4. Conclusions
It has previously been reported that rice grain yield under drought stress is a moderately heritable trait that is usually correlated with yield under nonstress conditions, but not necessarily highly so (Atlin et al., 2004; Bernier et al., 2007; Kumar et al., 2008; Venuprasad et al., 2007). It follows fromthis that direct selection for yield under stress is likely to be the most effective way to develop lines with improved drought tolerance. Our results provide strong support for the hypothesis that direct selection can improve yield under stress, without loss of yield potential. Two generations of pedigree selection for yield under stress conditions usually resulted in significantly greater gains under drought stress than did selection under nonstress conditions in the populations studied. Response was observed both under severe managed stress conditions in the dry season that were similar to the selection environment, and under naturally occurring stress in the wet season, a first reported confirmation that selection for yield under artificially imposed upland drought stress in rice results in yield improvement under natural stress in the wet season. There was also some evidence that selection for tolerance to upland drought stress resulted in small gains in yield under lowland stress; this point merits further attention in future studies. The more tolerant donor (Vandana) produced consistently greater gains under severe stress than the less-tolerant line Apo. In general, these results confirm that breeding rice for improved grain yield under drought stress is feasible via selection for yield under managed stress, and that the most efficient way to develop lines combining improved drought tolerance with high yield potential is to select in population derived from crosses between parents that complement each other for these traits. Further studies to test the efficacy of dry season screening as a surrogate for wet season screening at different sites are warranted.