PREDICTION OF ADAPTABILITY AND YIELD STABILITY OF DURUM-WHEAT GENOTYPES FROM YIELD RESPONSE IN NORMAL AND ARTIFICIALLY DROUGHT-STRESSED CONDITIONS

Adaptability, yield stability and yield reliability of durum wheat bre eding lines are usually assessed through regional testing. The opportu nity of partially substituting such testing by evaluation under normal and artificially drought-stressed rainfed conditions was investigated for a water-limited Italian region. Nine lines were grown at six site s for three seasons to assess adaptability across locations as Perkins and Jinks' slope of genotype regressions (beta), stability across env ironments as Shukla's stability variance (sigma(2)), mean yield (Y), a nd Eskridge's reliability (R) from Y plus sigma(2). Heterogeneity of g enotype regressions explained 54% of genotype-location interaction var iation. The beta values were strictly associated (r = -0.99) with geno type scores on the first genotype-location interaction principal compo nent (PC1), were not related to earliness of heading, and tended to ne gative correlation with plant stature that was hardly explainable in t erms of resistance to lodging. Mean yield, PC1 score and rainfall of s ites were correlated. The lines were also grown under normal and stres s conditions at four sites for two seasons. The stress was established by placing metal channels between the rows that evacuated a portion o f rainfall from the end of tillering stage onwards. Predictions of bet a, sigma(2), Y and R, attempted respectively from slope of genotype-st ress level interaction (beta(p)), beta(p)(2), mean yield across condit ions (Y-p), and Y-p plus beta(p)(2), were assessed as genetic correlat ion. Predictions based on beta(p) and Y-p computed over all test envir onments were all relatively good, whilst those based on data of indivi dual seasons or locations were mostly inaccurate for sigma(2), Y and R . High-yielding sites could better predict Y and R, Two seasons' data from one such site showed correlations of 0.60, 0.53, 0.72 and 0.75 fo r prediction respectively of beta, sigma(2), Y and R. Evaluation of ad vanced breeding lines under normal and artificially stressed condition s at a high-yielding site may prove useful for reducing the number of lines promoted to subsequent regional testing and/or restricting their regional testing to specific areas of adaptation.