for all traits, and broad sense heritabilities varied between 0.2 and 0.7 (ta ble I). n (2 to 16) ramets per clone were randomly extracted from the whole data sets to obtain simulated data. The simulations showed that accuracy of genotype evaluation, as measured by R(2) (or CD: average s quare of correlation between evaluated genotype mean and true genotype mean), increases with the number of ramets per clone and tends to a m aximum which is the R(2) obtained with the whole data sets (figs 1 and 2). But with six or eight ramets, the accuracy is only 3 to 10% less than the maximal accuracy (fig 2). This results in a few changes in th e ranking of clones (fig 2), but the correlation between the ranks obt ained with the whole data and those obtained with each simulation rema ins high (table III). Variations of genetic gain with the number of ra mets per clone was investigated As an example, figure 4 shows this var iation for height increment of wild cherry observed in one site. its m aximum is at four ramets per clone. For many characteristics, the maxi mum estimated genetic gain is obtained with two ramets per clone (tabl e III). With six to eight ramets per clone, accuracy of genotype means evaluation is sufficient and genetic gain is predicted to be 20 to 30 % higher than genetic gain obtained with 18 ramets per clone, the numb er of ramets currently used in clonal tests. Precision of genetic vari ances and covariances evaluation was also investigated: it increases w ith the number of ramets and tends to the precision obtained with the whole data sets (fig 5). With six ramets per clone or more, this preci sion is still quite good.